I EI.SEVIER

management,

and

Investment

opportunit:y

incentives* A. Scott Keatinga.*,

Jerold

L. Zimmermanb

aGraduate School of Business, University of Chicago, 1101 Eo 58th Street, Chicago, 1£ 60637, USA bWi//iam Eo Simon Graduate School of Business Administration, University of Rochester, Rochester, NY 14627, USA Received 29 March

1999; received in revised form 9 February 2000

Abstract

Contrary to previous studies, we find managers change depreciation policies in predictable ways. We identify three dimensions of depreciation-policy changes:whether it is a method change or an estimate revision; whether it is income-increasing or decreasing; and whether it applies to new assets only or both new and existing assets. This disaggregation leads to three findings: First, a 1981 tax law altered the frequency of estimate revisions and method changes. Second, firms adopting i1Jlcome-increasing method changes for all assetsexperience worse performance than those adopting such changes only for new assets. Finally, non-income-increasing policy changes are associated with changesin investment opportunities. (g) 2000 Elsevier ScienceB.V. All rights reserved. JEL

classification:

Keywords:

M41;

Contracting;

H25

Taxes;

Depreciation;

Policy

choice

*Financial support was provided by the John M. Olin Foundation and the Bradley JPolicy Research Center at the University of Rochester. We thank Ray Ball, Jim Brickl~y, John Core:,Dan Gode, S.P. Kothari, Andy Leone, Clifford Smith, Ross Watts (the editor), an anonymous referl:e and seminar participants at the University of Rochester for useful suggestionsand Michelle LOWTyfor research assistance. * Corresponding author. Tel.: + 1-773-834-4078;fax: + 1-773-702-0458. E-mail

address:

[email protected]

(A.S.

Keating).

I

0165-4101/00/$-see front matter @ 2000 Elsevier Science B.V. All rights reserved. PII: SOl 6 5 -4 1 O 1 ( 00) 0 00 04- 5

360

A.S. Keating,

J:L.

Zimmerman

/ Journal

of Accounting

and Economics

28 (2000)

359-389

Introduction Accounting depreciation affects firms' financial statements that are frequently used in contracts, disclosures to capital markets, internal decision-makin~~and control, and tax computations. Previous researchers examined just one dimension of accounting depreciation policy -the choice between straight-line and accelerated depreciation methods. We examine several dimensions: depreciation methods, asset life and salvage value estimates, and whether changes in these policies are applied only to new assetsor to both new and existing assets.We find that depreciation-policy changes are associated with changes in thc~tax treatment of depreciable assets,the firm's financial performance, and changes in the firm's investment opportunities. Previous researchers, notably Hagerman and Zmijewski (1979), Skinner (1993) and Bowen et al. (1995), find cross-sectional associations between depreciation methods and size, leverage, risk, and R & D, among others. However, these studies' explanatory power tends to be low. Holthausen (1981) and Sweeney (1994) find at best weak evidence of associations between changes in depreciation methods and hypothesized determinants of depreciation policies such as debt covenants and investment opportunities. The rather weak nature of these previous results is somewhat surprising since depreciation expense is one of the larger accruals over which managers exercise discretion. To increase the power of our tests, we study estimate revisions in addition to method changes,and also whether theseestimate revisions or method changesare applied to new assetsonly or instead to both new assetsand assetsalready in service.We document that policy changesapplied to both new and existing assets produce significantly larger earnings changesthan policy changesapplied only to new assets.We exploit this fact, along with differences between method changes and estimate revisions and differences between income-increasing, income..neutral, and income-decreasing changes,to document the following three findings: 1. A 1981 tax law changed the relative costs and benefits of estimate revisions versus method changes. Prior to 1981, if a firm estimated longer lives and higher salvage values (i.e. income-increasing estimates) for financial reporting than for taxes, the IRS could challenge the company's estimates for tax purposes and, if successful,increase the firm's tax liability. The 1981 tax law removed this indirect link between financial and tax reporting by mandating fixed depreciation schedulesfor taxes. Our evidence shows that the frequency of income-increasing estimate revisions increased following the 1981 tax law change, while the frequency of income-increasing method changes declined. This suggests that managers were reluctant to make estimate revisions prior to the 19811aw because such revisions might trigger higher taxes. 2. Firms making income-increasing method changes applied to both new and existing assets experience worse financial performance in the two ~lears

