THE FIRM “LIFE-CYCLE” HYPOTHESIS AND DIVIDEND POLICY: TESTS ON PROPENSITY TO PAY, DIVIDEND INITIATION, AND DIVIDEND GROWTH RATES

A dissertation submitted to: Kent State University Graduate School of Management in partial fulfillment of the requirements for the degree of Doctor of Philosophy

by Richard P. Hauser July, 2012

Dissertation written by Richard P. Hauser B.S., Purdue University, 1989 M.B.A., California Coast University, 2006 Ph.D., Kent State University, 2012

Approved by Chair, Doctoral Dissertation Committee Dr. John Thornton Members, Doctoral Dissertation Committee Dr. Jayaram X. Muthuswamy

Dr. Eric Johnson Accepted by

Dr. Murali Shankar

Doctoral Director, Graduate School of Management

Dr. Frederick W. Schroath

Associate Dean, Graduate School of Management

ii

iii

Table of Contents

CHAPTER 1……………………………………………………………………………….

Page 1

INTRODUCTION…………………………………...………………………..…………...

1

1.1 1.2 1.3 1.4

Overview…………………………………………………………………………… Hypotheses…………………………………………………………………………. Summary of Empirical Findings…………………………………………………… Contribution………………………………………………………………………...

1 7 10 15

CHAPTER 2……………………………………………………………………………….

19

REVIEW OF THE LITERATURE……………………………………..............................

19

2.1

Dividend Policy Irrelevance………………………………………………………... 2.1.1 Dividend Clienteles…………………………………………………………... 2.1.2 Optimal Dividend Policy……………………………………………………... 2.2 Dividend Policy and Taxes………………………………………………………… 2.3 Information Asymmetry and Dividend Signaling………………………………….. 2.4 Dividend Policy and Agency Costs………………………………………………… 2.5 Behavioral Models of Dividend Policy…………………………………………….. 2.6 Summary of Dividend Policy Theories…………………………………………….. 2.7 Propensity to Pay and “Disappearing Dividends”………………………………….. 2.7.1 Corporate Governance and “Disappearing Dividends”………………………. 2.7.2 Firm Maturity or “Life-Cycle” Hypothesis ………………………………….. 2.8 Dividend Initiation…………………………………………………………………. 2.9 Macroeconomics and Dividend Policy……………………………………………... 2.10 Dividend Policy and Firm Value…………………………………………………… 2.11 Dividend Growth…………………………………………………………………… 2.12 Dividend Policy and Complexity…………………………………………………...

19 19 20 21 22 23 24 26 27 29 30 32 33 35 36 38

CHAPTER 3……………………………………………………………………………….

39

MODELS……………………………………......................................................................

39

3.1 3.2 3.3

Dividend Growth Model…………………………………………………………… Sustainable Growth………………………………………………………………… A Statistical Model for the Dividend Growth Rate…………………………………

39 40 42

CHAPTER 4……………………………………………………………………………….

44

DATA AND METHODOLOGY………………………………………………………….. 4.1 Propensity to Pay……………………………………………………………………

44 44

iv

4.1.1 Data Sample………………………………………………………………….. 4.1.2 Control Variables…………………………………………………………….. 4.1.3 Other Control Variables……………………………………………………… 4.1.4 Maturity Hypothesis Variables……………………………………………….. 4.1.5 Industry Dummy Variables…………………………………………………... 4.1.6 Year Dummy Variables………………………………………………………. 4.1.7 Dependent Variable…………………………………………………………... 4.1.8 Fama and MacBeth Logit Model…………………………………………….. 4.1.9 Panel Logit Model……………………………………………………………. Dividend Initiation…………………………………………………………………. 4.2.1 Data Sample and Variables…………………………………………………... 4.2.2 Logit Models…………………………………………………………………. 4.2.3 Hazard Model of Dividend Initiation………………………………………… Valuation…………………………………………………………………………… 4.3.1 Regressions on M/B………………………………………………………….. Dividend Payout Policy…………………………………………………………….. 4.4.1 Dividend Growth Logits……………………………………………………… 4.4.2 Dividend Cut Logits………………………………………………………….. 4.4.3 Regressions on Dividend Payout Ratio………………………………………. 4.4.4 Regressions on Dividend Yield………………………………………………. 4.4.5 Regressions on Dividend Growth Rate……………………………………….

44 44 46 46 49 50 51 51 52 53 54 54 55 56 56 57 57 58 59 60 60

CHAPTER 5……………………………………………………………………………….

62

EMPIRICAL RESULTS…………………………………………………………………... 5.1 Descriptive Statistics……………………………………………………………….. 5.1.1 Overall Sample and Time Series……………………………………………... 5.1.2 Maturity Variables……………………………………………………………. 5.1.3 Valuation Parameters………………………………………………………… 5.1.4 Dividend Policy………………………………………………………………. 5.1.5 Economic Sector Analysis…………………………………………………… 5.2 Maturity and the Propensity to Pay Dividends……………………………………... 5.2.1 Probability of paying dividends-Fama and MacBeth method………………... 5.2.2 Probability of paying dividends-Panel logistic method……………………… 5.2.3 Implications of the Life-cycle Models……………………………………….. 5.2.4 Outlier Analysis………………………………………………………………. 5.2.5 Over-zealous payers and “disappearing dividends”………………………….. 5.2.6 Panel Logistic Regression with Year Effects………………………………… 5.3 Maturity and Life-cycle Valuation…………………………………………………. 5.3.1 M/B Regressions as a Function of Maturity Factor………………………….. 5.3.2 M/B Regressions in Maturity Factor Ranges………………………………… 5.3.3 M/B Regressions in Maturity Composite Ranges……………………………. 5.3.4 Life-cycle and Firm Value…………………………………………………… 5.3.5 Dividend Payout Policy and Firm Value……………………………………... 5.4 Maturity and Dividend Initiation…………………………………………………… 5.4.1 Time Series of Dividend Initiators……………………………………………

62 62 62 63 67 69 72 74 74 78 79 80 84 89 90 90 91 94 96 98 100 100

4.2

4.3 4.4

v

5.4.2 Probability of initiating a dividend- Fama and MacBeth method……………. 5.4.3 Probability of initiating a dividend- panel logistic method…………………... 5.4.4 Dividend initiation and the life-cycle………………………………………… 5.4.5 Dividend initiation and year effects………………………………………….. 5.4.6 Survival analysis of dividend initiation………………………………………. Maturity and Dividend Payout Policy……………………………………………… 5.5.1 Probability of dividend growth………………………………………………. 5.5.2 Probability of a dividend cut…………………………………………………. 5.5.3 Maturity and dividend payout ratio…………………………………………... 5.5.4 Maturity and dividend yield………………………………………………….. 5.5.5 Dividend growth rates………………………………………………………... 5.5.5.1 Maturity and the dividend growth rate……………………………….. 5.5.5.2 Estimating the dividend growth rate………………………………….

101 104 105 108 109 113 114 121 125 130 133 133 137

CHAPTER 6……………………………………………………………………………….

141

CONCLUSIONS, LIMITATIONS, FUTURE RESEARCH……………………………... 6.1 Summary and Conclusions…………………………………………………………. 6.1.1 Maturity and the propensity to pay a dividend……………………………….. 6.1.2 Maturity and firm value………………………………………………………. 6.1.3 Maturity and dividend initiation……………………………………………… 6.1.4 Maturity and dividend policy………………………………………………… 6.1.5 Hypotheses findings………………………………………………………….. 6.2 Limitations…………………………………………………………………………. 6.3 Future Research……………………………………………………………………..

141 141 141 143 146 148 151 152 153

Appendix A: Listing and Detailed Explanation of Variables Used………………………..

155

REFERENCES…………………………………………………………………………….

160

5.5

vi

LIST OF TABLES Table

Page

Table 1

Summary Statistics for the Sample, 1982-2010……………………………..

166

Table 2

Summary Statistics for Dividend Payers and Non-payers…………………..

167

Table 3

Summary Statistics for Dividend Payers by Maturity Factor Decile………..

168

Table 4

Summary Statistics for Non-payers by Maturity Factor Decile……………..

169

Table 5

Valuation and Returns for Dividend Payers and Non-payers, 1982-2010….

170

Table 6

Valuation and Returns by Maturity Factor Deciles for 1982-2010…………

171

Table 7

Summary Statistics for Dividend Growers and Dividend Cutters………......

172

Table 8

Summary Statistics by Economic Sector………………………………........

173

Table 9

Economic Sector Composition for Dividend Payers by Maturity Decile…..

174

Table 10

Economic Sector Composition for Non-payers by Maturity Decile………..

175

Table 11

Logit Analysis of the Decision to Pay Dividends-Fama and MacBeth approach, Models 1-6 (No control variables)……………………………….

176

Logit Analysis of the Decision to Pay Dividends-Fama and MacBeth approach, Models 7-12 (With control variables)………………………........

177

Logit Analysis of the Decision to Pay Dividends-Fama and MacBeth approach, Models 13-18 (With variable combinations and controls)………

179

Table 14

Average Partial Effects of Maturity Components…………………………...

181

Table 15

Logit Analysis of the Decision to Pay Dividends-Panel logistic method, Models 1-6 (No control variables)…………………………………………..

182

Logit Analysis of the Decision to Pay Dividends-Panel logistic method, Models 7-12 (With control variables)……………………………………….

183

Logit Analysis of the Decision to Pay Dividends-Panel logistic method, Models 13-18 (With variable combinations and controls)………………….

185

Analysis of Maturity Model Predictions…………………………………….

187

Table 12 Table 13

Table 16 Table 17 Table 18

vii

Table 19

Summary of Maturity Model Outlier Analysis……………………………...

188

Table 20

Residual Analysis of Dividend Paying Firms-Model 11………………........

189

Table 21

Residual Analysis of Dividend Paying Firms-Model 12………………........

190

Table 22

Residual Analysis of Dividend Paying Firms-Model 13………………........

191

Table 23

Residual Analysis of Non-paying Firms-Model 11……………………........

192

Table 24

Residual Analysis of Non-paying Firms-Model 12……………………........

193

Table 25

Residual Analysis of Non-paying Firms-Model 13……………………........

194

Table 26

Residual Analysis of Dividend Paying Firms-Model 11, 1982 vs. 2000.......

195

Table 27

Residual Analysis of Non-paying Firms-Model 11, 1982 vs. 2000…….......

196

Table 28

Logit Analysis of the Decision to Pay Dividends-Panel logistic method, Model11 (With year effects and controls)………………………………….

197

Table 29

M/B Regressions for Dividend Payers as a Function of Maturity Factor…..

199

Table 30

M/B Regressions for Non-payers as a Function of Maturity Factor………..

200

Table 31

Summary Statistics for Dividend Payers by Maturity Factor……………….

201

Table 32

Summary Statistics for Non-payers by Maturity Factor…………………….

202

Table 33

M/B Regression as a Function of Maturity Factor Range………………......

203

Table 34

Summary Statistics for Dividend Payers by Maturity Composite Range…..

204

Table 35

Summary Statistics for Non-payers by Maturity Composite Range………..

205

Table 36

M/B Regression as a Function of Maturity Composite Range………….......

206

Table 37

M/B Regressions for Non-payers by Maturity Composite Range………......

207

Table 38

M/B Regressions for Dividend Payers by Maturity Composite Range…......

208

Table 39

M/B Regressions with Dividend Policy by Maturity Composite Range........

209

Table 40

Summary Statistics for Non-payers that initiate a dividend…………….......

210

viii

Table 41

Logit Analysis of the Decision to Initiate Dividends-Fama and MacBeth method……………………………………………………………….............

211

Logit Analysis of the Decision to Initiate Dividends-Panel Regression method………………………………………………………………….........

212

Summary Statistics for Dividend Initiators by Maturity Composite Range…………………………………………………………………….......

213

Logit Analysis of the Decision to Initiate Dividends by Maturity Composite Range………………………………………………………........

214

Logit Analysis of the Decision to Initiate Dividends-Panel Regression method with year effects…………………………………………………….

215

Analysis of the Decision to Initiate Dividends-Cox proportional hazard model………………………………………………………………………...

217

Table 47

Summary Statistics for Dividend Growers versus Non-growers………........

218

Table 48

Logit Analysis of the Decision to Increase Dividends-Fama and MacBeth method……………………………………………………………………….

219

Logit Analysis of the Decision to Increase Dividends-Panel Regression method……………………………………………………………………….

221

Summary Statistics for Dividend Growers by Maturity Composite Range……………………………………………………………………......

223

Logit Analysis of the Decision to Increase Dividends by Maturity Composite Range……………………………………………………………

224

Table 52

Summary Statistics for Dividend Cutters versus Non-cutters………………

226

Table 53

Logit Analysis of the Decision to Decrease Dividends-Fama and MacBeth method……………………………………………………………………….

227

Logit Analysis of the Decision to Decrease Dividends-Panel Regression method……………………………………………………………………….

229

Summary Statistics for Dividend Payers by Dividend Payout Ratio Decile……………………………………………………………………......

231

Dividend Payout Ratio Regressions for Dividend Payers as a Function of Maturity……………………………………………………………...............

232

Table 42 Table 43 Table 44 Table 45 Table 46

Table 49 Table 50 Table 51

Table 54 Table 55 Table 56

ix

Table 57

Dividend Payout Ratio Regressions for Dividend Payers as a Function of Maturity with control variables – Fama and MacBeth method…………......

233

Table 58

Summary Statistics for Dividend Payers by Dividend Yield Decile……......

234

Table 59

Dividend Yield Regressions for Dividend Payers as a Function of Maturity with control variables – Fama and MacBeth method…………......................

235

Summary Statistics for Dividend Growers by Dividend Growth Rate Decile……………………………………………………………………......

236

Dividend Growth Rate Regressions for Dividend Growers as a Function of Maturity………………………………………………………………...........

237

Dividend Growth Rate Regressions for Dividend Growers as a Function of Maturity with control variables – Fama and MacBeth method……………..

238

Dividend Growth Rate Regressions with Sustainable Growth – Fama and MacBeth method…………………………………………………………….

239

Sustainable Growth Rate Regressions for Dividend Growers as a Function of Maturity…………………………………………………………………..

240

Dividend Growth Rate Regressions by Maturity Composite Range – Fama and MacBeth method………………………………………………………..

241

Time Series of Maturity and Dividend Growth Rates for Proctor & Gamble Co., 1982-2010…………………………………………………....................

242

Table 60 Table 61 Table 62 Table 63 Table 64 Table 65 Table 66

x

LIST OF FIGURES Figure

Page

Figure 1

Percentage of Dividend Paying Firms in Sample, 1982-2010………………...

243

Figure 2

Median CRSP Age for Dividend Payers and Non-payers, 1982-2010……….

244

Figure 3

Percentage of Dividend Paying Firms as a Function of Sample Median CRSP Age…………………………………………………………………………….

245

Figure 4

Median RE/TE for Dividend Payers and Non-payers, 1982-2010……………

246

Figure 5

Percentage of Dividend Paying Firms as a Function of Sample Median RE/TE………………………………………………………………………….

247

Median Standard Deviation of Monthly Returns for Dividend Payers and Non-payers, 1982-2010………………………………………………………..

248

Percentage of Dividend Paying Firms as a Function of Sample Median Standard Deviation of Monthly Returns………………………………………

249

Median Maturity Composite Score for Dividend Payers and Non-payers, 1982-2010……………………………………………………………………...

250

Figure 9

Median Maturity Factor for Dividend Payers and Non-payers, 1982-2010….

251

Figure 10

Median Maturity Factor for Sample Firms, 1982-2010……………………….

252

Figure 11

Percentage of Dividend Paying Firms as a Function of Sample Median Maturity Factor………………………………………………………………...

253

Figure 12

Percentage of Dividend Payers in Maturity Factor Deciles…………………...

254

Figure 13

Equal-Weighted M/B Ratio for Dividend Payers and Non-payers, 19822010……………………………………………………………………………

255

Average Monthly Returns for Dividend Payers and Non-payers, 19822010……………………………………………………………………………

256

Figure 15

Equal-Weighted M/B Ratio in Maturity Factor Deciles………………………

257

Figure 16

Equal-Weighted Average Monthly Returns in Maturity Factor Deciles……...

258

Figure 6 Figure 7 Figure 8

Figure 14

xi

Figure 17

Median M/B Ratio as a Function of Maturity Factor for Dividend Payers and Non-payers…………………………………………………………………….

259

Figure 18

Median Dividend Payout Ratio for Dividend Paying Firms, 1982-2010……..

260

Figure 19

Median Dividend Growth Rate for Dividend Paying Firms, 1982-2010……..

261

Figure 20

Percentage Dividend Growers, Cutters, and Initiators as a Function of Maturity Factor………………………………………………………………...

262

Median Percentage Dividend Payout and Growth Rate as a Function of Maturity Factor………………………………………………………………...

263

Figure 22

Percentage of Dividend Paying Firms by Economic Sector, 1982-2010……..

264

Figure 23

Percentage of Firms in Sample by Economic Sector, 1982-2010…………….

265

Figure 24

Median Maturity Factor by Economic Sector, 1982-2010……………………

266

Figure 25

Percentage of Dividend Paying Firms by Economic Sector as a Function of Sector Median Maturity Factor………………………………………………..

267

Percentage Dividend Paying Firms in Sample, 1982-2010-Models Versus Actual………………………………………………………………………….

268

Percentage Dividend Paying Firms in Sample, 1982-2010-Models Versus Actual………………………………………………………………………….

269

Figure 28

“Over-zealous” Dividend Payers, 1982-2010…………………………………

270

Figure 29

Disappearing M/B Ratio “Dividend Premium” for Model Outliers…………..

271

Figure 30

Percentage Dividend Paying Firms in Sample, 1982-2010-Model Versus Actual (Model 11 with Year Effects)………………………………………….

272

Figure 31

The Life-Cycle of Firm Value – Maturity Factor Scale……………………….

273

Figure 32

The Life-Cycle of Firm Value – Maturity Composite Scale…………………..

274

Figure 33

The Life-Cycle that Maximizes Firm Value…………………………………..

275

Figure 34

Percentage of Dividend Initiators, 1982-2010………………………………...

276

Figure 35

Empirical Hazard Function in “Life-Cycle” Time…………………………….

277

Figure 36

Empirical Survival Function in “Life-Cycle” Time…………………………...

278

Figure 21

Figure 26 Figure 27

xii

Figure 37

Empirical Hazard Function in Event Time……………………………………

279

Figure 38

Empirical Survival Function in Event Time…………………………………..

280

Figure 39

Growth Rates as a Function of Maturity Composite…………………………..

281

Figure 40

Dividend Growth Rate and Sustainable Growth Rate for Wal-Mart Stores, 1982-2010 ……………………………………………………………………..

282

xiii

CHAPTER 1 INTRODUCTION 1.1

Overview Dividend policy remains one of the great puzzles in Finance. According to Miller and

Modigliani ‘s (1961) seminal paper, dividend policy is irrelevant to firm value in a perfect world. Considering the tax disadvantage of dividends, Black (1976) proposes that investors should not want dividends, yet dividends are paid. Theoretical and empirical research on dividend policy tries to explain why firms pay dividends with market frictions and market imperfections such as taxes, agency costs, and information asymmetry. As if research in the field of dividend policy were not difficult enough, Fama and French (2001) report that the dividend policy of industrial firms in the United States has changed significantly over the period from 1978 to 1999. Specifically, Fama and French (2001) show that the propensity to pay dividends declines dramatically over the period. While 66.5% of listed firms paid dividends in 1978, only 20.8% of listed firms paid dividends in 1999. Furthermore, Fama and French (2001) find that the decline in the percentage of industrial firms that pay dividends is only partly explained by firm characteristics. Even after controlling for firm characteristics, the propensity to pay a dividend still declines over the 1978 to 1999 period. Fama and French (2001) refer to this phenomenon as “disappearing dividends”. Several research paths have been taken to resolve the “disappearing dividends” phenomenon. Grullon, Michaely, and Swaminathan (2002) refute prior signaling models that indicated dividend policy conveys information regarding future cash flow. Rather Grullon et al. (2002) show evidence that dividends convey information about the firm’s systematic risk.

1

2

Specifically, they propose the “maturity hypothesis” to describe the process where changes in dividend policy relate to a firm’s transition from a high growth phase to a lower growth phase. Grullon and Michaely (2002) show that stock repurchases may be substitutes for dividends. Julio and Ikenberry (2004) explore the extent to which stock repurchases have functioned as a substitute for dividends and show that the total percentage of earnings paid out as dividends and repurchases is relatively stable over the 1984-2004. Hence, to some extent, the reduction in dividends is simply the substitution to repurchases, which can be argued to be a more flexible distribution to shareholders (Brav el al. (2005)). While Julio and Ikenberry (2004) show that total payouts are stable, they still question whether the reduced fraction of firms paying dividends represents a fundamental shift in payout policy. Julio and Ikenberry (2004) test several hypotheses including Grullon et al.’s (2002) maturity hypothesis. Julio and Ikenberry (2004) follow Fama and French (2001) and argue that part of the declining percentage of dividend payers in the 1990’s has to be attributed to the increased number of IPO firms. Such newer and riskier firms are much less likely to pay dividends than large, mature firms with more stable cash flows. DeAngelo, DeAngelo, and Skinner (2004) show that indeed the reduction in payers occurs almost entirely among firms that paid very small dividends. Furthermore, Julio and Ikenberry (2004) show that firm age explains the decline in the propensity to pay a dividend in the 1990’s and further explains the subsequent increase in propensity to pay a dividend after 2001. A firm specific example of the maturity hypothesis is Microsoft. When Microsoft first traded on the NASDAQ in the late in 1980s, Microsoft did not pay a dividend, while its growth opportunities were large compared to its market values. By 2003, Microsoft was one of the largest publicly-traded corporations and had

3

lower growth opportunities. Julio and Ikenberry (2004) argue that Microsoft’s decision to initiate a dividend in early 2003 was consistent with a natural maturation process. Julio and Ikenberry (2004) also test Baker and Wurgler’s (2004a, 2004b) “catering” theory, where firms “cater” the dividend policy to investor preferences, but find no support. In order to explain “disappearing dividends”, Baker and Wurgler (2004a, 2004b) suggest that firms reduced dividend payouts since dividends were out of favor with investors. Hoberg and Prabhala (2009) also test the “catering” theory to explain the disappearing dividends. However, Hoberg and Prabhala (2009) find that Baker and Wurgler’s (2004) catering variables are insignificant when controlled for risk. In fact, Hoberg and Prabhala (2009) attribute 40% of the disappearing dividends to risk factors. Hoberg and Prabhala’s (2009) findings that risk explains dividend policy confirms prior empirical work of Grullon et al. (2002) and Rozeff (1982). DeAngelo, DeAngelo, and Stulz (2006) further the maturity hypothesis by showing that the firm’s financial maturity or “life-cycle”, which is characterized by its earned capital ratio, significantly explains a firm’s propensity to pay a dividend. DeAngelo et al. (2006) show that firms with high earned capital ratios are dividend payers while firms with low earned capital ratios tend not to pay dividends. While DeAngelo et al. (2006) explain that the earned capital ratio represents the firm’s financial maturity, they test the statistical significance of the earned capital ratio as an explanatory variable of a firm’s propensity to pay a dividend only using the Fama and French (2001) firm characteristics as control variables. Thus in review of the current literature regarding “disappearing dividends”, three separate versions of firm maturity hypothesis explanations exist. In each maturity hypothesis, there exists a different measure of firm maturity. Grullon et al. (2002) characterize the “firm maturity hypothesis” with risk variables, specifically beta and systematic risk. Hoberg and Prabhala

4

(2009) show that both systematic risk and idiosyncratic risk explain “disappearing dividends” although idiosyncratic risk has a greater marginal effect. Julio and Ikenberry (2004) regress the variable firm age to test the maturity hypothesis. Finally, DeAngelo et al. (2006) represent the firm’s financial maturity or “life-cycle” by the firm’s earned capital ratio.

Not only does the

prior literature use different measures of maturity, the conclusions of the different maturity variables on the “disappearing dividends” puzzle conflict each other. Both Julio and Ikenberry’s (2004) and Hoberg and Prabhala’s (2009) measures of maturity partially explain the “disappearing dividends” puzzle. However, DeAngelo et al. (2006) report that with the earned capital ratio as the measure of maturity the “disappearing dividends” phenomena is roughly twice as large as reported by Fama and French (2001). In the first part of the dissertation, I determine which maturity hypothesis variable (or combinations of variables) best explains the firm’s propensity to pay a dividend. Furthermore, this dissertation provides better definition of the concept of “firm maturity” and provides some further insights on the “disappearing dividends” puzzle. Since the decision to initiate a dividend is fairly similar to the decision to pay a dividend, the life-cycle hypothesis has been applied to dividend initiation. Baker and Wurgler (2004a, 2004b) study the rate of dividend initiation and find evidence of investor fads or “catering”. However, Julio and Ikenberry (2004) show that Baker and Wurgler’s (2004a, 2004b) “dividend premium” disappears when dividend initiation announcement returns are adjusted for firm size and firm age. DeAngelo et al. (2006) find that the earned capital ratio is statistically significant in the decision to initiate dividends. Hoberg and Prabhala (2009) find that risk is negatively related to the decision to initiate dividends and find the results consistent with the firm maturity view of dividend policy. As with the propensity to pay research, the dividend initiation research uses

5

different measures of firm maturity- firm age, earned capital ratio, and risk. Thus, I determine which maturity hypothesis variable (or combinations of variables) best explains the firm’s decision to initiate a dividend. In each case of the prior literature, the dividend policy time series have been studied with the Fama-Macbeth (1973) method. There are econometric advantages to analyzing the time series data using panel methods. One advantage of using a panel method is that it allows direct hypothesis testing of time effects on dividend policy with the base firm maturity or “life-cycle” model. Thus, the prior literature, which has developed the firm maturity hypothesis on dividend policy, has been tested with an econometric technique that has enabled only testing the microeconomics of the firm. It seems that time varying macroeconomic factors such as the growth rate of the gross domestic product have the potential to affect the growth prospects of the firm and thus impact corporate dividend policy. The only macroeconomic factor that has been extensively reviewed in the literature is the impact of taxes on dividend policy. However, recently Dittmar and Dittmar (2008) show evidence of financing waves or cycles that are related to economic growth. In as much as dividend policy should be related to corporate financing activity, it seems that economic growth may affect dividend policy. The panel regression methods employed in this research enable the testing of time effects on dividend policy. Since the life-cycle model of dividend policy is based on the trade-off between the costs of earnings retention and the costs of earnings distribution, this implies that a firm selects a dividend policy that maximizes the value of the firm. I consider the firms that have dividend policies that are in conflict with the dividend policy predicted by the life-cycle model, which contains the control variables and the maturity variables. The primary interest is the analysis of a conflicting dividend policy on firm value. In other words, testing the prediction of the life-cycle

6

model and analyzing the firms that do not fit the model, enables the study of the fundamental question does dividend policy matter? While there are many measures of firm value, Baker and Wurgler (2004) provide evidence in the literature on dividend policy that suggests that the market to book ratio (M/B) is a good firm valuation measure. They report that although the M/B ratio is time-varying, the M/B ratio of dividend payers is consistently lower than the M/B of nonpayers. The Baker and Wurgler (2004) results imply a systematic difference in M/B ratio between payers and non-payers. From the logit model on propensity to pay, there will be dividend payers that fit the model and non-dividend payers that fit the model. I first compare the M/B ratio of non-payers that do not fit the life-cycle model to the M/B ratio of non-payers that fit the life-cycle model. Then, I compare the M/B ratio of dividend payers that do not fit the lifecycle model to the M/B ratio of payers that fit the model to determine if the "market" places a valuation premium on the firms following the expected (or model predicted) dividend policy. After testing the maturity hypothesis on the propensity to pay dividends, I extend the firm maturity hypothesis to develop an empirical method for estimating dividend growth rates. It seems that a likely extension of the life-cycle hypothesis would be that the firm maturity continues to impact dividend policy even after the firm has decided to pay. As the firm’s tradeoff between the costs of retention of cash flow and the cost of distribution of cash flow evolves with the firm’s life-cycle, so should the firm’s dividend growth rate. As the dividend paying firm matures, it should have declining growth opportunities and increasing free cash flow, which suggests that dividend distributions should increase. Dividend growth rates are required inputs for dividend discount stock valuation models. Although dividend discount stock valuation models have existed since Williams (1938), stock analysts have done little to formalize the process of estimating dividend growth rates. In their popular textbook Fundamentals of

7

Investments, Jordan and Miller (2009) describe three methods that financial analysts use to estimate the dividend growth rate: (1) using the company’s historical average dividend growth rate, (2) using the industry median or average growth rate, or (3) using the sustainable growth rate based on ratio analysis. However, Jordan and Miller (2009) also note that “a historical average growth rate may or may not be a reasonable estimate of future dividend growth.” An empirical method to estimate the dividend growth rate based on the firm maturity variables could be useful for financial analysts. 1.2

Hypotheses In the first part of the dissertation, the emphasis involves integrating three existing yet

distinct versions of the firm maturity hypothesis regarding dividend policy and the different firm maturity variables. The firm maturity variables of risk, firm age, and earned capital ratio or combinations of the variables have not been tested jointly in the prior literature for significance in the propensity to pay dividends. Combinations of the maturity variables may describe firm “life-cycles” better than any single maturity variable. Consider the singular variable firm age. Some firms may quickly grow and reach a point of limited investment opportunity while others may continuously develop new investment opportunities. Likewise, the earned capital ratio variable does not capture whether the capital mix occurs on the ascent or descent of the firm’s “life-cycle”, which seems relevant to dividend policy. Therefore, I test the following hypothesis related to propensity to pay dividends: Hypothesis 1a. Total risk significantly determines a firm’s propensity to pay dividends when controlled for the Fama-French (2001) firm characteristics, firm age, earned capital ratio, and combinations of risk, firm age, and earned capital ratio.

8

Similarly, the measures of firm maturity have been only tested independently in the research on the decision to initiate a dividend. Therefore, I test the following hypothesis related to the decision to initiate dividends: Hypothesis 1b. Total risk significantly determines a firm’s decision to initiate dividends when controlled for the Fama-French (2001) firm characteristics, firm age, earned capital ratio, and combinations of risk, firm age, and earned capital ratio. As stated above, the prior research has utilized the Fama-Macbeth (1973) method to analyze the time series data. In this study, I analyze the time series data using panel methods. This enables the direct hypothesis testing of time effects on dividend policy.

Since

macroeconomic factors such as economic growth can alter the growth prospects of firms over time, these time varying macroeconomic factors may impact corporate dividend policy. In highly profitable periods, firms may generate abnormal free cash flows, which would influence dividend policy. Each year in the data panel can be represented by a dummy variable that is assigned 1 for data occurring in the given year or 0 if otherwise. The Fama-Macbeth (1973) method used in the prior literature is unable to discern a step change in dividend policy from a gradual shift in dividend policy. While the prior literature has focused on the microeconomics of the firm, I investigate the effects of both firm characteristics and (yearly) time effects on dividend policy. Based on the use of the panel method, I test the following hypothesis with the propensity to pay model: Hypothesis 2a. Relative to the base firm life-cycle model, yearly time effects captured by year dummy variables are significant determinants of the firm’s propensity to pay a dividend. Similarly, I propose to test the following hypothesis related to the decision to initiate dividends:

9

Hypothesis 2b. Relative to the base firm life-cycle model, yearly time effects captured by year dummy variables are significant determinants of the firm’s decision to initiate a dividend. Testing the predictions of the life-cycle model and analyzing the firms that do not fit the model enables the study of the fundamental question does dividend policy matter? The Baker and Wurgler (2004) results imply a systematic difference in M/B ratio between payers and nonpayers. The research question here, of course, is what about the firms that do not fit the model. I first compare the M/B ratio of non-payers that do not fit the life-cycle model to the M/B ratio of non-payers that fit the life-cycle model. The non-payers that are "immature" with high growth potential will fit the life-cycle model. The non- payers that seem mature and have less growth will not fit the model. If there is a difference in M/B ratio between these groups, it suggests that the "market" places a valuation premium on the firms following the expected (or model predicted) dividend policy. For example, if the non- payers that do not fit the model (against type1) have a lower M/B ratio than the non-payers that fit the life-cycle model, I would argue the "market" senses the agency costs of free cash flow and discounts the firm value of the againsttype, non-payers accordingly. Based on this, I test the following hypothesis: Hypothesis 3a. The non- payers that do not fit the model (against type) have a significantly lower M/B ratio than the non-payers that fit the life-cycle model.

Again based on the Baker and Wurgler (2004) results, I sort the dividend payers from the non-payers. I then compare the M/B ratio of dividend payers that do not fit the life-cycle model to the M/B ratio of payers that fit the model. The payers that are "mature" with limited growth potential should fit the model. However, dividend payers that are not “mature” and still have

1

In the empirical finance literature, firms that have characteristics contrary to the model classification are regarded to as “against type” firms. Here, against type firms have a dividend policy that is contrary to the model prediction. Jung, Kim, and Stulz (1996) use this terminology in an empirical study of capital structure.

10

growth potential should not fit the model. If there is a difference in the M/B ratio between these groups, it could again suggest the "market" places a difference on valuation for following the expected dividend policy. I then test the following hypothesis: Hypothesis 3b. The dividend payers that do not fit the model (against type) have a significantly lower M/B ratio than the dividend payers that fit the life-cycle model. The final part of the dissertation tests whether the firm maturity hypothesis can describe the firm’s dividend growth rate. It would be expected that if the firm maturity variables or combination of variables can describe a firm’s “life-cycle”, then these variables could be used to estimate the dividend growth rate. As the dividend paying firm matures, it should have declining growth opportunities and increasing free cash flow, which suggests that dividend distributions should increase. Although the maturity hypothesis implies that the dividend distributions should increase with firm maturity, the Law of Large Numbers2 applied to finance suggests that the probability of sustaining a large percentage growth rate declines. It seems that as a dividend paying firm matures, its dividend growth rate should decline. Based on the firm “life-cycle” model, I test the following hypothesis: Hypothesis 4. Total risk, firm age, earned capital ratio, and combinations of these firm maturity variables describe the cross section of firm dividend growth rates, and the dividend growth rate declines with firm maturity. 1.3

Summary of Empirical Findings The main premise of this dissertation is that firm maturity is related to dividend policy,

and in turn, dividend policy affects firm value throughout the firm’s life-cycle. While prior 2

Although the Law of Large Numbers generally refers to the statistical rule, which states that as the number of samples increases, the average of the samples approaches the mean of the population, it is often applied to financial growth rates. As related to finance, the Law of Large Numbers suggests that as a company grows the chances of sustaining a large percentage growth rate diminish. As the company size approaches the size of the economy, the growth rate approaches the growth rate of the economy. http://www.investopedia.com/terms/l/lawoflargenumbers.asp#axzz1uxxLGv9P

11

research advances a life-cycle or maturity hypothesis to explain corporate dividend policy, prior investigations utilize firm age, the earned capital ratio and risk independently to proxy firm maturity. Consistent with the prior literature, I show that firm maturity is positively related to the probability that a firm pays a dividend. The logit analysis of the probability of paying a dividend indicates that each individual definition of maturity reported in the prior literature captures a statistically significant dimension of firm maturity. Of the individual measures of maturity, the earned capital ratio, measured by the ratio of retained earnings to total equity (RE/TE), has the largest partial effect on the decision to pay a dividend. However, the combination of maturity variables provides the most complete definition of the life-cycle. The panel logistic regression analysis with year effects indicates that after controlling with the maturity model, the propensity to pay dividends is lower than 1982, and the reduction in propensity to pay each year after 1982 is statistically significant. Although this indicates that “disappearing dividends” and the reduced propensity to pay is statistically significant, the analysis failed to relate any specific macroeconomic factors to the year effects. Analysis of firms that do not follow the life-cycle model’s predictions for dividend policy provides further insight into the “disappearing dividends” phenomena. While the life-cycle models with combination maturity variables correctly classify about 85% of the observations, the majority of outliers or firms with a dividend policy contrary to the model are dividend paying firms in the 1982-2010 time series. I consider the dividend paying outliers to be “over-zealous” dividend payers as a group since they have low median maturity, low median profitability, and small size. My analysis shows that the decline in the propensity to pay dividends (or “disappearing dividends”) is due to the decline in these “over-zealous” dividend payers. Further investigation of the “over-zealous” dividend payers reveals that the decline in the “over-zealous”

12

dividend payers is related to the relative market valuation of the outliers. In the early 1980’s, there was no significant valuation difference between “immature” dividend payers and “mature” dividend payers that fit the life-cycle model. However, as market valuations became less favorable to “over-zealous” dividend payers, fewer “immature” firms paid dividends. With fewer over-zealous dividend payers since the 1980’s, the aggregate percentage of dividend payers declines. This dissertation resolves an empirical anomaly in the prior dividend literature. Prior empirical studies show that the median valuation of non-paying firms, as measured by the M/B ratio, is greater than the median valuation of dividend paying firms. If the median valuation of non-paying firms is greater than the median valuation of dividend paying firms, then why would a value maximizing firm ever pay a dividend? My research shows that firm valuation as measured by the M/B ratio is related to the firm maturity and the life cycle. Early in the lifecycle, non-paying firms have high growth potential and high M/B ratios. However, the M/B ratio of non-paying firms continues to decline monotonically as non-paying firms mature. On the other hand, the M/B ratio of dividend paying firms increases as dividend paying firms mature over the life-cycle. The opposing valuations with maturity set up a crossover point where eventually a maturity is reached where the M/B ratio of dividend payers is greater than the M/B ratio of non-payers. At this crossover maturity, a non-paying firm should begin a dividend payout as it matures further in order to maximize firm value. Otherwise, the firm’s value will continue to decline as it matures. Dividend payers do not become more valuable because they are more mature. Higher profitability is the major explanatory variable for the increase in valuation for dividend payers as they mature over the life-cycle. In summary, the life-cycle of firm value resolves the questions of when and why a firm should pay a dividend.

13

Consistent with the prior literature, the results from the logit analysis of the decision to initiate a dividend are very similar to the results from the logit analysis of the decision to pay a dividend. As a non-paying firm matures, its propensity to initiate a dividend increases. Furthermore, each separate definition of maturity captures a statistically significant dimension of firm maturity, but the combination maturity variables seem to provide the most complete definition of maturity in the life-cycle. As with the propensity to pay a dividend, the RE/TE percentile has the largest effect of the maturity variables on the decision to initiate. As firms become larger and more profitable, their propensity to initiate a dividend increases. Consistent with the maturity hypothesis, firms are less likely to initiate a dividend if they have significant growth potential as measured by the M/B ratio. In addition to firm characteristics, there are significant economic sector effects on the decision to initiate a dividend. As expected from the analysis of firm maturity and valuation, most firms initiate a dividend near the crossover maturity in the life-cycle. While a time series plot suggests a declining propensity to initiate from the early 1980’s to about 2003, the panel logit analysis with year effects confirms that the reduction in propensity to initiate is statistically significant even after controlling for maturity, firm characteristics, and economic sector. However, I find no significant negative year effect (from the base year of 1982) at the 5% level until the mid-1990’s, which is over a decade after the adoption of the safe-harbor rule (in 1982). Rather the panel logit results with year effects are consistent with my assertion that the macroeconomic environment shifted from one that favored distribution of earnings in the early 1980’s to one that then favored earnings retention in the mid 1990’s when the development of the internet and related new technologies provided corporations with new growth opportunities.

14

Firm maturity is positively related to the probability of a dividend increase and negatively related to the probability of a dividend cut. However, the relationship between dividend payout policy and maturity is more complex than with the propensity to pay and propensity to initiate analysis. With dividend payout policy analysis, the components of maturity are often inconsistent with the net effect of maturity. Interestingly, as the firm’s age increases, the probability of dividend growth declines and the probability of a dividend cut increases. However, standard deviation is the maturity variable with the largest effect, and as the firm matures and becomes less volatile, the probability of dividend growth increases and the probability of a dividend cut decreases. Therefore, the dominant effect of standard deviation cancels out the offsetting effect of firm age so that the net effect is still that the probability of dividend growth increases with maturity. The RE/TE ratio, which is the largest effect of maturity variables in the propensity to pay and propensity to initiate logits, is insignificant to the probability that a firm increases its dividend. Firm maturity is positively related to the dividend payout ratio. As a dividend paying firm matures, the dividend payout ratio increases. However, the significant maturity components that determine the dividend payout ratio are only age and volatility. The analysis also reveals that the sales growth rate is negatively and significantly related to the dividend payout ratio. These results are consistent with the maturity hypothesis. Young firms with large growth opportunities (high sales growth rates) will retain most of their earnings and have low distribution (low dividend payout ratios). Mature firms with lower growth opportunities (low sales growth rates) will retain less of their earnings and have higher distribution of earnings (higher dividend payout ratios).

15

Finally, the dividend growth rate for dividend growers declines as a firm matures. All of the components of maturity are significant and consistent with the overall effect of maturity on the dividend growth rate. Consistent with the maturity hypothesis and the “Law of Large Numbers”, the dividend distribution increases as the firm matures, but the dividend growth rate occurs at a diminishing rate. Further investigation of the dividend growth rate shows that the sustainable growth rate only approaches the dividend growth rate at the very mature stage of the life-cycle. 1.4

Contribution The firm maturity variables of risk, firm age, and earned capital ratio have not been tested

jointly in the prior literature for significance in the propensity to pay dividends and the probability of dividend initiation. I contribute to the dividend literature by finding that each maturity variable is significant even when tested jointly. Furthermore, the combinations of maturity variables provide the most complete and quantitative definition of the life cycle. My analysis indicates that the earned capital ratio has the largest partial effect on the decision to pay a dividend and the decision to initiate a dividend. The prior literature suggests industry effects on the decision to pay a dividend but no study reports comprehensive industry effects. I report significant industry effects on the decision to pay a dividend and the decision to initiate a dividend. Although the prior literature utilizes only the Fama and MacBeth method for the logit anlaysis, I demonstrate that the panel logistic regression is essentially equivalent to the Fama and MacBeth method. While the prior literature focuses on the relationship between maturity and the propensity to pay, my research investigates the implications of the life-cycle model. My analysis indicates that the valuation of firms that fit the life-cycle model is significantly greater than the valuation

16

of firms that have a dividend policy contrary to the life-cycle model. This provides strong empirical evidence that dividend policy does affect the value of the firm. Furthermore, my research demonstrates when and why a value-maximizing firm pays a dividend. The results also resolve an empirical anomaly in the prior dividend literature. I demonstrate that firm value, as measured by the M/B ratio, is related to firm maturity and the life-cycle. These valuation results are important to corporate boards of directors of U.S. industrial firms. Value-maximizing managers will want to ensure that they implement a dividend policy that follows the life-cycle model and consequently maximizes the value of the firm. Over the 1982-2010 time series, about 15% of the firm observations have dividend policies that are contrary to the life-cycle model. This implies about 15% of U.S. industrial firms have lower valuations than they would if they followed the life-cycle model. It should also be noted that one of the complexities of dividend policy is timing. My analysis of the crossover maturity indicates that paying a dividend “too soon” lowers valuation while not paying a dividend after the crossover maturity also lowers valuation. My analysis of valuation provides the literature further insight into the “disappearing dividends” phenomena. While DeAngelo et al. (2004, 2006) report that dividend policies change most with marginal dividend payers, they fail to explain the disappearance. My analysis reveals that the “over-zealous” dividend payers responded to shifts in the market valuation. As the relative M/B ratio of “over-zealous” dividend payers declined, fewer “immature” firms were “over-zealous” to pay dividends. This contribution not only offers an explanation to the “disappearing dividends” phenomena, my analysis indicates that “disappearing dividends” was beneficial to the economy as firms shifted dividend policy to maximize their valuations.

17

While the prior literature contains studies on the relationship between maturity and the propensity to pay a dividend as well as the relationship between the propensity to initiate a dividend and maturity, none explicitly studies dividend payout policy and the maturity hypothesis. I extend the maturity hypothesis and investigate the relationship between maturity and dividend payout policy. I show that the relationship between firm maturity and dividend payout policy is more complex than with the propensity to pay. With dividend payout policy, the individual maturity components are often inconsistent with the net effect of maturity. This may provide the best empirical evidence that the reported measures of maturity are independent and capture different dimensions of maturity. For example, the earned capital ratio (RE/TE) has the largest effect of the maturity variables in the propensity to pay logits, but it is insignificant to the probability that a firm increases its dividend or the dividend payout ratio. I extend the maturity hypothesis and show that firm maturity is positively related to the dividend payout ratio while the sales growth rate is negatively related to the dividend payout ratio. Young dividend paying firms with larger growth opportunities (or higher sales growth rates) will retain most of their earnings and have lower distributions (lower dividend payout ratios). Mature dividend paying firms with lower growth opportunities (or lower sales growth rates) will distribute most of their earnings and have higher distributions (higher dividend payout ratios). Finally, while the dividend distribution increases as the firm matures, the dividend growth rate declines with maturity. Further investigation of the dividend growth rate shows that the sustainable growth rate only approaches the dividend growth rate at the very mature stage of the life-cycle. In the early part of the life cycle, the dividend growth rate is significantly greater

18

than the sustainable growth rate. Furthermore, as firms progress in the life-cycle, the dividend growth rate continues to decline even in the most mature stage of the life-cycle. In addition to the academic contribution of providing empirical evidence for the maturity hypothesis with dividend payout policy, there is again much of interest to corporate boards of directors that seek to maximize firm value. My analysis reveals the complexity of dividend payout policy on the value of the firm. While positive dividend growth increases the firm’s valuation, the effect is most significant only in the most mature stage of the life-cycle. On the other hand, an excessively high dividend yield decreases firm value, especially for “immature” dividend payers. The investigation of the dividend growth rate offers investors and financial analysts a new technique for empirical estimates of the dividend growth rate. Investors and financial analysts often use estimates of the dividend growth rate in dividend discount models for firm valuation and in dividend discount models for the cost of equity. Financial analysts often estimate future dividend growth rates with the sustainable growth rate. I quantify the part of the life-cycle where the sustainable growth rate is a good estimate of the dividend growth rate. Since the dividend discount models often use the assumption of constant dividend growth rate, I define conditions where the assumption is valid. Life-cycle analysis of the dividend growth rate should enable financial analysts to replace ad hoc judgments regarding future dividend growth rates with more quantitative analysis.

CHAPTER 2 LITERATURE REVIEW 2.1 Dividend Policy Irrelevance The purpose of this chapter is to provide a review of the relevant research on corporate dividend policy and firm maturity. Sections 2.1 to 2.6 discuss the underlying theories for dividend policy. Sections 2.7 and 2.8 describe the current research relating the firm life-cycle hypothesis to dividend policy. Although I review recent studies regarding dividend policy and firm value in Section 2.10 and dividend growth in Section 2.11, I find no prior research relating firm maturity to firm value or dividend growth. The literature regarding corporate dividend policy is a study of how corporations should deliver value to shareholders. Thus while capital structure theory and dividend policy both relate to corporate valuations, dividend policy distinctly focuses on the delivery of value. The focal question of dividend policy is then how should a firm deliver value to shareholders in a way that maximizes the shareholders’ wealth? Miller and Modigliani (1961) show that the value of the firm is unaffected by dividend policy in a world without taxes, agency costs, information asymmetry, or other market imperfections. In this idealized world without taxes and market imperfections, it makes no difference whether (or even how) the firm’s cash flows are distributed to shareholders or reinvested in the firm because the shareholders’ wealth is the same in either case. 2.1.1 Dividend Clienteles Miller and Modigliani (1961) also show that even if different groups are attracted to stocks with different dividend payouts, the implication on firm value is the same as their “perfect

19

20

market case”.

Miller and Modigliani (1961) refer to these different groups who prefer a

particular dividend payout rate as “dividend clienteles.” It is possible to envision some investors having an incentive or demand for stocks with low dividend payouts. Other investors may desire high current income and have a demand for stocks with high dividend payouts. Thus, companies with low payouts will attract one group of investors while companies with high dividend payouts will attract another clientele. Then any further change in dividend policy is pointless as the dividend clienteles are satisfied in market equilibrium. Thus, the existence of dividend clienteles (without any taxes or market imperfections) still indicates that the dividend policy for any individual firm is irrelevant.

In general, the empirical evidence supports the existence of

dividend clienteles. Allen and Michaely (2003) report that in 1996 individual investors held 54% of all stocks by market value but only received 35% of all dividends paid. Graham and Kumar (2006) find evidence for dividend clienteles. Elton and Gruber (1970) measure clientele effects by observing average price declines when a stock goes ex-dividend. While Elton and Gruber’s (1970) study begins much research into the behavior of stock returns near the ex-dividend date, the existence of dividend clienteles fails to explain why a corporation pays a dividend. 2.1.2 Optimal Dividend Policy Despite the Miller and Modigliani (1961) theoretical argument that dividend policy is irrelevant to firm value, dividends are a major cash outlay for many corporations. Furthermore, managers of corporations do worry about dividend policy and maximizing shareholder value. As opposed to the “perfect market” case, the real world has taxes, usually both corporate and personal. Thus, in a real world with taxes, the possibility arises that dividends affect firm value because the taxation of the various distribution mechanisms is different. Furthermore, market imperfections such as transaction costs, agency costs and information asymmetry are possible

21

explanations for dividend policy. The “real” effects of taxes and such market imperfections imply that there may be an optimal dividend policy with respect to maximizing shareholder wealth. 2.2 Dividend Policy and Taxes Historically dividends have been taxed in the United States as ordinary income. However, recent tax law changes have lowered the maximum tax rate on dividends as well as long-term capital gains. The lower personal tax rate on dividends reduces the tax disincentive of dividends, but does not eliminate it. Because capital gain taxes are deferred, the effective capital gain tax rate is still much lower because the present value of the tax is less. Black (1976) then questions if stock repurchases and re-investment is more tax effective, why do firms pay any dividends? However, U. S. tax laws are very complex and affect dividend policy in a number of ways. Miller and Scholes (1978) show that dividend income could largely be sheltered from taxation, at least for some individuals and organizations. Masulis and Trueman (1988) then extend the Miller and Scholes (1978) arguments and model the interactions of the tax system and investment decisions. In their model, Masulis and Trueman (1988) show that the cost of deferring dividends could be large enough that firms would pay cash dividends. An implication of the Masulis and Trueman (1988) model is that shareholders with different personal tax rates will not all agree on the firm’s dividend policy. This model then sets the motivation that investors in different tax brackets would form dividend tax clienteles. In such case, investors in low tax brackets would purchase shares of high dividend paying firms. In fact Allen, Bernardo, and Welch (2000) propose such a theory of dividend policy based on tax clienteles. With regards to the present research, the Masulis and Trueman (1988) model provides a key implication that underlies the firm life-cycle model. Masulis and Trueman (1988) show that high-growth firms

22

with many profitable investment opportunities will use up their internally generated funds without paying dividends while older, more mature firms will pay dividends since their generated funds exceed their investment opportunities. 2.3 Information Asymmetry and Dividend Signaling Given information asymmetry as a market imperfection, the signaling concept of financial capital structure can be readily applied to dividend policy. According to the dividend signaling theory, a firm that increases its dividend payouts is signaling that it has expected future cash flows sufficient to meet expenses without increasing the probability of bankruptcy. Managers signal investors because financial managers have privileged information about the firm’s expected cash flows that outside investors cannot know. Bhattacharya (1979) proposes a dividend signaling model to explain why firms pay dividends despite the apparent tax disadvantage. In Bhattacharya’s (1979) dividend signaling model, investors believe that an unexpected dividend increase is a favorable signal. This assumes that the dividend contains information regarding firm value not conveyed in other public information, and that the dividend is a valid signal since it is expensive for less valuable firms to mimic. Then the signaling value of dividends is positive and can be traded off against the tax costs. The implication of such a dividend signaling model is that it suggests an optimal dividend policy where the signaling benefits of paying dividends offsets the tax disadvantages of paying dividends. Miller and Rock (1985) extend the dividend signaling model. Miller and Rock (1985) show that earnings, dividend, and financing announcements are closely related, and that announcement effects are implications of the basic firm valuation model.

A criticism of the

dividend signaling model is that the model fails to explain the cross section of dividend payouts across firms. Benartzi, Michaely, and Thaler (1997) and Grullon, Michaely, and Swaminathan

23

(2002) show that dividend changes do not signal future changes in cash flows. Rather, Grullon et al. (2002) propose that dividend changes are a sign of firm maturity. Using survey data, Brav, Graham, Harvey, and Michaely (2005) report that financial managers believe that while dividends do convey information, managers do not consciously use dividend payouts as costly signals. Furthermore, Li and Zhao (2008) show evidence against the basic dividend signaling premise of an information asymmetry. Li and Zhao (2008) find that firms with more transparent information actually pay out more dividends. Skinner and Soltes (2011) find that the reported earnings of dividend paying firms are more persistent than non-paying firms and that dividend payers are less likely to report losses. 2.4 Dividend Policy and Agency Costs A wealth-maximizing firm will seek monitoring policies that minimize costs, and it is likely that dividend payments serve as a means of monitoring management performance. A greater dividend payment implies that the firm will need some costly external financing. Thus the fact the firm must obtain external financing introduces outside suppliers of capital that help monitor management for the equity owners. Rozeff (1982) and Easterbrook (1984) propose an optimal dividend policy that is a trade-off between the flotation costs of raising external capital and the benefit of reduced agency costs. Rozeff (1982) shows that firms with higher growth potential have lower dividend payouts, while firms with diffuse outsider holdings have higher dividend payouts. Rozeff’s (1982) regressions confirm that riskier firms as measured by beta also have lower payouts. Jensen (1986) extends this idea of agency costs with the free cash flow hypothesis. Jensen (1986) argues that there are agency costs when firms have excessive cash and that managers are likely to use funds inefficiently by continuing money losing projects, pay excessive perks, or over-pay for acquisitions. Clearly, dividend policy is one way to reduce

24

excessive cash from the firm, although Kahle (2002) indicates that share repurchases do not eliminate all agency costs. Kahle (2002) shows results that indicate managers repurchase both to maximize their own wealth and to fund employee stock option exercises. The effect of corporate governance and the interaction with dividend policy is discussed below. However, it is interesting to note that as Rozeff (1982) found agency costs to explain dividend payouts in a cross section of U.S. firms, Fidrmuc and Jacob (2010) report an agency explanation for dividend payout policies across the world. Fidrmuc and Jacob (2010) explain that dividends are linked to cultural differences since culture influences the character of agency relations in different countries. To resolve empirical support between signaling and agency theories, Fuller and Blau (2010) develop a model that combines both rationales. They find that high quality firms pay dividends to eliminate the free cash flow problem, while firms that outsiders perceive as lower quality pay dividends to signal future earnings and reduce the free cash flow problem. 2.5 Behavioral Models of Dividend Policy Much of what we know about corporate dividend policy stems from interviews and survey data. Lintner (1956) interviews financial managers and reports results regarding the determination of dividend policy. His interviews suggest that U.S. corporations raise dividend payouts only after they are reasonably sure that they will be able to maintain them. Thus, managers seem to allow dividend payouts to lag increases in earnings since managers seem very hesitant to reverse dividend policy. Recently Brav et al. (2005) report survey data from financial executives and confirm much of Linter’s findings, but do note some changes. Brav et al. (2005) report that maintaining the dividend level is still an important consideration. However, now managers favor share repurchases since they are more flexible than dividends. Interestingly, the manager survey results provide little support for agency, signaling, and clientele theories of

25

dividend payout policy, while taxes are a secondary consideration. Baker et al.’s (2002) survey finds that managers of NASDAQ firms support the view of a dividend policy related to the firm life cycle. Unfortunately, none of the interview or survey data provide definitive reasons for paying dividends in the first place. In practice, the Lintner (1956) and Brav et al. (2005) observations indicate that dividend growth lags earnings growth and that dividend growth is much smoother than earnings growth. Hence in the finance literature, this practice of slow, steady changes to the dividend distribution is often referred to as “dividend smoothing”. Leary and Michaely (2011) take an empirical approach to document the observed cross-section of dividend smoothing policies. Leary and Michaely (2011) report that younger and smaller firms with low dividend yields smooth less. On the other hand, they find that firms with high free cash flow and low growth prospects smooth more. Michaely and Roberts (2012) find that private firms smooth less than public corporations. While Lintner (1956) develops his model from survey data, Lambrecht and Meyers (2012) derive a dynamic agency model where financing decisions are made by managers who attempt to maximize the rent they take from the firm. The Lambrecht and Meyers (2012) model indicates that managers smooth payout in order to smooth their flow of rents. While somewhat related to the clientele theory, Baker and Wurgler (2004a, 2004b) develop a behavioral theory where dividend policy is determined by investor demand. In this catering theory, Baker and Wurgler (2004a, 2004b) propose that managers “cater” to investors and initiate dividends when investors put a high valuation on dividend paying firms. Conversely, when investors’ sentiment or desire for dividends declines, firms reduce or omit their dividend payouts. However, Julio and Ikenberry (2004) test the catering theory with additional controls for firm size and age and find that shifts in payout policy do not appear to be driven by changes

26

in investor sentiment for dividends. Likewise, Hoberg and Prabhala (2009) find no support for firms’ catering to transient dividend fads when they control for risk. Although catering is refuted in the literature, investors demand for dividend paying stocks does seem to vary across time. Asem (2009) shows that overall momentum profits are lower for dividend paying stocks. Fuller and Goldstein (2011) show that dividends seem to matter more to shareholders in declining markets than advancing ones. They report that dividend paying stocks outperform nonpaying stocks by 1-2% more per month in declining markets than in advancing markets. 2.6 Summary of Dividend Policy Theories In an idealized world without taxes and market imperfections, Miller and Modigliani (1961) show that the value of the firm is unaffected by dividend policy. However, the real world has personal and corporate taxes as well as market imperfections such as transaction costs, agency costs, and information asymmetry. Much like capital structure theories, no single theory explains the cross section of corporate dividend policy. While there is strong evidence that taxes affect dividend policy, the complexities of the tax code seem to defer tax considerations to a “second-order” policy concern. There is also little debate that dividends convey information to the market and investors; however, there is little empirical evidence that corporate managers send costly signals to investors with dividends. Theories based on managerial behavior provide insights on payout philosophy but do not indicate why firms pay dividends in the first place. Other behavioral theories such as dividend “catering” have no empirical support when researchers control for risk or firm age. Although Brav et al. (2005) find that corporate managers do not believe (or are unwilling to admit) that agency costs effect dividend policy, there is much empirical evidence that suggests otherwise.

The agency costs of retention represent the

27

underlying trade-off in the “life-cycle” hypothesis for dividend policy. Baker (2009) provides a comprehensive summary of these dividend policy theories. 2.7 Propensity to Pay and “Disappearing Dividends” As if the “dividend puzzle” and why firms pay dividends were not hard enough to solve, Fama and French (2001) raise the issue that corporate dividend policy may be changing over time. Fama and French (2001) show that the propensity to pay dividends declines dramatically from 1978 where 66.5% of all listed firms paid dividends to 1999 where only 20.8% of all listed firms paid cash dividends. Fama and French (2001) note that population of publically traded firms changed over the period with many new listings of small firms with low profitability but high growth opportunities. Such characteristics are typical for firms that do not pay dividends. However, Fama and French (2001) indicate that regardless of their characteristics, firms have become less likely to pay dividends. Fama and French (2001) call this declining propensity to pay “disappearing dividends”. Grullon and Michaely (2002) note that stock repurchases have been increasing as cash dividends decline and suggest a substitution effect. Julio and Ikenbery (2004) show that when the total distribution (cash dividends plus stock repurchases) from U.S. corporations is considered over the 1984-2003 period, the total payout is remarkably stable. The substitution of repurchases for dividends, however, did not explain the apparent decline in the fraction of companies paying dividends. Julio and Ikenberry (2004) then test several hypotheses to explain U.S. corporations’ propensity to pay dividends. Based on the firm maturity hypothesis, Julio and Ikenberry (2004) find that a logistic regression model with firm size and firm age variables explains most of the decline in the propensity to pay dividends, as well as an increased propensity to pay dividends between 2000 and 2004.

28

DeAngelo, DeAngelo, and Skinner (2004) show that while the number or fraction of dividend-paying firms declined in the prior two decades, the aggregate real dividends paid by U.S. industrial corporations actually increased. They explain this phenomena by indicating that the reduction in dividend payers occurs mainly among firms that paid only small dividends while large dividend payers actually increased real dividends. Thus, DeAngelo et al. (2004) show a trend in consolidation and concentration of dividends in the U.S. Then, Denis and Osobov (2008) confirm the DeAngelo et al. (2004) results for the international case. Denis and Osobov (2008) also find that reduction in dividend payers across the international markets occurs mainly among firms that paid only small dividends. Kooli and L’Her (2010) report a reduction in dividend payers in Canada as well as the concentration of dividends in Canadian firms. DeAngelo, DeAngelo, and Stultz (2006) advance a “life-cycle” theory of dividends where the firm’s “life-cycle” is characterized by its earned capital ratio. DeAngelo et al. (2006) show that firms with low earned capital ratios have little propensity to pay dividends while firms with high earned capital ratios have a greater propensity to pay dividends. They also show that when controlling for the earned capital mix, those firms with negative retained earnings have no change in the propensity to pay dividends from the 1970s to 2002. At the same time, they find that firms whose earned capital ratios make them probable dividend payers have even a larger reduction in propensity to pay than Fama and French (2001) report. Baker and Wurgler (2004a, 2004b) claim that their catering theory of dividends explains the “disappearing dividend” phenomena. However, Hoberg and Prabhala (2009) find no support for firms’ catering to transient dividend fads when they control for risk, and they report that risk explains about 40% of the “disappearing dividends” puzzle.

29

2.7.1 Corporate Governance and “Disappearing Dividends” In addition to their discussion of new listings, Fama and French (2001) propose better corporate governance as a potential solution to the “disappearing dividends” puzzle. They postulate that better corporate governance reduces the benefits of dividends in controlling agency problems between shareholders and managers. While none of the recent research on corporate governance seems to explicitly attempt to solve the “disappearing dividends” puzzle, several recent works do examine the relationship between corporate governance and dividend policy. Grinstein and Michaely (2005) examine the relationship between institutional holdings and dividend payouts, and they find that institutional investors avoid firms that do not pay dividends. As there has been a trend in the U.S. toward more institutional holdings, this research does not explain the “disappearing dividend”. Furthermore, Grinstein and Michaely’s (2005) results do not support models that predict that institutional investors cause firms to increase their dividend payout. Amihud and Li (2006) propose that an explanation for “disappearing dividends” is the decline in the information content of dividend announcements, which reduces the propensity of firms to use dividends as signals. Amihud and Li (2006) claim that institutional investors exploit superior information and buy before dividend announcements. Rubin and Smith (2009) find that the correlation between institutional ownership and volatility depends on the firm’s dividend policy. Jo and Pan (2009) show that firms with entrenched managers are more likely to pay dividends. Jiraporn and Chintrakarn (2009) show that firms with staggered boards are more likely to pay dividends. They also show that among firms that pay dividends, firms with staggered boards pay larger dividends. Furthermore, Jiraporn and Chintrakarn (2009) show that this impact of staggered boards is substantially larger than the effect of all other corporate

30

governance provisions combined. Again, as there has been a trend toward more staggered boards in the U.S., this research also does not explain the “disappearing dividend”. Sharma (2011) finds a positive relation between the propensity to pay dividends and board independence. In a study of the relationship between corporate governance and dividend policy in Canada, Adjaoud and Ben-Amar (2010) find that Canadian listed firms with stronger corporate governance have higher dividend payouts. In summary, the recent literature of corporate governance does not seem to indicate that dividends substitute for corporate monitoring, rather the evidence indicates that dividend policy is a complementary part of corporate governance. Further investigation of the relationship between dividend policy and corporate governance is outside the scope of this dissertation. 2.7.2 Firm Maturity or “Life-Cycle” Hypothesis Fama and French (2001) discuss the impact of new listings on the population of firms and begin to define the characteristics of dividend payers. Although they imply a firm life-cycle with the discussion of new listings not having the characteristics of dividend payers, Fama and French (2001) do not discuss or test life-cycle variables. Rather, Grullon et al. (2002) formalize the discussion of the maturity hypothesis. Grullon et al. (2002) suggest that dividends convey information about changes in a firm’s life-cycle3. They postulate that changes in dividends indicate a firm’s transition from a high growth phase to a mature phase. The key variable that Grullon et al. (2002) utilize to define the firm maturity is systematic risk. When Julio and Ikenbeery (2004) test the maturity hypothesis and explain disappearing and reappearing dividends, they use firm age as the variable to define the firm maturity. While Julio and Ikenberry (2004) seem to solve the disappearing dividend puzzle with the firm age, DeAngelo et al. (2006) use a different variable to define the firm’s life cycle, the earned capital 3

The general concept of a firm life cycle with growth stages is generally attributed to Mueller (1972).

31

ratio. Based on controlling for the earned capital ratio, DeAngelo et al. (2006) report that the disappearing dividend puzzle is larger than reported by Fama and French (2001). Using the earned capital ratio variable, Denis and Osobov (2008) find evidence to support the life-cycle hypothesis in a cross section of international markets. Wang, Ke, Liu, and Huang (2011) essentially duplicate the DeAngelo et al. (2006) results for firms listed on the Taiwan Stock Exchange. Two other studies of international dividend policy actually test the propensity to pay dividends using the earned capital ratio and risk as determinants, but the focus of the research lies outside the maturity hypothesis. Ferris, Sen, and Unlu (2009) investigate the propensity to pay across common law and civil law countries and find that the global decline in the propensity to pay dividends is more pronounced in firms incorporated in common law jurisdictions. Likewise, Twu (2010) investigates the probability of paying dividends with firms in 34 countries and reports that prior dividend payers are more sensitive to the earned capital ratio while non-payers are more sensitive to risk. Neither Ferris, Sen and Unlu (2009) nor Twu (2010) control for firm age or interactions between firm maturity variables, however. DeAngelo and DeAngelo (2007) and DeAngelo, DeAngelo, and Skinner (2008) summarize the life-cycle model as a foundation for dividend payout policy. In the life-cycle model, an optimum dividend policy is achieved when firms trade-off flotation costs and other retention costs against the agency costs of free cash flow. They point out that the trade-off evolves over a firm’s life-cycle. In the early stages, firms have ample growth projects and relatively less ability to generate sufficient funds internally, so they avoid dividend payouts. In their mature phase, firms pay cash dividends since they generate sufficient internal funds as the investment opportunities decline. Distributions in the form of dividends and stock repurchases are important in the mature phase of the life-cycle because firms would face substantial agency

32

costs of free cash flow if cash accumulated internally. They assert that this model explains much of the empirical literature. Recent research indicates that the firm life-cycle has an impact on other corporate financing decisions. DeAngelo et al. (2010) report that a firm’s life-cycle stage influences the probability that it conducts a seasoned equity offering. Owen and Yawson (2010) find a positive relation between the firm life-cycle and the likelihood of becoming a bidder in M&A activity. However, the major focus of this dissertation is further investigation of the relationship between the firm life-cycle and dividend policy. 2.8 Dividend Initiation Closely related to the studies on propensity to pay dividends is research on dividend initiation as such research investigates the question of why firms pay dividends. Wansley and Lane (1987) study the financial characteristics of firms initiating dividends and report little evidence of a residual dividend policy and some evidence of dividend signaling. However, Deshmukh (2003) uses a hazard model to study dividend initiation and reports results that are consistent with a pecking order explanation but inconsistent with a signaling explanation. Baker and Wurgler (2004a, 2004b) study the rate of dividend initiation and find evidence of investor fads or “catering”. However, Julio and Ikenberry (2004) show that Baker and Wurgler’s (2004a, 2004b) “dividend premium” disappears when dividend initiation announcement returns are adjusted for firm size and firm age. DeAngelo et al. (2006) find that the earned capital ratio is statistically significant in the decision to initiate dividends. Hoberg and Prabhala (2009) find that risk is negatively related to the decision to initiate dividends, and they find that the results are consistent with the firm maturity view of dividend policy. Thus as with the propensity to pay

33

research, the dividend initiation research uses different measures of firm maturity- firm age, earned capital ratio, and risk. Bulan et al. (2007) report that life-cycle factors are fundamental to the initiation decision; dividend initiators are firms that have reached the mature stage of their life cycle. Jain et al. (2009) also report evidence that dividend initiations are driven by life cycle considerations; however, IPO firms demonstrate a preference for repurchases over dividends. Officer (2011) finds that firms with low Tobin’s Q and high cash flow have significantly more positive dividend initiation announcement returns than other firms. The prior literature also contains many studies of market price reactions to dividend initiations such as Michaely, Thaler, and Womack (1995), but market price reactions to dividend initiations are outside of the scope of this dissertation. 2.9 Macroeconomics and Dividend Policy The major macroeconomic factor studied with dividend policy is taxation. Recent studies have focused on dividend policy changes as a result of the 2003 Jobs and Growth Tax Relief Reconciliation Act where the top marginal tax rate on dividends was reduced to 15%. Julio and Ikenberry (2004) conclude that the 2003 dividend tax cut had an effect on corporate dividend policy, especially considering the increase in the number of firms that initiated dividends after the tax cut was enacted. While Julio and Ikenberry (2004) indicate that the 2003 tax cut had an effect on dividend policy, they consider that there are other factors such as firm maturity, which also affect dividend policy. Using a different data sample and methodology than Julio and Ikenberry, Chetty and Saez (2005, 2006) indicate that the reversal in the propensity to pay dividends, or “reappearing dividends” was a result of the 2003 dividend tax cut. While there is strong empirical evidence that the 2003 dividend tax cut did affect corporate dividend policy, there is some question of motive. Brown, Liang, and Weisbenner (2007) show that executives

34

with higher ownership were more likely to increase dividends after the tax cut in 2003. Brown et al. (2007) claim that while their work shows strong evidence of a tax effect on dividend policy, their results are also consistent with an agency theory perspective that managers may be inclined to consider their own financial incentives rather than the best interests of the shareholders. Until recently, little research has considered the relationship between other macroeconomic factors and dividend policy.

However, Dittmar and Dittmar (2008) study

corporate financing waves or patterns and find that stock repurchases follow the aggregate pattern of equity issuance and mergers. They conclude that these waves or trends in corporate financing decisions result from growth in GDP. Dittmar (2008) follows with additional research linking macroeconomics, corporate cash policy, and dividend policy. Especially relevant to this research, Dittmar (2008) shows that corporate cash policy (consequently then dividend policy) is highly determined by risk, and macroeconomic factors are a significant risk factor. Interestingly, Dittmar (2008) shows that over the 1980-2006 time period when dividends were “disappearing”, corporate cash holdings as a percent of total assets increased significantly.

She attributes the

increase in corporate cash holdings over the 25-year period to the increase over time in corporate risk. Dittmar (2008) discusses intense economy-wide competition and more “focused” core businesses as macroeconomic risk factors. Irvine and Pontiff (2009) also argue that economywide competition has increased volatility of firms’ cash flows. They find that over the past 40 years, the increase in volatility of firms’ cash flows has mirrored the increase in firms’ return volatility. In summary, there has been little research into the effects of macroeconomic factors on dividend policy, except taxes. However, some research on economic growth and competition suggest that these macroeconomic factors may influence dividend policy and deserve further

35

study. In this dissertation, I examine year effects on the propensity to pay dividends and the propensity to initiate dividends rather than specific macroeconomic variables. 2.10 Dividend Policy and Firm Value Most of the early literature relating dividend policy to firm value stems from asset pricing research. Black and Scholes (1974) add a dividend yield term to an empirical version of the CAPM and report that the dividend yield has no impact on the required return. This study has been criticized for low statistical power and the use of annual data. Litzenberger and Ramaswamy (1982) report that stocks with higher dividend yields have higher required returns. Similarly, Naranjo et al. (1998) investigate returns as a function of the Fama-French (1992) factors and a dividend yield term and also find that higher dividend yields have higher required returns. A criticism of these tests in the context of the effect of dividend policy on valuation is that stocks with high dividend yields could have (1) high dividend payouts or (2) low prices. A high yielding stock could have a low price due to any number of reasons unrelated to dividend policy; consequently, dividend yield is a poor proxy for overall dividend policy.4 Baker and Wurgler (2004a, 2004b) develop a catering theory and utilize the market to book ratio (M/B) for the valuation of firms. Fama and French (2001) claim that the M/B is a suitable proxy for Tobin’s Q (1969), which is a common measure of corporate valuation.5 Although Julio and Ikenberry (2004) and Hoberg and Prabhala (2009) refute the catering theory, Baker and Wurgler (2004a, 2004b) present empirical data which indicates that non-paying firms have higher equal weighted and value-weighted M/B ratios than dividend paying firms. The

4

The results in Section 5 confirm the prior literature that high dividend yields reduce firm value for dividend payers. Although a high dividend yield reduces firm value, this does not indicate that dividends are always undesirable. The results in Section 5 also show that dividend paying firms can have higher firm values than non-paying firms depending on the firm maturity. 5 In this dissertation, I utilize the Fama and French (2001) definition of the market value of assets to the book value of assets, M/B. See Section 4 for the calculation details.

36

explanation for the higher M/B ratio for the non-paying firms is that non-paying firms have more growth potential. DeAngelo et al. (2006) also report that non-paying firms have a higher median M/B than dividend paying firms. Given that the valuation (based on M/B) of non-paying firms is higher than the valuation of dividend paying firms, then why would a firm ever pay a dividend? Clearly, this is an anomaly in the dividend literature, and this dissertation resolves the dividend anomaly in Section 5. Most studies investigating the determinants of the market to book ratio (M/B) focus on capital structure research rather than valuation. For example, Chen and Zhao (2006) examine the relation between the M/B ratio and the leverage ratio. My approach closely follows Shin and Stulz (2000) who utilize the M/B ratio to proxy firm value. 2.11 Dividend Growth Based on his interviews with financial managers regarding dividend policy, Lintner (1956) proposed a dividend growth model based on dividend changes. His partial adjustment model is based on a target dividend payout and Lintner (1956) claims that the model explains 85% of the dividend changes in his sample. Fama and Babiak (1968) also find support for Lintner’s partial adjustment model when they test dividend policy in sample as well as with a validation sample. Brav et al. (2005) confirm Lintner’s (1956) findings using survey and field interview data. Managers continue to show a strong tendency to avoid dividend cuts and tend to increase dividends only after they are reasonably sure that they can be maintained. As a result for firms that pay a dividend, dividend payments tend to be smoothed from year to year, a practice that is called dividend smoothing. Aivazian, Booth, and Cleary (2006) show that firms that issue public debt are more likely to pay dividends and follow a dividend smoothing policy than firms that rely on bank debt. They argue that firms with public debt have a greater incentive

37

to adopt dividend polices that reduce agency problems. Aivazian et al. (2006) also estimate dividend changes using the Lintner partial adjustment model. Leary and Michaely (2011) report that corporate dividend smoothing policies are related to maturity. They find that younger and smaller firms smooth less, while firms with high free cash flow and low growth potential smooth more. Financial ratio analysis indicates that there are financial constraints on dividend growth and that dividend growth is interdependent on the firm’s operations, investment, and financing decisions. Higgins (1977) introduces the concept of sustainable growth for a restrictive case where the firm maintains a target payout ratio and capital structure without issuing new equity. Popular investments textbooks such as Jordan and Miller (2009) demonstrate the simple calculation of sustainable growth rate as a means to estimate the dividend growth rate in order to implement dividend discount models for valuation. However, as most equity analysts are aware, the sustainable growth assumptions are not realistic for most firms. Financial analysts can find better methods for estimating growth rates in popular textbooks from Damodaran (2011, 2012).6 The recent literature studying dividend growth rates has focused on time series methods. Lettau and Ludvigson (2005) use time series cointegration methods to investigate the relationship between returns and dividend growth using aggregate consumption, dividend, and income data from 1948-2001. Other studies such as Chen (2009) and Engsted and Pedersen (2010) also use time series models and aggregate data to investigate the predictability of dividend growth. Most studies find the relation between past growth and future growth to be

6

In addition to the textbooks Damodaran (2011, 2012), Damodaran posts valuation papers on his website, see http://pages.stern.nyu.edu/~adamodar/New_Home_Page/papers.html

38

weak. However, the focus on this research is on the cross-sectional characteristics of firms with regards to the dividend maturity hypothesis and not aggregate time series analysis. 2.12 Dividend Policy and Complexity Herbert Simon (1978) introduces the notion of complexity in Finance when he argued that the human brain is not capable of anything but very simple computations and called this “bounded rationality”. Given that humans have this “bounded rationality” it may not be possible to solve complex optimization problems. According to the premise of the maturity hypothesis, dividend policy is a complex optimization problem, where the optimal dividend policy maximizes the value of the firm by minimizing the costs of earnings retention and the cost of earnings distribution. One of the factors in establishing these costs is determining the investment opportunity set for the firm. Selecting such an “optimal investment portfolio” is one of the most common ways that complexity arises in Finance. Kao and Tate (2001) show that even when dealing with defined investments such as stock indices, the optimization of a proxy is formidable to the extent of being computationally intractable. Of course then for a firm, the problem is even more complex since the potential investment opportunity set for the firm is limited only by the “bounded rationality” of the firm’s management. Another factor in establishing the cost of retention of earnings retention is determining the agency costs, which also are likely to be computationally intractable. Furthermore, Welch (2010) argues that corporate behavior is so highly complex that researchers do not even understand the basics of managerial motivations. Given the complexity of the optimal dividend policy that maximizes firm value, this research does not attempt to find an analytical solution. Rather, this dissertation uses empirical observations to test the maturity hypothesis.

CHAPTER 3 MODELS 3.1

Dividend Growth Model With no prior research relating a firm’s dividend growth rate to the firm’s life-cycle, I

dervive a statistical model relating the dividend growth rate to maturity. A simple dividend growth model can be derived from financial ratio analysis and the definition of the dividend payout ratio. Commonly, the firm’s dividend payout ratio is defined as the fraction of its earnings or net income that the firm pays in common dividends. Then I can expand the dividend payout ratio definition to provide an equation for the total dividends per share paid, Div, as follows: Dividends per share = Net Income x Dividend Payout Ratio Number of common shares outstanding

Equation 1

or Div = NI x pay shares where: Div = common dividends paid per share NI = Net income shares = number of common shares outstanding pay = dividend payout ratio

Equation 2

To get to the dividend growth rate, I calculate the change in dividends per share by first taking the three partial derivatives, ∂Div = NI ∂pay shares

Equation 3a

∂Div = pay ∂NI shares

Equation 3b

39

40

∂Div = -pay x NI ∂shares shares2

Equation 3c

and then computing the total differential change in dividends per share, ΔDiv, as follows:

ΔDiv = NI Δpay + pay ΔNI - pay x NI Δshares shares shares shares2

Equation 3d

The dividend growth rate, g div is then: Dividend growth rate = change in dividends per share dividends per share

Equation 4

or g div = ΔDiv Div

Equation 5

Then, substituting Equation 3d and Equation 2 into Equation 5 yields: gdiv = Δpay + ΔNI - Δshares pay NI shares

Equation 6

or g div = g pay + g NI – g shares where:

3.2

Equation 7

g div = growth rate in dividends per share g pay = growth rate of dividend payout ratio g NI = growth rate of the firm’s net income g shares = growth rate of the firm’s shares outstanding Sustainable Growth In the sustainable growth case proposed by Higgins (1977), the firm is assumed to

maintain a target dividend payout ratio and maintain a target capital structure without issuing

41

equity so that g

pay

=g

shares=0.

Thus in the sustainable growth case, the dividend growth rate

equals the growth in net income. Sustainable dividend growth rate = growth rate in net income

Equation 8

or sustainable g div = g NI = ΔNI NI

Equation 8a

The usual assumption to model the growth in net income is to assume that the firm does not grow unless there is new investment. If the change in earnings or net income comes from new investment made with retained equity, then we have the following: Change in net income = New investment x Return on Equity

Equation 9

or ΔNI = ΔRE x ROE where:

Equation 10 Δ NI = change in net income Δ RE = addition to retained earnings, which equals new investment ROE = return on equity

Then for the sustainable growth case, I substitute Equation 10 into Equation 8a, such that: sustainable g div = ΔRE x ROE NI

Equation 11

or sustainable g div = b x ROE where

Equation 12

sustainable g div = dividend growth rate under sustainable growth assumptions b = retention ratio or plowback ratio b= ΔRE NI b= 1-pay ROE = return on equity

42

Equation 12 is the sustainable growth equation shown in investment textbooks such as Jordan and Miller (2009). However, this analysis implies that the sustainable growth assumptions are very restrictive and may not be realistic for many firms. From the derivation, it is clear that any firm that changes its payout ratio or issues new shares will violate the sustainable growth assumptions. 3.3

A Statistical Model for Dividend Growth Rate An objective of this research in Hypothesis 4 is to show that the variables of the dividend

maturity hypothesis explain the dividend growth rate in the cross-section of dividend paying firms with dividend growth. From Equation 7, the dividend growth rate depends on the growth rate of net income, the growth rate of the payout ratio, and the growth rate in shares outstanding. As stated, Equation 12 is the most often used approximation to estimate the dividend growth rate. I propose that a simple statistical model could improve the estimating accuracy of Equation 12 as well as provide the basis for testing the dividend maturity hypothesis variables. For the hypothesis testing, let’s suppose that the sustainable growth rate is an imperfect measure of the change in net income, even if the payout ratio is constant and the number of shares outstanding is constant because it is likely that the firm’s ROE will change with maturity. Furthermore, it is likely that the growth rate in payout ratio and growth rate in shares outstanding are also dependent on the firm’s maturity.

Summarizing these hypothesis statements in terms of

equations: g NI = f (sustainable g div, M)

Equation 13a

g pay =f (sustainable g div, pay, M)

Equation 13b

g shares =f (M)

Equation 13c

43

where

sustainable g div = dividend growth rate under sustainable growth assumptions g NI = growth rate of net income M = firm maturity variable g pay = growth rate in dividend payout ratio pay = dividend payout ratio g shares = growth rate in shares outstanding

Then from Equation 7, the dividend growth rate of a firm can be expressed as: g div = f (sustainable g div, pay, M)

Equation 14

The statistical model form of Equation 14 is then given in Model 1: g div = β0 + β1 sustainable g div + β2 pay +β3 M + ∑ βi x (Control variable i) + ε where

Model 1

g div = dividend growth rate sustainable g div = dividend growth rate under sustainable growth assumptions pay = dividend payout ratio M = firm maturity variable β0, β1, β2, β3, βi =regression coefficients ε = error term

Hypothesis tests on the estimated β3 coefficient then will indicate if the dividend growth rate depends on firm maturity. Section 5 reports the regression analysis of the relationship between the dividend growth rate, sustainable growth rate, and firm maturity.

CHAPTER 4 DATA AND METHODOLOGY 4.1 Propensity to Pay Dividends 4.1.1 Data Sample The data sampling procedure follows that of Fama and French (2001) and DeAngelo et al. (2006). As with the prior research, this study excludes financial firms and utilities by excluding those firms with Standard Industrial Classification (SIC) codes in the intervals of 4900-4949 and 6000-6999.

The analysis only considers NYSE, NASDAQ, and AMEX

industrial firms that have Center for Research in Security Prices (CRSP) share codes with 10 or 11 and that are incorporated in the United States according to Compustat. These restrictions eliminate ADRs, closed-end funds, ETFs, and real estate investment trusts (REITs). Given that Fama and French (2001) report 1978 as the reference point for “disappearing dividends”, this research focuses on the 1982-2010 time period.7 To be included in the sample, a firm must have non-missing annual data values for dividends and financials from Compustat, as well as return data from CRSP.

Following DeAngelo et al. (2006), firms with negative total equity are

removed from the sample. 4.1.2 Control Variables Fama and French (2001) report the firm characteristics for dividend paying firms and use logit regressions to provide evidence that the propensity to pay dividends depends on three fundamental factors- profitability, investment opportunities, and size. While Fama and French (2001) and Hoberg and Prabhala (2009) use earnings before interest expenses divided by assets, DeAngelo et al. (2006) simply use current and lagged return on assets, ROA, as the measure of 7

Since the models require prior dividend status and growth rates, the data series actually begins in 1981.