A.S. Keating, J:L. Zimmerman / Journal of Accounting and Economics 28 (2000) 359-389

361

preceding the change than firms making income-increasing method changes applied to new assetsonly. Poor financial performance adversely affects bond covenants, earnings-based compensation, and the likelihood of executive firings, and hence managers have incentives to change depreciation policies to offset poor performance. Depreciation method changes applied to both new and existing assetshave a larger earnings-impact than do method changes applied to new assets only since GAAP requires the cumulative effect of a retroactive method change be included in income in the year of the change. We predict that firms making income-increasing method changesfor all ~Lssets experience poorer performance than firms making income-increasing method changes applied to new assetsonly. The evidence supports this hypothesis. 3. Non-income-increasing depreciation-policy changes are in response to changes in investment opportunities. A firm's depreciation policies are part of its optimal organizational design given the operating environment. When operating environments change, optimal organizational design, including depreciation policies, likely change as well (Smith and Watts, 1992;Skinner, 1993; Brickley et al., 1997). For example, a firm experiencing an increase in investment opportunities can adopt a less-accelerated depreciation schedule to reduce the likelihood that managers forego positive NPV projects (Ball et al., 1999).Consistent with this hypothesis, we find that firms making non-incomeincreasing depreciation-policy changesexperiencesignificant changesin investment opportunities relative to firms not making such changes. As the above findings suggest, the process of disaggregating depreciationpolicy changes along their various dimensions allows us to draw stronger inferences about the determinants of such changes. Managers have discretion over all dimensions of accounting policies, not just whether or not to make a change. The choices managers make over each of these dimensions depend on their incentives. Thus, identifying the dimensions of a given accounting policy, and changes to that policy, allows richer, more powerful tests of the determinants of these policy changes. In Section 2, we describe the mechanics of depreciation-policy changes and managers' incentives to change those policies. In Section 3, the NAARS database is used to identify a sample of firms changing depreciation policies. Section 4 then presents our findings. Section 5 summarizes the paper's central points and discussesimplications for other studies of accounting method choice. 2. Depreciation rules and managers' incentives At the time a fixed asset is placed in service, managers establish its depreciation schedule by choosing a depreciation method -typically straight-line or accelerated -and estimating the asset's service life and salvage value. Once an asset is in service, managers can either change the depreciation method or revise

362

A.S. Keating,

J:L. Zimmerman

/ Journal

of Accounting

and Economics

28 (2000)

359-389

the estimated service life or salvagevalue of the asset.APB 20 requires that firms changing depreciation methods for assetsalready in service recalculate depreciation charges as if the new depreciation method had been used from the date the asset was originally placed in service. Prior financial statements are not restated, but the cumulative effect of the method change -the difference between what the depreciation expenseswould have been using the new method and what they were under the old method -must be included in earnings in the year of the method change. Thus, method changescan have a large impact on earnings in the year of the change. However, the magnitude of the effect of the method change depends on, among other things, the growth rate of the firm's depreciable assetbase and whether the method change applies only to new assets or to both new and existing assets.As discussedbelow, the empirical evidence indicates that method changes have on average a relatively small effect on depreciation expense and earnings, partly because many method changes are applied only to new assets. If managers revise life or salvage value estimates, there is no cumulative effect of the change. The remaining depreciable amount (gross book value less accumulated depreciation) is depreciated using the revised estimate of the asset's service life and/or the revised estimate of the asset's salvage value. While APB 20 requires disclosure of both changes in depreciation methods and revisions of depreciation estimates, the disclosure burden of the two types of changes differs. In their opinion letters auditors must identify a change in depreciation method as a change in accounting principles affecting the comparability of the financial statements to previous periods' financial statements. Moreover, Statement of Auditing Standards No.58 requires auditors to evaluate a change in depreciation methods and to be satisfied that management's reasons for adopting the new method are justified. Finally, the SEC requires auditors to submit to the SEC 'preferability letters' stating that the auditor is satisfied with management's justification for the depreciation method change. Auditors not satisfied with management's justifications can express a qualified opinion or, worse, an adverse opinion. These requirements likely reduce managers' incentives to change depreciation methods. By contrast auditors rarely evaluate or offer opinions on depreciation estimates or revisions of estimates and are not required to supply the SEC with any documentation concerning the estimate revision (O'Reilly et al., 1998, pp. 22-22).1 1In a random sample of 20 depreciation method changes from our total sample (described in Section 3) with a reported dollar effect of the change, 16 (80%) method changes had consistency qualifications in their audit reports. In all 16 cases, the auditor concurred in the consistency qualifications regarding the method change. In a random sample of 20 estimate revisions, only two (10%) revisions contained consistencyexceptions;the other 18 had clean audit reports. In every case where the auditor issued a consistency exception for a method change we found a preferability letter filed with the 10-K. We did not find any preferability letters on file when there was an estimate revision. Thus, depreciation method changes result in a higher frequency of consistency exceptions and preferability letters than depreciation estimate revisions.