44

45

profitability. This study follows DeAngelo et al. (2006) and uses current return on assets, ROA for the measure of profitability. Fama and French (2001) and Hoberg and Prabhala (2009) use asset growth rate as the measure of investment opportunities. DeAngelo et al. (2006) claim that asset growth rate is less than ideal, but also report that logit results are similar with asset growth rate, sales growth rate, or market to book ratio. This study follows DeAngelo et al. (2006) and uses sales growth rate and market to book ratio to measure investment opportunities. Julio and Ikenberry (2004) use the industry classifications to control for the profitability and investment opportunities. Although they indicate statistical significance of the industry classification codes, they do not report any coefficients. I discuss the use of industry codes in a later section. Fama and French (2001), DeAngelo et al. (2006), and Hoberg and Prabhala (2009) all use the NYSE market capitalization percentile as a measure of firm size. The analysis of time series with variables growing in time is problematic. To address this, Fama and French (2001) use the measure of size to be the NYSE percentile, that is, the percent of NYSE firms that have the same or smaller market capitalization. This size measure neutralizes the effects of the growth in a typical firm size through time. Julio and Ikenberry (2004) use the natural logarithm of the firm size decile. Since this study closely follows DeAngelo et al. (2006), this dissertation uses NYSE market capitalization percentile as the measure of firm size. In summary, I use the following control variables as measures of profitability, investment opportunities, and size: ROAt = return on assets in current year t SGR = sales growth rate, which equals (sales t / sales t-1) - 1 M/B = market to book ratio NYSE Percentile = NYSE market equity capitalization percentile.8 Note that the M/B ratio is as defined by Fama and French (2001):

8

I utilize the NYSE market equity capitalization percentile breakpoints provided at Dr. Kenneth R. French’s website, http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html

46

M/B = Book assets minus book equity plus market equity all divided by book assets. Market Equity= Year closing price times shares outstanding. Book Equity= Stockholders Equity minus Preferred Stock plus Balance Sheet Deferred Taxes and Investment Tax Credit minus Post Retirement Asset. If Stockholder’s Equity is not available, it is replaced by either Common Equity plus Preferred Stock Par Value or Assets minus Liabilities. Preferred Stock is Preferred Stock Liquidating Value or Preferred Stock Redemption Value or Preferred Stock Par Value. 4.1.3 Other Control Variables DeAngelo et al. (2006) introduce several other key control variables for the logit regressions. Given that their key life-cycle variable is the earned to contributed capital ratio, DeAngelo et al. (2006) control for the total equity to total asset ratio, or TE/TA. I include this control variable, as well as the cash to total asset ratio, Cash/TA used by DeAngelo et al. (2006). Finally, DeAngelo et al. find the lagged dividend status is highly significant in all logit regressions and I also include this indicator variable. Based on DeAngelo et al. (2006), I use the following additional control variables: TE/ TA = total equity to total asset ratio Cash/TA = cash to total asset ratio Divt-1 = lagged dividend status with value of 1 if the firm paid a dividend in the prior year and zero otherwise. 4.1.4 Maturity Hypothesis Variables Prior research shows that different maturity variables significantly explain a firm’s propensity to pay dividends when tested individually. The central focus of Hypothesis 1 is to jointly test the different maturity variables in order to determine which of the different maturity variables best explains the firm’s propensity to pay dividends. Julio and Ikenberry (2004) test

47

the maturity hypothesis with the logarithm of the firm age. They show that the firm’s propensity to pay dividends increases with the logarithm of the firm age; consequently, I follow Julio and Ikenberry (2004) and use the logarithm of the firm age for this dimension of maturity. The proxy for firm age is the number of years that the firm is in existence in the CRSP database. Although this is an imperfect measure since the CRSP database starts in 1925 and some firms have been in existence for longer periods, this measure has been used in the prior literature, for example Shumway (2001). Although this measure of firm age is subject to some truncation for firms in existence prior to 1925, the number of firms affected by the truncation is a very small fraction of the total number of observations. Finally in each year of the time series, I rank ages of each firm and compute the age percentile. DeAngelo et al. (2006) test the maturity or life-cycle hypothesis with the earned capital ratio with the retained earnings to total equity ratio (RE/TE) and with the retained earnings to total asset ratio (RE/TA), and I use these maturity variables. DeAngelo et al. (2006) show that the propensity to pay dividends increases with higher values of RE/TE or RE/TA as these variables characterize the firm’s “financial” stage in the life-cycle. In addition, I rank the RE/TE ratio of each firm in each year of the time series and compute the RE/TE percentile. Hoberg and Prabhala (2009) regard risk as a proxy for firm maturity and use idiosyncratic risk and systematic risk in the logit regressions. Since the firm’s risks decline with maturity, they show that the firm’s probability of paying a dividend is greater when the risk is lower. However, the use of risk measures can be problematic in regression since risk is difficult to observe and define. While systematic risk, as defined by Beta, is directly used as an independent variable in regressions on dividend payout rates in the research by Rozeff (1982), this violates the classical errors-in-variables assumption.

Hoberg and Prabhala (2009) resolve the errors-in-variables

48

problem by regressing the excess stock return on the excess market return as one normally computes Beta. However, rather than using Beta directly, Hoberg and Prabhala (2009) use the standard deviation of the residuals to measure each firm’s idiosyncratic risk. Then they use the standard deviation of the predicted values from the regressions as a measure of the systematic risk. I follow Hoberg and Prabhala (2009) and use standard deviations as measures of risk; however, I use the standard deviation of the firm’s monthly returns as done by Ferris, Sen, and Unlu (2009). I rank the standard deviation of monthly returns for each firm in each year of the time series and compute the standard deviation percentile. Inasmuch that each of these maturity variables captures a different perspective of a firm’s maturity, it seems likely that an interaction of these variables may also be statistically significant. I investigate the interaction between the firm maturity variables in two ways. First, I define a maturity factor variable, M, which is simply the product of multiplying the maturity variables together. Note, however, I multiply by the inverse of total risk since risk decreases with firm maturity. This variable definition of maturity follows the approach where the interaction of variables is modeled by multiplication of the factors. The maturity factor variable has an interesting interpretation at a value of zero. When the maturity factor has a value of zero, this represents the maturity at the time of incorporation. Note that the maturity factor takes on negative values when the RE/TE is negative. The other approach to investigate the interaction between maturity variables is to define a maturity composite, where the composite maturity is the sum of the percentile rankings of each factor. Thus to compute the composite maturity, the firms in the sample are ranked each year for the firm maturity variables-age, earned capital ratio, and risk. After ranking the firms, I add the percentile age, percentile earned capital ratio, and inverse percentile risk to compute the maturity

49

composite.

Due to the ranking definition, the maturity composite has a natural scale that

represents the firm life cycle. The minimum maturity composite score is zero, indicating the lowest level of maturity. The maximum maturity composite score is three (which corresponds to a firm that ranks at the 100th percentile of age, earned capital ratio, and inverse standard deviation), indicating the highest level of maturity. The regression of highly correlated variables, or multicollinearity, causes an increase in the variance of the estimated coefficients. While it is very likely that the three measures of firm maturity are indeed highly correlated, it is better to address the increase in variance than omitting a maturity variable, which could lead to bias. To mitigate this variance inflation, this study uses a very large data set with nearly 100,000 observations. In summary, I test the following maturity hypothesis variables: Log(firm age) = natural logarithm of the firm age (in years) Age Percentile = percentile rank by firm age RE/TE = ratio of retained earnings to total equity RE/TE Percentile = percentile rank by RE/TE RE/TA = ratio of retained earnings to total assets Total risk = standard deviation of firm’s monthly returns Standard Deviation Percentile =percentile rank by standard deviation of monthly returns Maturity Factor= Log(firm age) x RE/TE Total risk

= interaction variable of measures of firm maturity

Maturity Composite = the sum of the percentile rankings of age, earned capital ratio, and inverse of risk. 4.1.5 Industry Dummy Variables The prior research on the propensity to pay dividends is rather inconsistent with regards to industry controls. Julio and Ikenberry (2004) use the industry classifications to control for the profitability and investment opportunities. Hoberg and Prabhala (2009) control only for

50

Technology firms, while DeAngelo et al. (2006) use no industry controls. It seems that a firm’s economic sector should affect dividend policy as the firm’s growth potential is related to the overall industry. To control for the economic sector, I code each observation with a dummy variable representing the S&P economic sectors listed in Compustat. The dummy industry variable is assigned a value of one if the firm is contained in the S&P economic sector and zero otherwise. The following industry dummy variables are included for industry control variables: D Materials =

industry dummy variable assigned a value of one if the firm is included in the Materials economic sector, and zero otherwise. D Consumer Discretionary = industry dummy variable assigned a value of one if the firm is included in the Consumer Discretionary economic sector, and zero otherwise. D Consumer Staples = industry dummy variable assigned a value of one if the firm is included in the Consumer Staples economic sector, and zero otherwise. D Health Care = industry dummy variable assigned a value of one if the firm is included in the Health Care economic sector, and zero otherwise. D Energy = industry dummy variable assigned a value of one if the firm is included in the Energy economic sector, and zero otherwise. D Industrials = industry dummy variable assigned a value of one if the firm is included in the Industrials economic sector, and zero otherwise. D Technology = industry dummy variable assigned a value of one if the firm is included in the Technology economic sector, and zero otherwise. D Telecom = industry dummy variable assigned a value of one if the firm is included in the Telecommunication Services economic sector, and zero otherwise. Note that in order to have linear independence in the regression models, the industry dummy variable D Industrials is omitted. Then the coefficients on the remaining industry dummy variables represent the change from the Industrial economic sector. 4.1.6 Year Dummy Variables In order to test Hypothesis 2, I use year dummy variables to test for year effects. In strict econometric terms, the base year in the data panel is 1982, thus, all coefficients on the dummy variables represent the change from the base year of 1982. Each year dummy variable of year t is assigned a value of one if the observation occurs in year t, and zero otherwise. These variables

51

indicate whether the propensity to pay dividends, after controlling for the control variables and the maturity hypothesis variables, changes from the base year. Analysis of the regression coefficients will indicate if the propensity to pay changes gradually over time or if the change is even statistically significant. Thus, the year dummy variables are: D1983 to D2010 = year dummy variable of year t is assigned a value of one if the observation occurs in year t, and zero otherwise. 4.1.7 Dependent Variable In the research on the firm’s propensity to pay a dividend, the dependent variable is the firm’s dividend paying status in year t. Thus, the dependent variable is a dummy variable that is assigned a value of one if the firm paid a dividend in year t, and zero otherwise. For the logit model, the dependent variable is: D = dividend paying status in current year t. A dummy variable that is assigned a value of one if the firm paid a dividend in year t, and zero otherwise. 4.1.8 Fama and MacBeth Logit Model Consistent with the prior literature, I apply the Fama and French’s (2001) and Fama and MacBeth’s-based (1973) statistical methodology to determine whether the probability that a firm pays dividends depends on the maturity variables and the control variables. This procedure utilizes a multivariate logit model that treats the binary indicator of payment/nonpayment of the dividend as the dependent variable and the maturity variables and control variables as explanatory variables. Following Fama and MacBeth (1973), I run separate logit regressions for each of the 29 sample years (1982-2010) to obtain a time series of fitted coefficients. The mean coefficient, which I report, is the mean value from the 29 logit regressions (one for each year over 1982-2010). The t-statistics for each mean coefficient reported in the tables are based on the

52

hypothesis that the expected coefficient is zero. Consistent with Fama and French (2001) and DeAngelo et al. (2006), the tables report t-statistics unadjusted for serial correlation. 4.1.9 Panel Logit Model This study follows the prior dividend policy literature from Fama and French (2001) where a logit regression model is used to explain the probability that a firm pays a dividend. In the prior literature, Fama and French (2001), Julio and Ikenberry (2004), DeAngelo et al. (2006), and Hoberg and Prabhala (2009) use the Fama and MacBeth (1973) time series averages of the annual cross-sectional logit coefficients. In addition, I follow the method used by Al-Kuwari (2010) who studies the decision to pay dividends in emerging Middle East markets. Specifically, Al-Kuwari (2010) uses a random effects panel logistic regression to determine the probability that firms listed on the Gulf Cooperation Council (GCC) stock exchange pay a dividend. The pooling of multiple years of annual data in a data panel enables the inclusion of the time dummy variables in the logit model. With the time dummy variables in the logit model, I perform direct hypothesis tests for yearly changes in the probability that a firm pays dividends. While pooling data can lead to econometric issues, the panel logit accounts for the effects of clustering. Therefore the panel logit model provides a robustness test on the Fama and MacBeth logit model as well as the means to test Hypothesis 2 for year effects. Following Hoberg and Prabhala (2009), the average logit partial effects are computed to determine the magnitude of the significant maturity variables. In order to test Hypothesis 3, the outliers from the logit model must be determined. For this dissertation, a correct prediction from the logit model for a dividend payer is any predicted probability greater than or equal to 0.5. The selection of 0.5 as the cutoff for the prediction model is simply based on the premise that if the logit model predicts the probability of paying a

53

dividend as 0.5 or greater, the logit model is predicting that the firm is more likely to be a dividend payer. Likewise, a correct prediction for a non-payer is any predicted probability less than 0.5. Again, this is based on the idea that if the logit model predicts the probability of paying a dividend is less than 0.5, then the logit model is predicting that the firm is more likely to be a non-payer. After classifying the observations into those that fit the model and those that do not fit the model, I compare the median M/B ratio of non-payers that do not fit the logit model to the median M/B ratio of non-payers that fit the logit model. Then I compare the median M/B ratio of dividend payers that do not fit the logit model to the median M/B ratio of payers that fit the logit model. I perform a two-sided, median two-sample test to determine the level of significance for the differences in the medians. 4.2 Dividend Initiation Another perspective of dividend policy that is closely related to propensity to pay research is the study of dividend initiation. This is simply because changes in the dynamics of propensity to pay are rooted in dividend initiation. Prior research indicates that over 90% of dividend payers continue to pay dividends, thus omitting a dividend is rare and not a significant factor in the dynamics of propensity to pay. DeAngelo et al. (2006) study the life cycle impact on dividend initiations and find that the earned/contributed capital ratio is significant. Although DeAngelo et al. (2006) find that logit models with the earned capital ratio correctly classify about 95% of the observations as dividend initiators or not, they do not control for the other maturity variables of risk or firm age. Consistent with the maturity hypothesis, Hoberg and Prabhala (2009) report that the probability of dividend initiation is negatively related to risk, but they do not control for the firm age or earned capital ratio.

54

4.2.1 Data Sample and Variables I follow the method of Hoberg and Prabhala (2009) and start with all firms that do not pay a dividend in year t -1 and define dividend initiators as firms paying a dividend in year t. From the full data sample described in Section 4.1, I eliminate the dividend payers at year t-1 since they have already initiated dividends. The control variables, maturity variables, and dummy variables are the same as described in Section 4.1. Note that while the prior dividend status variable, Divt-1 is used to sort the firms that were already dividend payers in year t-1, the prior dividend status variable, Divt-1 is not used in the logit regression on the probability of dividend initiation. The only new variable requirement for the dividend initiation logit model is the dependent variable, which is a dummy variable indicating if the firm is a dividend initiator in year t or not. Thus, the dependent variable for the dividend initiator logit model is: DINT = dividend initiator status in current year t. A dummy variable that is assigned a value of one if the firm pays a dividend in year t after having not paid dividends in year t-1, and zero otherwise. 4.2.2 Logit Models This study follows the prior dividend policy literature from DeAngelo et al. (2006), and Hoberg and Prabhala (2009) where a logit regression model is used to explain the probability that a firm initiates a dividend. In the prior literature, DeAngelo et al. (2006) and Hoberg and Prabhala (2009) use the Fama and MacBeth (1973) time series averages of the annual crosssectional logit coefficients. I follow the prior literature and use the Fama and MacBeth logit model as described in Section 4.1.

55

In addition, I follow Al-Kuwari (2010) and utilize a random effects panel logistic regression model as described in Section 4.1. Therefore the panel logit model provides a robustness test on the Fama and MacBeth logit model as well as the means to test Hypothesis 2 for year effects. 4.2.3 Hazard Model of Dividend Initiation Deshmukh (2003) reports another econometric model to study dividend initiation, the hazard or duration model. Deshmukh (2003) explains that in the context of research on dividend initiation, the hazard rate is the rate at which a non-dividend paying firm initiates a dividend. The probability associated with a dividend initiation is inversely related to the time until the firm pays a dividend. Hence, the higher the hazard rate or probability, the sooner the firm will pay a dividend. Likewise, the time until the dividend initiation is the duration associated with the non dividend paying status. This shift over time can be attributed to the changes in the firm’s specific dividend policy, which I model with the maturity or life-cycle hypothesis. Bulan et al. (2007) also use hazard analysis to study the timing of dividend initiation over the firm life cycle. From the full data sample described in Section 4.1, I eliminate the dividend payers at year t-1 since they have already initiated dividends. Therefore, the hazard model for dividend initiation uses the same data set as the logit models for dividend initiation. The hazard in this part of the analysis is the risk of dividend initiation by all non-dividend paying firms. Note however, that once a firm initiates a dividend in a specific year, it no longer is part of the survivor set. While the control variables and the dummy variables then remain the same as used in the dividend initiator logit models, the firm age cannot be used as a maturity variable. In this hazard analysis, the variable firm age and the event history time are identical.

56

Following Deshmuh (2003), I use a Cox proportional hazards model to model the hazard rate as a function of the specified explanatory variables. Since the assumption of proportionality indicates that some of the explanatory variables are time dependent, I include the time dependent variables for the non-proportional predictors. 9 4.3 Valuation Following Baker and Wurgler (2004a, 2004b), I use the market to book ratio (M/B) as the proxy for firm value. The objective of the analysis is determine if there is a relationship between valuation and the firm life-cycle. Damodaran (2011) advocates valuation approaches based on the life-cycle, but his analysis of the life-cycle is qualitative rather than quantitative. In this analysis, I use the maturity variables to quantify the life-cycle. 4.3.1 M/B Regressions Based on the Baker and Wurgler (2004a, 2004b) observations, I first divide the full sample described in Section 4.1 and sort the dividend payers from the non-payers. To investigate the relationship between valuation and maturity, I consider three econometric techniques applicable for the analysis: cross-sectional using Fama and MacBeth’s (1973) procedure to break the serial correlation, random effects panel regression, and fixed effects panel regression. To begin the analysis, the regression model simply has the M/B ratio as the dependent variable and the maturity factor is the only explanatory variable. Based on Damodaran (2011), I expect the valuation to change over the firm life-cycle. In order to identify these potential changes in valuation over the life-cycle, I segment the life-cycle into several ranges for piecewise regression in each segment of the life-cycle.For robustness, I segment the life-cycle using the maturity factor variable as well as the maturity composite variable. 9

See http://www.ats.ucla.edu/stat/sas/seminars/sas_survival/default.htm for non-proportional predictors.

57

After the analysis with the simple regression models, I include the control variables and industry dummy variables described in section 4.1for the multivariate regressions with the M/B ratio as the dependent variable. 4.4 Dividend Payout Policy In this section of the dissertation, I extend the maturity hypothesis to dividend payout policy in order to determine the relationship of payout policy with maturity. I examine dividend growth, dividend cuts, the dividend payout ratio, the dividend yield, and the dividend growth rate. 4.4.1 Dividend Growth Logits In Hypothesis 4, I determine if the life-cycle or maturity hypothesis explains dividend growth. One way to gain some insight on the issue of dividend growth is to investigate the probability of a dividend increase. From the full data sample described in Section 4.1, I restrict the sample to dividend payers at year t. For this logit analysis of the probability of dividend growth, the control variables, maturity variables, and industry control variables are the same as discussed above in Section 4.1, with the exception of two additional control variables. First, I include the earnings growth rate to account for the observation that dividend payout changes follow earnings changes. Inclusion of the earnings growth rate variable eliminates observations with negative earnings. 10 Finally, I include the prior dividend payout ratio to account for the fact that the prior dividend payout ratio in year t-1 may affect the decision to increase/decrease the dividend in year t. However, for the case of dividend growth, I define a new dependent variable for the logit models. The dependent variable is a dummy variable with a value of one if the firm increased the

10

With the dividend paying firms, there is only a small loss of observations since most dividend payers are profitable. This variable is not used with the non-payers since the loss of observations would be significant.

58

dividends in year t and zero otherwise. Hence, the new variables introduced to model the probability that a firm increased its dividend are: Control variables EGR= earnings (before extraordinary items) growth rate Pay t-1 = dividend payout ratio in year t-1 Dependent variable DG = dividend growth status in current year t. A dummy variable that is assigned a value of one if the firm increased the dividend in year t, and zero otherwise. For the purposes of this research, the tolerance on a dividend increase is set such that a dividend increase is any annual dividend increase greater than $0.01/ share. I use the Fama and MacBeth logit model as described in Section 4.1. In addition, I utilize a random effects panel logistic regression model as described in Section 4.1, which provides a robustness test on the Fama and MacBeth logit model. 4.4.2 Dividend Cut Logits Since dividend cuts are so rare, it is of great theoretical and practical interest to investigate the probability of a dividend cut and determine if maturity is a factor. From the full data sample described in Section 4.1, I restrict the sample to dividend payers at year t. For this logit analysis of the probability of dividend cut, the control variables, maturity variables, and industry control variables are the same as discussed above in Section 4.4.1. However, for the case of a dividend cut, I define a new dependent variable for the logit models. The dependent variable is a dummy variable with a value of one if the firm cut the dividends in year t and zero otherwise. Hence, the new dependent variable introduced to model the probability that a firm cut its dividend is:

59

DC = dividend cut status indicator in current year t. A dummy variable that is assigned a value of one if the firm cut the dividend in year t, and zero otherwise. For the purposes of this research, the tolerance on a dividend cut is set such that a dividend cut is any annual dividend decrease greater than $0.01/ share. I use the Fama and MacBeth logit model as described in Section 4.1. In addition, I utilize a random effects panel logistic regression model as described in Section 4.1, which provides a robustness test on the Fama and MacBeth logit model. 4.4.3 Regressions on Dividend Payout Ratio In this section, I utilize regression analysis to study to investigate the dividend payout ratio and the life-cycle. However, before the regression analysis of the relationship between maturity and the dividend payout ratio, some constraints on the data are required. From the full data sample described in Section 4.1, I restrict the sample to dividend payers at year t. Nevertheless, the payout ratio is a continuous variable that extends without lower or upper theoretical bounds. However, there are limits to the payout ratio for sustainable growth of the firm. By definition the payout ratio can exceed 100%, but this indicates a situation where the firm distributes more than its earnings and the situation is clearly unsustainable. Furthermore, if the reported earnings are negative, then the payout ratio is negative (although this is still the situation where the firm distributes more than its earnings). In order to study sustainable firm growth, I eliminate observations with negative payout ratios and payout ratios above 100% from this analysis. After eliminating the observations with negative payout ratios and payout ratios over 100%, I consider three econometric techniques applicable for the analysis: cross-sectional using Fama and MacBeth’s (1973) procedure to break the serial correlation, random effects panel

60

regression, and fixed effects panel regression. To begin the analysis, the regression model simply has the dividend payout ratio as the dependent variable and a maturity variable as the only explanatory variable. After the analysis with the simple regression models, I include the control variables and industry dummy variables described in section 4.1 for the multivariate regressions with the dividend payout ratio as the dependent variable. 4.4.4 Regressions on Dividend Yield In this section, I utilize regression analysis to study to investigate the dividend yield and the life-cycle. From the full data sample described in Section 4.1, I restrict the sample to dividend payers at year t. After eliminating the observations with negative payout ratios and payout ratios over 100%, I consider three econometric techniques applicable for the analysis: cross-sectional using Fama and MacBeth’s (1973) procedure to break the serial correlation, random effects panel regression, and fixed effects panel regression. After the analysis with the simple regression models, I include the control variables and industry dummy variables described in section 4.1 for the multivariate regressions with the dividend yield as the dependent variable. In this research, the dividend yield is defined as: Dividend yield, DY =total dividends paid in year t Market equity at end of year t 4.4.5 Regressions on the Dividend Growth Rate While the logit models above in Section 4.4.1 provides useful insight into characteristics of a firm that increased the dividend, the logit model does not provide information on the magnitude of the dividend increase. In many financial applications such as valuations with the dividend discount model, the magnitude of the dividend increase or the dividend growth rate is

61

of primary interest. To investigate Hypothesis 4 and the relationship between the dividend growth rate and the life-cycle, I return to the statistical model, Model 1, derived in Chapter 3. Before the regression analysis of the relationship between maturity and the dividend growth rate, some constraints on the data are required. The growth rate extends without lower or upper bounds since it is a continuous variable. However, there are limits to the dividend growth rate for sustainable growth of the firm. Although the dividend growth rate can exceed 100%, this indicates a situation where the firm distributes more than twice its prior dividend and the situation is clearly unsustainable over the long run. If the firm cuts the dividend, then dividend growth is negative, and dividend cutting is again clearly not a sustainable growth strategy over the long run. Furthermore, firms with zero dividend growth are not of interest for this analysis. In order to study sustainable firm growth, I eliminate observations with non-growers and the observations with dividend growth rates above 100% for this analysis. For the regression analysis, I again consider the Fama and MacBeth (1973) method as well as random effects panel regression and fixed effects panel regression. In the regression model, the dividend growth rate is the dependent variable. I include the control variables and industry dummy variables described in section 4.1 for the multivariate regressions with the dividend growth rate as the dependent variable. In addition, I utilize the sustainable growth rate as an independent variable (as derived in Chapter 3) in the regressions in order to relate the actual dividend growth rate to the sustainable growth rate with a statistical model.

CHAPTER 5 EMPIRICAL RESULTS In this chapter, I report the findings of the empirical analysis for the hypotheses proposed in Chapter 1. The main theme of this dissertation is that firm maturity affects dividend policy, and in turn, dividend policy affects firm value throughout the firm’s life-cycle. I add to the literature on the “disappearing dividends”, the effect of dividend policy on firm value, and the effect of firm maturity on dividend payout policy. The results provide additional information for corporate boards and investors seeking to maximize the value of the firm. The reminder of Chapter 5 is organized into five main parts, with their accompanying tables and figures placed at the end of the dissertation. Section 5.1 provides the descriptive statistics for the sample over the 1982-2010 time series. Section 5.2 discusses the analysis of the firm’s propensity to pay dividends as well as the “disappearing dividends” phenomena. Section 5.3 presents the analysis of the effect of maturity on the valuation of firms while Section 5.4 presents the effect of firm maturity on dividend initiation. Finally, Section 5.5 demonstrates the effect of firm maturity on dividend payout policy. 5.1

Descriptive Statistics

5.1.1 Overall Sample and Time Series Table 1 reports the summary statistics for the sample over the 1982-2010 time series, which provides 95,996 observations with 11,576 firms. A review of the summary statistics indicates a significant skew in the sample and, as such, I will follow the prior literature and report sample medians. Table 1 also reports the breakdown on firms in the sample by the S&P economic sector, and some economic sector analysis is presented below. Consistent with the

62

63

prior literature, Table 2 details the characteristics of dividend paying firms compared to nonpayers. In Table 2, I report that dividend paying firms are larger, more profitable, have less growth opportunities, and have lower cash ratios similar to DeAngelo et al. (2006). Based on a two-sided, median two-sample test, the median differences between dividend payers and nonpayers are all significant at the 1% level. One can also see from Table 2 that there is significant “momentum” in the decision to pay or not pay dividends. About 95% of the dividend payers paid a dividend in the prior year, while about 97% of the non-payers did not pay a dividend in the prior year. Figure 1 shows the percentage of dividend paying firms in the sample over the 19822010 time period. Again, Figure 1 is consistent with the prior literature of Fama and French (2001) as the graph clearly illustrates the period prior to 2000 of “disappearing dividends” as well as the period after 2000 of “reappearing dividends” as reported by Julio and Ikenberry (2004). 5.1.2 Maturity Variables In this dissertation, I investigate several dimensions of maturity that have been reported in the prior literature-firm age, earned capital ratio or RE/TE, and risk (as measured by the standard deviation of monthly returns. Furthermore, I investigate combinations of these maturity variables. I define the maturity factor as the multiplicative combination of Log(firm age) times RE/TE times the inverse of total risk. I define the maturity composite as the sum of the percentile ranking of firm age plus the percentile ranking of RE/TE plus the percentile ranking of the inverse of total risk. Inspection of Table 2 confirms the basic premise of the maturity hypothesis in that indeed dividend payers are much more mature than non-payers. In terms of median firm age, dividend payers are almost three times as old. For the median earned capital ratio, the difference is even larger as dividend payers have almost ten times the RE/TE of non-payers. The

64

median risk is non-payers is almost twice as high as the risk of dividend payers. The maturity factor stretches the scale and is thirty times larger for dividend payers. The maturity factor has an interesting reference point of zero when a firm is incorporated. On the other hand, the maturity composite has a condensed scale and is about two times larger for dividend paying firms. The maturity composite has an interesting percentile interpretation with its percentile ranking basis. Table 2 indicates that the median maturity composite of dividend payers is 2.22, and the maturity composite consists of three maturity percentile components. Therefore, the maturity composite implies that the median dividend paying firm averages about the 75th percentile ranking (2.22/3) in age, RE/TE, and risk. Likewise, Table 2 shows that the median maturity composite of nonpayers is 1.18. Therefore, the median non-paying firm averages about the 40th percentile ranking (1.18/3) in age, RE/TE, and risk. After confirming that dividend payers are indeed more mature firms, I investigate the components of maturity over the sample study period. Figure 2 shows that the median firm age has increased over the 1982-2010 time series, but the relative ratio that dividend payers are about three times older than non-payers is relatively stable. Figure 3 illustrates a scatter plot of the median age of dividend paying firms as a function of the percentage of dividend paying firms in the sample. The scatter plot in Figure 3 indicates some relation that as firms get older, there is a higher probability of paying a dividend. Figure 4 shows that the median RE/TE for dividend paying firms has been relatively steady over the 1982-2010 time series, but the RE/TE of nonpayers has declined significantly from the 1982 values. The scatter plot in Figure 5 indicates a rather strong relation that as the median firm has more earned equity; there is a higher probability of paying a dividend. While Figure 6 shows that the median risk for dividend paying firms is lower than the median risk of non-payers, there are periods when risk spikes for both payers and

65

non-payers. For example, the median standard deviation of monthly returns rises significantly for both dividend payers and non-payers in 2000 near the “NASDAQ Crash” and in 2008 during the financial crisis. While Figure 6 indicates the lower median risk of dividend payers, the scatter plot in Figure 7 indicates only a weak relation between median risk and the probability of paying a dividend. Figure 8 shows that the median maturity composite score is relatively stable over the 1982-2010 time series and that dividend payers consistently have a maturity composite score twice as large as non-payers. Figure 9 illustrates that the median maturity factor of dividend paying firms stays in the range of 20-30 for most of the time series, while the median maturity factor of non-payers hovers near zero. In Figure 10, I note that the median maturity factor in the sample follows a trend in the time series very similar to the percentage of payers. The scatter plot in Figure 11 indicates that the maturity factor is strongly related to the probability of paying a dividend. To investigate the relationships with firm maturity, I first sort the sample each year into ten maturity factor deciles. Figure 12 illustrates the percentage of dividend payers in maturity factor decile 1, decile 5, and decile 10, where decile 1 represents the ten percent of the sample with the lowest maturity and decile 10 represents the ten percent of the sample with the highest maturity. As I would expect, decile 1 with the lowest maturity factors have the lowest percentage of dividend payers. Decile 10 with the highest maturity factors has the highest percentage of dividend payers. The percentage of dividend payers in decile 1 stays relatively constant at about 3% for the entire time series. In decile 10, the percentage of dividend payers hovers around 90% until about 1995. After 1995, the percentage of dividend payers in decile 10 bounces between 78% and 88%. First, this indicates that the maturity factor can rather accurately classify firms into dividend payers and non-payers as dividend payers have high maturity factors, and non-

66

payers have low maturity factors. Next, this indicates that changes in propensity to pay over the period do not occur with the most mature (Decile 10) or least mature firms (Decile 1). However, the percentage of dividend payers in Decile 5 with the median maturity factor declines dramatically from about 60% in 1982 to under 10% in 2000. This implies that firms with median levels of maturity (Decile 5) drive the “disappearing dividends” phenomena. I divide the sample first by dividend payers and non-payers in order to study the relationship with maturity. I then sort the dividend payers into ten maturity factor deciles and sort the non-payers into ten maturity factor deciles. Table 3 reports the summary statistics for the dividend payers sorted by maturity factor decile, and Table 4 reports the summary statistics for the non-payers sorted by maturity factor decile. Inspection of the dividend payers in Table 3 provides a few surprises. In decile 1, the dividend payers have a low median RE/TE and other low measures of maturity. In fact, the decile 1 paying firms have low measures of profitability, size, and negative earnings growth, but they did pay a dividend. Similarly, inspection of the nonpayers in Table 4 also provides a few surprises. In decile 10, the non-payers have a high median RE/TE and other high measures of maturity. In fact, the decile 10 non-paying firms high measures of profitability, large earnings growth, and lower sales growth, but they did not pay a dividend. This indicates that there are both dividend payers and non-payers that have dividend policies that are “against type”. Although maturity increases as one moves from decile 1 to decile 10 in either Table 3 or 4, there are some key differences in the median characteristics of dividend payers and non-payers as they mature. First let’s discuss the impact of maturity on size as measured by the NYSE percentile. Non-payers grow as they mature, but the most mature nonpayers only reach the 15th NYSE percentile at the median. The most mature dividend payers reach the 80th NYSE percentile at the median. Although profitability increases as both dividend

67

payers and non-payers mature, about half of all non-payers are not profitable. The median cash ratio for dividend paying firms is stable as the firms mature. However, for non-paying firms the cash ratio declines significantly as the non-payers mature. In the following sections, I discuss other key differences in the firm characteristics as dividend payers and non-payers mature. 5.1.3 Valuation Parameters Following the corporate finance literature, I use the M/B ratio as a measure of firm value. Fama and French (2001) report that non-payers have a generally have a higher M/B ratio than dividend payers. Furthermore, Baker and Wurgler (2004a, 2004b) report that the M/B ratio difference between non-payers and dividend payers varies over time. Since the M/B ratio of nonpayers is generally higher than payers, Baker and Wurgler propose a dividend “catering” fad. Although Hoberg and Prabhala (2009) show that “catering” is not significant when controlled for risk, they do not contradict the empirical finding that the M/B ratio of non-payers is generally higher than the M/B ratio of dividend payers. In Table 5, I report the Equal-Weighted M/B ratio for dividend payers and non-payers for the sample over the time series and find results consistent with Baker and Wurgler (2004a, 2004b). Over the 1982-2010 time series, the Equal-Weighted M/B ratio of the non-payers is greater than the Equal-Weighted M/B ratio of dividend payers. Figure 13 illustrates how the Equal-Weighted difference in M/B ratio between non-payers and dividend payers increases and decreases over time. Over the 1982-2010 time series, one might suggest a recession explanation to the time variation rather than a “catering” explanation since the equal-weighted M/B ratios of payers and non-payers converge during the recessions in the early 1990’s, in the early 2000’s, and in 2008.

Table 5 also shows that the median monthly

return for dividend payers is greater than the median monthly return on non-payers for the 19822010 time series. Figure 14 illustrates the higher returns for dividend-paying firms are achieved

68

with lower volatility than with nonpaying firms, which suggests that investing in dividend paying firms would result in higher Sharpe ratios. To investigate the relationship of firm value with firm maturity, I sort the sample each year into ten maturity factor deciles. Table 6 reports the Equal-Weighted M/B ratio and EqualWeighted monthly return in maturity factor decile 1, decile 5, and decile 10, where decile 1 represents the ten percent of the sample with the lowest maturity and decile 10 represents the ten percent of the sample with the highest maturity. In Table 6, decile 1 with the lowest maturity factor has the highest M/B ratio while decile 10 with the highest maturity factor, has the lowest M/B ratio. Figure 15 also illustrates that decile 1 has the highest M/B ratio over the 1982-2010 time series. Figure 16 illustrates that decile 1 has the lowest monthly returns with the highest volatility. Decile 10 has lower monthly returns than decile 5, but also much lower volatility than decile 5. The results in Table 6 indicate a relationship between firm value and maturity, but the relationship is not clear due to the complication of dividend paying status. In order to remove the complication of dividend paying status, I first return to Table 3 and Table 4, where I divide the sample into dividend payers and non-payers before I sort into maturity deciles. In addition, I return to reporting median properties of the M/B ratio rather than the Equal-Weighted M/B ratio. Although the scale is different with the median M/B ratio, DeAngelo et al. (2006) and the results in Table 2 again show that the median M/B ratio of nonpayers is higher than the median M/B of dividend payers. Since the prior literature consistently finds the M/B ratio of non-payers to be greater than dividend payers, there seems to be an apparent anomaly. That anomaly is, if the median valuation of non-paying firms is greater than the median valuation of dividend paying firms, then why would a value maximizing firm ever pay a dividend? In a broad sense, this is a re-statement of the dividend puzzle. The results in

69

Table 3 and 4 offer a potential explanation to the apparent anomaly and the dividend puzzle. Inspection of the dividend payers in Table 3 reveals that the median M/B ratio first declines with maturity. Then from decile 5 to decile 10, as a dividend paying firm matures, the median M/B ratio increases. Analysis of the non-payers in Table 4 reveals that the median M/B ratio begins at a large value but rather consistently declines with maturity. Figure 17 illustrates the consistent decline in median M/B ratio of non-payers as they mature. Since dividend payers eventually reach a maturity where the M/B ratio increases as they mature, there is a crossover point as can be seen in Figure 17. This crossover point in Figure 17 represents an explanation to the apparent anomaly of valuation differences between non-payers and payers. When a firm is young and its maturity factor is low, it has a high valuation based on M/B ratio. The high valuation when it is young explains the high Equal-Weighted M/B or median M/B ratio of non-payers. However, as the non-paying firm matures, its value continues to decline. Past the crossover point, the firm has an incentive to pay a dividend because the valuation of dividend paying firms will increase with maturity. Thus, there is a rational explanation to why a firm pays a dividend. A firm pays a dividend when it reaches a maturity where its value can be maximized by paying the dividend. Further analysis of the relationships between maturity and firm value are presented in later sections. 5.1.4 Dividend Policy The prior literature indicates that firm maturity is related to dividend initiation. In this dissertation, I also investigate the effect of maturity on dividend payout policy as characterized by the dividend payout rate and dividend growth rate. Figure 18 illustrates that the median dividend payout ratio for dividend paying firms has somewhat followed the trend of disappearing and reappearing dividends. The median payout ratio declines over the period from 1982-2008

70

and then spikes upward in 2009-2010. An alternate interpretation on Figure 18 is that the payout ratio seems to spike upward during recessions implying that the payout rises because net income falls during the recessionary periods. Figure 19 shows the median dividend growth rate for dividend paying firms over the time series. The median dividend growth rate seems to cycle between 2% and 10% around and average of 6%. Peaks in the median dividend growth rate seem to correspond with economic expansions while troughs in the median dividend growth rate seem to correspond to economic contractions.11 In the prior section, I find evidence that the valuation of dividend paying firms increases with maturity. I now discuss how dividend policy affects the positive relationship between firm value and maturity. Table 7 reports the summary statistics for dividend growers and dividend cutters. Inspection of Table 7 reveals that median maturity factor of dividend growers is larger than all dividend payers, while the median maturity factor of dividend cutters is significantly lower than all payers. Table 7 clearly exposes that dividend cutters have lower median financial maturity with lower RE/TE and negative earnings growth. On the other hand, dividend growers have higher median financial maturity with high RE/TE and high earnings growth. The implication here then is that dividend cutters were not mature enough to begin a consistent dividend policy, while dividend growers have the maturity to maintain a consistent policy of dividend growth. Consistent with prior literature, dividend cutters have a very low median monthly return in the year of the dividend cut. Dividend growers have median monthly returns that are higher than the median monthly return of all dividend payers. Thus, I can now link the prior literature on dividend policy with firm maturity. Prior literature indicates that investors do not like dividend cuts and punish the market value of those cutting firms, whereas, investors do

11

U.S. business cycle expansions and contractions are reported by the National Bureau of Economic Research at http://www.nber.org/cycles.html

71

like dividend growth and reward those firms. Consequently, as seen in Table 7, the median M/B ratio of dividend cutters is well below all dividend payers, and the median M/B ratio of dividend growers is well above all dividend payers. This result implies an explanation between the positive relationship between maturity and firm value for dividend payers. When a firm pays a dividend and the market determines the firm to be too young (and fears a dividend cut), the M/B will be low. When a firm pays a dividend and the market determines the firm to be financially mature (and with a growing dividend), the M/B will be high. To further investigate the effects of maturity on dividend policy, I return to Table 3 where the dividend paying firms are sorted into 10 maturity factor deciles. Review of Table 3 indicates that the probability of firm increasing its dividend (i.e. being a dividend grower) increases with maturity. On the other hand, the probability of a firm decreasing its dividend (i.e. being a dividend cutter) decreases with maturity. Consistent with the prior literature, the probability of dividend initiation decreases with maturity. Figure 20 illustrates that while the probability of dividend cuts and dividend initiation both decrease with maturity, dividend initiation falls more rapidly with maturity. Figure 21 shows that the dividend payout ratio increases monotonically with maturity. On the other hand, dividend growth seems to peak early then decline with maturity. Interestingly, decile 1 firms with the lowest maturity have the lowest median dividend growth rate. In this decile it seems that the higher incidence of dividend cuts reduces the median dividend growth rate. The dividend growth rate then appears to peak in decile 2 before it slowly declines with maturity. The inference here is that firm maturity affects dividend payout policy, and I investigate this relationship further with regression analysis in a section that follows below.

72

5.1.5 Economic Sector Analysis Although Fama and French (2001), Julio and Ikenberry(2004), and Hoberg and Prabhala (2009) indicate that the industry sector affects dividend policy, the prior literature has little comprehensive data demonstrating the effect of industry sector on dividend policy. I divide the sample into the S&P Economic Sectors, which represent the broad economic industry groups. Table 8 reports the summary statistics by S&P Economic Sector and inspection of Table 8 reveals the impact of economic sector on dividend policy. In the Materials and Consumer Staples sectors, about 60% of firms paid a dividend. However, in the Technology and Health Care sectors, only about 15% of firms paid a dividend. Figure 22 illustrates that over the 1982-2010 period, the economic sectors all track the overall sample and show the “disappearing dividends” and “reappearing dividends” trend. However, the relationship that firms in the Consumer Staples sector are much more likely to pay a dividend than firms in the Technology sector is always true over the period. Another factor that makes the impact of the economic sector difficult to quantify is that the composition of the sample changes over time. Figure 23 shows that the Consumer Staples sector is larger than the Health Care Sector in 1982, but by 2010, the Health Care sector is three times larger than the Consumer Staples sector. Likewise, the Industrial sector is larger than the Technology sector in 1982, but by 2010, the Technology sector is larger than the Industrial sector. Thus, some of the decline in the percentage of dividend paying firms can be attributed to the rise of economic sectors (Health Care and Technology) that have a low propensity to pay dividends and the decline of economic sectors (Consumer Staples and Industrials) that have a high propensity to pay dividends. A review of Table 8 indicates that firm maturity explains much of the dividend policy of the economic sectors. The economic sectors with the highest median maturity factor (Consumer

73

Staples and Materials) are the sectors with the highest percentage of dividend paying firms. Likewise, the economic sectors with the lowest median maturity factor (Health Care and Technology) are the sectors with the lowest percentage of dividend paying firms. It is interesting to note that the Health Care and Technology sectors are “young” by any definition of maturity. The median RE/TE of the Health Care sector is negative and the median CRSP age of the Health Care sector is half the median CRSP age of the Materials sector. The median monthly standard deviation of the Health Care sector is 50% higher than the median monthly standard deviation of the Consumer Staples sector. Figure 24 illustrates a similar time trend for the median maturity factor by economic sector and the percentage of firms paying a dividend by sector. The scatter plot in Figure 25 indicates that the maturity factor is strongly related to the probability of paying a dividend in any economic sector. To further investigate the relationship between economic sector and maturity, I sort the dividend payers and non-payers into the 10 maturity factor deciles. In Table 9, I report the economic sector composition in each maturity factor decile for dividend paying firms. Decile 1 firms with the lowest maturity factor are comprised mostly of firms in the Consumer Discretionary and Industrial sectors. Since Health Care and Technology sector firms have a low propensity to pay dividends, they are under-weighted in all deciles of dividend paying firms. Hence, despite the fact that over 21% of the entire sample consists of Technology sector firms, they comprise only 3% of Decile 10 dividend payers with the highest maturity factors. On the other hand, about 6% of the sample consists of firms in the Consumer Staples sector, but they comprise almost 21% of Decile 10 dividend payers. Inspection of Table 10 reveals that firms in the Technology and Health Care sectors comprise most of Decile 1 non-payers. Interestingly, Technology and Health Care sectors are under-weighted in the Decile 10 non-payers with the

74

highest maturity. This indicates that few firms in the Health Care and Technology sectors even become “mature” non-payers in the 1982-2010 time series, let alone mature to become dividend payers. 5.2

Maturity and the Propensity to Pay Dividends

5.2.1 Probability of paying dividends-Fama and MacBeth method The prior literature reports that maturity variables age, earned capital ratio, and risk all significantly relate to the probability that a firm pays a dividend. I test Hypothesis 1 proposed in Chapter 1 to determine which of the maturity variables or combination of variables best describes the propensity to pay a dividend. While the prior literature shows that each definition of maturity is significant individually, no prior tests show which maturity variable is most significant or even if the prior definitions measure the same effect. Another complication of the prior literature is that each research group uses different control variables. Julio and Ikenberry (2004) control for industry sector and firm size, but do not control for growth potential, profitability, RE/TE, or risk. DeAngelo et al. (2006) control for firm size, growth potential, profitability, cash ratio, and equity to asset ratio, but do not control for risk, firm age, or industry sector. Hoberg and Probhala (2009) control for firm size, growth potential, profitability, and technology sector, but do not control for RE/TE, firm age, and other industry sectors. Of course, I utilize a consistent, comprehensive set of control variables based on the prior literature for this study. Consistent with the prior literature, I apply the Fama and French’s (2001) and Fama and MacBeth’s-based (1973) statistical methodology to determine whether the probability that a firm pays dividends depends on the maturity variables and the control variables. This procedure utilizes a multivariate logit model that treats the binary indicator of payment/nonpayment of the

75

dividend as the dependent variable and the maturity variables and control variables as explanatory variables. Following Fama and MacBeth (1973), I run separate logit regressions for each of the 29 sample years (1982-2010) to obtain a time series of fitted coefficients. The mean coefficient, which I report, is the mean value from the 29 logit regressions (one for each year over 1982-2010). The t-statistics for each mean coefficient reported in the tables are based on the hypothesis that the expected coefficient is zero. Consistent with Fama and French (2001) and DeAngelo et al. (2006), the tables report t-statistics unadjusted for serial correlation. In Table 11, I report the results of the logit analysis for models with only a single explanatory variable. In Models 1-3, each maturity variable is statistically significant at the 1% level as would be expected. RE/TE and Log(Age) are positively related to the probability of paying a dividend and the standard deviation of monthly returns is negatively related to the probability of paying a dividend as expected. In terms of model fit statistics, Model 1 with RE/TE has a higher pseudo R2 value and higher percent predicted correct than Model 2 with Log(Age) or Model 3 with standard deviation. For the purposes here, a correct prediction for a dividend payer is any predicted probability greater than or equal to 0.5. The selection of 0.5 as the cutoff for the prediction model is simply based on the premise that if the model predicts the probability of paying a dividend as 0.5 or greater, the model is predicting that the firm is more likely to be a dividend payer. Likewise, a correct prediction for a non-payer is any predicted probability less than 0.5. Again, this is based on the idea that if the model predicts the probability of paying a dividend is less than 0.5, then the model is predicting that the firm is more likely to be a non-payer. Model 4 provides an interesting benchmark with the single explanatory variable of the prior dividend indicator dummy. Model 4 has the highest pseudo R2 value and itself correctly predicts 96.1% of the observations. In Model 5 and 6, I investigate the combination

76

maturity variables, the maturity composite and maturity factor, respectively. Both Model 5 with the maturity composite and Model 6 with the maturity factor have a higher pseudo R 2 value and higher percent correct than any individual maturity variable. At this point with no control variables, the maturity combination variables seem to capture “maturity” better than any individual maturity definition reported in the prior literature. Table 12 reports the logit analysis for models with a single maturity explanatory variable as well as control variables for size, growth potential, profitability, cash ratio, equity to asset ratio, and economic sector. Model 7 has only controls and can be considered an expansion of Fama and French’s (2001) base model. Consistent with Fama and French (2001) size and profitability are positively related to the probability that a firm pays dividends, while growth potential (as measured by both sales growth rate and M/B ratio) are negatively related to the probability that a firm pays dividends. For the industry controls, the Industrial sector is selected as the base economic sector. Only firms in the Materials and Consumer Staples sectors are more likely to pay a dividend than a firm in the Industrial sector, but all economic sectors are significant in Model 7. It is interesting to note that Model 5 with the single explanatory variable of maturity composite has a higher pseudo R2 value and higher percent correct than Model 7 with 13 control variables. Model 8 is essentially the DeAngelo et al. (2006) model with expanded control variables. Consistent with DeAngelo et al. (2006), RE/TE is highly significant and positively related to the probability that a firm pays a dividend. Model 9 replicates Julio and Ikenberry (2004) and finds that Log (Age) is highly significant and positively related to the probability that a firm pays a dividend. Model 10 duplicates Hoberg and Prabhala (2009) and finds that standard deviation is highly significant and negatively related to the probability that a firm pays a dividend. In Model 11 and 12, I investigate the combination maturity variables, the

77

maturity composite and maturity factor, respectively, with the control variables. Both Model 11 with the maturity composite and Model 12 with the maturity factor have a higher pseudo R 2 value and higher percent correct than any individual maturity variable model. Even with control variables, the maturity combination variables seem to capture “maturity” better than any individual maturity definition reported in the prior literature. Furthermore, many of the economic sector control variables remain statistically significant in the logit regressions. Table 13 provides the results from the logit analysis with variable combinations as well as the control variables. Model 13 includes RE/TE, Log (Age), and standard deviation in the same model. Each individual maturity variable retains its sign and statistical significance. Model 13 and Model 11, which have combinations of maturity variables, have a higher pseudo R 2 value and a higher percent correct than any individual maturity variable model. This agrees with the results of the logit regressions with combination maturity variables. The individual definitions of maturity reported in the prior literature seem to each capture a different dimension of maturity, and the combinations provide the best complete definition of “maturity”. It is difficult to analyze the coefficients from the logit regression and quantify which component of maturity drives the decision to pay dividends due to the different scales on RE/TE, Log (Age), and standard deviation. Therefore, Model 14 re-scales each maturity component on a percentile basis so that the explanatory maturity variables are RE/TE percentile, Age percentile, and standard deviation percentile. Inspection of the coefficients on Model 14 indicates that RE/TE has the largest coefficient (of the maturity variables) and therefore the largest maturity effect on the probability that a firm pays a dividend. In Table 14, I report the results of the average partial effect on the probability that a firm in the Industrial sector pays a dividend. Based on a change from the 50 th percentile to the 75th percentile in the RE/TE percentile, the average partial effect in the sample

78

is an increase of about 11% in the probability of an Industrial firm to pay a dividend. Since the percentiles are combinations of the maturity composite, Models 15, 16, and 17 examine individual maturity variables with the combination maturity variables. Although these models likely have significant multicollinearity issues, the combination maturity variable remains highly significant. In the case of Models 15 and 16, the maturity composite coefficient is hardly affected by the addition of the individual maturity variable percentiles.

Finally, in Model 18 the prior

dividend indicator variable is included. Even with the inclusion of the prior dividend indicator, the maturity composite variable remains highly significant. In addition, the Materials, Consumer Discretionary, Consumer Staples, Health Care, and Technology economic sector control variables remain significant even with the prior dividend indicator variable in Model 18. In summary, the logit analysis of the probability of paying dividends indicates that each individual definition of maturity reported in the prior literature captures a statistically significant dimension of maturity. However, the combinations of maturity variables provide the most complete definition of firm maturity. Of the components of maturity, the RE/TE has the largest effect on the decision to pay a dividend. Finally, economic sector control variables are statistically significant factors in the decision to pay a dividend. 5.2.2 Probability of paying dividends-Panel logistic method The prior literature research on propensity to pay dividends utilizes the Fama and MacBeth approach as demonstrated in the previous section. One disadvantage of the Fama and MacBeth approach is the inability to model year effects. In order to model year effects, a panel logistic method is required. In this section, I demonstrate that the panel logistic method is essentially equivalent to the Fama and MacBeth approach. I repeat the logit analysis of the decision to pay a dividend from the previous section using a random effects panel logistic model.

79

Tables 15, 16, and 17 report the panel logistic results for Models 1-6, Models 7-12, and Models 13-18, respectively. Comparison of the panel logistic method results in Tables 15, 16, and 17 to the Fama and MacBeth method results in Tables 11, 12, and 13 reveals that the methods yield similar results. Since both methods yield similar coefficients and statistical significance, the percent correct predicted by both methods are essentially the same. Based on the analysis of Section 5.2, Hypothesis 1a is supported. The finding is robust in both the Fama and MacBeth propensity to pay logit models and the panel logit models. Hypothesis 1a. Total risk significantly determines a firm’s propensity to pay dividends when controlled for the Fama-French (2001) firm characteristics, firm age, earned capital ratio, and combinations of risk, firm age, and earned capital ratio. Finding: Supported 5.2.3 Implications of the Life-cycle Models Figure 26 illustrates the predicted percentage of dividend paying firms from Models 1113 compared to the actual percentage of dividend paying firms in the sample. Models 11-13 all can be considered maturity models (with control variables), where the maturity is a combination of RE/TE, age, and risk. Figure 26 shows that all of the models fail to match the high percentage of dividend paying firms observed in the early 1980’s period. Model 11, which models the maturity with the maturity composite variable, fails to match the decline in dividend paying firms in the 1995 - 2002 period.

Models 12 and 13 follow the actual percentage of dividend

paying firms fairly well after the 1995 period. Figure 27 illustrates that a full model (Model 18) which includes maturity, control variables, and the prior dividend status essentially matches the actual data. This implies that there is substantial “momentum” in the decision to pay a dividend. Model 18 seems to very accurately model how firms decide whether to pay a dividend or not. The dominant factor in the decision of whether to pay a dividend or not is maintaining the prior

80

dividend policy.

In following sections, I investigate whether simply maintaining the prior

dividend policy is a firm value-maximizing strategy. Table 18 provides an analysis of the predictions from maturity Model 11. In this analysis, I consider two sets of observations: those observations where maturity Model 11 predicts greater than a 90% probability of paying a dividend and those observations where Model 11 predicts less than a 10% probability of paying a dividend. Inspection of Table 18 clearly reveals that the maturity Model 11 classifies over 96% of the observations in each set correctly; thus, these firms follow the maturity model. As one would expect from a maturity model, the median maturity in the set of observations with greater than 90% predicted probability is much larger than the median maturity in the set of observations with less than 10% predicted probability of paying a dividend.

Further analysis of Table 18 reveals some interesting implications for these sets of

firms that “follow” the maturity model. First, firms with greater than 90% probability of paying a dividend have 50% greater median monthly return than firms with less than a 10% probability of paying a dividend. Furthermore, the firms with 90% probability of paying a dividend attain the higher median monthly returns with about one-third of the risk. Although it is outside the scope of this dissertation, this suggests that investors could use the maturity model to select stocks with superior risk adjusted returns. Finally, the median M/B ratios for the observations that “follow” the maturity model are both higher than the median M/B ratios of either all dividend payers or all non-payers. This implies that the valuation of firms that “follow” the maturity model is higher than the valuation of firms that do not “follow” the model. 5.2.4 Outlier Analysis To investigate Hypothesis 3, I further analyze the valuation of firms that follow the maturity model and firms that do not follow the maturity models. For this analysis, a correct

81

model prediction for a dividend payer is any predicted probability greater than or equal to 0.5. Likewise, a correct model prediction for a non-payer is any predicted probability less than 0.5. Then for maturity Models 11, 12, and 13, I sort the observations into four categories: non-payers that the model correctly predicts as non-payers, non-payers that the model incorrectly predicts as payers, payers that the model correctly predicts as payers, and payers that the model incorrectly predicts as non-payers. Table 19 summarizes some critical characteristics of these four categories. First, Table 19 reveals that the median M/B ratio of non-payers that fit Models 11 and 13 is greater than the M/B ratio of non-payers that do not fit Models 11 and 13. For all models, the median M/B ratio of dividend payers that fit the models is greater than the median M/B ratio of dividend paying firms that do not fit the model. This analysis indicates that the valuation of firms that “follow” the maturity model is higher than the valuation of firms that do not “follow” the model. Further analysis of Table 19 provides some explanation to the valuation premium for firms that correctly follow the maturity model. The median maturity factor of non-payers that fit the model is the lowest of the four categories. As one would expect, the median maturity factor of dividend payers that fit the model is the highest of the four categories. The results are surprising for the categories that are “outliers” from the models or firms that have a dividend policy contradictory to the maturity models. The non-paying firms that do not fit the maturity models actually have higher maturity factors than the dividend paying firms that do not fit the maturity models. Again, this analysis indicates that the valuation of dividend paying firms increases as dividend paying firms mature. However, the valuation of non-payers decreases as non-payers mature.

82

Tables 20, 21, and 22 provide additional residual analysis on the dividend paying firms. As stated, the dividend paying outliers have a lower median maturity factor and a lower M/B ratio. Examination of Tables 20, 21, and 22 reveals that the dividend paying outliers have lower median maturity, lower median profitability, smaller median size, and higher sales growth rates. I show the outliers from Models 11, 12 and 13 in Tables 20, 21, and 22 respectively, to show that the results are robust to the particular maturity combination variable. In summary, the dividend paying outliers do not have the firm characteristics of dividend payers. Among the dividend paying outliers, the percentage of firms cutting the dividend is more than twice the percentage of dividend cutters from payers that fit the models. Furthermore, the dividend paying outliers have a lower median dividend growth rate. It seems that these dividend paying outliers do not have the financial maturity to be consistent dividend payers. It follows that investors would view these “immature” dividend paying outliers as riskier (and they do have a higher median standard deviation), discount the outliers more, and assign the outliers a lower M/B valuation. Tables 23, 24, and 25 provide additional residual analysis on the non-paying firms. Recall, the non-paying outliers have a higher median maturity factor and a lower M/B ratio. Examination of Tables 23, 24, and 25 reveals that the non-paying outliers have higher median maturity, higher median profitability, larger median size, and lower sales growth rates. An economic sector analysis of the non-paying outliers reveals that most non-paying outliers are in the Industrials and Consumer Discretionary sectors. Again, I show the outlier analysis results are robust to any particular maturity model. In summary, the non-paying outliers have the firm characteristics of dividend payers, but do not pay. It seems that these non-paying outliers do have the financial maturity to be consistent dividend payers. Since these non-paying outliers have a lower M/B ratio, it seems that investors view these “mature” non-payers as firms with low

83

growth potential. In addition, the lower valuation may also be the manifesting agency costs of excess free cash flow. Since this analysis shows the valuation of firms that “follow” the maturity models is higher than the valuation of firms that do not “follow” the maturity models, I consider the maturity models to be normative economic models for maximizing the value of a firm. Over 15% of firms in the sample then fail to maximize firm value with their corporate dividend policy. It appears that these outlier firms simply follow the prior dividend policy rather than maximize firm value. These results are completely in line with the theoretical nature of the maturity hypothesis of paying a dividend. Based on the theoretical framework of the maturity hypothesis, the firm trades-off the costs of distribution of retained earnings versus the costs of retention of the earnings. The theoretical implication of the maturity hypothesis then is that the firm is a value-maximizing firm and finds the optimal trade-off. On the other hand, simply following the prior dividend policy has no theoretical framework for maximizing firm value. I analyze the dividend policy that maximizes firm value in Section 5.3. In summary, I find support for both Hypothesis 3a and Hypothesis 3b since the analysis of Section 5.2.4 shows the valuation of firms that “follow” the life-cycle model is significantly higher than the valuation of firms that do not “follow” the life-cycle model. Hypothesis 3a. The non- payers that do not fit the model (against type) have a significantly lower M/B ratio than the non-payers that fit the life-cycle model. Finding: Supported Hypothesis 3b. The dividend payers that do not fit the model (against type) have a significantly lower M/B ratio than the dividend payers that fit the life-cycle model. Finding: Supported

84

5.2.5 Over- Zealous Payers and “disappearing dividends” In Table 19, I show that maturity Models 11, 12, and 13 all correctly classify about 85% of the observations. However, the maturity models all classify the non-payers better than payers. The maturity models correctly classify about 90% of the non-payer observations, but correctly classify only about 70% of the dividend payers. As a consequence, the majority of the “outliers” or firms that have a dividend policy contradictory to the maturity model, are dividend paying firms. In fact, there are about 50% more dividend paying outliers than non-paying outliers from each maturity model. Based on the outlier analysis from Section 5.2.4, I classify the outliers into two categories: over-zealous payers, and non-payers that should pay. While the prior literature focuses on non-payers that should pay, my results show that over-zealous dividend payers are more numerous and have lower median valuations. Fama and French (2001) focus on the aggregate decline in percentage dividend payers as Figure 1 illustrates, and they famously label the phenomena “disappearing dividends”.

Figure 26 shows that maturity models cannot

“explain” the high percentage dividend payers in the early 1980’s. While the actual percentage dividend payers in the sample during 1982-1984 is about 50%, the maturity models predict only about 35% - 40% in the sample. However, recall my claim that the maturity models are normative economic models that describe a dividend policy that maximizes the value of the firm. Thus, I offer a completely different description of the phenomena and call it “over-zealous” dividend payers. I consider the dividend paying outliers to be “over-zealous” as a group since they have low median maturity, lower median profitability, and smaller median size. Over the 1982-2010 time series, “over-zealous” dividend payers are more numerous than non-payers that should pay. Furthermore, Figure 28 illustrates that amount of “over-zealous” dividend payers is extremely high in the early 1980’s, and “over-zealous” dividend payers represent almost 90% of

85

the outliers from the maturity models. In the period between 2000 and 2010, there seems to be much more balance in the model outliers, where about half the model outliers are dividend payers. Incidentally, my observation that the propensity of the “over-zealous” dividend payers declines over the 1982-2010 is in agreement with the prior literature. I have just divided the decline in the propensity to pay into two groups. For the group of dividend payers that fit the maturity models, the decline in propensity to pay is explained by the model. In fact, DeAngelo et al. (2006) even note that the propensity to pay reduction is lower in the highest RE/TE group (most mature). However, I show the greatest decline in the propensity to pay dividends occurs with the “over-zealous” payers that do not fit the maturity model. My results are also in agreement with the prior literature in regards to the rise in non-payers that should pay. Nonpayers that should pay represent less than 10% of the outliers of the maturity models in 1982, but represent about half of the outliers in the 2000-2010 time period. My analysis here shows that decline in the propensity to pay dividends (or “disappearing dividends”) is due to the decline in the propensity of “over-zealous” dividend payers. Based on the normative nature of the maturity models, the “disappearing dividends” phenomena due to a decline in “over-zealous” payers is positive to shareholders because “over-zealous” payers destroy firm value. On the other hand, the rise in non-payers that should pay is negative for shareholders because non-payers that should pay also destroy firm value. If the decline in the “over-zealous” dividend payers explains “disappearing dividends”, then what explains the decline in the “over-zealous” dividend payers? Examination of Table 19 reveals that there is a valuation incentive for a maturing non-payer to stay a non-payer if it is not fully mature. Thus, it seems logical that a maturing non-payer would continue not paying until it is fully mature because outlier non-payers have a higher median M/B ratio than outlier

86

dividend payers. However, I find that the M/B of outlier dividend payers is not always lower over the 1982-2010 time series. Figure 29 illustrates that the M/B of outlier dividend payers is actually greater than outlier non-payers throughout the 1982-1990 time period. During this period, the market rewards dividend payers with a higher median M/B ratio even if they are “over-zealous” dividend paying outliers.

As Figure 29 shows, the valuation advantage of

dividend paying outliers disappears after the mid-1990’s. Table 26 reports the shift in the decline of the “over-zealous” dividend payers. In 1982, about 41% of all dividend payers are outliers of maturity Model 11. The 1982 “over-zealous” outliers are less mature, less profitable, and much smaller than the 1982 dividend payers that fit the model as one would expect. Furthermore, 1982 “over-zealous” outliers have a greater percentage of dividend cutters and a lower median dividend growth rate, which is consistent with the overall time series for outliers. However, the 1982 “over-zealous” outliers have a slightly higher median M/B than the 1982 dividend payers that fit the maturity model, which is anomalous with the overall time series for outliers. By 2000, the anomalous M/B ratio for the dividend outliers is reversed. Table 26 shows that the 2000 “over-zealous” outliers are still less mature, less profitable, and much smaller than the 2000 dividend payers that fit the maturity model. In 2000, the “over-zealous” dividend payers have a lower median M/B than the dividend payers that fit the model. Only about 20% of all dividend payers are “over-zealous” in 2000. This implies that decline in “over-zealous” dividend payers is linked to the relative market valuation of these outliers. In the early 1980’s, the market seemed to value “immature” dividend payers equal to “mature” dividend payers. The favorable valuations in the early 1980’s seemed to provide positive feedback for “over-zealous” dividend payers. However, as the market valuations became less favorable to “over-zealous” dividend payers, fewer firms became “over-zealous” in paying dividends. With less firms being “over-

87

zealous’ dividend payers, the aggregate percentage of dividend payers declines; hence one observes “disappearing dividends”. The outlier explanation also accounts for the rise of non-payers that should pay. Table 27 reports the rise in non-paying outliers, or non-payers that should pay. In 1982, non-paying outliers represent only 6% of the total non-payers in 1982. The 1982 non-paying outliers, are more mature, more profitable, and have less sales growth than the 1982 non-payers that fit the maturity model. The 1982 non-paying outliers have the firm characteristics of dividend payers, but just do not pay, which is consistent with the overall time series. However, in 1982 the nonpaying outliers face a severe median M/B discount compared to non-payers that fit the model. In the overall time series, non-paying outliers face a 5% median M/B discount to model non-payers. In 1982, the non-paying outliers faced a 23% median M/B discount to model non-payers. The severe M/B discount on non-paying outliers seems to provide a strong signal to avoid being nonpaying outliers, which explains the low occurrence of non-payers that should pay in 1982. By 2000, the severe M/B discount on non-paying outliers disappears. Given the more favorable valuation in 2000, it follows that the non-paying outliers increase to over 14% of all non-payers in 2000. Therefore, as the market valuations became more favorable to non-paying outliers, more firms became non-payers that should pay. With more firms being non-paying outliers, the aggregate percentage of dividend payers declines; hence one observes “disappearing dividends”. The phenomena that Fama and French (2001) describe as “disappearing dividends” can be attributed to the decline in “over-zealous” payers and the rise of non-payers that should pay. The reason for the decline in “over-zealous” payers is a decline in favorable M/B valuations placed on these “over-zealous” payers. Similarly, the rise in payers that do not pay is the disappearance of the severe M/B discounts placed on these non-paying outliers. Analysis of the

88

“shift” in market valuation is outside the scope of this dissertation, although I offer an explanation that is consistent with the maturity hypothesis and the empirical evidence. The early 1980’s period with a large percentage of dividend payers is near the end of a long secular bear market. It is likely that corporate managers as well as investors see fewer growth opportunities in such circumstances. In that “secular bear market,” macroeconomic setting, the cost of retaining earnings would seem higher and shift firms to paying dividends. Market valuations would confirm the costs of retention and reward dividend payers.

In the 1990’s when this shift

reverses, the U.S. is in the middle of a great bull market. New industries are created; thus, corporate managers and investors see abundant growth opportunities.

In this “great bull

market,” macroeconomic setting, the cost of distributing earnings would seem higher and shift firms to retaining earnings. Market valuations would confirm the costs of earnings distribution and reward non-payers. It is perhaps important at this point to distinguish the key point of this analysis from Baker and Wurgler’s (2004a, 2004b) catering theory. Baker and Wurgler (2004a, 2004b) do not discriminate dividend payers, while my analysis specifically considers the maturity of dividend payers. My explanation of “disappearing dividend” considers the immature dividend payers and the decline in the occurrence of these “over-zealous” dividend payers as the primary explanation. Furthermore, Baker and Wurgler (2004a, 2004b) assert that firms “cater” to behavioral fads. My analysis is based on the maturity hypothesis, where firms trade-off the costs of retention and the costs of distribution. By choosing an optimal dividend policy, firms should maximize firm value. Markets then recognize the firm value and assign an appropriate M/B valuation. The differences between maximizing firm value and “catering” are discussed further in Section 5.3.

89

5.2.6 Panel Logistic Regression with Year Effects An advantage of the panel logistic regression over the Fama and MacBeth method is that year effects can be added to the panel logistic regression via year dummy variables. In this section, I utilize maturity Model 11, where the maturity combination variable is represented by the maturity composite term. Based on the analysis in the prior sections, the results indicate that the maturity models are robust to the various combination variables so that only results for Model 11 are reported. In the maturity Model 11 with year effects, 1982 is considered the base the year so that year effects are actually the change from 1982. Table 28 reports the results of the logit analysis of the decision to pay a dividend for maturity Model 11 with year effects. Inspection of the parameter estimates indicates that the maturity composite remains highly significant, as do the control variables. Interestingly, the year effects from 1983 to 2010 are all negative and significant. This indicates that after controlling with the maturity model, the propensity to pay dividends is lower than 1982, and the reduction in propensity to pay each year after 1982 is statistically significant. Although maturity Model 11 with year effects improves the model fit, the year effects are difficult to attribute to any specific macroeconomic factor. Figure 30 displays the predictions from Model 11 with year effects against the actual percentage of dividend payers in the sample, 1982-2010. With the year effects, the model follows the actual percentage much better, especially in the 1982-1985 period. However, this technique of using year effects is better left to short time panels where the year effect could be better defined, such as in an event study. Overall, I find support for Hypothesis 2a since year effects are significant. Hypothesis 2a. Relative to the base firm life-cycle model, yearly time effects captured by year dummy variables are significant determinants of the firm’s propensity to pay a dividend. Finding: Supported

90

5.3

Maturity and Life-cycle Valuation

5.3.1 M/B Regressions as a Function of Maturity Factor Based on the theoretical framework of the maturity hypothesis, the firm trades-off the costs of distribution of retained earnings versus the costs of retention of the earnings. The theoretical implication of the maturity hypothesis then is that when the firm achieves this optimal trade-off, the value of the firm is maximized. I use the M/B ratio as a measure of firm value based on Fama and French (2001) who assert that the M/B is a suitable proxy for Tobin’s Q. In this section, I build the regression analysis from observations of Figure 17, which shows that the median M/B ratio of non-payers declines as the median maturity factor increases. Figure 17 also indicates that the median M/B ratio of dividend payers increases as the median maturity factor increases. Figure 17 also provides the intuition that the relationships may not be truly linear and may require piecewise regression. Based on these observations, I first divide the dividend payers from the non-payers. I further separate each of these categories into two halves where the first half has the lowest half of maturity factors and the second half has the highest half of maturity factors. I consider three econometric techniques applicable for the analysis: crosssectional using Fama and MacBeth’s procedure to break the serial correlation, random effects panel regression, and fixed effects panel regression. Here, the regression model simply has the M/B ratio as the dependent variable and the maturity factor is the only explanatory variable. Table 29 reports the results of the M/B regressions for the dividend payers. In the lower half of maturity factors, the slope on the maturity factor is slightly negative, but not statistically significant in any of the regression methods. In the upper half of the maturity factors, the slope on the maturity factor is positive and statistically significant in all of the regression methods. For all dividend payers, the slope on the maturity factor is positive and statistically significant in all

91

the regression methods. This confirms the prior observation that the M/B ratio of dividend payers increases with maturity factor. The analysis also indicates that the M/B ratio of dividend payers rises most sharply with the upper half (most mature) of maturity factors. The M/B ratio does not increase as dividend paying firms mature in the lower half (least mature) of maturity factors. Table 30 reports the results of the M/B regressions for the non-payers. In the lower half of maturity factors, the slope on the maturity factor is slightly negative, but is only statistically significant with the Fama and MacBeth regression method. In the upper half of the maturity factors, the slope on the maturity factor is negative and statistically significant in all of the regression methods. This confirms the prior observation that the M/B ratio of non-payers decreases with maturity factor. Since of all the econometric techniques provide very similar results and statistical inferences, I proceed with the Fama and MacBeth method in the further regression analysis. 5.3.2 M/B Regressions in Maturity Factor Ranges After confirming that dividend payers increase in value with maturity and non-payers decrease in value with maturity, I sort the dividend payers and non-payers into maturity factor ranges. The objective here is to compare the piecewise regression line of the dividend payers to the non-payers in the same maturity factor range. I group the dividend payers into three ranges of maturity factor and the non-payers into four ranges of maturity factor. The intention is to have a direct comparison of the dividend payers and non-payers over as wide a range of maturity factor as possible; however, there are very few dividend payers with a maturity factor less than zero and very few non-payers with a maturity factor over 25. The maturity factor ranges of 0 to 10 and 10-25 overlap both dividend payers and non-payers. At the same time, I need to be sure that each maturity factor range had sufficient observations for the regression analysis. Since there are

92

more non-paying firms in the sample, the observations can be divided among four groups with sufficient observations in each range of maturity factor. The dividend payers are divided into only three ranges in order to have sufficient observations for the regressions. Although the division of the maturity factor ranges for the piecewise regression is arbitrary, I show that the conclusions over the life-cycle are robust in the following section. Table 31 summarizes the median firm characteristics for dividend payers while Table 32 summarizes the median firm characteristics for non-payers. These tables continue to show that dividend payers increase in value with maturity and non-payers decrease in value with maturity. It is important to note that in the maturity factor range of 0-10, non-payers have a slightly higher median M/B ratio than dividend payers. However, as the firm maturity increases to the maturity factor range of 10-25, dividend payers have a higher median M/B ratio than non-payers. Just as in Figure 17, this crossover suggests that the maturity hypothesis indicates the actual maturity at which the costs of earnings retention exceed the costs of distribution. At that point, if the firm does not distribute the earnings (pay a dividend), the excessive costs of earnings retention will decrease firm value. The market observes the decrease in firm value and applies a lower M/B ratio. Here, the piecewise regression occurs over each range of maturity factor, but a separate regression analysis is completed for dividend payers and non-payers. Again, the regression model simply has the M/B ratio as the dependent variable and the maturity factor is the only explanatory variable. Table 33 reports the results of the M/B regressions for the dividend payers in Panel A and the non-payers in Panel B. Examination of the dividend payers in Panel A indicates that the M/B ratio slightly decreases in the maturity factor ranges of 0-10 and 10-25, but the slightly negative slope is not statistically significant. When the maturity factor of dividend payers is greater than 25, the slope of the maturity factor is positive and statistically

93

significant. Panel B shows that the M/B ratio decreases with maturity in all maturity factor ranges, and the negative slope is statistically significant. The decrease in M/B ratio is most severe in the maturity factor range of 0-10. In the range where the maturity factors overlap for dividend payers and non-payers, the negative slope on the maturity factor of non-payers is four times larger in magnitude than the negative slope on payers. Figure 31 displays the piecewise regression lines for dividend payers and non-payers. Figure 31 illustrates that non-payers with very low maturities (less than -20) have very high M/B valuations. However, as non-payers continue to mature, the M/B ratio continues to decline. For dividend payers with low maturities, the M/B tends to decline somewhat as maturity increases, but the slope is relatively flat compared to non-payers. Then, as the dividend payers mature further, the M/B ratio increases with maturity. The opposing slopes set up a crossover point where eventually a maturity is reached where the M/B ratio of dividend payers is greater than the M/B on non-payers. At this crossover point, a non-paying firm should begin a dividend policy as it matures in order to maximize firm value. Otherwise, the firm’s value will continue to decline as the non-paying firm matures. In Figure 31, the crossover maturity occurs at a maturity factor of about 8. If a firm pays dividends and its maturity factor is less than 8, Figure 31 shows that the firm will have a lower M/B ratio than if it did not pay a dividend. Likewise, non-paying firms with maturity factors over 8 have a lower M/B ratio than if they paid a dividend. Note that this analysis is consistent with the analysis of outliers from the propensity to pay maturity models in 5.2.4. Figure 31 shows the “over-zealous” payers as payers that have a lower M/B ratio than if they did not pay a dividend. They pay a dividend before the crossover point. The “over-zealous” payers are not “mature” and would have a higher valuation if they retained their earnings and pursued growth opportunities. Likewise, the non-payers that should pay have maturity beyond the

94

crossover point. The non-payers that should pay would increase firm value if they paid a dividend. Figure 31 provides an easy illustration to distinguish the value-maximizing strategy from behavioral “catering”. The empirical evidence shows that most firms choose a dividend policy that maximizes the value of the firm. In addition, a high growth firm with a negative maturity factor could not simply “cater” to dividend investors to obtain a higher market valuation as the evidence shows investors value high growth non-payers more than immature payers. In summary, Figure 31 illustrates the life-cycle of firm value on the maturity factor scale and resolves the question of why (and when) should a firm pay a dividend. When a firm is most immature, it has tremendous growth potential and the market values that enormous growth potential with extremely high M/B ratios. To take advantage of the large growth opportunities, the firm requires capital and retains any earnings. However, as the firm matures the growth opportunities decline and the market values the declining growth potential with declining M/B ratios. As the non-paying firm further matures to the “crossover” point, the firm sees fewer reinvestment opportunities and the costs of earnings retention exceed the costs of earnings distribution. Thus, the firm begins a dividend policy in order to maximize firm value. The market observes the firm’s value and assigns the appropriate M/B valuation. The firm then matures as a dividend-payer, and its value rises as the financial characteristics of the mature firm (such as profitability) further improve. 5.3.3 M/B Regressions in Maturity Composite Ranges To test the robustness of the relationship between maturity and firm value, I repeat the analysis of Section 5.3.2 with the maturity composite variable. Table 34 reports the median firm characteristics for dividend payers sorted by the maturity composite. Despite the different combination variable of maturity, the median M/B ratio of dividend payers still increases with

95

the increasing maturity composite. Table 35 summarizes the median firm characteristics for nonpayers sorted by the maturity composite. Likewise, the median M/B ratio of the non-payers still decreases with increasing maturity. This indicates that the general relationships between maturity and firm value are robust to the definition of maturity. The piecewise regression now occurs over each range of maturity composite, but a separate regression analysis is completed for dividend payers and non-payers. Once more, the regression model simply has the M/B ratio as the dependent variable, but now the maturity composite is the only explanatory variable. Table 36 reports the results of the M/B regressions for the dividend payers in Panel A and the non-payers in Panel B. Examination of the dividend payers in Panel A indicates that the M/B ratio slightly decreases in the maturity composite range of 1-2. When the maturity composite of dividend payers is in the range of 2-3, the slope of the maturity factor is positive and highly significant. Panel B shows that the M/B ratio decreases with maturity in all maturity composite ranges, and the negative slope is statistically significant. The decrease in M/B ratio is most severe in the maturity composite range of 0-1. In the range where the maturity composite overlaps (in a range of 1-2) for dividend payers and non-payers, the negative slope on the maturity factor of non-payers is about four times larger in magnitude than the negative slope on payers. Figure 32 displays the piecewise regression lines for dividend payers and non-payers on the maturity composite scale. Again, the opposing slopes set up a crossover point where eventually a maturity is reached where the M/B ratio of dividend payers is greater than the M/B on non-payers. At this crossover point, a non-paying firm should begin a dividend policy as it matures in order to maximize firm value. Otherwise, the firm’s value will continue to decline as the non-paying firm matures. In Figure 32, the crossover maturity occurs at a maturity composite of about 1.5. The maturity composite variable provides a rather

96

interesting interpretation of the crossover point. At the maturity composite of 1.5, a firm is on average at the 50th percentile of RE/TE, age, and risk. The crossover point in Figure 32 indicates that firms, which average lower than the 50th percentile of RE/TE, age, and risk, should not pay a dividend. Meanwhile firms that average higher than the 50th percentile of RE/TE, age, and risk, should pay a dividend. This analysis confirms that the relationships between maturity and firm value are robust to the definition of maturity. 5.3.4 Life-cycle and Firm Value While the relationship between maturity and firm value provides a vivid illustration of the firm life-cycle, it is important to recognize that maturity is in itself not a determinant of firm value for dividend payers. The true determinants of firm value are strongly correlated with maturity. For example, firm size and firm profitability increase with firm maturity. Therefore, to recognize the determinants of firm value, the regression equation must be expanded with further control variables. I include firm size, sales growth, profitability, cash ratio, equity to asset ratio, and economic sector variables. I further separate the maturity composite into RE/TE percentile, age percentile, and standard deviation percentile. The dependant variable of the regression model is still the M/B ratio as the proxy for relative firm value. The piecewise regression now occurs over each range of maturity composite, and again, a separate regression analysis is completed for dividend payers and non-payers. Based on Section 5.3.3, I follow the life-cycle of a non-paying firm starting in the maturity composite range of 0-1, then 1-2. At that maturity composite range, a firm should begin to pay a dividend. I then follow the life-cycle of a dividend paying firm in the maturity composite range of 1-2, then finally 2-3. Table 37 reports the results of the M/B regressions for the non-payers, while Table 38 reports the results for the dividend payers.

97

In the maturity composite range of 0-1 for non-payers, the RE/TE percentile is negatively related to firm value. At this stage of the life-cycle, value seems to be maximized with contributed capital. As the non-payer matures to a maturity composite range of 1-2, the relation between the RE/TE percentile and M/B ratio is still negative, but much less negative. As the firm moves to a dividend payer in the maturity composite range of 1-2, surprisingly, the RE/TE percentile is still negatively related to firm value, but much less negative. Only for the most mature dividend payers in the maturity composite range of 2-3 does the relationship between the RE/TE percentile and M/B ratio turn positive, but it is not statistically significant. Across the entire life-cycle, the age percentile is negatively related to firm value. It is interesting that again while always negative, the coefficient is much less negative on the dividend payers in maturity composite range 1-2 than the non-payers in maturity composite range 1-2. This implies that paying a dividend at this point in the life cycle counters the negative effects of firm age. However, the positive impact of the dividend is short-lived, as the most mature dividend payers again have the strong negative relation between firm age and value. In all stages of the life-cycle, the standard deviation percentile is positively related to firm value. This indicates that volatility adds value to firm, much like an option. While the standard deviation percentile is always positively related to the M/B ratio, the effect is greatest at the lowest ranges of maturity. Similarly, the positive effect of volatility on firm value is much larger on non-payers than on dividend payers. Firm size is also always positively related to firm value in all stages of the life-cycle. However, the positive effect is much larger for the least mature firms. Likewise, the positive effect of firm size is much larger on non-payers than on dividend payers.

98

In the maturity composite range of 0-1 for non-payers, profitability is negatively related to firm value. At this stage of the life-cycle, value seems to be maximized if the firm grows at the expense of profitability. As the non-payer matures to a maturity composite range of 1-2, the relation between profitability and M/B ratio turns positive. As the firm moves to a dividend payer in the maturity composite range of 1-2, profitability is even more positively related to firm value. For the most mature dividend payers in maturity composite range 2-3, profitability is the key determinant of firm value. The cash ratio is also always positively related to firm value in all stages of the life-cycle. However, the positive effect is much larger for the least mature firms. Likewise, the positive effect of cash ratio is much larger on non-payers than on dividend payers. Although difficult to generalize, many economic sectors significantly affect firm value, which indicates that these are important control variables. 5.3.5 Dividend Payout Policy and Firm Value To investigate the effects of dividend payout policy on firm value, I expand the regression equation with further variables from a firm’s dividend payout policy. I include the dividend growth rate, the dividend payout ratio, and the dividend yield as well as the prior control variables. The dependant variable of the regression model is still the M/B ratio. I then follow the life-cycle of a dividend paying firm in the maturity composite range of 1-2, then finally 2-3. Table 39 reports the results of the M/B regressions with dividend payout policy variables. Examination of the results in Table 39 reveals that the dividend payout ratio has no effect on firm value. Over the life-cycle of a dividend paying firm, the dividend growth rate is positively related to firm value. However, the relationship is not statistically or economically significant in

99

the maturity composite range of 1-2, or when the firm is an “immature” payer. It seems that these “young” payers still have some growth opportunities, thus firm value is maximized by balancing dividend growth with retention of earnings. Then when the paying firm matures to the maturity composite range of 2-3, it is a “mature” payer, and there are fewer growth opportunities. In this case, firm value is maximized by providing dividend growth with less retention of earnings. In all stages of the life-cycle, the dividend yield is negatively related to firm value. Furthermore, the negative effect of dividend yield is much larger for the least mature payers. Again, this indicates that “immature” payers still have some growth opportunities, thus firm value would be maximized by balancing dividend payout with retention of earnings. These results indicate that dividend payout policy does affect firm value. However, the analysis also reveals the complexity of the payout policy. While dividend growth adds value to the firm, the effect is only significant for the most mature payers. Finally, a high dividend yield decreases firm value, especially for “immature” payers. Figure 33 summarizes the life-cycle that maximizes firm value. When the maturity composite is less than 1, the firm should not pay a dividend. At this stage of the life-cycle, growth of the market capitalization is required over profitability to maximize value. As the nonpaying firm matures to a maturity composite range of 1-2, growth of both the market capitalization and profitability are required to maximize value. As the growth in market capitalization slows and the growth of profitability increases, a firm will increase its value by paying a dividend since dividend payers are rewarded more for profitability than non-payers. As the firm continues to mature into the maturity composite range of 2-3, profitability is the key driver of firm value. A mature, dividend paying firm is also rewarded for dividend growth, but the dividend yield must be not be excessive.

100

5.4

Maturity and Dividend Initiation

5.4.1 Time series of dividend initiators The prior literature indicates a positive relation between firm maturity and the probability of initiating a dividend; however, the prior studies use different proxy variables for firm maturity, just as the case with the propensity to pay a dividend. Thus, the research question I investigate in this section is what maturity variable or combination of variables best determines a firm’s probability of initiating a dividend. In this dissertation, I follow the method of Hoberg and Prabhala (2009) and start with all firms that do not pay a dividend in year t -1 and define dividend initiators as firms paying a dividend in year t. Based on this definition for dividend initiation, only a small percentage of non-payers initiate a dividend in year t. For the 1982-2010 time series, Figure 34 illustrates the percentage of non-payers that initiate a dividend. From the early 1980’s to 2002, the declining trend in dividend initiation follows the downward trend in propensity to pay. This downward trend in dividend initiation supports my conclusions on “disappearing dividends”, where I assert that the numbers of marginal dividend payers declined. The fact that a declining percentage of firms initiate a dividend in this period confirms my premise that the macroeconomic environment shifted from one that favored distribution of earnings in the early 1980’s to one that then favored retention until 2002. Another factor influencing a firm’s decision to distribute earnings is the safe-harbor rule adopted in 1982, which allows open market stock repurchases. After 2003, the percentage of dividend initiations rises rapidly, and this spike in dividend initiations is usually attributed to the Bush Tax Cut on dividends in 2003. Finally, note that dividend initiation again drops precipitously during the Financial Crisis in 2008 and 2009 and then rises rapidly in 2010.

101

Since the investigation of dividend initiation depends on binary models, it helps to analyze the characteristics of initiators and non-initiators. Table 40 compares the summary statistics for non-payers that initiate to non-payers that continue as non-payers. Consistent with the prior literature, dividend initiators are older, have lower risk, and have higher RE/TE.