A.S. Keating, J.L. Zimmerman / Journal of Accounting and Economics 28 (2000) 359-389

363

Depreciation method changes and estimate revisions are substitutes in the sensethat they provide managers with alternate ways of affecting the time series of depreciation expense. Managers can control the magnitude of the change on earnings for both method changes and estimate revisions by applying the change either to all existing assets,some subset of existing assets,or new assets only. However, for a given impact on earnings, method changes are likely more costly because they require auditors to issue qualified audit report opinions and SEC preferability letters.2 Thus, to achieve a given dollar change in depreciation expense, earnings, or book values, managers are more likely to prefer estimate revisions over method changes. As documented below, there is a greater incidence of estimate revisions than method changes, arguably due in part to the greater disclosure conditions placed on method changes. However, if management wants to achieve large changes in income statement or balance sheet numbers, method changes are the only way to accomplish this. Prior findings support a variety of reasons for earnings management (Watts and Zimmerman, 1986, 1990; Warner et al., 1988; Weisbach, 1988).A decline in accounting performance (a) increases the likelihood of technical violation of bond covenants, (b) lowers the expected payouts from executive compensation plans linked to earnings, and (c) increases the likelihood of management turnover. Declining accounting performance creates incentives to adopt incomeincreasing accounting changes, including depreciation method changes and estimate revisions. Managers seeking to offset large declines in accounting earnings will be inclined to use depreciation method changes applied to both new and existing assetsbecause such method changes have a larger impact on earnings than those applied to new assets only. Thus, we predict and report consistent evidence below that firms making income-increasing depreciation method changes applied to both new and existing assetshave significantly worse financial performance than those making income-increasing method changes for new assets only. Firms also change depreciation policy to better align managers' incentives with those of shareholders whenever the firm's operating environment changes (Skinner, 1993). In addition, firms change financial accounting depreciation policies if those policies affect tax depreciation and also if tax laws are revised. These internal incentive-based and tax-based motives for depreciation-policy changes are discussed in greater detail in Section 4.

2 In a private conversation with an auditor, the auditor indicated that consistency exceptions and preferability letters add to the time required for the engagement,but not significantly in most cases. Choi and Jeter (1992) find that earnings response coefficients are lower after auditors issue both 'subject to' and consistency qualifications. One interpretation of these findings is that the market's perception of earnings noise or earnings persistenceis altered by the qualification. If this is true, the firm bears a cost from the consistency exception.

364

A.S. Keating,

J.L. Zimmerman

I Journal

of Accounting

and Economics

28 (2000)

359-389

3. Sample selection The NAARS database contains annual reports for approximately 4000 companies each year. From 1972 through 1994,we identify firms changing depreciation methods or revising depreciation estimates by using a keyword search of the annual report footnotes, locating all instances of the word 'depreciation' that occur within 10 words of various combinations of words associated with accounting policy changes (i.e. 'change', 'changed', 'revise', 'revised', 'method', or 'estimate'). We then examine the full text of each keyword instance to confirm that the firm had in fact changed depreciation methods or revised depreciation estimates. This search yields 252 instances of depreciation method changes and 394 instances of depreciation estimate revisions -a total of 646 changes-representing 0.28% of the NAARS firm-years -made by 531 different firms for the period 1972 through 1994.3 As a test of the comprehensiveness of our sample, we compare our NAARS sample to Compustat firms whose footnote codes indicate a change in depreciation policies. Compustat footnote code # 5 specifies whether a firm changed depreciation policies. Additionally, Compustat footnote code # 15 specifies the type of depreciation method used; thus we can detect depreciation method changes by comparing a firm's current year's footnote code # 15 to its prior year footnote code # 15. Four hundred and fifty eight firm-years (0.3% of all Compustat firm-years between 1972 and 1993)have a footnote code # 5 indicating a change in depreciation policies, and 3364 firm-years (2.2%) have a change in footnote code # 15 indicating a change in depreciation methods.4 Somewhat surprisingly, there is a less than perfect overlap between the NAARS-listed firms we identify as making depreciation-related policy changes and the firms identified in Compustat as making depreciation-related policy changes. Of the 440 NAARS-listed firm-years we identify making depreciationrelated policy changes and which are also in Compustat, only 95 (22%) have a Compustat footnote code # 5 indicating a change in depreciation-related 3 A matched control sample is not used because of the nature of NAARS. NAARS, an on-line word-searchable database, contains large, primarily NYSE and AMEX firms (Mutchler and Shane, 1995).Companies included one year need not be included the next year. The result is a sample biased towards larger, more successfulfirms. NAARS does not provide annually a list of firms in its files. Therefore, there is no easy way to select a control sample without searching NAARS by company name to determine if a firm is present. If a matched sample is drawn from Compustat, it is difficult to find 'control' firms that are as large and as profitable as the NAARS firms. 4 Changes in footnote code # 15 overstate the frequency of method changes since a firm's depreciation method footnote code in a given year can differ from its depreciation method footnote code in the previous year for reasons other than changesin depreciation methods. For example, an acquisition of a firm using the straight-Iine method by a firm using the accelerated method with neither firm changing depreciation methods will result in a change in the combined firm's method footnote.

A.S. Keating, J.L. Zimmerman / Journal of Accounting and Economics 28 (2000) 359-389

365

policies in the same year. Moreover, only 74 (42%) of the 175 NAARS-listed firm-years changing depreciation methods and which are also listed in Compustat have a change in Compustat footnote code # 15 indicating a method change in the same year. Hence, less than half of our NAARS-detected depreciationpolicy changes are identified in Compustat footnote codes. We next examine the extent to which Compustat-coded policy changes are discussed in the financial statements of NAARS-listed firms. Twenty firm-years with Compustat-detected method changes and that are also listed in the NAARS database were randomly selected.We search the NAARS-listed annual report in the year of and prior to the date of the method change detected by Compustat. Of these 20 firms, nine (45%) show no evidence in their annual report of any change in depreciation methods, which thus must reasonably be considered as coding errors by Compustat. The remaining 11 firms (55%) appear to change methods since the method used in one year differs from that used in the prior year. In all of these 11 cases,there is no discussion of the method change in either the annual report or the auditor's letter. Of these 11 changes, only one is from a single method to a different single method, suggesting that the change was applied to all assets. In the absence of a complete disposition of existing assets(depreciated under the old method) and acquisition of all new assets(depreciated under the new method), we are at a loss to explain how a firm changes depreciation method for all assetswithout noting the policy change in its financial statement footnotes. The remaining ten changes were either from a mix of methods to one single method or vice versa. These latter ten changes could be due to mergers, assetdispositions, or method changes applied to new assetsonly, hence not requiring mention in the financial statements. We cannot explain why Compustat fails to detect depreciation-policy changes accurately, but these findings suggest that researchersshould be cautious when using Compustat footnote codes. The higher frequency of method changesin the Compustat sample than in the NAARS sample highlights the sample selection bias underlying our results. Firms choose not to disclose in their annual report footnotes depreciationpolicy changes (and hence are not in our NAARS-based study) for at least four reasons. First, GAAP does not require firms to report the effect of an accounting policy change unless the change has a material effect on earnings. Second, some firms change depreciation policies -say from all straight-line to a mix of accelerated and straight-line -due to mergers or acquisitions. Third, asset acquisitions or dispositions may result in changes in depreciation methods. For example, consider a firm using both the straight-line and units-of-production methods. If it disposes of the assets being depreciated using the units-ofproduction method, then in the year following the disposition of those assetsit reports using only the straight-line method. Fourth, a firm changing depreciation policies for new assets only, and not for assetsalready in service, will not necessarily report the policy changes and their impact because the policy