In

terms of the combined maturity variables, dividend initiators have a higher median maturity factor and a higher median maturity composite. Furthermore, dividend initiators have higher profitability than non-initiators. However, notice that dividend initiators do not have all the median properties of mature dividend payers. Note that dividend initiators have about the same median sales growth rate as non-initiators. Although the median size is larger for dividend initiators, the median size of initiators is much smaller than the median size of dividend payers. Consistent with the analysis of Section 5.3, non-payers that initiate a dividend have a higher median M/B ratio than non-payers that continue as non-payers. Likewise, in the year of dividend initiation, initiators have a higher median monthly return than non-initiators. The economic sector analysis is similar to the results from the propensity to pay investigation. The Technology sector is over-weighted in the non-initiator population and under-weighted in the initiator population, thus, firms in the Technology sector have a lower propensity to initiate a dividend and pay a dividend. 5.4.2 Probability of initiating a dividend-Fama and MacBeth method Consistent with the prior literature, I apply the Fama and French’s (2001) and Fama and MacBeth’s-based (1973) statistical methodology to determine whether the probability that a firm initiates a cash dividend depends on the maturity variables and the control variables. This procedure utilizes a multivariate logit model that treats the binary indicator of initiation/noninitiation of the dividend as the dependent variable and the maturity variables and control

102

variables as explanatory variables. Following Fama and MacBeth (1973), I run separate logit regressions for each of the 29 sample years (1982-2010) to obtain a time series of fitted coefficients. The mean coefficient, which I report, is the mean value from the 29 logit regressions (one for each year over 1982-2010). The t-statistics for each mean coefficient reported in the tables are based on the null hypothesis that the expected coefficient is zero. Consistent with Fama and French (2001) and DeAngelo et al. (2006), the tables report t-statistics unadjusted for serial correlation. For consistency with the prior literature, I use the same comprehensive control variables that were used in the probability of paying a dividend logit analysis. In Table 41, I report the results of the logit analysis for models with the maturity explanatory variables as well as the control variables. As would be expected from the prior literature, models that use the combination maturity variables (the maturity factor and the maturity composite) indicate that the combination maturity variables are positively related to dividend initiation and are statistically significant at the 1% level. In the model with the separate maturity variables, each maturity variable is statistically significant at the 5% level. The RE/TE percentile and age percentile are positively related to the probability of initiating a dividend while the risk percentile is negatively related to the probability of initiating a dividend. Just as with the propensity to pay logit analysis, this seems to indicate that each separate maturity variable contributes to the overall concept of firm maturity. In the model with the separate maturity variables, each maturity variable is scaled on the same percentile range. Therefore, the largest effect on the probability of dividend initiation from the maturity variables is attributed to the RE/TE percentile since it has the largest coefficient, which again is the same result as the propensity to pay analysis.

103

As the prior literature reports, the same determinants that are significant in the propensity to pay logit analysis are also significant in the probability of paying a dividend logit analysis. Thus, Table 41 also reveals that size and profitability are positively related to the probability of dividend initiation, while the sales growth rate and M/B ratio are negatively related to the probability of paying a dividend. Inclusion of the economic sectors in the logit analysis indicates that some economic sectors have significant effects on the probability of dividend initiation. In the model, the Industrial sector is the base sector; therefore, the sign of the coefficients on the remaining sectors indicates whether the propensity to initiate is greater than or less than the propensity of the Industrial sector. The only economic sector that has a greater propensity to initiate a dividend than the Industrial sector is the Consumer Staples sector. The Telecom, Technology, Energy, and Health Care sectors all have a significantly lower propensity to initiate a dividend than the Industrial sector. Consistent with the prior literature, I find the propensity to initiate a dividend logit regression models have much less explanatory power than the propensity to pay a dividend logit regression models. Despite the fact that the models can correctly classify an observation as an initiator or not with over 97% accuracy, the pseudo R2 values are very low. The problem with the models is that there are very few initiators relative to non-initiators, and the relative frequency distorts the maximum likelihood estimates. The infrequency of initiators leads to another econometric issue when one considers the small economic sectors. In a year with only 25 dividend initiators, none is likely to be from the Telecom sector as that sector accounts for less than 2% of the sample population. Using annual logits per the Fama and MacBeth method seems to cause a problem with the estimated coefficients on the Telecom sector due to this issue. For a

104

robustness check of the Fama and MacBeth method (as well as the flexibility to add year effects to the models), I analyze dividend initiation with the panel logistic method in the next section. In addition, the event analysis in Section 5.4.5 provides another robustness test on dividend initiation from a very different perspective. 5.4.3 Probability of initiating a dividend-Panel logistic method In this section, I repeat the logit analysis of the decision to pay a dividend from the previous section using a random effects panel logistic model. I demonstrate that the panel logistic method is essentially equivalent to the Fama and MacBeth approach. Table 42 reports the results from the panel logistic regression on the decision to initiate a dividend. Comparison of the panel logistic method results in Table 42 to the Fama and MacBeth method results in Table 41 reveals that the methods yield similar results for the maturity variables. Both methods indicate that each individual definition of maturity reported in the prior literature captures a statistically significant dimension of maturity. However, the combinations of maturity variables provide the most complete definition of firm maturity in regards to dividend initiation. Of the components of maturity, the RE/TE percentile has the largest effect on the decision to initiate a dividend. Both the Fama and MacBeth method and the panel regression logits indicate that firm size and profitability are positively related to the probability that a firm initiates a dividend, while the M/B ratio is negatively related to the probability that a firm initiates a dividend. While both methods show that sales growth is negatively related to the probability of dividend initiation, the relation is not significant in the panel logistic regression. This result is actually more consistent with the univariate results in Table 40 where there is no statistical difference in median sales growth rate between initiators and non-initators.

105

Both the Fama and MacBeth method and the panel regression logits indicate that there are significant economic sector effects on the decision to initiate a dividend. The results for the more populous sectors are very similar for the two methods. For the Telecom sector, the estimated coefficient from the panel regression logits seems most consistent. Both methods confirm that firms in the Health Care, Energy, Technology, and Telecom sectors are less likely to initiate a dividend, while firms in the Consumer Staples sector are more likely to initiate a dividend. In summary, the results from the logit analysis of the decision to initiate a dividend are very similar to the results from the logit analysis of the decision to pay a dividend. As a nonpaying firm matures, its propensity to initiate a dividend increases. Each separate definition of maturity captures a statistically significant dimension of firm maturity, but the combination maturity variables seem to provide the most complete definition of maturity. As with the propensity to pay a dividend, the RE/TE percentile has the largest effect of the maturity variables on the decision to initiate. As firms become larger and more profitable, their propensity to initiate a dividend increases. Consistent with the maturity hypothesis, firms are less likely to initiate a dividend if they have significant growth potential as measured by the M/B ratio. In addition to firm characteristics, there are significant economic sector effects on the decision to initiate a dividend. 5.4.4 Dividend initiation and the life-cycle The analysis in Section 5.3 indicates a relationship between firm value and the life-cycle. The major implication of the maturity hypothesis is that a firm maximizes value by paying a dividend when the costs of earnings retention exceed the costs of earnings distribution. Based on that empirical analysis, Figure 32 illustrates that the “crossover” point in the life-cycle where the

106

costs of earnings retention exceed to costs of earnings distribution occurs at a maturity composite value of about 1.5. Assuming that most firms attempt to maximize value, I would then expect that most firms would initiate dividends at a maturity composite of about 1.5. Indeed, a review of Table 40 reveals that the median maturity composite for dividend initiators is 1.63. To investigate the distribution of the maturity composite values for the dividend initiators, I sort the dividend initiators into three ranges of maturity composite value: from 0-1, 12, and 2-3. The summary statistics for the dividend initiators sorted by maturity composite range are presented in Table 43. Consistent with my expectations, over 60% of dividend initiators begin paying dividends at the maturity composite range of 1-2. Note that only a small number of dividend initiators (less than 15%) commence dividend payments in the maturity composite range of 0-1. Based on the analysis of dividend payers, these “early initiators” lack the maturity of median dividend payers. The “early initiators” have low median RE/TE ratios, low median profitability, low median size, high sales growth rates, and high risk - none of the usual characteristics of dividend payers. Consistent with their growth potential, the “early initiators” have a high M/B ratio. These “early initiators” do see a short term gain as the median monthly return is very high during the year in which they initiate a dividend. In the maturity composite range of 1-2, these “expected initiators” have most of the characteristics of dividend payers but not all. The “expected initiators” are more mature with higher RE/TE, less risk, and higher profitability. However, unlike mature dividend payers, the “expected initiators” still have a high growth rate and lower median size. As reflected by the lower M/B ratio, the “expected initiators” have less growth potential than the “early initiators”. Finally, “late initiators” in the maturity composite range of 2-3 are significantly more mature. These “late initiators”, which are about 25% of all initiators, have the characteristics of mature

107

dividend payers except size. As a group, it seems that the “late initiators” spend about a decade trying to grow, but do not. After realizing their actual limited growth potential, the “late initiators” finally begin to pay a dividend. The “late initiators” have the lowest M/B ratio and have the lowest median monthly return during the year in which they initiate a dividend. After sorting the dividend initiators and non-initiators by maturity composite range, I repeat the logit analysis on the decision to initiate a dividend for each maturity composite range. Table 44 summarizes the results from the logit analysis by maturity composite range. In the maturity composite range of 0-1 where the “early initiators” begin paying dividends, maturity is positively related to the probability of dividend initiation. However, at the early stage of the lifecycle only the RE/TE percentile and risk percentile are statistically significant. This indicates that age is not a determinant of dividend initiation for the “early initiators”. While size and profitability are positively related to dividend initiation for the “early initiators”, growth potential is not. In the middle of the life-cycle when most firms are expected to initiate, maturity is still positively related to the probability of dividend initiation. Again, at this stage of the life-cycle only the RE/TE percentile and risk percentile are statistically significant. This indicates that age is not a determinant of dividend initiation for the “expected initiators”. While size and profitability are positively related to dividend initiation for the “expected initiators”, the emphasis has shifted in this stage of the life-cycle. The size of the positive coefficient on profitability has increased while the size of the positive coefficient on size has decreased for the “expected initiators”. Also for the “expected initiators”, growth potential (as measured by M/B ratio) is significant and negatively related to the propensity to initiate a dividend.

108

In the mature part of the life cycle, the “late initiators” finally begin to pay a dividend but maturity is only a weak factor. At the mature part of the life-cycle, the RE/TE percentile and risk percentile are not statistically significant. This indicates that age is the only weak determinant of maturity of dividend initiation for the “late initiators”. Again, the size of the positive coefficient on profitability increases while the size of the positive coefficient on size decreases for the “late initiators”. Also for the “late initiators”, growth potential (as measured by M/B ratio) is weakly significant but still negatively related to the propensity to initiate a dividend. The logit analysis results in Table 44 also reveals that significant economic sector effects occur in all stages of the life-cycle. The Health Care and Technology sectors show a reduced propensity to initiate a dividend in all stages of the life-cycle. Firms in the Materials and Consumer Staples sectors show an increased propensity to initiate a dividend, but only in the mature stage of the life-cycle. 5.4.5 Dividend initiation and year effects An advantage of the panel logistic regression over the Fama and MacBeth method is that year effects can be added to the panel logistic regression via year dummy variables. In this section, I utilize a panel logistic regression model for dividend initiation with year dummy variables. In the dividend initiation model with year effects, 1982 is considered the base the year so that year effects are actually the change from 1982. Table 45 reports the results of the logit analysis of the decision to initiate a dividend for the model with year effects. Inspection of the parameter estimates indicates that the maturity remains highly significant, as do the control variables. Examination of the year effects reveals that significant changes in the propensity to initiate dividends from 1982 levels do occur. This indicates that in some years the propensity to initiate changes by more than can be captured by maturity and the control variables. It is

109

interesting to point out that while the safe-harbor rule allowing open market stock repurchases is often cited as a factor in the decline in dividend initiation, these logit results do not support that assertion. The logit results in Table 45 show no significant negative year effect at the 5% level until the mid-1990’s, which is over a decade after the adoption of the safe-harbor rule. The logit results are consistent with my assertion that the macroeconomic environment shifted from one that favored distribution of earnings in the early 1980’s to one that then favored earnings retention. By the mid 1990’s, the development of the internet and related new technologies provides corporations with growth opportunities previously unseen; hence, corporations needed more retained earnings for reinvestment. Finally, the logit results in Table 45 provide strong support that the Financial Crisis in 2008-2009 significantly reduces the propensity to initiate a dividend. Overall, the logit analysis with the year effects indicates that significant changes in the propensity to initiate a dividend do occur over the 1982 to 2010 time period. Figure 34, which displays the percentage of dividend initiators over time, suggests a declining propensity to initiate from the early 1980’s to about 2003. The logit analysis with year effects confirms that the reduction in propensity to initiate is statistically significant even after controlling for maturity, firm characteristics, and economic sector. Based on the analysis of Section 5.4.5, Hypothesis 2b is supported. The panel logit analysis indicates significant year effects in the propensity to initiate a dividend. Hypothesis 2b. Relative to the base firm life-cycle model, yearly time effects captured by year dummy variables are significant determinants of the firm’s decision to initiate a dividend. Finding: Supported 5.4.6 Survival Analysis of dividend initiation Another method to analyze the probability of dividend initiation is to utilize time to event analysis or survival analysis. In regards to dividend initiation, time to event analysis places the

110

focus on observing the time that it takes a non-paying firm to initiate a dividend. The fundamental dependent variable in time to event analysis is the hazard rate. For dividend initiation, the hazard rate is the probability that a corporation will initiate a dividend in year t while the firm is still at risk for having the event (that is a non-payer at year t-1). The hazard rate or probability of initiating a dividend is negatively related to the time to initiate a dividend. Thus, the higher the hazard rate, the shorter the time to dividend initiation. Based on these definitions, the data set for the hazard model is the same as the data set used for the logit analysis. Deshmukh (2003) shows a hazard model for dividend initiation confirms the results from a logit analysis for dividend initiation. Deshmukh’s (2003) hazard model of dividend initiation is consistent with the prior literature in that size and profitability are positively related to the probability of dividend initiation. In this section, I extend Deshmukh’s (2003) hazard model to include maturity variables and the consistent, comprehensive set of control variables used in the logit anlaysis. Before I develop the hazard model for dividend initiation, it is useful to examine the empirical hazard function from the data. Although time to event analysis focuses on linear time, the analysis of this dissertation shows that dividend policy dynamics follow “life-cycle” time rather than linear time. Furthermore, the maturity composite variable best captures this “lifecycle” time, and dividend initiation is expected at a maturity composite value of about 1.5. Figure 35 displays the empirical hazard function as a function of the maturity composite, which represents the life-cycle time. As expected, the empirical hazard function is very low at values of maturity composite values less than 1. Then as the maturity composite increases, the hazard rate increases rapidly. At maturity composite values above 2.0 (the mature life-cycle stage), the hazard rate increases exponentially. Figure 35 also confirms the economic sector effects seen in the logit analysis. Firms in the Materials and Consumer Staples sectors have a higher hazard rate

111

than firms in the Technology sector, which indicates that firms in the Materials and Consumer Staples sectors initiate dividends sooner than firms in the Technology sector. Figure 36 displays the corresponding empirical survival distribution function of the data. In regards to dividend initiation, non-payers that continue as non-payers are considered survivors that are still at risk of initiating. As survivors mature and the maturity composite increases, some of the non-payers initiate and the percent of surviving non-payers in the population decreases. As expected, the empirical survival distribution function is nearly 1.0 at values of maturity composite values less than 1.0. At this early stage of the life-cycle, there are very few initiators. Then as the remaining survivors continue to mature, the survival rate begins to decrease rapidly. In summary, Figures 35 and 36 confirm the prior observations with time to event analysis. As the maturity composite increases above 1.5, the hazard rate or probability of initiating a dividend increases rapidly. While Figures 35 and 36 clearly illustrate that the probability of dividend initiation is related to firm maturity and the life-cycle, the dividend initiation hazard model has limitations on the ability to statistically test the maturity hypothesis. First, in order to have hypothesis tests on the maturity variables in the hazard model, one must model the hazard rate as a function of linear time rather than “life-cycle” time. Otherwise, the maturity variables would be part of both the dependent and independent variables. Figures 37 and 38 show the empirical hazard function and survival distribution function, respectively, for the data in terms of event time. Review of Figures 37 and 38 indicates that dividend initiation is difficult to define in event time. In Figure 37, the hazard rate does not increase sharply until the censoring time. Likewise, in Figure 38 the survival rate declines in an almost linear trend, which makes the prediction of dividend initiation difficult in event time. Since the firm matures in event time, one can conclude from Figures 37 and 38 that the hazard (probability of initiation) does increase with maturity. However, life-cycle time

112

clearly describes dividend initiation better than event time. The next problem is that the hazard model event time is identical to one of the maturity variables, the CRSP age variable. Since the CRSP age variable is part of the combination maturity variables, the combination variables cannot be tested in the hazard model. However, the RE/TE and risk maturity variables can be separated from the maturity composite variable and can be tested in the hazard model. Following Deshmukh (2003), I utilize the Cox proportional hazard model to estimate the hazard rate of dividend initiation. Deshmukh (2003) finds no evidence of violation of the proportional hazard assumptions for the 1990-1997 period of the study. If a time-dependent covariate is significant in the Cox proportional hazard model, this indicates a violation of the proportionality assumption for that specific variable. However in the 1982-2010 sample, I find the RE/TE ratio, size, ROA, M/B ratio, and TE/TA ratio to be significant time-dependent covariates. This is consistent with the logit analysis in Section 5.4.4 which indicates that the determinants of dividend initiation change over the life-cycle. A solution to the violation of the proportional hazard assumption is to include the time-dependent variables for the nonproportional variables in the model. Table 46 reports the results of the Cox proportional hazard regression for the hazard rate of dividend initiation. Consistent with the logit analysis, the hazard rate (probability of initiation) increases with the time-dependent RE/TE ratio. Likewise, the hazard rate decreases with increasing standard deviation (risk). This indicates that the hazard rate for dividend initiation increases with increased maturity. As the time-dependent size and profitability increase, the probability of initiating a dividend increases. Finally, the Cox proportional hazard regression indicates significant economic sector effects consistent with the logit analysis. The Health Care and Technology sectors show lower hazard rates while the Materials and Consumer Staples sectors show higher hazard rates.

In summary, the Cox

113

proportional hazard regression indicates that the hazard rate for dividend initiation increases with firm maturity. The hazard model for dividend initiation provides results consistent with the logit analysis. In conclusion, I find strong support for Hypothesis 1b with the analysis in Section 5.4. The finding is robust in the Fama and MacBeth logit models, the panel logit models, and the hazard model. Hypothesis 1b. Total risk significantly determines a firm’s decision to initiate dividends when controlled for the Fama-French (2001) firm characteristics, firm age, earned capital ratio, and combinations of risk, firm age, and earned capital ratio. Finding: Supported 5.5

Maturity and Dividend Payout Policy The prior literature investigations of the maturity hypothesis for dividend policy study the

propensity to pay dividends as well as dividend initiation/ omission. None investigate dividend policy and the maturity hypothesis. The most significant related empirical work is Rozeff’s (1982) study of optimal dividend payout ratios. Although Rozeff (1982) investigates the tradeoff between growth potential and agency costs, he does not consider the firm maturity. In this dissertation, I investigate the relationship between maturity and dividend payout policy. There are several important theoretical as well as practical implications. First as seen in Section 5.3, dividend payout policy impacts firm value. Dividend growth increases firm value while an excessive dividend yield decreases firm value. Corporate managers need to understand the relation between dividend growth and maturity in order to maximize firm value. Next, to many investors in dividend paying stocks, a dividend cut may be one of the largest sources of firm risk. Investors need to understand what factors are related to a dividend paying firm’s decision to cut a dividend. Finally, financial analysts utilize the dividend growth rate to value firms according to

114

dividend discount models. Analysts need to understand how the dividend growth rate changes with maturity in order to properly value the firm. In Sections 5.5.1 and 5.5.2, I present the logit analysis for the probability of dividend growth and the probability of a dividend cut, respectively. I report the regression analysis in Section 5.5.3 of the relationship between firm maturity and the dividend payout ratio. Section 5.5.4 details the link between firm maturity and dividend yield. Finally in Section 5.5.5, I present the regression analysis of the dividend growth rate. 5.5.1 Probability of dividend growth The data set for the studies on dividend payout policy is the subset of all dividend payers from the full sample. Consistent with the prior sections, I apply the Fama and French’s (2001) and Fama and MacBeth’s-based (1973) statistical methodology to determine whether the probability that a firm increases the cash dividend depends on the maturity variables and the control variables. This procedure utilizes a multivariate logit model that treats the binary indicator of increase/non-increase of the dividend as the dependent variable and the maturity variables and control variables as explanatory variables. The binary indicator of a dividend growing firm is equal to 1 if the firm’s dividend per share in year t is greater than the dividend per share in year t-1; otherwise, the indicator is zero. I add two variables to the comprehensive control variables that were used in the probability of paying a dividend logit analysis. First, I include the earnings growth rate to account for the observation that dividend payout changes follow earnings changes. Inclusion of the earnings growth rate variable eliminates observations with negative earnings. Finally, I include the prior dividend payout ratio to account for the fact that the prior dividend payout ratio in year t-1 may affect the decision to increase/decrease the dividend in year t.

115

It is helpful to examine the median properties of dividend growers to see if there are significant differences between the dividend growers and non-growers. Table 47 compares the median properties of dividend growers to non-growers. Dividend growers appear more mature by every definition of maturity as dividend growers are older, have less risk, and have higher RE/TE. Since firm size is also correlated to maturity, dividend growers are also larger than nongrowers. While dividend growers are more mature than non-growers, the largest differences appear in the operating performance. Dividend growers have higher sales growth, higher earnings growth, and higher profitability than non-growers. In fact, the ROA of dividend growers is almost twice the ROA of non-growers. Dividend growers have subtle differences in their payout policy. The dividend growers increase their median payout ratio from 24.72% in year t-1 to 27.74% in year t, while non-growers decrease their median payout ratio. Dividend growers have a higher median dividend per share, but a slightly lower dividend yield. The market rewards the stellar operating performance of the dividend growers as the dividend growers have a significantly higher M/B ratio and median monthly return. Following Fama and MacBeth (1973), I run separate logit regressions for each of the 29 sample years (1982-2010) to obtain a time series of fitted coefficients. The mean coefficient, which I report, is the mean value from the 29 logit regressions (one for each year over 19822010). The t-statistics for each mean coefficient reported in the tables are based on the null hypothesis that the expected coefficient is zero. Consistent with Fama and French (2001) and DeAngelo et al. (2006), the tables report t-statistics unadjusted for serial correlation. In Table 48, I report the results of the Fama and MacBeth logit analysis for the decision to grow a dividend. In the first regression, there are only maturity variables (RE/TE percentile, age percentile, and standard deviation percentile) with no control variables. Immediately, one notes a departure in

116

the relationship between the maturity variables. In the propensity to pay and probability to initiate logits, each individual component of maturity is consistent with the overall effect of maturity. However, with the probability to grow a dividend logit, the CRSP age variable has the opposite effect of the standard deviation variable. As the CRSP age variable increases (and the firm increases in maturity), the firm has a lower probability of growing the dividend. As the standard deviation variable decreases (and the firm increases in maturity), the firm has a higher probability of growing its dividend.

Not surprising then, the combination maturity variables

(maturity factor and maturity composite) are positive but insignificant when the control variables are included in the logit regression. Thus to interpret the relationship between maturity and the probability of dividend growth, the maturity variables must be separated into the components. In the model where the component maturity variables are included with the control variables, the standard deviation has the largest effect of the maturity variables and it is statistically significant at the 1% level. As the standard deviation percentile variable decreases, the firm has a higher probability of growing its dividend. This indicates that stability is a key determinant in the probability that a firm increases its dividend. On the other hand, as the firm’s CRSP age increases, it is less likely to increase its dividend, although this is only significant at the 10% level. Interestingly, the RE/TE percentile has no significant effect on the probability that a firm grows its dividend. Consistent with the univariate analysis, profitability is positively and significantly related to the probability that a firm grows its dividend. As the firm size increases, the probability that a firm grows its dividend also increases. The prior dividend payout ratio in year t-1, is negatively related to the probability that a firm grows its dividend. This indicates that a high payout ratio lowers the probability of future dividend growth. Finally, there are significant economic sector

117

effects. Firms in the Consumer Staples sector are more likely to grow their dividends, while firms in the Energy and Technology sectors are less likely to grow their dividends. For robustness, I repeat the logit analysis of the decision to increase a dividend using a random effects panel logistic model. I demonstrate that the panel logistic method is essentially equivalent to the Fama and MacBeth approach. Table 49 reports the results from the panel logistic regression on the decision to increase a dividend. Comparison of the panel logistic method results in Table 49 to the Fama and MacBeth method results in Table 48 reveals that the methods yield similar results for the maturity variables. Both methods indicate that the relationship between firm maturity and the probability of dividend growth is complex. As the firm matures and is more stable (with less volatility), the probability of dividend growth increases. However, the probability of dividend growth declines with CRSP age, and the RE/TE ratio has no effect on the probability of dividend growth. Of the components of maturity, the standard deviation percentile has the largest effect on the decision to increase a dividend. Both the Fama and MacBeth method and the panel regression logits indicate that firm size and profitability are positively related to the probability that a firm increases a dividend. The M/B ratio is negatively related to the probability that a firm increases a dividend. While both methods show that the prior dividend payout ratio is negatively related to the probability of a dividend increase, the relation is not significant in the panel logistic regression. Both the Fama and MacBeth method and the panel regression logits indicate that there are significant economic sector effects on the decision to increase a dividend. Firms in the Consumer Staples sector are more likely to increase the dividend, while firms in the Energy and Technology sectors are less likely to increase the dividend.

118

To investigate the probability of a dividend increase over the life-cycle, I sort the dividend growers into three ranges of maturity composite value: from 0-1, 1-2, and 2-3. These ranges of maturity composite values correspond to the early life-cycle stage, the transition period, and the mature life-cycle stage, respectively. Table 50 reports the median properties of the dividend growers by the maturity composite range. Based on the analysis in prior sections, very few dividend growers would be expected in the early life-cycle stage with the maturity composite less than 1.0 and indeed only 2% of dividend growers are in this stage. The dividend growers in the early life-cycle stage have low median RE/TE, low median CRSP age, and high median standard deviation. Furthermore, the dividend growers in the early life-cycle stage have high median sales growth and small median firm size. Firms with these characteristics in the early life-cycle stage would not even be expected to pay a dividend much less grow the dividend. These early life-cycle dividend growers have low median per share dividend, low median dividend payout ratios, but very high median dividend growth rates. Since the largest effect of the maturity variables on the probability of increasing a dividend is standard deviation, the number of dividend growers is expected to increase in the later life-cycles as standard deviation declines. Indeed, as the median standard deviation declines in the life-cycle, the number of dividend growers increases. Almost 70% of dividend growers are in the mature life-cycle stage with a maturity composite value over 2.0. In this mature life-cycle stage, the dividend growers have all the established characteristics of dividend payers. The mature life-cycle dividend growers have high median RE/TE, high median CRSP age, and low median standard deviation. Furthermore, the dividend growers in the mature life-cycle stage have low median sales growth, large median firm size, and high median profitability. The mature life-cycle dividend growers have a high median per share dividend, high median dividend payout

119

ratios, but lower median dividend growth rates. Consistent with the prior results, the most mature dividend growers have the highest M/B ratio. After sorting the dividend growers and non-growers by maturity composite range, I repeat the logit analysis on the decision to increase a dividend for each maturity composite range. Table 51 summarizes the results from the logit analysis by maturity composite range. In the maturity composite range of 0-1 with the early life-cycle dividend growers, maturity is not related to the probability of a dividend increase. At the early stage of the life-cycle, none of the maturity variables are statistically significant. This indicates that maturity is not a determinant of dividend growth in the early life-cycle stage. In fact, it is difficult to find any significant variables that explain the probability of dividend growth in the early stage of the life-cycle. In the early stage of the life-cycle, profitability is positively related to the probability that a firm increases a dividend, while the sales growth rate is negatively related to the probability that a firm increases a dividend. Interestingly, firm size is not significant in the early life-cycle stage. In the maturity composite range of 1-2, firm maturity is a significant determinant of the probability that a firm grows its dividend, but again the relation is complex. In this transition period of the life-cycle, the probability of dividend growth increases as standard deviation declines. However, the probability of dividend growth declines with CRSP age, and the RE/TE ratio has no significant effect on the probability of dividend growth. Profitability and firm size are positively related to the probability that a firm increases a dividend in the maturity composite range of 1-2. The M/B ratio is negatively related to the probability that a firm increases a dividend in the maturity composite range of 1-2. In the mature life-cycle stage when the maturity composite is in the range of 2-3, firm maturity is a significant determinant of the probability that a firm grows its dividend, but the

120

relation remains complex. In this mature period of the life-cycle, the probability of dividend growth still increases as standard deviation declines. However, the CRSP age now has no significant effect on the probability of dividend growth in the mature life-cycle stage. Although not significant in any of the other dividend growth logits, the RE/TE is positively related to the probability of dividend growth in the mature life-cycle stage. Of the components of maturity, the standard deviation percentile has the largest effect on the decision to increase a dividend in any life-cycle stage. Profitability and firm size are positively related to the probability that a firm increases a dividend in the mature life-cycle stage. Across the life-cycle, there are some significant economic sector effects. Firms in the Technology sector are less likely to increase the dividend in all stages of the life-cycle. Firms in the Consumer Staples sector are more likely to increase the dividend, but only in the later stages of the life-cycle. In summary, the logit analysis for the decision to increase the dividend indicates that the relationship between firm maturity and probability of a dividend increase is complex. As the CRSP age variable increases (and the firm increases in maturity), the firm has a lower probability of growing the dividend. As the standard deviation variable decreases (and the firm increases in maturity), the firm has a higher probability of growing its dividend. The standard deviation has the largest effect of the maturity variables, and it is the largest maturity effect in all stages of the life-cycle. This indicates that stability is a key determinant in the probability that a firm increases its dividend. On the other hand, as the firm’s CRSP age increases, it is less likely to increase its dividend, and the RE/TE percentile has no significant effect on the probability that a firm grows its dividend except in the mature life-cycle stage. Firm size and profitability are positively related to the probability that a firm increases a dividend, while the M/B ratio is negatively related to the probability that a firm increases a dividend. Increasing profitability increases the

121

probability of a dividend increase in all stages of the life-cycle although the effect of profitability increases in the mature life-cycle stage. Overall, these results are consistent with the maturity hypothesis. As the dividend paying firm matures and becomes more stable (less standard deviation), dividend growth is more likely if there is less growth potential (lower M/B ratio). As the firm size and profitability increase, one expects to find higher agency costs of free cash flow so that distribution of earnings should increase. The results of the logit analysis confirm that the probability of dividend growth increases with firm size and profitability. 5.5.2 Probability of a dividend cut Again, the data set for this analysis is the subset of all dividend payers from the full sample. I apply the Fama and French’s (2001) and Fama and MacBeth’s-based (1973) statistical methodology to determine whether the probability that a firm decreases the cash dividend depends on the maturity variables and the control variables. This procedure utilizes a multivariate logit model that treats the binary indicator of decrease/non-decrease of the dividend as the dependent variable and the maturity variables and control variables as explanatory variables. The binary indicator of a dividend cutting firm is equal to 1 if the firm’s dividend per share in year t is less than the dividend per share in year t-1; otherwise, the indicator is zero. I utilize the comprehensive control variables that were used in the probability of increasing a dividend logit analysis. As a point of clarity, this analysis only includes dividend cutters that still pay a dividend since the data set is all dividend payers in the sample. Therefore, dividend omitters are not included in this analysis of dividend cuts. It is helpful to examine the median properties of dividend cutters to see if there are significant differences between the dividend cutters and non-cutters. Table 52 compares the median properties of dividend cutters to non-cutters. Dividend cutters appear less mature by

122

every definition of maturity as dividend cutters are younger, have more risk, and have lower RE/TE than non-cutters. While dividend cutters are less mature than non-cutters, the largest differences appear in the operating performance. Dividend cutters have lower sales growth, lower earnings growth, and lower profitability than non-cutters. In fact, the median sales growth rate and median earnings growth rate of dividend cutters is negative. Dividend cutters also have significant differences in their payout policy. The dividend cutters decrease their median payout ratio from 39.68% in year t-1 to 17.46% in year t, while non-cutters increase their median payout ratio. Dividend cutters have a lower median dividend per share and a higher dividend yield. The market punishes the poor operating performance of the dividend cutters as the dividend cutters have a significantly lower M/B ratio and median monthly return. Due to the negative market response for dividend cutters, dividend cuts are infrequent and represent less than 9% of the dividend paying observations. In Table 53, I report the results of the Fama and MacBeth logit analysis for the decision to cut a dividend. The results indicate that firm maturity is negatively related to the probability of a dividend cut. In regressions where the maturity is modeled with the combination maturity variable (maturity factor or maturity composite), the parameter estimate on the combination maturity variable is negative and significant at least at the 5% level. This indicates that as a firm matures there is a lower probability of a dividend cut. However, as is the case with dividend growth, the details of the maturity relationship are more complex. In order to study the maturity relationship in more detail, one must examine the separate maturity components. In the regression model with the separate maturity components, the RE/TE percentile, age percentile, and standard deviation percentile are all statistically significant. However, as is the case with dividend growth, the maturity components do not all follow the overall effect of maturity.

123

Specifically, as the RE/TE percentile increases (and the firm matures) the probability of a dividend cut declines and as the standard deviation percentile decrease (and the firm matures) the probability of a dividend cut decreases. However, as the firm’s CRSP age increases, the probability of a dividend cut increases. Since the separate maturity variables are all on the same percentile scale, the largest coefficient has the largest effect. Here, the standard deviation has the largest effect of the maturity variables and it is statistically significant at the 1% level. As the standard deviation percentile variable decreases, the firm has a lower probability of cutting its dividend. This indicates that stability is a key determinant in the probability that a firm cuts its dividend. Consistent with the univariate analysis, profitability is negatively and significantly related to the probability that a firm cuts its dividend. As the firm size increases, the probability that a firm cuts its dividend also decreases. The logit analysis further confirms the dismal median operating performance of dividend cutters. As the sales growth rate and earnings growth rate increase, the probability of a dividend cut decreases. The prior dividend payout ratio in year t-1, is positively related to the probability that a firm cuts its dividend. This indicates that a high payout ratio increases the probability of a future dividend cut. Both the univariate and logit analysis indicate that dividend cutting is strongly related to poor operating performance. Quite simply, the poor operating performance with lower (or even negative) sales growth, earnings growth, and profitability places the firm in the position where it has little free cash flow to distribute in the form of dividends. The net effect of firm maturity is that a more mature firm will be more stable, have a higher RE/TE ratio, and larger in size. These attributes provide the firm with other financing options in lieu of a dividend cut, thus more

124

mature firms have a lower probability of a dividend cut. In terms of dividend safety, an excessive payout ratio increases the probability of a future dividend cut. For robustness, I repeat the logit analysis of the decision to cut a dividend using a random effects panel logistic model. I demonstrate that the panel logistic method is essentially equivalent to the Fama and MacBeth approach. Table 54 reports the results from the panel logistic regression on the decision to decrease a dividend. Comparison of the panel logistic method results in Table 54 to the Fama and MacBeth method results in Table 53 reveals that the methods yield similar results for the maturity variables. Both methods indicate that the relationship between firm maturity and the probability of a dividend cut is complex. As the firm matures and is more stable (with less volatility), the probability of a dividend cut decreases. However, the probability of a dividend cut increases with CRSP age. As the RE/TE ratio increases (and the firm matures), the probability of a dividend cut decreases. Of the components of maturity, the standard deviation percentile has the largest effect on the decision to decrease a dividend. Both the Fama and MacBeth method and the panel regression logits indicate that firm size and profitability are negatively related to the probability that a firm decreases a dividend. The sales growth rate is negatively related to the probability that a firm cuts a dividend. While both methods show that the earnings growth rate is negatively related to the probability of a dividend cut, the relation is not significant in the panel logistic regression. Both the Fama and MacBeth method and the panel regression logits indicate that the prior dividend payout ratio in year t-1 is positively related to a probability of a dividend cut. In summary, firm maturity is positively related to the probability of a dividend increase and negatively related to the probability of a dividend cut. Interestingly, as the firm’s CRSP age increases, the probability of dividend growth declines and the probability of a dividend cut

125

increases. However, standard deviation is the maturity variable with the largest effect, and as the firm matures and becomes less volatile, the probability of dividend growth increases and the probability of a dividend cut decreases. Firm size and profitability are positively related to the probability of a dividend increase and negatively related to the probability of a dividend cut. Poor operating performance is strongly related to the probability of a dividend cut, as lower sales growth and lower earnings growth increase the probability of a dividend cut. An excessive dividend payout ratio reduces the probability of future dividend growth and increases the probability of a future dividend cut. 5.5.3 Maturity and the dividend payout ratio In this section, I investigate the relationship between firm maturity and the dividend payout ratio. By extension of the research of Rozeff (1982), one would expect that the dividend payout ratio is related to firm maturity. Furthermore, based on the maturity hypothesis, one expects more mature firms to distribute more earnings as their growth opportunities decline and the costs of earnings retention increase.

Indeed, the univariate summary statistics reported in

Section 5.1 suggest a positive relationship between firm maturity and the dividend payout ratio. I sort the dividend paying firms into maturity factor deciles and report the median characteristics for Deciles 1-10 in Table 3. A review of Table 3 indicates that as the maturity factor of dividend paying firms increases from Decile 1 to Decile 10 then the median dividend payout also increases. Figure 21 illustrates that the median dividend payout ratio increases monotonically with the increasing median maturity factor of Decile 1 to Decile 10. In this section, I utilize regression analysis to study this apparent relationship. However, before the regression analysis of the relationship between maturity and the dividend payout ratio, some constraints on the data are required. As a continuous variable, the

126

payout ratio extends without lower or upper bounds. However, there are limits to the payout ratio for sustainable growth of the firm. By definition the payout ratio can exceed 100%, but this indicates a situation where the firm distributes more than its earnings and the situation is clearly unsustainable. Furthermore, if the reported earnings are negative, then the payout ratio is negative (although this is still the situation where the firm distributes more than its earnings). In order to study sustainable firm growth, I eliminate observations from with negative payout ratios and payout ratios above 100% for this analysis. After eliminating the observations with negative payout ratios and payout ratios over 100%, it is informative to examine some properties of the remaining sample. First I rank the remaining sample by dividend payout ratio and sort the sample into ten deciles ranked by dividend payout ratio. The summary statistics for the dividend payers sorted by dividend payout ratio decile are reported in Table 55. Decile 1 contains the 10% of the sample with the lowest payout ratios while Decile 10 contains the 10% of the sample with the highest dividend payout ratios. Review of Table 55 indicates that again the median maturity increases with increasing payout ratio until Decile 9. Further inspection of Table 55 reveals some other interesting relationships. In Decile 1 with the lowest dividend payout ratios, the firms have the highest median sales growth rate. In fact, this inverse relationship between median sales growth rate and median dividend payout ratio continues across all deciles. Thus, in Decile 10 with the highest dividend payout ratios, the firms have the lowest sales growth rate. This relationship is consistent with the maturity hypothesis in that firms with large growth opportunities (high sales growth rates) will retain most of their earnings and have low distribution (low dividend payout ratios). By definition, the payout ratio is inversely related to the earnings growth rate. Thus in Decile 1 with the lowest payout ratios, the firms have the highest earnings growth rate. Likewise in

127

Decile 10 with the highest payout ratios, the firms have the lowest earnings growth rate. In fact, in Decile 10, the firms have a negative median earnings growth rate, which implies that the high median payout ratios in Decile 10 result from a decline in earnings rather than an increase in the dividend payment. The median M/B ratio in each decile then reflects this picture of the growth potential. Firms in Decile 1 with the lowest payout ratios have the highest sales growth rates and growth potential; consequently, those firms in Decile 1 have the highest M/B ratio. Likewise firms in Decile 10 with the highest payout ratios have the lowest growth rates and growth potential; consequently, the firms in Decile 10 have the lowest M/B ratios. Interestingly, the operating performance of the low payout/high growth firms is better than the operating performance of the high payout/low growth firms. Thus, firms in Decile 1 with the lowest payout ratios (but highest growth rates) also have the highest ROA and ROE. Investors reward the high growth potential and best operating performance firms in Decile 1 with the highest median monthly returns. Likewise, firms in Decile 10 with the lowest growth potential and lowest operating performance have the lowest median monthly returns. For the regression analysis, I consider three econometric techniques applicable for the investigation: cross-sectional using Fama and MacBeth’s procedure to break the serial correlation, random effects panel regression, and fixed effects panel regression. Here, the regression model simply has the dividend payout ratio as the dependent variable and the maturity variables are the only explanatory variables. Table 56 reports the results of the dividend payout ratio regressions for the dividend payers. In the first model, I utilize the maturity factor as the combination maturity variable, while I employ the maturity composite as the maturity combination variable in the second model. All three econometric techniques have similar results which indicate that maturity is positively related to the dividend payout ratio, and that each

128

maturity variable is significant at the 1% level. The components of maturity are separated in the third model to see if each component of maturity is significant. Again all three econometric techniques have similar results and find that each component of maturity is significant at the 1% level. Interestingly, the RE/TE percentile component of maturity is negative, but there are no control variables in this simple regression. At this point, I continue the analysis and add the comprehensive control variables and report the results using Fama and MacBeth’s method. (In unreported results, the panel methods again yield similar results and inferences as the Fama and MacBeth method). Table 57 reports the results from the dividend payout regressions when the control variables are included. Again, in the first two models one finds that the maturity factor and maturity composite are positively related to the dividend payout ratio. In the third model where the components of maturity are separated, I find that the RE/TE percentile is insignificant after controlling for size, profitability, growth, and industry. The age component is positive, and the standard deviation component is negative, which both correspond to increased maturity. This indicates that firm maturity is positively related to the dividend payout ratio, but the significant maturity components are only age and volatility. Thus as a firm matures and the CRSP age increases, the dividend payout increases. Likewise, as a firm matures and the standard deviation decreases, the dividend payout increases. Further inspection of Table 57 reveals that the sales growth rate is negatively and significantly related to the dividend payout ratio, as expected. Hence, as the sales growth rate increases, the dividend payout ratio decreases. Again, these results are consistent with the maturity hypothesis. By definition of the payout ratio, a negative relation between profitability (ROA) and earnings with the dividend payout ratio is expected. Indeed, the results in Table 57 show that the profitability and earnings growth rate are negatively and significantly related to the dividend

129

payout ratio. As the earnings increase, the payout ratio decreases. Interestingly, the Fama and MacBeth method regression analysis indicates that size is not a significant determinant of the dividend payout ratio. This seems to agree with the univariate analysis in Table 55, where the median size of Decile 1 with the lowest decile of payout ratios is the same median size as the Decile 10 with the highest decile of payout ratios. Finally, there are significant economic sector effects on the dividend payout ratio. This also seems consistent with the maturity hypothesis as growth opportunities could be determined by the macroeconomic conditions for an entire industry sector. The regression indicates that firms in the Telecom, Energy, and Consumer Staples sector have higher payout ratios, while firms in the Technology sector have lower dividend payout ratios. In summary, the regression analysis indicates that firm maturity is positively related to the dividend payout ratio. As a dividend paying firm matures, the dividend payout ratio increases. However, the significant maturity components that determine the dividend payout ratio are only age and volatility. Thus as a firm matures and the CRSP age increases, the dividend payout increases. In addition, the standard deviation decreases, and the dividend payout increases. This confirms the result from the dividend growth logits, which indicated that low volatility increases the probability of a dividend increase. The analysis also reveals that the sales growth rate is negatively and significantly related to the dividend payout ratio. Therefore, as the sales growth rate increases, the dividend payout ratio decreases. Again, these results are consistent with the maturity hypothesis. Young firms with large growth opportunities (high sales growth rates) will retain most of their earnings and have low distribution (low dividend payout ratios). Mature firms with lower growth opportunities (low sales growth rates) will retain less of their earnings and have higher distribution of earnings (higher dividend payout ratios).