366

A.S. Keating,

J:L. Zimmerman

/ Journal

of Accounting

and Economics

28 (2000)

359-389

changes do not affect earnings (except in the sensethat earnings might have been different if the firm had applied the same policies to new assetsthat it applied to assets already in service). Thus, our sample of depreciation-policy changes consists of firms where the change is material and hence not representative of all firms making such policy changes. However, this selection bias should increase the power of our NAARS-based tests relative to a sample drawn using Compustat footnote codes. We exclude firms taking fixed-asset write-downs, which is arguably another type of depreciation policy, for two reasons. First, fixed-asset write-downs are often not mentioned explicitly but rather included as part of restructurings or plant shut-downs, and thus we are unable to use a keyword search of the NAARS database to identify all, or even most, instances of fixed-asset writedowns. Second, many write-downs include write-downs of inventories, receivables, and other asset accounts in addition to depreciable assets,and thus it is not possible to discern the dollar impact of the fixed-asset write-down alone.

4. Findings In this section we report that depreciation changes are relatively rare events; estimate revisions have a larger impact on earnings than do method changes;the time series of the frequency of estimate revisions and method changes was altered by a 1981 tax law change; firms making income-increasing method changes applied to both new and existing assets experience worse financial performance than firms making income-increasing method changes for new assetsonly; and non-income-increasing changes applied only to new assetsare associated with changes in firms' investment opportunities. 4.1.

Frequency

of depreciation

changes

Panel A ofTable 1 shows the distribution ofdepreciation-policy changes and the time series of those changes. (We defer discussion of the time series to the next sub-section.) NAARS-reported depreciation changes are relatively infrequent, occurring in only 0.71% of the firm-years (Column 3). If managers exercise discretion over reported earnings, they do so infrequently. Panel B reports the frequency of income-increasing, income-neutral, and income-decreasing depreciation-policy changes. Income-increasing policy changes (Column 3) are the most frequent in the sample, occurring in 0.35% of all NAARS firmyears followed by income-decreasing changes (0.27%) in Column 1 and incomeneutral changes (0.08%) in Column 2. Panel A of Table 2 reports the incidences of firms making multiple changes. Of the 531 firms in the sample, 444 (69%) make only one depreciation change in the 23-year period, 69 firms (21%) make two changes, 12 (6%) make three

A.S. Keating, J.L. Zimmerman / Journal of Accounting and Economics 28 (2000) 359-389

367

changes and six (4% ) make four or more changes. Panels Band C of Table 2 report the incidences of firms making multiple changes of the same type. The vast majority of firms (over 80% in both cases)make only one such change. Only 20 of the 232 firms adopting new depreciation methods do so more than once (Panel B). Forty three of the 338 firms revising depreciation estimates do so more than once (Panel C). The low rate of repeated changes of one type suggests that managers are not switching back and forth to manipulate earnings. The numbers we report on repeat depreciation-policy changes are likely biased downwards from the frequency of such changes in the population. This is again becausethe NAARS sample does not consist of a constant set offirms acrosstime. Table 3 reports the incidence offirms making more than one type of depreciation-related change. Again, the vast majority (93%) of firms makes only one type of change. The remaining 7% change a method and revise an estimate. Of the 39 firms making both changes,the mean (median) time between changes is 2.07 (0) years. Tables 2 and 3 suggest that among those firms changing depreciation policies, few exercise discretion over more than one depreciation policy. 4.2. Time series of depreciation changes Panel A of Table 1 shows the distribution of both method changes and estimate revisions across time (Columns 1 and 2). The highest yearly incidence of a depreciation method change in any year (Column 1) is 0.53% of the population (1972 and 1980), and the highest incidence of a depreciation estimate revision (Column 2) is 1.05% (1994). Method changes decline in frequency over the period 1972-1994, while estimate changes increase. However, the combined frequency of both policy changes (Column 3) remains relatively constant over time.5 Table 1, Panel B, reports the time series of the frequency of incomeincreasing, income-neutral, and income-decreasing depreciation-policy changes. Income-decreasing (Column 1) and income-increasing (Column 3) policy changes show no time trend, while income-neutral changes (Column 2) decline in frequency over time. As Panel A of Table 1 shows, the relatively constant frequency of depreciation-policy changes over time hides the fact that the frequency of method changes has declined noticeably over the period, while the frequency of estimate changes has increased. These time trends in frequencies suggest asystematic change in the relative costs and/or benefits of method changes versus estimate revisions. We next examine the effect of changes in the tax treatment of

5 Because the two policy compositions

vary over time and differ from each other, all financial data

used in this paper are restated to 1992 year-end dollars to control effects of inflation.

for the possible confounding

(\)

~

368

~ ~ ~ .g

~-o ~
,

§ ~ ~ ..E

CI) Q

8 .s Q o Q)

-. '.-.oQ) "' e

~

E ~ ~ ~ ..E

CI)..S .-~ Q) u Q) Q)

..s Q) ~0

Q) ~ r:~ Q)

370

Table

A.S. Keating,

.JL. Zimmerman

/ Journal

of Accounting

and Economics

359-389

I I

2

NAARS-listed

28 (2000)

firms

making

depreciation

method

changes

and/~r

estimate

revisions

in the period

1972-1994"

Number

of

changes

Panel A: Depreciation

Number firms

of

Number of changes made

method changes and depreciation estimate revisions 1 444 444 2 69 138 3 12 36 4 4 16 5 1 5 6 0 0 7 1 I 7

Total

%

of

firms

69 21 6 2 1 O

100

531

646

212 20

212 40

90 10

Total

232

252

100

Panel C: Estimate revisions Number of revisions

Number firms

Number of changes made

% of firms

Panel B: Method changes 2

2 3 4 5 Total

of

295

295

87

34

68

10

6

18

2

2

8

1

1

5

0

338

394

100

aIn Panel A, 531 firms changed at least one depreciation policy. ,PanelsBand C indicate that 232 firms changed methods and 338 firms revised estimates, for a ~otal of 570 firms. The difference between 531 and 570, namely 39 firms, is due to some firms maIqing more than one type of policy change.

depreciation in 1981 on the relative frequency of depreciation method changes and estimate revisions. 4.2.1. 1981 ta-xlaw changes Prior to 1981 taxpayers were entitled, subject to IRS challenge, to set depreciation methods and make useful life and salvage value estimates for tax purposes based on their experience and judgment of all facts and circumstances of the asset being depreciated. However, firms using life and salvage value estimates outside of the IRS-specified ranges risked having those estimates

A.S. Keating, J.L. Zimmerman / Journal of Accounting and Economics 28 (2000) 359-389

Table 3 NAARS-listed firms making both a depreciation period 1972-1994 .

method change and an estimate revision in thl

Combination

Number

Firms making method change only Firms making estimate revision only Firms making method change and estimate revision