130

5.5.4 Maturity and the dividend yield In this section, I investigate the relationship between firm maturity and the dividend yield. Since firm maturity is related to the dividend payout ratio, one might expect that the dividend yield is related to firm maturity. Furthermore, based on the maturity hypothesis, one expects mature firms to distribute more earnings as their growth opportunities decline and the costs of earnings retention increase. Before the regression analysis, I rank the sample of dividend payers by dividend yield and sort the sample into ten deciles ranked by dividend yield. The summary statistics for the dividend payers sorted by dividend yield decile are reported in Table 58. Decile 1 contains the 10% of the sample with the lowest dividend yields while Decile 10 contains the 10% of the sample with the highest dividend payout yields. Review of Table 58 indicates that the median maturity increases with increasing payout ratio until Decile 8. Further inspection of Table 58 reveals some other interesting relationships. In Decile 1 with the lowest dividend yields, the firms have the highest median sales growth rate. In fact, this inverse relationship between median sales growth rate and median dividend payout ratio continues across all deciles. Thus, in Decile 10 with the highest dividend payout ratios, the firms have the lowest sales growth rate. This relationship is consistent with the maturity hypothesis in that firms with large growth opportunities (high sales growth rates) will retain most of their earnings and have low distribution (low dividend yield). Furthermore, in Decile 1 with the lowest dividend yields, the firms have the highest earnings growth rate. Likewise, in Decile 10 with the highest dividend yields, the firms have the lowest earnings growth rate. In fact, in Deciles 9 and 10, the firms have a negative median earnings growth rate. The median M/B ratio in each decile also reflects this picture of the growth potential. Firms in Decile 1 with the lowest dividend yields have the

131

highest sales growth rates and growth potential; consequently, those firms in Decile 1 have the highest M/B ratio. Likewise, firms in Decile 10 with the highest yields have the lowest growth rates and growth potential; consequently, the firms in Decile 10 have the lowest M/B ratios. Interestingly, the operating performance of the low yield/high growth firms is better than the operating performance of the high yield/low growth firms. Thus, firms in Decile 1 with the lowest dividend yields (but highest growth rates) also have the highest ROA and ROE. Investors reward the high growth potential and best operating performance firms in Decile 1 with the highest median monthly returns. Likewise, firms in Decile 10 with the lowest growth potential and lowest operating performance have the lowest median monthly returns. Overall, the univariate analysis based on a sort of the dividend payers by dividend yield deciles reveals results similar to the sort of the dividend payers ranked by dividend payout ratio. For the regression analysis, I again consider three econometric techniques applicable but only report the results from the analysis by the Fama and MacBeth method. (In unreported results, the panel methods again yield similar results and inferences as the Fama and MacBeth method). In the regression model, the dividend yield is the dependent variable. Again, the maturity variables are the explanatory variables while I utilize the prior control variables for growth, size, profitability, earnings growth rate, and economic sector. Table 59 reports the results of the dividend yield regressions for the dividend payers. In the first model, the maturity factor is the combination maturity variable, and the regression analysis indicates that maturity is not significant. In the second model, the maturity composite is the combination maturity variable, and again the regression analysis indicates that maturity is not significant. This indicates that dividend yield is not related to firm maturity. In the third model, the components of maturity are separated so that one can better understand the complicated relation with maturity. Table 59

132

shows that the parameter estimate for the RE/TE percentile is negative and significant at the 5% level. This indicates that the dividend yield decreases as the RE/TE percentile increases (as the firm matures). However, examination of the parameter estimate for the age percentile reveals that the coefficient is positive and significant at the 1% level. This indicates that the dividend yield increases as the age percentile increases (as the firm matures). These two components of maturity essentially cancel each other, leaving the net effect of the combination maturity variable to be insignificant. To better understand this somewhat surprising result, it helps to review both the prior results with the dividend payout ratio as well as the definition of dividend yield. Based on the prior regression analysis, the dividend payout ratio is positively related to firm maturity. However, examination of the maturity components indicates that as the age increases the dividend payout ratio increases, but the RE/TE percentile has no effect on the payout ratio. The regression on dividend yield still shows this positive relation with age. As a firm ages the dividend payout ratio increases, and all else held constant, the dividend yield increases. However, the dividend yield is the per share dividend distribution divided by the price per share. Therefore, there is an effect from the dividend distribution, but there is also a price effect on the dividend yield. The regression results indicate that the dividend distribution increases with maturity in accordance with the maturity hypothesis. On the contrary, there is no relation between firm maturity and the price per share in accordance with the efficient market hypothesis. In summary, the market determines dividend yield, and dividend yield is not related to firm maturity. Other significant variables in the regression are generally consistent with the firm lifecycle trade-offs between growth and earnings distribution. Sales growth rate and earnings growth rate are negatively related to the dividend yield, therefore, the dividend yield declines as sales

133

and earnings growth increase. Interestingly, the only significant economic sector effect is the Telecom sector, and firms in the Telecom sector have a significantly higher dividend yield. 5.5.5 Dividend growth rates 5.5.5.1 Maturity and the dividend growth rate Finally, I investigate the relationship between firm maturity and the dividend growth rate. The univariate summary statistics reported in Table 50 suggest a negative relationship between firm maturity and the dividend growth rate for dividend growers. A review of Table 50 indicates that as the maturity composite of dividend paying firms increases, the median dividend growth rate decreases. In this section, I utilize regression analysis to study this apparent relationship. However, before the regression analysis of the relationship between maturity and the dividend growth rate, some constraints on the data are required. The growth rate extends without lower or upper bounds since it is a continuous variable. However, there are limits to the dividend growth rate for sustainable growth of the firm. Although the dividend growth rate can exceed 100%, this indicates a situation where the firm distributes more than twice its prior dividend and the situation is clearly unsustainable over the long run. If the firm cuts the dividend, then dividend growth is negative, and dividend cutting is again clearly not a sustainable growth strategy over the long run. Furthermore, firms with zero dividend growth are not of interest for this analysis. In order to study sustainable firm growth, I eliminate observations with nongrowers and the observations with dividend growth rates above 100% for this analysis. After eliminating the observations without positive dividend growth rates as well as dividend growth rates over 100%, it is informative to examine some properties of the remaining sample. First, I rank the remaining sample by dividend growth rate and sort the sample into ten deciles ranked by dividend growth rate. The summary statistics for the dividend payers sorted by

134

dividend growth rate decile are reported in Table 60. Decile 1 contains the 10% of the sample with the lowest dividend growth rates while Decile 10 contains the 10% of the sample with the highest dividend growth rates. Review of Table 60 again indicates the generally negative relationship between firm maturity and dividend growth rate. Thus, Decile 10 with the highest dividend growth rates has the lowest median maturity. Table 60 reveals that dividend growth seems positively related to the sales and earnings growth rates. Decile 10 with the highest dividend growth rates also has the highest median sales and earnings growth rates. Interestingly the higher dividend growth rates correspond to the higher levels of operating performance. Decile 10 with higher dividend growth rates than Decile 1 also has a higher ROA% and ROE% than Decile 1. As expected, higher dividend growth rates correspond to lower median dividend payout ratios. This continues to reaffirm my assertion that the maturity hypothesis determines dividend policy in the distribution stages of the life-cycle. Table 60 shows a general decline in the per share dividend as the firm maturity declines. Furthermore, the growth rate of the dividend seems to depend on the firm maturity. For the regression analysis, I again consider the Fama and MacBeth method as well as random effects panel regression and fixed effects panel regression. In the regression model, the dividend growth rate is the dependent variable. Again, the maturity variables are the explanatory variables while I utilize the prior control variables for growth, size, profitability, earnings growth rate, and economic sector. Table 61 reports the results of the dividend growth rate regressions for the dividend growers with only maturity variables in the regressions. In the first model, the maturity factor is the combination maturity variable, and the regression analysis indicates that maturity is indeed negatively related to the dividend growth rate and significant. In the second model, the maturity composite is the combination maturity variable, and again the regression

135

analysis indicates that maturity is negatively related to the dividend growth rate and significant. This indicates that the dividend growth rate decreases as firm maturity increases. In the third model, the components of maturity are separated so that one can better understand the relation of the dividend growth rate with maturity. Table 61 shows that the parameter estimates for all the components of maturity are significant at the 1% level. Furthermore, all the components of maturity correspond to the same overall effect of maturity as with the propensity to pay and propensity to initiate analysis. Since the Fama and MacBeth method provides similar results and inferences as the panel regressions, I report the remaining regression results based on the Fama and MacBeth method. Table 62 reports the results from the dividend growth rate regressions when the control variables are included. Again, in the first two models one finds that the maturity factor and maturity composite are negatively related to the dividend growth rate even after adding the control variables. In the third model where the components of maturity are separated, I again find that all components of maturity are significant at the 1% level after controlling for size, profitability, growth, and industry sector. This indicates that firm maturity is negatively related to the dividend growth rate, and all of the components of maturity are significant. Furthermore, all of the components of maturity are consistent with the overall effect of maturity. Thus as a firm matures and the CRSP age increases, the dividend growth rate decreases. If the RE/TE percentile increases, the dividend growth rate declines. Likewise, as a firm matures and the standard deviation decreases, the dividend growth rate falls. Further inspection of Table 62 reveals that the sales growth rate and earnings growth rate is positively related to the dividend growth rate, which confirms the univariate analysis of Table 60. This indicates that an increasing sales (and earnings) growth rate increases the dividend

136

growth rate, while a decreasing sales growth rate decreases the dividend growth rate. As expected by the maturity hypothesis, the dividend distribution increases at low sales growth rates and low growth potential as the firm matures. Although the magnitude of the dividend distribution increases as the firm matures, the dividend grows at a diminishing rate. This declining rate of dividend growth as the firm matures is the “Law of Large Numbers” applied to corporate growth. Mature firms with large dividend distributions have a lower probability of increasing the rate of dividend growth than young firms with lower dividend distributions. Furthermore, as the firm matures and becomes a larger proportion of the economy, the dividend growth rate declines asymptotically to the growth rate of the economy. Likewise, a high prior dividend payout ratio (large dividend distribution) decreases the dividend growth rate. The regression analysis also indicates that profitability and the cash ratio are positively related to the dividend growth rate. This supports the concept that firms that generate cash flow and have excess cash are more likely to increase the dividend distribution. Finally, the only significant economic sector effect is the Telecom sector. Firms in the Telecom sector have higher dividend payout ratios and high dividend yields; consequently, firms in the Telecom sector have a significantly lower dividend growth rate. In summary, the dividend growth rate declines as a firm matures. Furthermore, all of the components of maturity are significant and consistent with the overall effect of maturity. Consistent with the maturity hypothesis and the “Law of Large Numbers”, the dividend distribution increases as the firm matures, but the dividend growth rate occurs at a diminishing rate.

137

5.5.5.2 Estimating the dividend growth rate Financial analysts often estimate a firm’s future dividend growth rate in order to utilize dividend discount models to value a firm’s equity. A common method to estimate the future dividend growth rate is to compute the firm’s sustainable growth rate. In theory, the sustainable growth rate does not depend on firm maturity. In order to test the relationship between the actual dividend growth rate and the sustainable growth rate, I include the sustainable growth rate as an explanatory variable in the dividend growth rate regressions. Table 63 reports the results of the dividend growth rate regressions with sustainable growth as an explanatory variable. In the simple regression with no control variables, the sustainable growth rate is positively related to the dividend growth rate, but the explanatory power is poor. The second model in Table 63 indicates that dividend growth rate depends on maturity even when the sustainable growth rate is included in the regression. In the third model, again the dividend growth rate depends on maturity even when the sustainable growth and control variables are included in the regression. Finally, when the components are separated in the model all of the components of maturity are still significant and negatively related to the dividend growth rate. These results indicate that over the cross section of firms, the sustainable growth rate is only one of many determinants of the dividend growth rate. In other words, the sustainable growth rate by itself is a rather poor predictor of the dividend growth rate. Furthermore, the dividend growth rate depends on maturity even when the sustainable growth rate is included in the regression model. Although the sustainable growth rate is not a function of maturity in theory, it is informative to regress the sustainable growth rate on maturity. The results of the regressions of the sustainable growth rate as a function of maturity are reported in Table 64. Examination of Table 64 indicates that maturity is not a significant determinant of the sustainable growth rate as

138

expected. Figure 39 shows the linear models for the dividend growth rate and sustainable growth rate as a function of the maturity composite. Figure 39 clearly illustrates several problems estimating the dividend growth rate. First, the sustainable growth rate only approaches the dividend growth rate at the very mature stage of the life-cycle. In the early part of the life-cycle, the dividend growth rate is significantly above the sustainable growth rate. Thus, the sustainable growth rate is only a reasonable estimate of the dividend growth rate when the maturity composite is greater than 2.5. Figure 39 indicates that the dividend growth rate declines monotonically with maturity. Therefore, the dividend growth rate crosses the sustainable growth rate, and the dividend growth rate is lower than the sustainable growth rate at the maximum level of maturity. In order to further investigate the monotonic decline of the dividend growth rate, I regress the dividend growth rate by maturity composite range and report the results in Table 65. Over all ranges of the life-cycle, firm maturity is negatively related to the dividend growth rate. In fact, the parameter estimate on maturity has the largest negative coefficient in the most mature lifecycle stage with the maturity composite between 2.5 and 3.0. This indicates that as firms progress in the life-cycle, the dividend growth rate continues to decline, even in the most mature part of the life-cycle. The following examples highlight the implications that this research has on the methods financial analysts utilize to estimate dividend growth rates. First, Figure 40 shows the dividend growth rate and sustainable growth rate as a function of the maturity composite for Wal-Mart Stores over the 1982-2010 data series. Similar to the cross-sectional data in Figure 39, the dividend growth rate and the sustainable growth rate of Wal-Mart Stores decline as the firm matures. Likewise, Figure 40 shows that the dividend growth rate of Wal-Mart Stores is

139

significantly larger than the sustainable growth rate when the maturity composite is less than about 2.5. Thus, the sustainable growth rate is only a reasonable estimate of the dividend growth rate for Wal-Mart Stores when the maturity composite is greater than 2.5. At best, the computed sustainable growth rate is a point estimate of the dividend growth rate in the later stage of the firm life-cycle. In the case of Wal-Mart Stores, Figure 40 illustrates that the dividend growth rate declined from about 50% in the early 1980’s to about 10% in 2010. At first glance, this research seems to refute the underlying assumption of constant dividend growth (at some time) in the dividend discount models. However, that is not a correct implication of these results. The research indicates that the dividend growth rate declines as the firm matures in the life-cycle. Therefore, the research actually provides a condition on the assumption of constant dividend growth, and that condition for constant dividend growth is that the firm must remain fixed in the life-cycle. It is important to recall that linear time is not life-cycle time. In the case of Wal-Mart Stores, its maturity composite increased from about 2.1 in 1982 to about 2.8 in 2010. Wal-Mart matured on the life cycle scale; consequently, its dividend growth rate declined. However, consider Table 66, which contains the time series data for Procter & Gamble over the 1982-2010 study. Obviously, Procter & Gamble’s CRSP age increased from 1982 to 2010, but note that Procter & Gamble’s maturity composite (the proxy for life-cycle) remains fairly constant. Although a peculiarity of the study end points, the maturity composite is 2.84 in 1982 and 2.84 in 2010. Since Procter & Gamble remained fixed in its stage of the life-cycle, the dividend growth rate was fairly constant over the period. The correct implication of this dividend grow rate research is that the constant dividend growth rate assumption requires a firm to remain fixed in its stage of the life-cycle.

140

In summary, the sustainable growth rate only approaches the dividend growth rate at the very mature stage of the life-cycle. In the early part of the life-cycle, the dividend growth rate is significantly above the sustainable growth rate. Thus, the sustainable growth rate is only a reasonable estimate of the dividend growth rate when the maturity composite is greater than 2.5. Furthermore, as firms progress in the life-cycle, the dividend growth rate continues to decline, even in the most mature part of the life-cycle. Constant dividend growth rate implies that a firm remains fixed in its life-cycle stage. Based on the analysis of Section 5.5, I find strong support for Hypothesis 4. Firm maturity is significant and negatively related to the dividend growth rate. Hypothesis 4. Total risk, firm age, earned capital ratio, and combinations of these firm maturity variables describe the cross section of firm dividend growth rates, and the dividend growth rate declines with firm maturity. Finding: Supported

CHAPTER 6 CONCLUSIONS, LIMITATIONS, FUTURE RESEARCH 6.1

Summary and Conclusions The main theme of this dissertation is that firm maturity affects dividend policy, and in

turn, dividend policy affects firm value throughout the firm’s life-cycle. While prior research advances a “life-cycle” or maturity hypothesis to explain the corporate dividend policy for industrial firms, the prior investigations of the life-cycle hypothesis utilize different measures of maturity to capture different dimensions of a firm’s life-cycle. These prior studies show that firm age, the earned capital ratio, and risk are statistically significant measures of maturity when tested independently. This dissertation investigates all these measures of firm maturity jointly in order to determine which maturity variable or combination of maturity variables best explains a firm’s dividend policy. 6.1.1 Maturity and the propensity to pay a dividend Consistent with the prior literature, I show that firm maturity is positively related to the probability that a firm pays a dividend. The logit analysis of the probability of paying dividends indicates that each individual definition of maturity reported in the prior literature captures a statistically significant dimension of maturity.

Furthermore, the combinations of maturity

variables provide the most complete definition of the life-cycle. Of the components of maturity, the RE/TE ratio has the largest effect on the decision to pay a dividend. Although the prior literature utilizes only the Fama and MacBeth method for the logit analysis, I demonstrate that the panel logistic method is essentially equivalent to the Fama and MacBeth approach for robustness. Finally, the analysis reveals that a firm’s economic sector is also a statistically

141

142

significant factor in the decision to pay a dividend. Life-cycle models with combination maturity variables correctly classify about 85% of the observations as dividend payers or not. There are significant implications for firms that have a dividend policy contrary to the life-cycle model. The valuation of firms that “follow” the maturity model is higher than the valuation of firms that do not “follow” the model. Including the prior dividend status into the life-cycle model improves the correct classification of observations as dividend payers or not to over 96%. This indicates that there is a substantial amount of “dividend momentum” in the decision to pay a dividend, but following the prior dividend status may not be a value-maximizing strategy. Analysis of the firms that do not follow the life-cycle model’s prediction for dividend policy provides further insight in the “disappearing dividends” phenomena. The majority of the “outliers” or firms that have a dividend policy contradictory to the maturity model are dividend paying firms. While the prior literature focuses on non-payers that should pay, my results show that “over-zealous” dividend payers are more numerous in the 1982-2010 time series. I consider the dividend paying outliers to be “over-zealous” as a group since they have low median maturity, lower median profitability, and smaller median size. My analysis shows that the decline in the propensity to pay dividends (or “disappearing dividends”) is due to the decline in the propensity of “over-zealous” dividend payers. My analysis reveals that the decline in “over-zealous” dividend payers is related to the relative market valuation of these outliers. In the early 1980’s, the market seemed to value “immature” dividend payers equal to “mature” dividend payers. However, as the market valuations became less favorable to “overzealous” dividend payers, fewer firms became “over-zealous” in paying dividends. With fewer “over-zealous’ dividend payers, the aggregate percentage of dividend payers declines; hence one

143

observes “disappearing dividends”. The panel logistic regression analysis with year effects indicates that after controlling with the maturity model, the propensity to pay dividends is lower than 1982, and the reduction in propensity to pay each year after 1982 is statistically significant. Although this indicates that “disappearing dividends” and the reduced propensity to pay is statistically significant, the analysis failed to relate any specific macroeconomic factors to the year effects. 6.1.2 Maturity and firm value This study provides evidence that resolves an empirical anomaly in the prior dividend literature. Prior empirical studies show that the median valuation of non-paying firms, as measured by the M/B ratio, is greater than the median valuation of dividend paying firms. This result leads to a re-statement of the dividend puzzle-Why would a value-maximizing firm ever pay a dividend if it leads to a lower valuation? My research shows that firm value, as measured by the M/B ratio, is related to firm maturity and the life-cycle. Early in the life-cycle, non-paying firms have high growth potential and high valuations. However, the M/B ratio of non-paying firms continues to decline as the non-paying firms mature. On the other hand, the M/B ratio of dividend paying firms increases as the dividend paying firms mature in the life-cycle. The opposing slopes in the life-cycle set up a crossover point where eventually a maturity is reached where the M/B ratio of dividend payers is greater than the M/B on non-payers. At this crossover point, a non-paying firm should begin a dividend policy as it matures in order to maximize firm value. Otherwise, the firm’s value will continue to decline as the non-paying firm matures. The valuation analysis is also consistent with the analysis of outliers from the propensity to pay maturity models. The “over-zealous” payers have a lower M/B ratio than if they did not pay a dividend because they pay a dividend prior the crossover point in the life-cycle. Likewise,

144

the non-payers that should pay have a maturity beyond the crossover point. The non-payers that should pay would maximize firm value if they paid a dividend. In summary, the life-cycle of firm value (on the maturity scale) resolves the question of why (and when) should a firm pay a dividend. When a firm is most immature, it has tremendous growth potential and the market values that enormous growth potential with extremely high M/B ratios. To take advantage of the large growth opportunities, the firm requires capital and retains any earnings. However, as the firm matures the growth opportunities decline and the market values the declining growth potential with declining M/B ratios. As the non-paying firm further matures to the “crossover” point, the firm sees fewer re-investment opportunities and the costs of earnings retention exceed the costs of earnings distribution. Thus, the firm begins a dividend policy in order to maximize firm value. The market observes the firm’s value and assigns the appropriate M/B valuation. The firm then matures as a dividend-payer, and its value rises as the financial characteristics of the mature firm (such as profitability) further improve. I find that the relationships between maturity and firm value are robust to the definition of maturity. The maturity composite variable provides a rather interesting interpretation of the crossover point since the empirical data indicates that a firm should become a dividend payer at a maturity composite value of about 1.5. At the maturity composite of 1.5, a firm is on average at the 50th percentile of RE/TE, age, and risk. This indicates that firms, which average lower than the 50th percentile of RE/TE, age, and risk, should not pay a dividend. Meanwhile firms that average higher than the 50th percentile of RE/TE, age, and risk, should pay a dividend in order to maximize firm value. Across the entire life-cycle, the age percentile is negatively related to firm value. The regression results imply that paying a dividend in the middle of the life-cycle counters the negative effects of firm age. In all stages of the life-cycle, the standard deviation percentile is

145

positively related to firm value. This indicates that volatility adds value to firm, much like an option. While the standard deviation percentile is always positively related to the M/B ratio, the effect is greatest at the lowest ranges of maturity. Similarly, the positive effect of volatility on firm value is much larger on non-payers than on dividend payers. Firm size is also always positively related to firm value in all stages of the life-cycle. However, the positive effect is much larger for the least mature firms. Likewise, the positive effect of firm size is much larger on non-payers than on dividend payers. In the early part of the life-cycle, profitability is negatively related to firm value. At this stage of the life-cycle, value seems to be maximized if the firm grows at the expense of profitability. As the non-payers mature, the relation between profitability and M/B ratio turns positive. However, profitability is even more positively related to firm value for dividend payers. This indicates that at the crossover point, a firm with less growth potential and more profitability will have a higher value if it pays a dividend. Finally, for the most mature dividend payers in the life-cycle, profitability is the key determinant of firm value. Higher profitability is the major explanation for the increase in valuation for dividend payers as they mature over the life-cycle. Dividend payers do not become more valuable simply because they are more mature. Further regression results indicate that dividend payout policy does affect firm value of dividend payers. However, the analysis also reveals the complexity of the payout policy. Over the life-cycle of a dividend paying firm, the dividend growth rate is positively related to firm value. On the other hand, the analysis reveals that the dividend payout ratio has no effect on firm value. While dividend growth adds value to the firm, the effect is only significant for the most mature payers. Finally, a high dividend yield decreases firm value, especially for “immature” payers.

146

6.1.3 Maturity and dividend initiation Consistent with the prior literature, the results from the logit analysis of the decision to initiate a dividend are very similar to the results from the logit analysis of the decision to pay a dividend. As a non-paying firm matures, its propensity to initiate a dividend increases. Furthermore, each separate definition of maturity captures a statistically significant dimension of firm maturity, but the combination maturity variables seem to provide the most complete definition of maturity in the life-cycle. As with the propensity to pay a dividend, the RE/TE percentile has the largest effect of the maturity variables on the decision to initiate. As firms become larger and more profitable, their propensity to initiate a dividend increases. Consistent with the maturity hypothesis, firms are less likely to initiate a dividend if they have significant growth potential as measured by the M/B ratio. In addition to firm characteristics, there are significant economic sector effects on the decision to initiate a dividend. The Health Care and Technology sectors show a reduced propensity to initiate a dividend in all stages of the lifecycle. Firms in the Materials and Consumer Staples sectors show an increased propensity to initiate a dividend but only in the mature stage of the life-cycle. Based on the analysis of firm maturity and valuation, one would expect most firms to initiate a dividend near the crossover point in the life cycle at a maturity composite value of about 1.5. Indeed the analysis reveals that most dividend initiation occurs in the maturity composite range of 1-2. Although infrequent, firms that initiate a dividend early in the life-cycle or “early initiators” have low median RE/TE ratios, low median profitability, low median size, high sales growth rates, and high risk - none of the usual characteristics of dividend payers. Consistent with their growth potential, these “early initiators” have a high M/B ratio and experience high median monthly returns during the year in which they initiate a dividend.

147

Finally, “late initiators” in the maturity stage of the life-cycle have the characteristics of mature dividend payers except size. As a group, it seems that the “late initiators” spend about a decade trying to grow, but do not. After realizing their actual limited growth potential, the “late initiators” finally begin to pay a dividend. Consistent with the valuation analysis, “late initiators” have the lowest M/B ratio and have the lowest median monthly return during the year in which they initiate a dividend. While the prior literature utilizes only the Fama and MacBeth method for the logit analysis, again I demonstrate that the panel logistic method is essentially equivalent to the Fama and MacBeth approach for robustness. The panel logistic regression analysis with the year effects indicates that significant changes in the propensity to initiate a dividend do occur over the 1982 to 2010 time period. While a time series plot suggests a declining propensity to initiate from the early 1980’s to about 2003, the panel logit analysis with year effects confirms that the reduction in propensity to initiate is statistically significant even after controlling for maturity, firm characteristics, and economic sector.

I find no significant negative year effect (from the

base year of 1982) at the 5% level until the mid-1990’s, which is over a decade after the adoption of the safe-harbor rule (in 1982). Rather the panel logit results with year effects are consistent with my assertion that the macroeconomic environment shifted from one that favored distribution of earnings in the early 1980’s to one that then favored earnings retention in the mid 1990’s when the development of the internet and related new technologies provided corporations with growth opportunities previously unseen. Then corporations need more retained earnings for reinvestment, and the propensity to initiate a dividend declines. For robustness, I show that event analysis and a hazard model for dividend initiation provides results consistent with the logit analysis. The Cox proportional hazard regression

148

indicates that the hazard rate (or probability) for dividend initiation increases with firm maturity. In addition, the Cox proportional hazard regression indicates significant economic sector effects consistent with the logit analysis. The Health Care and Technology sectors show lower hazard rates (or lower probabilities to initiate a dividend) while the Materials and Consumer Staples sectors show higher hazard rates. Finally, the hazard plot in life-cycle time clearly describes dividend initiation better than hazard plot in event time. The hazard plot in life-cycle time shows the hazard rate (or probability of dividend initiation) increases dramatically in the maturity composite range of the crossover point between non-payers and dividend payers. 6.1.4 Maturity and dividend policy While the prior literature contains studies on the relationship between maturity and the propensity to pay a dividend as well as maturity and the propensity to initiate a dividend, none investigate dividend payout policy and the maturity hypothesis. I investigate the relationship between maturity and dividend payout policy and report that firm maturity is positively related to the probability of a dividend increase and negatively related to the probability of a dividend cut. However, the relationship between dividend payout policy and maturity is more complex than with the propensity to pay and propensity to initiate analysis. With dividend payout policy analysis, the components of maturity are often inconsistent with the net effect of maturity. Interestingly, as the firm’s CRSP age increases, the probability of dividend growth declines and the probability of a dividend cut increases. However, standard deviation is the maturity variable with the largest effect, and as the firm matures and becomes less volatile, the probability of dividend growth increases and the probability of a dividend cut decreases. Therefore, the dominant effect of standard deviation cancels out the offsetting effect of firm age so that the net effect is still that the probability of dividend growth increases with maturity. The RE/TE ratio,

149

which is the largest effect of maturity variables in the propensity to pay and propensity to initiate logits, is insignificant to the probability that a firm increases its dividend. Firm size and profitability are positively related to the probability of a dividend increase and negatively related to the probability of a dividend cut. Poor operating performance is strongly related to the probability of a dividend cut, as lower sales growth and lower earnings growth increase the probability of a dividend cut. An excessive dividend payout ratio reduces the probability of future dividend growth and increases the probability of a future dividend cut. The net effect of firm maturity is that a more mature firm will be more stable, have a higher RE/TE ratio, and larger in size. These attributes provide the firm with other financing options in lieu of a dividend cut, thus more mature firms have a lower probability of a dividend cut. For robustness, I repeat the Fama and MacBeth method logit analysis of the decision to increase (or decrease) a dividend using a random effects panel logistic model and demonstrate that the panel logistic method results are essentially equivalent to the results from the Fama and MacBeth approach. My analysis indicates that firm maturity is positively related to the dividend payout ratio. As a dividend paying firm matures, the dividend payout ratio increases. However, the significant maturity components that determine the dividend payout ratio are only age and volatility. Thus as a firm matures and the CRSP age increases, the dividend payout increases. In addition, the standard deviation decreases, and the dividend payout increases. This confirms the result from the dividend growth logits, which indicated that low volatility increases the probability of a dividend increase.

The analysis also reveals that the sales growth rate is negatively and

significantly related to the dividend payout ratio. Therefore, as the sales growth rate increases, the dividend payout ratio decreases. Again, these results are consistent with the maturity

150

hypothesis. Young firms with large growth opportunities (high sales growth rates) will retain most of their earnings and have low distribution (low dividend payout ratios). Mature firms with lower growth opportunities (low sales growth rates) will retain less of their earnings and have higher distribution of earnings (higher dividend payout ratios). Also firms in the Telecom, Energy, and Consumer Staples sector have higher payout ratios, while firms in the Technology sector have lower dividend payout ratios. The regression analysis indicates that dividend yield is not related to firm maturity. When the components of maturity are separated so that one can better understand the complicated relation with maturity, the significant components of maturity essentially, cancel each other leaving the net effect of the combination maturity variable on dividend yield to be insignificant. Since the dividend yield is the per share dividend distribution divided by the price per share, there is an effect from the dividend distribution, but there is also a price effect on the dividend yield. The regression results indicate that the dividend distribution increases with maturity in accordance with the maturity hypothesis. However, there is no relation between firm maturity and the price per share in accordance with the efficient market hypothesis. The market price largely determines dividend yield; consequently, dividend yield is not related to firm maturity. The only significant economic sector effect on dividend yield is the Telecom sector, and firms in the Telecom sector have a significantly higher dividend yield. Investigation of the dividend growth rate for dividend growers reveals that the dividend growth rate declines as a firm matures. Furthermore, all of the components of maturity are significant and consistent with the overall effect of maturity. Consistent with the maturity hypothesis and the “Law of Large Numbers”, the dividend distribution increases as the firm matures, but the dividend growth rate occurs at a diminishing rate.

151

Further investigation of the dividend growth rate shows that the sustainable growth rate only approaches the dividend growth rate at the very mature stage of the life-cycle. In the early part of the life-cycle, the dividend growth rate is significantly greater than the sustainable growth rate. Furthermore, as firms progress in the life-cycle, the dividend growth rate continues to decline, even in the most mature part of the life-cycle. Constant dividend growth rate then implies that a firm remains fixed in its life-cycle stage. 6.1.5 Hypotheses findings I find that all hypotheses in this dissertation are supported as summarized below. Hypothesis

Finding

Hypothesis 1a. Total risk significantly determines a firm’s propensity to pay Supported dividends when controlled for the Fama-French (2001) firm characteristics, firm age, earned capital ratio, and combinations of risk, firm age, and earned capital ratio. Hypothesis 1b. Total risk significantly determines a firm’s decision to initiate Supported dividends when controlled for the Fama-French (2001) firm characteristics, firm age, earned capital ratio, and combinations of risk, firm age, and earned capital ratio. Hypothesis 2a. Relative to the base firm life-cycle model, yearly time effects Supported captured by year dummy variables are significant determinants of the firm’s propensity to pay a dividend. Hypothesis 2b. Relative to the base firm life-cycle model, yearly time effects Supported captured by year dummy variables are significant determinants of the firm’s decision to initiate a dividend. Hypothesis 3a. The non- payers that do not fit the model (against type) have a Supported significantly lower M/B ratio than the non-payers that fit the life-cycle model. Hypothesis 3b. The dividend payers that do not fit the model (against type) have a Supported significantly lower M/B ratio than the dividend payers that fit the life-cycle model. Hypothesis 4. Total risk, firm age, earned capital ratio, and combinations of these Supported firm maturity variables describe the cross section of firm dividend growth rates, and the dividend growth rate declines with firm maturity.

152

6.2

Limitations One of the key limitations of this study is the restrictions on the data set. By design of the

study, financial firms and utilities are removed from the analysis. Therefore, the conclusions of this investigation are restricted to the “industrial” firms in the study. Another significant data limitation set by the research design is the inclusion of only firms with positive equity so that the RE/TE ratio can be computed. The requirement of positive equity eliminates mostly non-paying firms from the sample. A limitation of the firm age variable is the use of the CRSP data set for determining the age since the CRSP data set begins in 1925, and some firm were trading as public corporations before 1925. Another important limitation is to realize the weakness of using specific variables to proxy broad financial or economic concepts. This issue is evident in the major topic of research – firm maturity and the life-cycle. I expand the definition of firm maturity to include the three components of firm maturity reported in the prior literature; however, there may be additional components of that capture some part of the wide concept of firm maturity. Other expansive financial concepts in the study that utilize proxy variables are risk, growth potential, and valuation. In each case, I utilize proxy variables that are widely reported in the literature of corporate finance, but again one must acknowledge the limitations of using simple proxy variables for wide-ranging financial concepts. In addition to the above limitations with the data set and proxy variables, a common problem with using regression analysis for empirical research is that the models may be misspecified, leading to biased estimates and/or incorrect inferences. To mitigate this issue, a comprehensive set of control variables based on the previous literature is utilized in the regression models. Nevertheless, variables may still be omitted or the functional relationship

153

may be incorrect. Further robustness tests then included panel regressions to corroborate the results. Although the results are robust, even the panel regressions have econometric limitations. An important issue of this study is the applicability of the results to out of sample periods. An advantage of this investigation is that the time series spans 1982-2010. As the analysis shows, however, the macroeconomic conditions can and do shift from conditions that favor retention of earnings to circumstances that favor distribution of earnings. Finally, this investigation extends the life-cycle analysis to dividend payout policy of cash distributions. Share repurchases and equity distributions are not considered. 6.3

Future Research An extension of the present research would be to remove the data restrictions on

industrial and include the financial and utility firms in the sample to investigate if financial and utility firms also have a dividend policy that follows a life-cycle. Of course, that analysis would surely require the industry sector controls that I used in order to account for the different regulatory environment for those industry sectors. One of the major contributions of this research is demonstrating that the value of nonpaying firms continues to decline with maturity. The fact that non-paying firms decline more than dividend paying firms strongly suggests the increasing agency costs of free cash flow for the non-payers. Future research could further investigate the decline in value to better define the loss of value to agency costs. Since the focus of this research was dividend policy, it is only reported that the cash management and capital structure change with firm maturity in this dissertation. Future research could further investigate cash management and capital structure over the life-cycle and determine

154

if the maturity hypothesis can be incorporated with existing cash management and capital structure theories. This dissertation utilizes the standard deviation of monthly returns as the proxy variable for risk, but does not otherwise consider the monthly returns. Although the median or equalweighted returns are reported in tables, formal analysis of the monthly return data is outside the scope of the present study. However, it is clearly noted in several tables that monthly returns appear to be related to firm maturity. This strongly suggests that research in efficient portfolios would find that the firm life-cycle and dividend policy is relevant. Finally, this research shows that the distribution of earnings increases with firm maturity over the life cycle. Since the ultimate source of the distribution is free cash flow, it is likely that free cash flow increases with firm maturity. Extending the life-cycle analysis to free cash would be valuable research since the analysis of free cash flow generation is more widely applicable in valuation. Dividend discount models are a subset of valuation models, whereas free cash models can be used to value any corporation. An interesting issue to resolve with the free cash flow analysis is the agency problem. This research shows that the valuation of non-payers declines over the life-cycle regardless of the firm’s free cash flow generation. On the other hand, general valuation models only consider free cash flow and ignore the dividend paying status. Future research of free cash flow and the life-cycle might resolve this valuation issue.

155

Appendix A: Listing and Detailed Explanation of Variables Used Variable

Definition

Age Percentile =

Percentile rank by firm age

b=

Firm reinvestment ratio

Balance Sheet Deferred Taxes and Investment Tax Credit

This item represents the accumulated tax deferrals due to timing differences between the reporting of revenues and expenses for financial statements and tax forms and investment tax credit. [Compustat Mnemonic TXDITC, Compustat Data Item A35]

Book Equity=

Stockholders Equity minus Preferred Stock plus Balance Sheet Deferred Taxes and Investment Tax Credit minus Post Retirement Asset. If Stockholder’s Equity is not available, it is replaced by either Common Equity plus Preferred Stock Par Value or Total Assets minus Liabilities. Preferred Stock is Preferred Stock Liquidating Value or Preferred Stock Redemption Value or Preferred Stock Par Value

Cash (and Equivalents)

[Compustat Mnemonic CHE, Compustat Data Item A1]

Cash/TA =

Cash to total asset ratio

Common Equity

[Compustat Mnemonic CEQ, Compustat Data Item A60] Equivalent to total common equity, TE12

CRSP Age

Proxy variable for firm age. The CRSP age is the time that the firm entity (Permno) has had Price data available in the CRSP database.

D=

Dividend paying status in current year t. A dummy variable that is assigned a value of one if the firm paid a dividend in year t, and zero otherwise.

D1983 to D2010 =

Year dummy variable of year t is assigned a value of one if the observation occurs in year t, and zero otherwise.

D Materials =

Industry dummy variable assigned a value of one if the firm is included in the Materials economic sector, and zero otherwise.

D Consumer Discretionary =

Industry dummy variable assigned a value of one if the firm is included in the Consumer Discretionary economic sector, and zero otherwise.

12

TE as defined by DeAngelo et al. (2006)

156

D Consumer Staples =

Industry dummy variable assigned a value of one if the firm is included in the Consumer Staples economic sector, and zero otherwise.

D Health Care =

Industry dummy variable assigned a value of one if the firm is included in the Health Care economic sector, and zero otherwise.

D Energy =

Industry dummy variable assigned a value of one if the firm is included in the Energy economic sector, and zero otherwise.

D Industrials =

Industry dummy variable assigned a value of one if the firm is included in the Industrials economic sector, and zero otherwise.

D Technology =

Industry dummy variable assigned a value of one if the firm is included in the Technology economic sector, and zero otherwise.

D Telecom =

Industry dummy variable assigned a value of one if the firm is included in the Telecommunication Services economic sector, and zero otherwise.

DC =

Dividend cut status indicator in current year t. A dummy variable that is assigned a value of one if the firm cut the dividend in year t, and zero otherwise.

DINT =

Dividend initiator status in current year t. A dummy variable that is assigned a value of one if the firm pays a dividend in year t after having not paid dividends in year t-1, and zero otherwise.

Divt-1 =

Lagged dividend status with value of 1 if the firm paid a dividend in the prior year and zero otherwise. Also referred to as prior dividend status.

Dividends-Common

This item represents the total dollar amount of dividends (cash) declared on the common stock of the company in year t. [Compustat Mnemonic DVC, Compustat Data Item A21]

Dividend growth rate

Dividend growth rate, which equals (DPS t / DPS t-1) – 1

Dividends per Share (by Ex-date), DPSt

This item represents the cash dividends per share for which the exdividend dates occurred during the year t. [Compustat Mnemonic DVPSX, Compustat Data Item A26]

Dividend yield

total dividends paid in year t Market equity at end of year t

157

DG=

Dividend growth status in current year t. A dummy variable that is assigned a value of one if the firm increased the dividend in year t, and zero otherwise.

Economic sector code

This item contains the code that identifies all companies in the broad S&P economic industry groups. Code Economic Sector 1000 Materials 2000 Consumer Discretionary 3000 Consumer Staples 3500 Health Care 4000 Energy 6000 Industrials 8000 Technology 8600 Telecommunication Services

EGR=

Earnings (before extraordinary items) growth rate, which equals (IB t /IB t-1) -1, where IB is net income before extraordinary items

GVKEY

Global Vantage Key, unique firm identifier for Compustat database

Income before extraordinary items

[Compustat Mnemonic IB, Compustat Data Item A18]

Liabilities

Total Liabilities, [Compustat Mnemonic LT, Compustat Data Item A181]

Log(firm age) =

Natural logarithm of the firm age (in years)

M/B =

Market to book ratio. Book assets minus book equity plus market equity all divided by book assets.

Market Equity=

Year closing price times shares outstanding.

Maturity Composite =

The sum of the percentile rankings of age, earned capital ratio, and inverse of risk.

Maturity Factor=

Log(firm age) x RE/TE , Total risk = interaction variable of measures of firm maturity

Monthly returns

Total monthly returns from CRSP database

158

NYSE Percentile=

NYSE market equity capitalization percentile. The fraction of NYSE firms having equal or smaller capitalization than firm I in year t.