193 299 39

Total

531

Time between changesof different types Mean Median Maximum Minimum

371

of firms

%

of total

firms

36 57 7

100

Years' 2.07 0 12 0

challenged by the IRS.6 One of the factors the IRS considered when evaluating the appropriateness of a firm's depreciation life and salvage value estimates for tax purposes was whether the firm used the same estimates for financial reporting purposes.7 Thus, for assetsplaced in service prior to 1981,financial reporting depreciation estimates were not independent of tax depreciation estimates. Taxpayers had incentives to estimate relatively short useful asset lives for financial reporting to justify short estimated lives for tax purposes. Moreover, managers likely were reluctant to make income-increasing estimate revisions for financial reporting since such revisions might trigger similar revisions for tax purposes and thus higher taxes. By contrast, the IRS did not challenge taxpayers' choices of depreciation methods, allowing managers to change book depreciation methods without any potential tax consequences.Hence, prior to 1981,income-increasing estimate revisions were a potentially more costly substitute to method changes. The Accelerated Cost Recovery System (ACRS) became law in 1981 and applied to assets placed in service after 1980. Under ACRS and its successor MACRS (Modified Accelerated Cost Recovery System), asset tax lives are specified by law and not subject to taxpayer discretion.8 Thus, for assetsplaced 6 See, for example, The adoption of the asset depreciation range (ADR) system, US Treasury Department Release,June 22, 1971, and Jones (1998, p. 13). 7 United States Treasury Revenue Procedure 62-21, 1962-2 CB 418. 8 For some types of assetsplaced in service after 1980,taxpayers can extend the recovery period for an assetby opting out of the ACRS and MACRS and into the Alternative Depreciation System (ADS). The ADS specifieslonger recovery periods for some (but nOt all) classesof assets,as well as a less-accelerated depreciation schedule (150% declining balance rather than 200% declining balance) for some asset classes.As with ACRS and MACRS, under ADS taxpayers can choose to depreciate the asset straight-line over the recovery period in lieu of using the specified accelerated schedule.

A.S. Keating, J:L. Zimmerman / Journal of Accounting and Economics 28 (2000) 359- 389

in service after 1980, a firm's depreciation policies and estimates for non-tax purposes have no effect on tax depreciation or the amount of taxes the firm pays (unless this triggers the Alternative Minimum Tax introduced in 1986).The 1981 tax law had the effect of reducing the relative cost of income-increasing estimate revisions versus income-increasing method changes. Assuming a constant demand for depreciation-related policy changes over the 1972-1994 period -as evidenced by the relatively constant frequency of depreciation-policy changes over the period (Table 1) -we predict more income-increasing estimate revisions relative to method changes following the 1981 tax law.9 The incidence of income-increasing estimate revisions in our sample increased from 0.16% of firm-years in the 1972-1981 period to 0.29% in the 1982-94 period, while the incidence of income-increasing method changes declined from 0.15% of firm-years prior to 1982 to 0.11% of firm-years after 1981 (not tabulated). A chi-squared test rejects the independence of these frequencies at the 1% level. Note that a critical assumption underlying these conjectures is that the tax laws prior to 1981 allowed the IRS to challenge taxpayers' tax depreciation estimates if shorter life estimates and larger salvagevalue estimates were used for tax purposes rather than for financial reporting purposes. We explore this assumption in more detail later. While the difference in the frequency of method changes and estimate revisions between the 1972-1981 and the 1982-1994 periods are significant, Table 1 reveals that the frequency of method changes declined more in 1984 than in 1982, and estimate revisions did not increase substantially until the mid-1980s. However, these patterns are not inconsistent with our tax hypothesis. The 1981 tax law applied only to assetsplaced in service after 1980. Thus, firms' incentives to change depreciation policies for assets placed in service before 1981 did not change. Moreover, in the early years of the new tax depreciation regime, the income effect of a depreciation policy change applied only to assetsplaced in service after 1980 would be relatively small- since the fraction of the firm's asset base placed in service after 1980 was small. Thus, we would not expect to observe changes in the relative frequency of method changes and estimate revisions until the post-1980 fraction of a firm's assets became significant, and this likely did not occur until several years after the tax law change. 9 For the purposes of our tax law-related tests,we consider the year 1981 to be part of the pre-tax law change period. Recall that the 1981 tax law applied only to new assets;assetsthat were already in service continued to be depreciated under the old tax rules. A firm placing an assetin service in 1981would be unlikely to make an estimate revision or method changefor that assetin 1981.Hence, any estimate revisions or method changes in 1981 can reasonably be expected to apply to assets placed in service prior to 1981 and thus subject to pre-1981 tax laws. The first depreciation policy changes applying to assetsplaced in service after 1980 occurred in 1982.