Pay

Dividend payout ratio, [Compustat Mnemonic DVPOR]

Permno

Unique firm identifier for CRSP database

Post Retirement Asset

Indicates the funded status of a postretirement plan as either overfunded or underfunded. [Compustat Mnemonic PRBA, Compustat Data Item A330]

Preferred Stock

Preferred Stock is Preferred Stock Liquidating Value or Preferred Stock Redemption Value or Preferred Stock Par Value

Preferred Stock Liquidating Value

[Compustat Mnemonic PSTKL, Compustat Data Item A10]

Preferred Stock Redemption Value

[Compustat Mnemonic PSTKRV, Compustat Data Item A56]

Preferred Stock Par Value

[Compustat Mnemonic PSTK, Compustat Data Item A130]

Retained Earnings

[Compustat Mnemonic RE, Compustat Data Item A36]

RE/TA =

Ratio of retained earnings to total assets

RE/TE =

Ratio of retained earnings to total equity

RE/TE Percentile =

Percentile rank by RE/TE

ROAt =

Return on assets in current year t, [Compustat Mnemonic ROA]

ROE =

Return on equity, [Compustat Mnemonic ROE]

Sales

[Compustat Mnemonic SALE, Compustat Data Item A12]

SIC

Standard Industry Classification Codes

SGR =

Sales growth rate, which equals (sales t / sales t-1) – 1

Standard deviation of monthly returns

Firm standard deviation of monthly returns, measure of total risk

159

Standard Deviation Percentile =

Percentile rank by firm standard deviation of monthly returns

Stockholder’s Equity

This item represents the common and preferred shareholder’s interest in the company. [Compustat Mnemonic SEQ, Compustat Data Item A216]

Sustainable growth rate

The theoretical growth rate under the sustainable growth rate assumptions. Sustainable growth rate = b x ROE

TA =

Total Assets, [Compustat Mnemonic AT, Compustat Data Item A6]

TE / TA =

Total equity to total asset ratio

Total risk =

Standard deviation of firm’s monthly returns

160

REFERENCES Adjaoud, F., and W. Ben-Amar, 2010. Corporate governance and dividend policy: Shareholders’ protection or expropriation? Journal of Business Finance & Accounting 37, 648-667. Aivazian, V.A., L. Booth, and S. Cleary, 2006. Dividend smoothing and debt ratings. Journal of Financial and Quantitative Analysis 41, 439- 453. Al-Kuwari, D., 2010. To pay or not to pay: Using emerging panel data to indentify factors influencing corporate payout decisions. International Research Journal of Finance and Economics 42, 19-36. Allen, F., A. E. Bernardo, and I. Welch, 2000. A theory of dividends based on tax clienteles. Journal of Finance 55, 2499-2536. Allen, F., and R. Michaely, 2003. Payout policy. In Constantinides, G., M. Harris, and R. Stulz, (Eds.), Handbook of the Economics of Finance: Corporate Finance Volume 1A. Elsevier North-Holland, 337-430. Amihud,Y., and K. Li, 2006. The declining information content of dividend announcements and the effects of institutional holdings. Journal of Financial and Quantitative Analysis 41, 637 – 660. Asem, E., 2009. Dividends and price momentum. Journal of Banking & Finance 33, 486 – 494. Baker, H. K., 2009. Dividends and Dividend Policy, John Wiley & Sons. Baker, H. K., G. E. Powell, and E. T. Veit, 2002. Revisting managerial perspectives on dividend policy. Journal of Economics and Finance 26, 267 – 283. Baker, M., and J. Wurgler, 2004a. Appearing and disappearing dividends: The link to catering incentives. Journal of Financial Economics 73, 271-288. Baker, M., and J. Wurgler, 2004b. A catering theory of dividends. Journal of Finance 59, 1125-1165. Benartzi, S., R. Michaely, and R. H. Thaler,, 1997. Do changes in dividends signal the future or the past? Journal of Finance 52, 1007- 1034. Bhattacharya, S., 1979. Imperfect information, dividend policy, and “the bird in the hand” fallacy. Bell Journal of Economics 10, 259-270. Black, F., 1976. The dividend puzzle. Journal of Portfolio Management, 5-8. Black, F. and M. Scholes, 1974. The effects of dividend yield and dividend policy on common stock prices and returns. Journal of Financial Economics 1, 1-22.

161

Brav, A., J. R. Graham, C. R. Harvey, and R. Michaely, 2005. Payout policy in the 21st century.Journal of Financial Economics 77, 483-527. Brown, J.R., N. Liang, and S. Weisbenner, 2007. Executive financial incentives and payout policy:Firm response to the 2003 dividend tax cut. Journal of Finance 62, 1935-1965. Bulan, L., N. Subramanian, and L. Tanlu, 2007. On the timing of dividend initiations. Financial Management 36, 31 – 65. Chen, L., 2009. On the reversal of return and dividend growth predictability: A tale of two periods. Journal of Financial Economics 92, 128-151. Chen, L., and X. Zhao, 2006. On the relation between market to book ratio, growth opportunity, and leverage ratio. Finance Research Letters 3, 253 – 266. Chetty, R., and E. Saez, 2005. Dividend taxes and corporate behavior: Evidence from the 2003 dividend tax cut. Quarterly Journal of Economics 120, 791- 833. Chetty, R., and E. Saez, 2006. The effects of the 2003 dividend tax cut on corporate behavior: Interpreting the evidence. American Economic Review 96, 124-129. Damodaran, A., 2011. The Little Book of Valuation: How to Value a Company, Pick a Stock, and Profit, John Wiley & Sons. Damodaran, A., 2012. Investment Valuation: Tools and Techniques for Determining the Value of any Asset 3rd ed., John Wiley & Sons. Dittmar, A., 2008. Corporate cash policy and how to manage it with stock repurchases. Journal of Applied Corporate Finance 20, 22-34. Dittmar, A. K., and R. F. Dittmar, 2008. The timing of financing decisions: An examination of the correlation in financing waves. Journal of Financial Economics 90, 59-83. DeAngelo, H., and L. DeAngelo, 2007. Payout Policy Pedagogy: What matters and Why. European Financial Management 13, 11-27. DeAngelo, H., L. DeAngelo, and D. Skinner, 2004. Are dividends disappearing? Dividend concentration and the consolidation of earnings. Journal of Financial Economics 72, 425-456. DeAngelo, H., L. DeAngelo, and D. Skinner, 2008. Corporate Payout Policy. Foundations and Trends in Finance 3, 95 – 287. DeAngelo, H., L. DeAngelo, and R. Stulz, 2006. Dividend policy and the earned/contributed capital mix: a test of the life-cycle theory. Journal of Financial Economics 81, 227- 254.

162

DeAngelo, H., L. DeAngelo, and R. Stulz, 2010. Seasoned equity offerings, market timing, and the corporate lifecycle. Journal of Financial Economics 95, 275 - 295. Deshmukh, S., 2003. Dividend initiations and asymmetric information: A hazard model. The Financial Review 38, 351 – 368. Easterbrook, F., 1984. Two agency cost explanations of dividends. American Economic Review 74, 650-659. Elton , E. J., and M. J. Gruber, 1970. Marginal stockholders’ tax rates and the clientele effect. Review of Economics and Statistics, 68-74. Engsted, T., and T. Q. Pedersen, 2010. The dividend price ratio does predict dividend growth: International evidence. Journal of Empirical Finance 17, 585 – 605. Fama, E., and H. Babiak, 1968. Dividend policy: An empirical analysis. Journal of the American Statistical Association 63, 1132 – 1161. Fama, E., and K. French , 1992. On the cross-section of expected stock returns. Journal of Finance 47, 427 – 466. Fama, E., and K. French , 2001. Disappearing dividends: Changing firm characteristics or lower propensity to pay? Journal of Financial Economics 60, 3-44. Fama, E., and J. MacBeth, 1973. Risk, return, and equilibrium: empirical tests. Journal of Political Economy 81, 607 – 636. Ferris, S. P., N. Sen, and E. Unlu, 2009. An international analysis of dividend payment behavior. Journal of Business Finance & Accounting 36, 496 – 522. Fidrmuc, J. P., and M. Jacob, 2010. Culture, agency costs, and dividends. Journal of Comparative Economics 38, 321-339. Fuller, K., and B. M. Blau, 2010. Signaling, free cash flow and “nonmonotonic” dividends. Financial Review 45, 21 – 56. Fuller, K., and M. A. Goldstein, 2011. Do dividends matter more in declining markets? Journal of Corporate Finance 17, 457 – 473. Graham J., and A. Kumar, 2006. Do dividend clienteles exist? Evidence on dividend preferences Of retail investors. Journal of Finance 61, 1305 – 1336. Grinstein, Y., and R. Michaely, 2005. Institutional holdings and payout policy. Journal of Finance 60, 1389- 1426.

163

Grullon, G., and R. Michaely, 2002. Dividends, share repurchases, and the substitution hypothesis. Journal of Finance 62, 1649-1684. Grullon, G., R. Michaely, and B. Swaminathan, 2002. Are dividend changes a sign of firm maturity? Journal of Business 75, 387- 424. Higgins, R. C., 1977. How much growth can a firm afford? Financial Management , 7-16. Hoberg, G., and N. R. Prabhala, 2009. Disappearing dividends, catering, and risk. The Review of Financial Studies 22, 79-116. Irvine, P. J., and J. Pontiff, 2009. Idiosyncratic return volatility, cash flows, and product market competition. Review of Financial Studies 22, 1149- 1177. Jain, B. A., C. Shekhar, and V. Torbey, 2009. Payout initiation by IPO firms: The choice between dividends and share repurchases. Quarterly Review of Economics and Finance 49, 1275 – 1297. Jensen, M. C., 1986. Agency costs of free cash flow, corporate finance, and takeovers. American Economic Review 76, 323 – 329. Jo, H., and C. Pan, 2009. Why are firms with entrenched managers more likely to pay dividends? Review of Accounting and Finance 8, 87 – 116. Jordan, B. D., and T. W. Miller, 2009. Fundamentals of Investments 5th ed., McGraw-Hill. Jiraporn, P., and P. Chintrakarn, 2009. Staggered boards, managerial entrenchment, and dividend Policy. Journal of Financial Services Research 36, 1-19. Julio, B., and D. Ikenberry, 2004. Reappearing dividends. Journal of Applied Corporate Finance 16, 89-100. Jung, K., Y-C Kim, and R. M. Stulz. 1996. Timing, investment opportunities, managerial discretion, and the security issue decision. Journal of Financial Economics 42, 159- 185. Kahle, K. M., 2002. When a buyback isn’t a buyback: open market repurchases and employee options. Journal of Financial Economics 63, 235 – 261. Kao, M-Y., and S. R. Tate, 2001. Designing proxies for stock market indices is computationally hard. Quantitative Finance 1, 361 – 371. Kooli, M. and J.-F. L’Her, 2010. Dividends versus share repurchases evidence from Canada: 1985-2003. Financial Review 45, 57 – 81. Lambrecht, B.M., and S. C. Myers, 2012. A Lintner model of payout and managerial rents. Journal of Finance, forthcoming.

164

Leary, M.T., and R. Michaely, 2011. Determinants of dividend smoothing: Empirical evidence. Review of Financial Studies 24, 3197 – 3249. Lettau, M., and S. C. Ludvigson, 2005. Expected returns and expected dividend growth. Journal of Financial Economics 76, 583 – 626. Li, K., and X. Zhao, 2008. Asymmetric information and dividend policy. Financial Management 37, 673-694. Lintner, J., 1956. Distribution of incomes of corporations among dividends, retained earnings, and taxes. American Economic Review 46, 97-113. Litzenberger, R., and K. Ramaswamy, 1982. The effects of dividends on common stock prices: Tax effects or information effects. Journal of Finance 37, 429-444. Masilus, R., and B. Trueman, 1988. Corporate investment and dividend decisions under different personal taxation. Journal of Financial and Quantitative Analysis 23, 369 – 385. Michaely, R., and M. R. Roberts, 2012. Corporate dividend policies: Lessons from private firms. Review of Financial Studies 25, 711 – 746. Michaely, R., R. H. Thaler, and K. L. Womack, 1995. Price reactions to dividend initiations and omissions: Overreaction or drift? Journal of Finance 50, 573 – 608. Miller, M. H., and F. Modigliani, 1961. Dividend policy, growth, and the valuation of shares. Journal of Business 34, 411-433. Miller, M. H., and K. Rock, 1985. Dividend policy under asymmetric information. Journal of Finance 40, 1031-1051. Mueller, D., 1972. A life cycle theory of the firm. Journal of Industrial Economics 20, 199-219. Naranjo, A., M. Nimalendran, and M. Ryngaert, 1998. Stock returns, dividend yields and taxes. Journal of Finance 53, 2029-2057. Officer, M. S., 2011. Overinvestment, corporate governance, and dividend initiations. Journal of Corporate Finance 17, 710 – 724. Owen, S., and A. Yawson, 2010. Corporate life cycle and M&A activity. Journal of Banking and Finance 34, 427 – 440. Rozeff, M., 1982. Growth, beta, and agency costs as determinants of dividend payout ratios. Journal of Financial Research 5, 249-259. Rubin, A., and D. R. Smith, 2009. Institutional ownership, volatility and dividends. Journal of Banking and Finance 33, 627- 639.

165

Sharma, V., 2011. Independent directors and the propensity to pay dividends. Journal of Corporate Finance 17, 1001-1015. Shin, H., and R. Stulz, 2000. Firm value, risk , and growth opportunities. Working Paper No. 7808. NBER. Shumway, T., 2001. Forecasting bankruptcy more accurately: A simple hazard model. Journal of Business 74, 101 – 124. Simon, H. A., 1978. Rational decision-making in business organizations. Nobel Memorial Lecture, December 8, 1978. Skinner, D. J., and E. Soltes, 2011. What do dividends tell us about earnings quality? Review of Accounting Studies 16, 1 – 28. Tobin, J., 1969. A general equilibrium approach to monetary policy. Journal of Money, Credit and Banking 1, 15 – 29. Twu, M., 2010. Prior payment status and the likelihood to pay dividends: International evidence. The Financial Review 45, 785 – 802. Wang, M., M. Ke, D. Liu, and Y. Huang, 2011. Dividend policy and the life cycle hypothesis: Evidence from Taiwan. International Journal of Business and Finance Research 5, 33-52. Wansley, J. W., and W. R. Lane, 1987. A financial profile of the dividend initiating firm. Journal of Business Finance & Accounting 14, 425 – 436. Welch, I., 2010. A critique of quantitative structural models in corporate finance. Working Paper. Williams, J. B., 1938. The Theory of Investment Value. Harvard University Press.

166

TABLE 1 Summary Statistics for Sample, 1982-2010

N

Mean

Median

Std. Dev.

Variable CRSP Age-years Std Dev of Returns-% RE/TE Maturity Factor Maturity Composite NYSE Size Percentile

95996 95996 95996 95996 95996 95996

14.72 15.32% -1.6574 -5.45 1.53 27.02%

10.00 12.86% 0.3375 4.46 1.50 15.00%

14.23 10.51% 96.93 743.07 0.69 27.38%

Sales Growth Rate, % Earnings Growth Rate, % ROA-% ROE-%

95996 67660 95996 95996

52.32% -66.60% -2.99% -37.46%

9.37% 7.11% 3.31% 7.27%

3182.00% 2482.00% 27.24% 2983.00%

CA/TA RE/TA TE/TA

95996 95996 95996

0.1713 -0.2155 0.5238

0.0858 0.1496 0.5189

0.2043 1.805 0.2197

M/B Avg. Monthly Return, %

95996 95996

1.94 1.33%

1.37 1.14%

2.13 5.36%

Dividend Growth Rate, % Dividend Payout, % Dividend Yield, %

31420 95996 95996

19.83% 18.66% 0.93%

4.12% 0.00% 0.00%

622.00% 729.00% 6.78%

% of Firms in Sample by Economic Sector Materials Consumer Discretionary Consumer Staples Health Care Energy Industrials Technology Telecom

6.65% 23.35% 6.48% 13.56% 5.89% 20.08% 21.47% 1.55%

167

Table 2 Summary Statistics for Dividend Payers and Nonpayers

Dividend Payers

Nonpayers

20.0 8.99% 0.762 22.7 2.22 40.0%

7.0 15.30% 0.078 0.66 1.18 10.0%

126.3 *** -127.9 *** 153.3 *** 166.6 *** 163.9 *** 114.7 ***

Median Sales Growth Rate, % Median Earnings Growth Rate, % Median ROA-% Median ROE-%

7.18% 8.50% 5.67% 12.27%

11.50% 4.89% 1.53% 3.28%

-33.4 *** 7.1 *** 88.0 *** 96.2 ***

Median CA/TA Median RE/TA Median TE/TA

0.0552 0.3473 0.4811

0.1116 0.0353 0.5444

-48.8 *** 138.6 *** -31.2 ***

Median M/B Median Monthly Return, %

1.36 1.29%

1.38 1.01%

-3.5 *** 10.3 ***

94.68% 30978

3.22% 65018

Dividend Payers

Nonpayers

12.5% 25.7% 11.3% 6.3% 5.4% 27.2% 8.7% 1.9%

3.9% 22.2% 4.2% 17.0% 6.1% 16.7% 27.6% 1.4%

Variable Median CRSP Age-years Median Std Dev of Returns-% Median RE/TE Median Maturity Factor Median Maturity Composite Median NYSE Percentile

% Paid Prior Dividend N

% of Firms by Economic Sector

Materials Consumer Discretionary Consumer Staples Health Care Energy Industrials Technology Telecom

Statistical significance at the 1%, 5%, and 10% levels are indicated by ***, **, and *.

Z Value

168 Table 3 Summary Statistics for Dividend Payers by Maturity Factor Decile Maturity Factor Deciles Decile 4 Decile5 Decile6 Decile7 16.0 18.0 20.0 23.0 11.24% 10.13% 9.17% 8.32% 0.6857 0.7467 0.7988 0.8526 16.01 20.39 25.09 30.93 1.97 2.13 2.27 2.40 30.0% 35.0% 40.0% 45.0%

Variable Median CRSP Age-years Median Std Dev of Returns-% Median RE/TE Median Maturity Factor Median Maturity Composite Median NYSE Percentile

Decile1 8.0 12.49% 0.028 0.3904 1.30 25.0%

Decile2 9.0 12.57% 0.4198 6.57 1.56 25.0%

Decile3 13.0 12.01% 0.5778 11.46 1.79 30.0%

Median Sales Growth Rate Median Earnings Growth Rate Median ROA-% Median ROE-%

7.72% -3.42% 2.95% 7.77%

10.35% 8.70% 5.00% 10.92%

8.24% 8.85% 5.27% 11.54%

8.46% 10.45% 5.46% 11.69%

7.80% 8.81% 5.51% 11.75%

7.28% 8.28% 5.84% 12.28%

Median CA/TA Median RE/TA Median TE/TA

0.0502 0.0096 0.4053

0.0500 0.1925 0.4793

0.0557 0.2713 0.4859

0.0508 0.3242 0.4878

0.0535 0.3604 0.4920

% Paid Prior Dividend % Increased Dividend % Decreased Dividend % Dividend Initiators

80.53% 44.46% 15.18% 19.47%

89.70% 49.77% 11.52% 10.30%

93.35% 49.23% 10.72% 6.65%

95.29% 50.94% 9.72% 4.71%

Median Dividend Growth Rate Median Sustainable Growth Median Dividend Payout Median Dividend Yield

1.82% 3.41% 15.09% 1.55%

7.22% 7.73% 17.98% 1.42%

5.85% 8.29% 19.68% 1.58%

1.31 1.17%

1.31 1.42%

3097

3098

Median M/B Median Monthly Return N

Decile8 26.0 7.28% 0.8836 38.46 2.52 55.0%

Decile9 Decile10 31.0 38.0 6.28% 4.99% 0.9224 1.125 50.19 80.2 2.65 2.80 65.0% 80.0%

6.43% 8.53% 5.98% 12.41%

6.82% 10.08% 6.08% 12.97%

6.09% 8.74% 6.32% 13.42%

5.46% 8.27% 7.38% 16.65%

0.0566 0.3879 0.4998

0.0567 0.4191 0.5046

0.0571 0.4337 0.5000

0.0605 0.4626 0.4898

0.0601 0.5467 0.4458

96.38% 52.66% 7.23% 3.62%

97.00% 54.68% 8.42% 3.00%

97.93% 55.97% 7.36% 2.07%

98.32% 59.26% 6.29% 1.68%

99.06% 63.91% 5.10% 0.94%

99.19% 71.01% 5.04% 0.81%

6.06% 8.29% 21.35% 1.70%

5.56% 8.09% 24.21% 1.91%

5.62% 8.36% 26.32% 2.02%

5.43% 8.25% 29.36% 2.20%

5.55% 8.53% 31.39% 2.29%

5.37% 8.42% 34.41% 2.36%

5.13% 10.34% 38.13% 2.34%

1.31 1.42%

1.29 1.44%

1.29 1.41%

1.32 1.35%

1.34 1.31%

1.39 1.25%

1.45 1.30%

1.68 1.03%

3098

3098

3097

3098

3098

3098

3098

3098

169 Table 4 Summary Statistics for Nonpayers by Maturity Factor Decile

Variable Median CRSP Age-years Median Std Dev of Returns-% Median RE/TE Median Maturity Factor Median Maturity Composite Median NYSE Percentile

Decile1 9.0 17.57% -7.11 -75.4 0.8 5.0%

Decile2 8.0 17.81% -2 -21.2 0.84 5.0%

Decile3 6.0 18.52% -0.8969 -8.17 0.83 5.0%

Maturity Factor Deciles Decile 4 Decile5 Decile6 Decile7 5.0 4.0 4.0 6.0 19.15% 18.98% 16.76% 14.97% -0.3478 -0.0153 0.1868 0.3246 -2.72 -0.12 1.48 3.52 0.82 0.87 1.06 1.29 5.0% 10.0% 10.0% 10.0%

Decile8 9.0 13.66% 0.45098 6.51 1.53 15.0%

Decile9 Decile10 12.0 17.0 11.88% 8.64% 0.5934 0.7876 11.52 23.45 1.81 2.21 15.0% 15.0%

Median Sales Growth Rate Median Earnings Growth Rate Median ROA-% Median ROE-%

2.86% -126.00% -24.59% -76.06%

6.26% -73.50% -13.22% -29.17%

10.40% -61.41% -7.76% -16.34%

13.35% -71.38% -4.26% -8.96%

20.59% -29.22% 1.00% 1.94%

21.35% 17.22% 4.24% 8.23%

15.90% 16.65% 4.60% 9.51%

12.22% 14.98% 4.50% 9.24%

9.75% 12.90% 4.68% 9.69%

7.75% 10.71% 5.08% 10.50%

0.2147 -2.4800 0.3851

0.1629 -0.9380 0.5268

0.1404 -0.4426 0.5566

0.1313 -0.1811 0.5921

0.1250 -0.0069 0.6121

0.0974 0.0982 0.5775

0.0847 0.1685 0.5536

0.0821 0.2310 0.5465

0.0788 0.3045 0.5397

0.0744 0.4025 0.5279

0.91%

1.23%

1.48%

1.94%

4.49%

5.17%

3.91%

3.29%

4.20%

5.55%

1.82 -0.38%

1.56 0.62%

1.40 0.84%

1.37 1.04%

1.37 1.26%

1.38 1.44%

1.35 1.50%

1.27 1.27%

1.26 1.13%

1.26 1.08%

6501

6502

6502

6502

6502

6502

6502

6502

6502

6502

Median CA/TA Median RE/TA Median TE/TA % Paid Prior Dividend Median M/B Median Monthly Return N

170

Table 5 Valuation,Volatility, and Returns of Dividend Payers and Nonpayers for 1982-2010

EW M/B Ratio Dividend NonYEAR Payers Payers

t

Median Monthly Return Dividend NonPayers Payers Z

1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

1.34 1.50 1.36 1.51 1.54 1.45 1.53 1.60 1.37 1.66 1.71 1.81 1.65 1.78 1.87 2.05 1.98 1.84 1.73 1.73 1.59 1.85 1.99 1.98 2.00 1.86 1.40 1.70 1.88

2.01 9.6 *** 2.19 6.7 *** 1.62 6.0 *** 1.86 7.0 *** 1.80 4.6 *** 1.54 2.1 ** 1.67 3.0 *** 1.80 3.9 *** 1.50 3.1 *** 2.24 6.4 *** 2.12 7.2 *** 2.25 6.9 *** 1.94 6.8 *** 2.39 9.7 *** 2.28 7.2 *** 2.27 4.1 *** 2.13 2.4 ** 3.13 11.7 *** 2.01 4.1 *** 2.03 5.5 *** 1.55 -0.9 2.37 8.6 *** 2.39 6.8 *** 2.31 5.7 *** 2.28 5.2 *** 2.17 5.5 *** 1.37 -0.7 1.76 1.4 2.03 2.7 ***

2.50% 2.62% 0.05% 2.32% 1.23% 0.61% 1.73% 1.43% -0.86% 2.36% 1.22% 1.16% 0.11% 1.82% 1.56% 2.20% 0.25% -0.12% 0.60% 1.36% 0.04% 2.62% 1.69% 0.56% 1.33% 0.27% -2.76% 3.26% 2.17%

Mean

1.70

2.03 4.30 ***

1.15%

1.72% 3.8 *** 2.76% 0.9 -1.87% 14.6 *** 1.37% 7.7 *** 0.23% 7.1 *** -0.28% 7.2 *** 1.64% 0.8 0.89% 3.3 *** -1.70% 5.0 *** 3.65% -7.8 *** 1.45% -1.7 * 1.55% -3.1 *** -0.37% 5.4 *** 2.07% -1.8 * 1.31% 1.6 1.40% 6.5 *** -0.20% 3.1 *** 1.92% -10.7 *** -0.82% 7.9 *** 2.21% -6.3 *** -1.12% 9.2 *** 4.75% -15.1 *** 1.51% 1.6 0.33% 1.3 1.10% 2.6 *** -0.30% 3.9 *** -4.72% 9.7 *** 4.72% -8.4 *** 2.37% -1.2 0.95%

0.5

Median Std. Dev. of Returns Dividend NonPayers Payers Z 10.23% 9.42% 8.43% 8.20% 8.95% 12.87% 7.94% 7.27% 9.62% 9.27% 8.08% 7.91% 7.31% 7.32% 8.02% 8.59% 10.93% 11.14% 12.15% 10.52% 9.44% 8.14% 7.17% 7.49% 7.33% 7.43% 13.13% 12.35% 9.28%

15.38% 16.01% 13.04% 14.02% 14.10% 17.91% 13.40% 13.17% 15.82% 16.76% 15.54% 13.93% 12.77% 13.87% 15.22% 15.47% 17.92% 18.54% 21.81% 21.99% 17.72% 14.98% 12.83% 11.98% 11.70% 11.08% 17.46% 18.08% 12.95%

-21.1 -27.2 -23.4 -28.9 -26.2 -25.2 -26.1 -27.4 -25.5 -28.6 -28.8 -28.2 -26.8 -29.0 -28.6 -29.5 -25.4 -24.8 -25.3 -25.6 -24.0 -23.1 -24.4 -20.4 -18.9 -18.1 -13.2 -13.8 -15.8

*** *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** ***

9.17% 15.36% -10.1 ***

Statistical significance at the 1%, 5%, and 10% levels are indicated by ***, **, and *.

171 Table 6 Valuation, Volatility and Returns by Maturity Factor Deciles for 1982-2010

Maturity Factor Decile 1 EW EW EW

Maturity Factor Decile 5 EW EW EW

Maturity Factor Decile 10 EW EW EW

YEAR

Monthly Std. Dev. M/B Return Returns

Monthly Std. Dev. M/B Return Returns

Monthly Std. Dev. M/B Return Returns

1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

2.52 3.05 2.20 2.43 2.42 2.01 2.36 2.33 1.95 3.51 3.06 3.20 2.75 3.58 3.01 3.24 2.73 4.91 2.52 3.08 2.17 3.62 3.59 3.00 2.97 2.55 1.72 2.37 2.89

0.96% 2.76% -3.05% -0.16% -0.75% -1.74% -0.30% -0.26% -2.60% 4.42% 1.07% 1.22% -1.76% 2.12% -0.15% -0.07% -1.42% 4.33% -4.30% 2.36% -3.83% 7.78% 0.24% -1.59% 0.46% -2.27% -6.41% 6.52% 1.77%

20.56% 22.25% 16.29% 19.30% 19.75% 21.22% 17.85% 19.09% 20.45% 23.29% 19.88% 18.37% 15.99% 18.56% 18.30% 19.39% 22.46% 25.48% 28.17% 30.80% 24.54% 23.91% 17.14% 15.18% 15.79% 14.64% 20.49% 24.77% 16.20%

1.49 1.63 1.47 1.58 1.58 1.42 1.58 1.62 1.31 1.75 1.87 1.95 1.77 2.15 2.29 2.19 2.07 2.94 2.43 1.97 1.39 2.03 2.26 2.11 2.11 2.05 1.18 1.44 1.85

3.00% 3.32% -1.08% 2.49% 0.59% 0.36% 2.41% 1.47% -1.35% 4.80% 2.33% 2.38% -0.02% 2.55% 2.53% 2.27% 0.65% 2.97% 1.95% 4.29% -0.26% 6.04% 3.13% 1.61% 2.21% 0.40% -4.49% 5.95% 2.86%

13.98% 14.18% 12.19% 12.97% 12.96% 17.32% 13.04% 12.56% 15.80% 17.40% 16.37% 14.95% 13.22% 14.82% 16.55% 17.03% 20.49% 21.11% 28.44% 27.44% 21.80% 16.94% 16.48% 13.71% 13.38% 12.43% 20.33% 23.29% 13.97%

1.16 1.31 1.27 1.44 1.49 1.42 1.55 1.65 1.50 1.72 1.76 1.79 1.67 1.79 1.82 2.04 2.00 1.90 1.81 1.85 1.85 2.01 2.07 2.04 2.13 2.06 1.17 1.93 2.08

2.00% 2.22% 0.52% 2.26% 1.15% 0.48% 1.41% 1.31% -0.48% 2.00% 0.92% 0.98% 0.20% 1.70% 1.30% 2.00% 0.25% -0.35% 0.45% 1.21% 0.23% 2.23% 1.35% 0.26% 1.14% 0.28% -2.25% 2.49% 1.84%

6.55% 6.17% 5.69% 5.69% 6.70% 10.10% 5.58% 5.31% 6.84% 6.43% 5.53% 5.52% 5.41% 5.28% 5.87% 6.62% 8.53% 8.72% 9.41% 8.29% 7.66% 6.44% 5.28% 5.50% 5.34% 5.62% 10.49% 9.37% 6.97%

Mean

2.82

0.18%

20.35%

1.84

1.91%

16.73%

1.73

1.00%

6.79%

172 Table 7 Summary Statistics for Dividend Growers and Dividend Cutters

Growers Z-Value 1

Cutters Z-Value 2

All Payers

Variable Median CRSP Age-years Median Std Dev of Returns-% Median RE/TE Median Maturity Factor Median Maturity Composite Median NYSE Percentile

20.0 8.45% 0.7949 25.28 2.28 50.0%

0.9 -11.9 9.1 11.1 9.2 15.8

*** *** *** *** ***

18.0 11.63% 0.6068 13.12 1.91 15.0%

-1.6 15.7 -10.1 -12.8 -13.6 -18.5

*** *** *** *** ***

20.0 8.99% 0.762 22.7 2.22 40.0%

Median Sales Growth Rate Median Earnings Growth Rate Median ROA-% Median ROE-%

8.47% 11.16% 7.13% 14.78%

10.7 8.2 28.1 29.5

*** *** *** ***

-0.84% -37.62% 0.86% 2.12%

-19.8 -19.4 -27.9 -28.2

*** *** *** ***

7.18% 8.50% 5.67% 12.27%

Median CA/TA Median RE/TA Median TE/TA

0.0602 0.3747 0.4965

4.9 *** 10.1 *** 6.4 ***

0.0505 0.2386 0.4325

-2.9 *** -10.2 *** -6.7 ***

0.0552 0.3473 0.4811

Median M/B Median Monthly Return

1.53 1.39%

22.1 *** 4.1 ***

1.05 0.74%

-22.7 *** -8.7 ***

1.36 1.29%

13.13% 27.74% 1.92% 17097

50.5 *** 4.6 *** -2.1 ***

-50.29% 0.00% 1.22% 3993

-44.1 *** -8.8 *** 8.5 ***

5.41% 26.52% 1.96% 30978

Median Dividend Growth Rate Median Dividend Payout Median Dividend Yield N

% of Firms by Economic Sector

Materials Consumer Discretionary Consumer Staples Health Care Energy Industrials Technology Telecom

Growers

Cutters

All Payers

12.33% 25.35% 13.28% 7.21% 4.21% 26.97% 7.33% 2.28%

12.95% 28.02% 7.71% 4.86% 6.31% 27.67% 9.74% 1.33%

12.47% 25.72% 11.30% 6.32% 5.38% 27.21% 8.69% 1.87%

Statistical significance at the 1%, 5%, and 10% levels are indicated by ***, **, and *. 1. Median Two Sample Test Z value is between growers and all payers. 2. Median Two Sample Test Z value is between cutters and all payers.

Note: Dividend Cutters includes dividend omitters.

173 Table 8 Summary Statistics by Economic Sector

Variable Median CRSP Age-years Median Std Dev of Returns-% Median RE/TE Median Maturity Factor Median Maturity Composite Median NYSE Percentile

Consumer Consumer Materials Discretionary Staples 16.0 10.0 14.0 10.44% 12.21% 9.81% 0.6181 0.4603 0.6102 13.75 6.91 14.68 1.96 1.62 1.96 25.0% 15.0% 20.0%

Economic Sector Health Care Energy Industrials Technology Telecom 8.0 10.0 13.0 8.0 8.0 14.99% 13.09% 11.43% 16.00% 11.92% -0.1805 0.1924 0.5168 0.1248 0.1135 -1.56 2.29 9.64 1.05 1.61 1.1 1.41 1.77 1.18 1.4 10.0% 15.0% 15.0% 10.0% 30.0%

Median Sales Growth Rate Median Earnings Growth Rate Median ROA-% Median ROE-%

6.14% 3.32% 3.84% 9.12%

8.62% 7.33% 3.67% 8.51%

7.27% 9.61% 4.83% 11.43%

14.45% 11.99% 0.99% 1.87%

10.61% -3.16% 1.95% 4.39%

8.37% 7.63% 4.01% 9.11%

11.21% 4.59% 2.53% 4.39%

12.41% 4.94% 2.07% 5.43%

Median CA/TA Median RE/TA Median TE/TA

0.0359 0.2469 0.4492

0.0531 0.1886 0.4730

0.0486 0.2466 0.4438

0.2223 -0.1033 0.6377

0.0474 0.0874 0.4821

0.0594 0.2135 0.4759

0.2298 0.0724 0.6405

0.0498 0.0407 0.3743

60.96% 60.48%

36.39% 35.54%

56.15% 56.28%

14.89% 15.03%

30.12% 29.48%

44.37% 43.72%

13.40% 13.06%

39.43% 39.03%

Median Dividend Growth Rate Median Dividend Payout-% Median Dividend Yield

3.83% 8.13% 1.09%

3.23% 0.00% 0.00%

6.96% 7.85% 0.70%

7.89% 0.00% 0.00%

1.55% 0.00% 0.00%

4.09% 0.00% 0.00%

1.83% 0.00% 0.00%

5.79% 0.00% 0.00%

Median M/B Median Monthly Return

1.24 1.16%

1.23 1.03%

1.4 1.32%

2.02 1.11%

1.21 1.10%

1.26 1.15%

1.61 1.17%

1.4 1.52%

6389

22412

6217

13021

5651

19278

20614

1486

% Paid Prior Dividend % Paid Dividend

N

174

Table 9 Economic Sector Composition for Dividend Payers by Maturity Factor Decile

Variable Median Maturity Factor

Decile1 0.3904

Decile2 6.57

Maturity Factor Deciles Decile3 Decile 4 Decile5 Decile6 Decile7 11.46 16.01 20.39 25.09 30.93

Economic Sector Percentage Materials Consumer Discretionary Consumer Staples Health Care Energy Industrials Technology Telecommunication

13.24% 24.44% 9.01% 7.14% 9.49% 21.70% 11.17% 2.62%

10.36% 25.60% 8.80% 6.75% 6.00% 26.34% 12.07% 2.87%

11.94% 26.50% 7.91% 7.04% 5.55% 25.55% 12.23% 2.16%

11.33% 27.40% 9.42% 4.52% 4.94% 28.05% 10.78% 2.78%

10.62% 27.45% 10.01% 4.81% 4.97% 28.83% 10.24% 2.10%

12.46% 28.41% 10.04% 4.71% 4.71% 28.34% 8.30% 1.84%

14.04% 26.18% 10.75% 5.97% 3.62% 29.31% 7.30% 1.74%

Decile8 38.46

Decile9 Decile10 50.19 80.2

15.07% 24.69% 10.65% 6.75% 4.62% 29.50% 6.20% 1.32%

13.88% 23.60% 15.62% 6.92% 4.26% 28.41% 5.62% 0.77%

11.78% 22.89% 20.76% 8.59% 5.62% 26.11% 3.00% 0.52%

175

Table 10 Economic Sector Composition for Nonpayers by Maturity Factor Decile

Variable Median Maturity Factor

Decile1 -75.4

Decile2 -21.2

Maturity Factor Deciles Decile3 Decile 4 Decile5 Decile6 Decile7 -8.17 -2.72 -0.12 1.48 3.52

Economic Sector Percentage Materials Consumer Discretionary Consumer Staples Health Care Energy Industrials Technology Telecommunication

3.38% 13.34% 2.75% 30.81% 5.15% 10.66% 31.06% 2.31%

3.68% 15.84% 3.25% 27.75% 6.51% 12.38% 28.19% 1.97%

4.01% 17.52% 3.55% 23.44% 7.06% 14.16% 27.73% 1.83%

3.81% 18.72% 3.72% 19.62% 7.78% 13.58% 30.08% 2.11%

3.68% 22.58% 4.26% 14.63% 6.87% 13.92% 31.33% 1.88%

3.32% 24.50% 4.15% 12.41% 7.00% 17.54% 29.07% 0.80%

3.45% 26.15% 4.21% 11.73% 5.71% 18.39% 28.87% 0.55%

Decile8 6.51

Decile9 Decile10 11.52 23.45

3.72% 26.48% 4.48% 10.49% 5.71% 19.46% 27.58% 0.80%

4.24% 27.73% 4.78% 9.35% 5.57% 22.09% 24.52% 0.74%

5.54% 29.33% 6.66% 9.94% 3.94% 24.68% 17.23% 0.95%

176 Table 11 Logit Analysis of the decision to pay dividends-Fama and MacBeth approach.

Variable Intercept RE/TE

Model 1 -1.5869 (-21.89) 2.2733 (12.39)

Ln(Age)

Mean Parameter Estimate (t-Value in parentheses) Model 2 Model 3 Model 4 Model 5 Model 6 *** -4.3653 *** 2.0978 *** -3.681 *** -5.8906 *** -1.6675 *** (-33.99) (16.12) (-42.63) (-41.04) (-31.10) *** 1.4654 *** (66.98)

Std Dev Returns

-23.029 *** (-22.29)

Maturity Composite

3.009 *** (62.67)

Maturity Factor

0.0872 *** (14.35)

Prior Dividend

6.3879 *** (95.68)

NYSE Percentile SGR ROA M/B CA/TA TE/TA Model Fit Statistics

% Correct Log Likelihood Psuedo R2 Predicted % Dividend Payers

Model 1

Model 2

Model 3

Model 4

Model 5

Model 6

79.08 -1550

75.89 -1612.5

74.72 -1607.3

96.10 -520.65

81.54 -1249

81.49 -1405.4

0.2484

0.2226

0.2211

0.6000

0.3748

0.3110

25.95

24.49

26.09

32.73

30.16

23.63

Statistical significance at the 1%, 5%, and 10% levels are indicated by ***, **, and *.

177 Table 12 Logit Analysis of the decision to pay dividends-Fama and MacBeth approach.

Mean Parameter Estimate (t-Value in parentheses) Variable Model 7 Model 8 Model 9 Model 10 Model 11 Model 12 Intercept -0.9856 *** -1.4107 *** -3.6506 *** 0.775 *** -5.0248 *** -1.4131 *** (-13.31) (-17.16) (-29.83) (9.48) (-30.91) (-17.75) RE/TE 1.4131 *** (10.72) Ln(Age) 1.0006 *** (42.66) Std Dev Returns -13.319 *** (-20.36) Maturity Composite 2.3921 *** (58.21) Maturity Factor 0.0571 *** (13.23) Prior Dividend NYSE Percentile SGR ROA M/B CA/TA TE/TA Economic Sector Materials Cons. Discretionary Consumer Staples Health Care Energy Technology Telecom

4.1989 (21.29) -1.3055 (-13.75) 0.0756 (21.65) -0.3956 (-10.75) -1.5746 (-20.17) 0.6793 (7.26)

***

0.4494 (8.72) -0.4176 (-12.76) 0.298 (10.67) -1.0191 (-24.96) -0.8388 (-17.28) -1.4441 (-43.38) -0.5052 (-6.85)

***

*** *** *** *** ***

*** *** *** *** *** ***

3.5592 (20.50) -1.0889 (-11.12) 0.0424 (12.68) -0.282 (-10.14) -1.1618 (-14.22) 0.2422 (2.30)

***

0.5208 (8.80) -0.3986 (-14.75) 0.3045 (9.94) -0.8838 (-18.32) -0.5142 (-8.31) -1.2627 (-42.13) -0.1319 (-1.99)

***

*** *** *** *** **

*** *** *** *** *** *

3.3579 (20.56) -0.8779 (-9.66) 0.0727 (19.21) -0.2537 (-8.32) -1.4512 (-17.71) 0.758 (6.49)

***

0.4829 (7.22) -0.2676 (-9.48) 0.3767 (13.05) -0.9746 (-23.48) -0.659 (-15.32) -1.3282 (-53.86) -0.0813 (-0.81)

***

*** *** *** *** ***

*** *** *** *** ***

3.3956 (18.48) -1.1273 (-10.83) 0.063 (17.53) -0.2959 (-10.08) -1.3418 (-16.12) 0.3692 (3.68)

***

0.4808 (8.82) -0.3436 (-11.04) 0.2308 (7.17) -1.0129 (-19.14) -0.6723 (-12.35) -1.1543 (-29.42) -0.506 (-6.33)

***

*** *** *** *** ***

*** *** *** *** *** ***

2.4034 (14.49) -0.5761 (-6.39) 0.0388 (10.75) -0.1191 (-5.36) -1.0038 (-9.54) 0.0315 (.22)

***

0.5856 (7.41) -0.2396 (-9.33) 0.2962 (7.96) -0.7543 (-13.51) -0.2713 (-5.73) -0.9605 (-35.06) 0.0836 (0.82)

***

*** *** *** ***

*** *** *** *** ***

3.0409 (20.26) -0.9257 (-9.59) 0.0466 (14.44) -0.2079 (-9.64) -1.2514 (-13.57) 0.0082 (.08)

***

0.5322 (7.88) -0.3367 (-13.07) 0.2347 (7.57) -0.8433 (-16.34) -0.3855 (-6.62) -1.0815 (-38.2) -0.1198 (-1.35)

***

*** *** *** ***

*** *** *** *** ***

178 Table 12 (continued) Logit Analysis of the decision to pay dividends-Fama and MacBeth approach.

Model 7

Model 8

Model 9

Model 10

Model 11

Model 12

79.66 -1362.3

82.00 -1259.8

81.51 -1243.5

81.24 -1276.5

83.37 -1111

83.46 -1198.9

0.3326

0.3708

0.3786

0.3651

0.4257

0.393

25.78

27.61

27.46

28.00

30.17

27.28

Model Fit Statistics % Correct Log Likelihood Psuedo R2 Predicted % Dividend Payers

Statistical significance at the 1%, 5%, and 10% levels are indicated by ***, **, and *.

179 Table 13 Logit Analysis of the decision to pay dividends-Fama and MacBeth approach.

Variable Intercept RE/TE

Model 13 -2.0696 (-20.31) 0.9007 (8.89)

Mean Parameter Estimate (t-Value in parentheses) Model 14 Model 15 Model 16 Model 17 Model 18 *** -3.0336 *** -5.0324 *** -5.2384 *** -2.9299 *** -6.3337 *** (-26.31) (-32.06) (-32.74) (-27.43) (-30.12) ***

RE/TE Percentile Ln(Age)

0.030 *** 0.010 *** 0.008 *** 0.0277 *** (67.44) (11.07) (10.22) (47.51) 0.8134 *** (29.97)

Age Percentile Std Dev Returns

0.022 *** (33.31)

0.0021 ** (2.09)

0.0216 *** (32.73)

-9.8519 *** (-20.17)

Std. Dev. Percentile

-0.020 *** (-24.62)

Maturity Composite Maturity Factor

0.0021 ** -0.0192 *** (2.09) (-24.04) 1.9988 *** 2.2049 *** 1.4629 *** (24.62) (33.31) (22.32) 0.004 *** (4.47)

Prior Dividend NYSE Percentile SGR ROA M/B CA/TA TE/TA

2.4877 (16.83) -0.6976 (-7.50) 0.0424 (12.18) -0.1233 (-5.25) -1.0343 (-11.83) 0.1134 (0.96)

*** *** *** *** ***

2.4569 (15.53) -0.569 (-6.41) 0.0344 (10.10) -0.1226 (-5.49) -0.971 (-8.89) -0.0015 (-0.01)

*** *** *** *** ***

2.4569 (15.53) -0.569 (-6.41) 0.0344 (10.10) -0.1226 (-5.49) -0.971 (-8.89) -0.0015 (-0.01)

*** *** *** *** ***

2.4569 (15.53) -0.569 (-6.41) 0.0344 (10.10) -0.1226 (-5.49) -0.971 (-8.89) -0.0015 (-0.01)

*** *** *** *** ***

2.4491 (15.51) -0.5771 (-6.53) 0.0345 (10.16) -0.1194 (-5.39) -0.9758 (-8.96) -0.0279 (-0.20)

*** *** *** *** ***

5.7056 (68.93) 1.6685 (14.10) -0.3375 (-3.90) 0.0520 (8.81) -0.0832 (-3.65) -0.1671 (-0.92) 0.3403 (2.15)

*** *** *** *** ***

**

180

Table 13 (continued) Logit Analysis of the decision to pay dividends-Fama and MacBeth approach.

Variable

Model 13

Economic Sector Materials Cons. Discretionary Consumer Staples Health Care Energy Technology Telecom

0.5609 (7.60) -0.2291 (-8.59) 0.315 (9.53) -0.7848 (-14.96) -0.3534 (-7.38) -1.031 (-36.84) 0.0788 (0.78)

Model 14

*** *** *** *** *** ***

0.5844 (7.33) -0.2603 (-10.89) 0.3059 (8.88) -0.7235 (-14.00) -0.2571 (-5.50) -0.9593 (-38.59) 0.1564 (1.50)

Model 15

*** *** *** *** *** ***

0.5844 (7.33) -0.2603 (-10.89) 0.3059 (8.88) -0.7235 (-14.00) -0.2571 (-5.50) -0.9593 (-38.59) 0.1564 (1.50)

Model 16

*** *** *** *** *** ***

0.5844 (7.33) -0.2603 (-10.89) 0.3059 (8.88) -0.7235 (-14.00) -0.2571 (-5.50) -0.9593 (-38.59) 0.1564 (1.50)

Model 17

*** *** *** *** *** ***

0.5873 (7.34) -0.2596 (-10.90) 0.3017 (8.80) -0.725 (-14.05) -0.2541 (-5.44) -0.9558 (-38.72) 0.1556 (1.48)

Model 18

*** *** *** *** *** ***

0.3292 (2.98) -0.1681 (-2.45) 0.2892 (3.31) -0.3874 (-5.09) 0.0804 (-0.89) -0.6261 (-10.6) 0.0524 (0.38)

Model Fit Statistics % Correct Log Likelihood Psuedo R2 Predicted % Dividend Payers

83.59 -1141.05

83.28 -1105.3

83.28 -1105.3

83.28 -1105.3

83.35 -1104.2

96.17 -415.2

0.4146

0.4277

0.4277

0.4277

0.4280

0.6244

29.01

30.28

30.28

30.28

30.20

32.41

Statistical significance at the 1%, 5%, and 10% levels are indicated by ***, **, and *.

*** *** *** ***

***

181

Table 14 Average Partial Effects of Maturity Components

Average Partial Effect on the probability to pay a dividend for a US Industrial based on a change in each maturity component from the 50th percentile to the 75th percentile.

Variable RE/TE Percentile Age Percentile Std. Dev. Percentile

Average Partial Effect

t Value

11.03%

583.29 ***

7.48%

506 ***

-6.26%

-460.28 ***

Statistical significance at the 1%, 5%, and 10% levels are indicated by ***, **, and *.

182 Table 15 Logit Analysis of the decision to pay dividends-Panel logistic method.

Variable Intercept RE/TE Ln(Age) Std Dev Returns Maturity Composite Maturity Factor Prior Dividend

Parameter Estimates (Z-Value in parentheses) Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 -1.5512 *** -4.0413 *** 1.7873 *** -3.6418 *** -5.6334 *** -1.6252 *** (-33.14) (-52.89) (35.21) (-135.25) (-74.46) (-48.91) 2.2752 *** (28.22) 1.3418 *** (45.36) -19.893 *** (-55.52) 2.8575 *** (66.16) 0.0823 *** (33.88) 6.2827 *** (142.70)

NYSE Percentile SGR ROA M/B CA/TA TE/TA Model Fit Statistics

% Correct Predicted % Dividend Payers

Model 1

Model 2

Model 3

Model 4

Model 5

Model 6

79.16

75.92

74.75

96.10

81.57

81.28

27.14

22.87

25.24

32.73

29.54

22.89

Statistical significance at the 1%, 5%, and 10% levels are indicated by ***, **, and *.

183 Table 16 Logit Analysis of the decision to pay dividends-Panel logistic method.

Parameter Estimates (Z-Value in parentheses) Model 7 Model 8 Model 9 Model 10 Model 11 Model 12 -0.8924 *** -1.2402 *** -3.167 *** 0.6451 *** -4.4927 *** -1.2312 *** (-10.95) (-13.41) (-27.82) (6.72) (-37.61) (-14.39) RE/TE 1.3771 *** (16.64) Ln(Age) 0.8857 *** (28.30) Std Dev Returns -11.321 *** (-33.30) Maturity Composite 2.1662 *** (42.58) Maturity Factor 0.0525 *** (20.78) Prior Dividend Variable Intercept

NYSE Percentile SGR ROA M/B CA/TA TE/TA Economic Sector Materials Cons. Discretionary Consumer Staples Health Care Energy Technology Telecom

3.7212 (33.01) -1.1777 (-16.73) 0.0726 (30.23) -0.3667 (-13.91) -1.5088 (-8.44) 0.6479 (5.04)

***

0.4482 (4.50) -0.4072 (-5.58) 0.335 (3.23) -1.2161 (-10.66) -0.8683 (-7.18) -1.4534 (-13.56) -0.4691 (-2.16)

***

*** *** *** *** ***

*** *** *** *** *** **

3.0782 (26.68) -0.8753 (-10.62) 0.0397 (14.41) -0.2686 (-10.64) -1.027 (-5.67) 0.1489 (1.04)

***

0.5392 (5.23) -0.3925 (-5.39) 0.3379 (3.14) -0.962 (-8.29) -0.5702 (-4.69) -1.2343 (-11.65) 0.2124 (0.07)

***

*** *** *** ***

*** *** *** *** ***

2.9133 (24.73) -0.6657 (-8.97) 0.0694 (28.09) -0.2622 (-10.67) -1.4401 (-8.12) 0.6973 (5.30)

***

0.4799 (4.64) -0.2739 (-3.73) 0.4073 (3.76) -1.1339 (-9.91) -0.738 (-5.77) -1.3587 (-12.76) -0.0871 (-0.38)

***

*** *** *** *** ***

*** *** *** *** ***

3.0903 (27.13) -1.0088 (-12.07) 0.059 (24.08) -0.2904 (-11.20) -1.2885 (-7.37) 0.3889 (3.00)

***

0.4797 (4.81) -0.3531 (-4.84) 0.2733 (2.64) -1.1602 (-10.04) -0.7263 (-6.08) -1.2032 (-11.48) -0.4469 (-2.03)

***

*** *** *** *** ***

*** *** *** *** *** **

2.0336 (16.95) -0.2735 (-2.28) 0.0377 (14.88) -0.1492 (-6.54) -0.8847 (-4.95) 0.0389 (0.28)

***

2.6241 (21.53) ** -0.7309 (-9.59) *** 0.0441 (18.51) *** -0.215 (-8.97) *** -1.1391 (-6.32) -0.0461 (-0.33)

***

0.581 (5.41) -0.2535 (-3.38) 0.3423 (2.99) -0.9499 (-7.90) -0.3857 (-3.05) -1.0025 (-9.57) 0.1115 (0.49)

***

***

*** *** *** *** ***

0.5484 (5.28) -0.3281 (-4.51) 0.2772 (2.51) -0.9769 (-8.24) -0.4782 (-3.94) -1.0795 (-10.42) 0.0281 (0.13)

*** *** *** ***

*** *** *** *** ***

184

Table 16 (continued) Logit Analysis of the decision to pay dividends-Panel logistic method. Model 7

Model 8

Model 9

Model 10

Model 11

Model 12

79.82

82.24

81.64

81.37

83.46

83.54

25.34

27.7

27.31

27.77

29.58

26.59

Model Fit Statistics % Correct Predicted % Dividend Payers

Statistical significance at the 1%, 5%, and 10% levels are indicated by ***, **, and *.

185 Table 17 Logit Analysis of the decision to pay dividends-Panel logistic method.

Parameter Estimates (Z-Value in parentheses) Variable Model 13 Model 14 Model 15 Model 16 Model 17 Model 18 Intercept -1.6832 *** -2.8365 *** -4.4935 *** -5.0156 *** -2.6681 *** -6.0478 *** (-13.42) (-23.07) (-37.22) (-26.23) (-19.62) (-60.51) RE/TE 0.8474 *** (11.42) RE/TE Percentile 0.0267 *** 0.0101 *** 0.0049 *** 0.0228 *** (23.64) (7.51) (2.93) (12.79) Ln(Age) 0.7052 *** (22.31) Age Percentile 0.0218 *** 0.0052 *** 0.0210 *** (21.18) (4.23) (19.63) Std Dev Returns -8.6349 *** (-26.05) Std. Dev. Percentile -0.0166 *** 0.0052 *** -0.0153 *** (-23.75) (4.23) (-18.33) Maturity Composite 1.657 *** 2.1791 *** 1.3918 *** (23.75) (21.18) (34.40) Maturity Factor 0.0063 *** (2.49) Prior Dividend

NYSE Percentile SGR ROA M/B CA/TA TE/TA

5.51 *** (125.16) 2.1357 *** (17.74) -0.4795 *** (-6.13) 0.0382 *** (13.81) -0.1457 *** (-6.16) -0.944 *** (-5.19) 0.0678 (0.47)

2.0879 (17.27) -0.2621 (-2.21) 0.0353 (13.71) -0.1555 (-6.74) -0.8713 (-4.83) 0.0186 (0.13)

***

2.0879 *** (17.27) ** -0.2621 ** (-2.21) *** 0.0353 *** (13.71) *** -0.1555 *** (-6.74) *** -0.8713 *** (-4.83) 0.0186 (0.13)

2.0879 *** (17.27) -0.2621 ** (-2.21) 0.0353 *** (13.71) -0.1555 *** (-6.74) -0.8713 *** (-4.83) 0.0186 (0.13)

2.0759 *** (17.08) -0.2757 ** (-2.29) 0.0351 *** (13.48) -0.1506 *** (-6.52) -0.8706 *** (-4.80) -0.0135 (-0.09)

1.416 *** (16.04) -0.0191 (-0.77) 0.0452 *** (13.71) -0.0827 *** (-4.25) 0.0282 (0.19) 0.2724 ** (2.34)

186

Table 17 (continued) Logit Analysis of the decision to pay dividends-Panel logistic method.

Variable

Model 13

Economic Sector Materials Cons. Discretionary Consumer Staples Health Care Energy Technology Telecom

Model 14

0.5682 *** (5.35) -0.2511 *** (-3.41) 0.3469 *** (3.10) -0.9345 *** (-7.90) -0.4464 *** (-3.52) -1.0416 *** (-9.98) 0.169 (0.71)

Model 15

0.5854 *** (5.37) -0.2665 *** (-3.53) 0.3611 *** (3.12) -0.9183 *** (-7.60) -0.3692 *** (-2.87) -1.0229 *** (-9.69) 0.1844 (0.81)

Model 16

0.5854 *** (5.37) -0.2665 *** (-3.53) 0.3611 *** (3.12) -0.9183 *** (-7.60) -0.3692 *** (-2.87) -1.0229 *** (-9.69) 0.1844 (0.81)

Model 17

Model 18

0.5854 *** 0.5908 *** 0.342 *** (5.37) (5.42) (4.14) -0.2665 *** -0.2646 *** -0.1695 *** (-3.53) (-3.51) (-3.14) 0.3611 *** 0.3543 *** 0.3323 *** (3.12) (3.04) (3.99) -0.9183 *** -0.9187 *** -0.4512 *** (-7.60) (-7.58) (-5.39) -0.3692 *** -0.3662 *** -0.1641 * (-2.87) (-2.85) (-1.85) -1.0229 *** -1.016 *** -0.6035 *** (-9.69) (-9.63) (-8.83) 0.1844 0.1942 0.0911 (0.81) (0.85) (0.60)

Model Fit Statistics % Correct Predicted % Dividend Payers

83.69

83.46

83.46

83.46

83.55

96.16

29.47

29.67

29.67

29.67

29.39

32.44

Statistical significance at the 1%, 5%, and 10% levels are indicated by ***, **, and *.

187 Table 18 Analysis of Maturity Model Predictions Model 11 > 90% Probability of paying dividend

< 10% Probability of paying a dividend

Z

Variable Median CRSP Age-years Median Std Dev of Returns-% Median RE/TE Median Maturity Factor Median Maturity Composite Median NYSE Percentile

37.0 6.70% 0.9193 48.25 2.66 85.0%

5.0 18.86% -0.4859 -3.52 0.85 5.0%

89.6 -88.1 88.4 88.4 88.4 91.1

*** *** *** *** *** ***

Median Sales Growth Rate Median Earnings Growth Rate Median ROA-% Median ROE-%

6.53% 10.49% 6.88% 16.00%

13.51% -29.01% -4.32% -8.65%

-33.2 34.2 84.6 84.3

*** *** *** ***

Median CA/TA Median RE/TA Median TE/TA

0.0401 0.4134 0.4329

0.1787 -0.2271 0.5810

-62.1 *** 88.4 *** -45.2 ***

Median M/B Median Monthly Return

1.59 1.26%

1.52 0.84%

6.3 *** 11.2 ***

6.53% 35.34% 2.28% 96.64% 96.86% 6634

0.00% 0.00% 0.00% 4.74% 96.41% 37054

-25.0 *** 6332.3 *** 183.1 ***

> 90% probability 24.10% 16.51% 21.62% 3.41% 5.25% 24.79% 0.71% 2.38%

< 10% probability 1.80% 16.79% 2.30% 23.83% 6.04% 10.62% 36.69% 1.39%

Median Dividend Growth Rate Median Dividend Payout Median Dividend Yield % Paid Prior Dividend % Correct N % of Firms by Economic Sector

Materials Consumer Discretionary Consumer Staples Health Care Energy Industrials Technology Telecom

Statistical significance at the 1%, 5%, and 10% levels are indicated by ***, **, and *.

188

Table 19 Summary of Outlier Analysis

Non-payers that fit the model

Payers that should not pay

Number of Observations, N Model 11 Model 12 Model 13

58368 59846 58532

9228 10625 9170

6650 5172 6486

21750 20353 21808

Median Maturity Factor, M Model 11 Model 12 Model 13

-0.08 0.07 -0.05

8.29 9.69 8.37

20.22 22.45 20.26

30.44 31.97 30.25

40988.4 *** 37853.9 *** 40255.4 ***

1.39 1.38 1.39

1.26 1.24 1.25

1.32 1.39 1.34

1.4 1.42 1.41

322.0 *** 478.4 *** 348.6 ***

Median NYSE Percentile Model 11 Model 12 Model 13

10.0% 10.0% 10.0%

15.0% 15.0% 15.0%

35.0% 45.0% 35.0%

55.0% 60.0% 55.0%

19675.6 *** 21528.4 *** 20495.4 ***

Median Profitability, ROA% Model 11 Model 12 Model 13

0.82% 0.96% 0.85%

4.40% 4.36% 4.37%

5.30% 5.66% 5.25%

6.04% 6.16% 6.04%

10820.9 *** 10895.9 *** 10692.6 ***

Median Monthly Return, % Model 11 Model 12 Model 13

0.97% 0.97% 0.96%

1.35% 1.25% 1.23%

1.18% 1.25% 1.26%

1.27% 1.30% 1.30%

114.90 *** 137.80 *** 145.40 ***

Variable

Median Market to Book ratio, M/B Model 11 Model 12 Model 13

Non-payers Payers that that should fit the model pay

Statistical significance at the 1%, 5%, and 10% levels are indicated by ***, **, and *.

ChiSquare

189 Table 20 Residual Analysis of Dividend Paying Firms-Model 11 Model 11 Outliers

Fit Model

Z

Variable Median CRSP Age-years Median Std Dev of Returns-% Median RE/TE Median Maturity Factor Median Maturity Composite Median NYSE Percentile

11.0 12.21% 0.5119 8.29 1.63 15.0%

24.5 7.97% 0.8389 30.44 2.41 55.0%

-77.6 67.3 67.3 -100.9 -105.6 -76.9

*** *** *** *** *** ***

Median Sales Growth Rate Median Earnings Growth Rate Median ROA-% Median ROE-%

8.73% 3.91% 4.40% 9.49%

6.81% 9.38% 6.04% 13.19%

8.2 -6.4 -20.6 -27.6

*** *** *** ***

Median CA/TA Median RE/TA Median TE/TA

0.0734 0.2321 0.5057

0.0503 0.3898 0.4741

13.5 *** -43.0 *** 8.6 ***

Median M/B Median Monthly Return

1.26 1.35%

1.4 1.27%

-16.1 *** 1.5

Median Dividend Growth Rate Median Dividend Payout Median Dividend Yield

3.92% 17.46% 1.59%

5.67% 29.68% 2.10%

-4.8 *** -31.4 *** -19.1 ***

% Dividend Cutters % Paid Prior Dividend % Dividend Initiators Median Predicted Probability N

13.32% 87.53% 12.47% 28.34% 9228

6.68% 97.71% 2.29% 82.34% 21750

Outliers 4.93% 28.36% 5.78% 9.15% 6.25% 23.16% 20.03% 1.26%

Fit Model 15.67% 24.59% 13.64% 5.12% 5.01% 28.93% 3.88% 2.13%

% of Firms by Economic Sector Materials Consumer Discretionary Consumer Staples Health Care Energy Industrials Technology Telecom

Statistical significance at the 1%, 5%, and 10% levels are indicated by ***, **, and *.

190 Table 21 Residual Analysis of Dividend Paying Firms-Model 12 Model 12 Outliers

Fit Model

Z

Variable Median CRSP Age-years Median Std Dev of Returns-% Median RE/TE Median Maturity Factor Median Maturity Composite Median NYSE Percentile

12.0 11.69% 0.5310 9.69 1.74 15.0%

24.0 7.86% 0.8477 31.97 2.43 60.0%

-66.6 66.0 -68.8 -103.5 -99.9 -90.7

*** *** *** *** *** ***

Median Sales Growth Rate Median Earnings Growth Rate Median ROA-% Median ROE-%

8.53% 4.14% 4.36% 9.36%

6.77% 9.53% 6.16% 13.48%

8.2 -7.3 -24.5 -33.3

*** *** *** ***

Median CA/TA Median RE/TA Median TE/TA

0.0711 0.2440 0.5099

0.0498 0.3934 0.4707

13.6 *** -41.6 *** 11.8 ***

Median M/B Median Monthly Return

1.24 1.25%

1.42 1.30%

-21.6 *** -1.4

Median Dividend Growth Rate Median Dividend Payout Median Dividend Yield

3.39% 18.33% 1.67%

5.92% 29.90% 2.09%

-8.9 *** -30.7 *** -16.3 ***

% Dividend Cutters % Paid Prior Dividend % Dividend Initiators Median Predicted Probability N

13.19% 88.75% 11.25% 30.69% 10625

6.29% 97.77% 2.23% 82.57% 20353

Outliers 4.39% 30.17% 5.68% 8.92% 5.85% 24.23% 18.56% 1.29%

Fit Model 16.70% 23.39% 14.23% 4.96% 5.13% 28.77% 3.54% 2.18%

% of Firms by Economic Sector Materials Consumer Discretionary Consumer Staples Health Care Energy Industrials Technology Telecom

Statistical significance at the 1%, 5%, and 10% levels are indicated by ***, **, and *.

191 Table 22 Residual Analysis of Dividend Paying Firms-Model 13 Model 13 Outliers

Fit Model

Z

Variable Median CRSP Age-years Median Std Dev of Returns-% Median RE/TE Median Maturity Factor Median Maturity Composite Median NYSE Percentile

10.0 12.13% 0.5320 8.37 1.65 15.0%

25.0 8.03% 0.8328 30.25 2.40 55.0%

-80.1 62.7 59.6 96.0 96.4 78.9

*** *** *** *** *** ***

Median Sales Growth Rate Median Earnings Growth Rate Median ROA-% Median ROE-%

8.77% 3.38% 4.37% 9.34%

6.81% 9.48% 6.04% 13.22%

8.3 -7.4 -20.7 -28.3

*** *** *** ***

Median CA/TA Median RE/TA Median TE/TA

0.0736 0.2440 0.5106

0.0503 0.3838 0.4725

13.8 *** -36.7 *** 10.9 ***

Median M/B Median Monthly Return

1.25 1.23%

1.41 1.30%

-17.9 *** 1.6

Median Dividend Growth Rate Median Dividend Payout Median Dividend Yield

3.26% 17.18% 1.60%

5.83% 29.65% 2.10%

-7.9 *** -31.5 *** -18.7 ***

% Dividend Cutters % Paid Prior Dividend % Dividend Initiators Median Predicted Probability N

13.26% 87.74% 12.26% 30.67% 9170

6.72% 97.59% 2.41% 80.14% 21808

Outliers 4.36% 28.59% 5.10% 9.67% 6.32% 22.54% 21.12% 1.22%

Fit Model 15.88% 24.50% 13.90% 4.91% 4.98% 29.17% 3.46% 2.15%

% of Firms by Economic Sector Materials Consumer Discretionary Consumer Staples Health Care Energy Industrials Technology Telecom

Statistical significance at the 1%, 5%, and 10% levels are indicated by ***, **, and *.

192 Table 23 Residual Analysis of Nonpaying Firms-Model 11

Model 11 Outliers

Fit Model

Z

Variable Median CRSP Age-years Median Std Dev of Returns-% Median RE/TE Median Maturity Factor Median Maturity Composite Median NYSE Percentile

18.0 9.11% 0.7108 20.22 2.19 35.0%

7.0 16.13% -0.0106 -0.08 1.11 10.0%

70.4 -71.4 77.6 79.3 86.0 49.8

*** *** *** *** *** ***

Median Sales Growth Rate Median Earnings Growth Rate Median ROA-% Median ROE-%

8.49% 14.58% 5.30% 11.50%

12.22% 0.87% 0.82% 1.76%

-15.9 16.0 54.2 56.1

*** *** *** ***

Median CA/TA Median RE/TA Median TE/TA

0.0559 0.3304 0.4963

0.1220 -0.0045 0.5526

-29.0 *** 76.6 *** -16.3 ***

Median M/B Median Monthly Return

1.32 1.18%

1.39 0.97%

-7.4 *** 4.0 ***

65.13% 5.43% 6650

6.17% 2.96% 58368

Outliers 10.74% 27.55% 9.76% 6.33% 4.86% 28.41% 8.78% 1.59%

Fit Model 3.10% 21.61% 3.54% 18.23% 6.27% 15.35% 29.70% 1.37%

Median Predicted Probability % Paid Prior Dividend N

% of Firms by Economic Sector Materials Consumer Discretionary Consumer Staples Health Care Energy Industrials Technology Telecom

Statistical significance at the 1%, 5%, and 10% levels are indicated by ***, **, and *.

193 Table 24 Residual Analysis of Nonpaying Firms-Model 12

Model 12 Outliers

Fit Model

Z

Variable Median CRSP Age-years Median Std Dev of Returns-% Median RE/TE Median Maturity Factor Median Maturity Composite Median NYSE Percentile

16.0 8.92% 0.7406 22.45 2.18 45.0%

7.0 15.92% 0.0092 0.07 1.13 10.0%

52.7 -60.2 70.1 70.8 71.7 53.0

*** *** *** *** *** ***

Median Sales Growth Rate Median Earnings Growth Rate Median ROA-% Median ROE-%

8.35% 15.05% 5.66% 12.39%

12.12% 1.57% 0.96% 2.08%

-15.2 15.5 50.5 52.7

*** *** *** ***

Median CA/TA Median RE/TA Median TE/TA

0.0559 0.3398 0.4842

0.1195 0.0040 0.5517

-26.3 *** 69.2 *** -16.9 ***

Median M/B Median Monthly Return

1.39 1.25%

1.38 0.97%

0.7 5.7 ***

63.81% 5.22% 5172

9.79% 3.04% 59846

Outliers 0.1295 0.2753 0.1036 0.0611 0.0425 0.2647 0.0829 0.0174

Fit Model 0.031 0.2176 0.0365 0.1796 0.0629 0.1584 0.2923 0.0136

Median Predicted Probability % Paid Prior Dividend N

% of Firms by Economic Sector Materials Consumer Discretionary Consumer Staples Health Care Energy Industrials Technology Telecom

Statistical significance at the 1%, 5%, and 10% levels are indicated by ***, **, and *.

194 Table 25 Residual Analysis of Nonpaying Firms-Model 13

Model 13 Outliers

Fit Model

Z

Variable Median CRSP Age-years Median Std Dev of Returns-% Median RE/TE Median Maturity Factor Median Maturity Composite Median NYSE Percentile

18.0 9.15% 0.6935 20.26 2.16 35.0%

7.0 16.13% -0.0069 -0.05 1.11 10.0%

72.9 -72.5 75.1 77.1 84.4 52.0

*** *** *** *** *** ***

Median Sales Growth Rate Median Earnings Growth Rate Median ROA-% Median ROE-%

8.30% 14.75% 5.25% 11.45%

12.27% 1.08% 0.85% 1.83%

-17.0 15.7 52.4 54.6

*** *** *** ***

Median CA/TA Median RE/TA Median TE/TA

0.0588 0.3190 0.4903

0.1213 -0.0031 0.5536

-27.4 *** 73.9 *** -17.9 ***

Median M/B Median Monthly Return

1.34 1.26%

1.39 0.96%

-4.6 *** 6.3 ***

63.40% 5.66% 6486

6.45% 2.95% 58532

Outliers 0.1175 0.264 0.1041 0.0572 0.0476 0.2857 0.0848 0.0179

Fit Model 0.0301 0.2176 0.0349 0.1827 0.0628 0.1537 0.2968 0.0135

Median Predicted Probability % Paid Prior Dividend N

% of Firms by Economic Sector Materials Consumer Discretionary Consumer Staples Health Care Energy Industrials Technology Telecom

Statistical significance at the 1%, 5%, and 10% levels are indicated by ***, **, and *.

195 Table 26 Residual Analysis of Dividend Paying Firms-Model 11, 1982 vs. 2000 1982 Outliers Fit Model Variable Median CRSP Age-years Median Std Dev of Returns-% Median RE/TE Median Maturity Factor Median Maturity Composite Median NYSE Percentile

11.0 0.1308 0.6555 11.24 1.58 10.0%

19.0 0.0893 0.8471 26.42 2.25 50.0%

Median Sales Growth Rate Median Earnings Growth Rate Median ROA-% Median ROE-%

0.0351 -0.0833 0.0476 0.1028

0.0107 -0.0835 0.0596 0.1247

2000 Outliers Fit Model

Z -16.1 18.2 -16.7 -28.3 -28.9 -19.1

Z

*** *** *** *** *** ***

8.50 0.1822 0.3055 3.47 1.65 20.0%

29.00 -12 *** 0.1147 9.3 *** 0.8581 -9.8 *** 22.88 -14.1 *** 2.47 -14.3 *** 50.0% -7.1 ***

2.8 *** 0.0 -3.9 *** -3.5 ***

0.1036 -0.0242 0.0293 0.0658

0.0765 0.0697 0.0609 0.1459

1.7 * -0.7 -5.5 *** -6.6 ***

Median CA/TA Median RE/TA Median TE/TA

0.0573 0.2926 0.4993

0.0563 0.2 0.4212 -11.7 *** 0.5163 -1.4

0.0423 0.1264 0.4605

0.0282 0.372 0.4222

2.4 ** -8.1 *** 1.5

Median M/B Median Monthly Return

1.12 2.73%

1.10 2.36%

1.11 0.39%

1.32 0.62%

2.0 ** -0.5

0.00% 9.73% 1.45%

1.11% 25.25% 1.95%

18.18% 84.09% 15.91% 31.27% 176

7.44% 98.46% 1.54% 83.74% 712

0.4 1.7 *

Median Dividend Growth Rate Median Dividend Payout Median Dividend Yield

5.26% 19.54% 1.98%

7.14% -1.3 35.42% -10.5 *** 3.76% -11.9 ***

% Dividend Cutters % Paid Prior Dividend % Dividend Initiators Median Predicted Probability N

16.74% 94.52% 5.48% 26.63% 657

11.06% 99.79% 0.21% 75.88% 949

% of Firms by Economic Sector Materials Consumer Discretionary Consumer Staples Health Care Energy Industrials Technology Telecom

Outliers Fit Model 4.57% 16.12% 32.12% 23.92% 6.70% 18.23% 5.78% 3.37% 6.70% 5.06% 27.70% 27.40% 14.76% 2.95% 0.76% 1.48%

Outliers Fit Model 10.23% 14.61% 19.89% 26.12% 4.55% 12.64% 9.09% 5.48% 10.23% 4.35% 17.61% 29.63% 25.00% 4.21% 1.14% 1.83%

Statistical significance at the 1%, 5%, and 10% levels are indicated by ***, **, and *.

-1.1 -4.4 *** -1.4

196 Table 27 Residual Analysis of Non-paying Firms-Model 11, 1982 vs. 2000

1982 Outliers Fit Model Variable Median CRSP Age-years Median Std Dev of Returns-% Median RE/TE Median Maturity Factor Median Maturity Composite Median NYSE Percentile

15.0 8.97% 0.8613 24.26 2.16 10.0%

5.0 15.94% 0.2564 2.01 0.92 5.0%

Median Sales Growth Rate Median Earnings Growth Rate Median ROA-% Median ROE-%

-0.21% 1.17% 4.15% 8.29%

Median CA/TA Median RE/TA Median TE/TA Median M/B Median Monthly Return Median Predicted Probability % Paid Prior Dividend N

2000 Outliers Fit Model

Z 8.8 -7.9 8.8 8.8 9.0 3.9

*** *** *** *** *** ***

16.0 12.34% 0.6956 13.79 2.19 30.0%

6.0 23.87% -0.1354 -0.72 1.14 10.0%

8.32% -14.58% 1.62% 3.44%

-2.5 ** 0.8 2.7 *** 3.2 ***

10.76% 12.53% 5.27% 12.53%

18.12% -6.10 *** -7.34% 4.30 *** -0.96% 17.30 *** -2.16% 17.30 ***

0.0704 0.3646 0.4452

0.0575 0.1142 0.4829

0.8 7.4 *** -1.1

0.0333 0.3051 0.4799

0.1294 -10.80 *** -0.0681 21.00 *** 0.5663 -5.50 ***

0.96 1.50%

1.24 1.76%

-5.8 *** -0.8

1.19 0.82%

1.16 -1.23%

60.75% 11.69%

5.08% 6.41%

65.93% 5.91%

6.43% 2.11%

77

1202

423

2557

% of Firms by Economic Sector Materials Consumer Discretionary Consumer Staples Health Care Energy Industrials Technology Telecom

Z

Outliers Fit Model 23.38% 5.41% 31.17% 23.04% 15.58% 5.41% 0.00% 7.90% 3.90% 13.98% 16.88% 20.80% 3.90% 21.13% 2.60% 0.67%

Outliers Fit Model 8.04% 1.96% 31.91% 20.49% 9.46% 3.01% 5.44% 18.15% 4.26% 4.65% 30.50% 12.63% 7.80% 36.33% 0.95% 2.15%

Statistical significance at the 1%, 5%, and 10% levels are indicated by ***, **, and *.

18.10 -20.20 21.00 21.00 22.20 8.90

*** *** *** *** *** ***

0.40 5.90 ***

197 Table 28 Logit Analysis of the decision to pay dividends-Panel logistic method.

Model 11 with Year Effects Variable Parameter Estimate Z Statistic Intercept -3.5044 -28.48 RE/TE Ln(Age) Std Dev Returns Maturity Composite 2.3595 43.39 Maturity Factor Prior Dividend NYSE Percentile 2.2265 17.06 SGR -0.2286 -1.93 ROA 0.0321 12.27 M/B -0.0996 -4.47 CA/TA -0.9353 -4.9 TE/TA 0.1023 0.7 Materials 0.5901 5.1 Consumer Discretionary -0.238 -3.04 Consumer Staples 0.3031 2.51 Health Care -0.8536 -6.83 Energy -0.3272 -2.48 Technology -0.964 -8.96 Telecom 0.1858 0.7 1983 -0.2276 -5.66 1984 -0.4796 -9.11 1985 -0.6395 -11.78 1986 -0.8128 -13.9 1987 -1.0811 -17.18 1988 -1.2127 -18.48 1989 -1.165 -17.38 1990 -1.1882 -17.52 1991 -1.2303 -17.79 1992 -1.268 -17.85 1993 -1.3769 -19.23 1994 -1.6455 -22.82 1995 -1.6882 -23.38 1996 -1.8509 -25.33 1997 -2.0346 -27.59 1998 -2.0469 -27.17 1999 -2.2121 -28.82 2000 -2.3446 -29.73 2001 -2.4402 -30.68 2002 -2.4686 -30.31 2003 -2.1337 -25.77

***

***

*** * *** *** *** *** *** ** *** ** *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** ***

198

Table 28 (continued) Logit Analysis of the decision to pay dividends-Panel logistic method.

Variable 2004 2005 2006 2007 2008 2009 2010

Model 11 with Year Effects Parameter Estimate Z Statistic -1.8467 -22.1 -1.6618 -19.96 -1.6098 -19.23 -1.61 -18.69 -1.6314 -18.21 -1.751 -19.53 -1.5984 -18.03

*** *** *** *** *** *** ***

Statistical significance at the 1%, 5%, and 10% levels are indicated by ***, **, and *.

Model Fit Statistics % Correct Predicted % Dividend Payers

84.62 29.79

199 Table 29 M/B Regressions for Dividend Payers as a Function of Maturity Factor The dependent variable is the M/B ratio. The data set includes only dividend paying firms from the sample. The only explanatory variable in the regression is the maturity factor.

RANGE OF MATURITY FACTOR, M 0-50% PERCENTILE 50-100% PERCENTILE

Variable

Intercept Maturity Factor, M

Intercept Maturity Factor, M

Intercept Maturity Factor, M

OVERALL

Parameter Estimates (t values in parentheses) Panel A. Fama and MacBeth Method 1.6948 *** 1.5882 *** (45.05) (45.25) -0.002 0.0027 *** (-1.87) * (5.38)

1.657 *** (50.49) 0.0012 ** (2.37)

Panel B. Random Effects Panel Regression 1.631 *** 1.6261 *** (39.14) (33.32) 0.0000 0.0004 *** (-0.07) (5.05)

1.6205 *** (38.10) 0.0002 ** (2.54)

Panel C. Fixed Effects Panel Regression 1.5798 *** 2.4671 *** (4.32) (5.61) 0.0000 0.0004 *** (0.18) (5.41)

1.5788 *** (4.57) 0.0002 *** (3.08)

Statistical significance at the 1%, 5%, and 10% levels are indicated by ***, **, and *.

200 Table 30 M/B Regressions for Non-payers as a Function of Maturity Factor The dependent variable is the M/B ratio. The data set includes only non-paying firms from the sample. The only explanatory variable in the regression is the maturity factor.

RANGE OF MATURITY FACTOR, M 0-50% PERCENTILE 50-100% PERCENTILE

Variable

Intercept Maturity Factor, M

Intercept Maturity Factor, M

Intercept Maturity Factor, M

OVERALL

Parameter Estimates (t values in parentheses) Panel A. Fama and MacBeth Method 2.3136 *** 1.8462 *** (26.42) (32.55) -0.0007 *** -0.0103 *** (-3.84) (-4.67)

Panel B. Random Effects Panel Regression 2.2691 *** 1.7867 *** (22.98) (34.94) 0.0000 -0.0051 *** (-0.98) (-8.73)

Panel C. Fixed Effects Panel Regression 2.5247 * 1.2436 (1.89) (1.50) 0.0000 -0.0067 *** (-0.37) (-10.83)

Statistical significance at the 1%, 5%, and 10% levels are indicated by ***, **, and *.

2.0221 *** (30.26) -0.0012 *** (-4.14)

201

Table 31 Summary Statistics for Dividend Payers by Maturity Factor Range

Variable Median CRSP Age-years Median Std Dev of Returns-% Median RE/TE Median Maturity Factor Median Maturity Composite Median NYSE Percentile

Maturity Factor Range 0-10 10-25 M > 25 Chi-Square 8.0 16.0 28.0 6469.3 *** 12.60% 10.63% 7.01% 8321.7 *** 0.3516 0.6982 0.9069 9442.3 *** 5.65 17.47 40.74 26664.2 *** 1.51 2.05 2.56 18612.8 *** 25.0% 30.0% 55.0% 2123.5 ***

Median Sales Growth Rate Median Earnings Growth Rate Median ROA-% Median ROE-%

10.55% 9.52% 4.79% 10.67%

7.98% 9.31% 5.45% 11.76%

6.28% 8.77% 6.42% 13.51%

270.6 *** 0.9 383.4 *** 363.1 ***

Median CA/TA Median RE/TA Median TE/TA

0.0513 0.1581 0.4749

0.0531 0.3324 0.4886

0.0583 0.4546 0.4884

23.0 *** 4909.8 *** 11.1 ***

Median M/B Median Monthly Return

1.32 1.45%

1.30 1.43%

1.44 1.20%

282.4 *** 40.5 ***

Median Dividend Growth Rate Median Dividend Payout-% Median Dividend Yield

6.34% 17.56% 1.38%

5.85% 22.81% 1.78%

5.36% 32.44% 2.28%

8.2 *** 1283.0 *** 872.5 ***

% Paid Prior Dividend % Increased Dividend % Decreased Dividend % Dividend Initiators

86.99% 49.60% 11.43% 13.01%

95.50% 51.52% 9.13% 4.50%

98.42% 61.77% 6.12% 1.58%

5381

10191

13998

N

Statistical significance at the 1%, 5%, and 10% levels are indicated by ***, **, and *.

202 Table 32 Summary Statistics for Nonpayers by Maturity Factor Range

Variable Median CRSP Age-years Median Std Dev of Returns-% Median RE/TE Median Maturity Factor Median Maturity Composite Median NYSE Percentile

Maturity Factor Range M Maturity Composite < 2

Maturity Composite > 2

276

Figure 34 Percentage of Dividend Initiators, 1982-2010 5.0%

4.5%

4.0%

Percentage

3.5%

3.0%

2.5%

2.0%

1.5%

1.0%

0.5%

0.0%

1980

1985

1990

1995

2000

2005

2010

2015

277

Figure 35 Empirical Hazard Function in Life-Cycle Time

The hazard function is the probability that a non-payer initiates a dividend. The time (X-axis) scale is the maturity composite (MC) score. Compustat S&P economic sectors are as follows: 1000=Materials; 2000=Consumer Discretionary; 3000=Consumer Staples; 4000=Energy; 6000 = Industrials; 8000=Technology.

278

Figure 36 Empirical Survival Function in Life-Cycle Time

The survival distribution is the percentage of non-payers that continue as non-dividend paying firms. The time (X-axis) scale is the maturity composite (MC) score. Compustat S&P economic sectors are as follows: 1000=Materials; 2000=Consumer Discretionary; 3000=Consumer Staples; 4000=Energy; 6000 = Industrials; 8000=Technology.

279

Figure 37 Empirical Hazard Function in Event Time

(Years)

The hazard function is the probability that a non-payer initiates a dividend. The time (X-axis) scale is the time to dividend initiation in years.

280

Figure 38 Empirical Survival Function in Event Time

(Years)

The survival distribution is the percentage of non-payers that continue as non-dividend paying firms. The time (X-axis) scale is the time to dividend initiation in years.

281

Figure 39 Growth Rates as a Function of Maturity Composite 40%

35%

Percentage Growth Rate

30%

25%

20%

Dividend Growth Sustainable Growth

15%

10%

5%

0% 0

0.5

1

1.5

Maturity Composite

2

2.5

3

282

Figure 40 Dividend Growth Rate and Sustainable Growth Rate for Wal-Mart Stores, 1982-2010 70%

Percentage Growth Rate

60%

50%

40% Dividend Growth Rate Sustainable Growth Rate

30%

Linear (Dividend Growth Rate) Linear (Sustainable Growth Rate)

20%

10%

0% 2

2.2

2.4 2.6 Maturity Composite

2.8

3