A.S. Keating, J:L. Zimmerman I Journal of Accounting and Economics 28 (2000) 359-389

4.2.2. The effect of marginal tax rates We examine the relation between depreciation-policy changes and firms' marginal tax rates as additional tests of the tax hypothesis.1° If firms were reluctant to increase book life estimates of pre-1981 assets because of the potential increased tax liability, then those firms with high marginal tax rates -thereby facing a larger increase in tax liability from an IRS challenge -were least likely to increase book estimates. By contrast, for all post-1980 assetsthere were no tax consequencesof estimate revisions for financial reporting purposes, so marginal tax rates likely did not affect firms' decisions to revise depreciation estimates for assetsplaced in service after 1980. Also, firms with high marginal tax rates have stronger incentives to reduce taxes than those with low marginal tax rates, suggesting that pre-1982 high marginal tax rate firms were more likely to adopt income-decreasing book estimate revisions to reduce taxes than pre1982 low marginal tax rate firms. Finally, over the entire 1972-1994 period, tax and financial depreciation methods were independent, so a method change for financial reporting purposes had no effect on tax liabilities. While marginal tax rates likely affected firms' decisions to revise estimates in the pre-1982 period, they did not affect their decisions to change methods in that period. Hence, we make the following predictions. The marginal tax rates of pre-1982 incomeincreasing estimate revision firm-years are: (i) lower than the marginal tax rates of post-1981 income-increasing estimate revision firm-years; (ii) lower than the marginal tax rates of pre-1982 income-decreasing estimates revision firm-years; and (iii) lower than the marginal tax rates of pre-1982 income-increasing method change firm-years. To test these predictions we use the Shevlin (1990) trichotomous method for estimating marginal tax rates: Firm-years with positive net operating loss carryforwards (Compustat item # 52) and zero taxes (Compustat item # 63) are assigned a marginal tax rate of zero.ll Firm-years with zero net operating loss carryforwards and positive taxes are assigned a marginal tax rate equal to one. Firm-years with either positive net operating loss carryforwards and positive taxes or zero net operating loss carryforwards and zero taxes are assigned

10Our thanks to the referee for suggesting these tests. 11Graham (1996a, b) develops a methodology for estimating marginal tax rates using a simulation, and makes his simulated marginal rates for firm years between 1980and 1995 available on the internet at www.duke.edu/ , jgraham/taxform.html. Graham (1996b) shows that the Shevlin (1990) trichotomous marginal tax rate estimates are highly correlated with Graham's simulated rates, and are less costly to obtain. Since Graham only has simulated rates for 1'980-1995,and our sample contains a considerable number offirm-years prior to 1980,we are unable to use Graham's rates to test the robustness of our results.

A.S. Keating, J:L. Zimmerman / Journal of Accounting and Economics 28 (2000) 359-389

Table 4 Deviations from average Compustat marginal tax rates for firms making depreciation estimate revisions and method changes,1971-1994.Marginal tax rate is 1 iffirm-year has positive taxes and zero net operating loss carryforwards, 0 if the firm-year has zero taxes and positive net operating loss carryforwards, and 0.5 if taxes are positive and net operating loss carryforwards are positive, or if taxes are zero and net operating loss carryforwards are zero (Shevlin, 1990).Average Compustat marginal tax rate based on marginal tax rates of all Compustat-listed firms with available data for that year. Real property-intensive sub-sample (Panel B) consists of those firm-years whose ratio of depreciable real property (Compustat items 155 and 73) to total property, plant and equipment (item 8) is in the upper two quintiles of the sample; equipment-intensive sub-sample (Panel C) consists of those firm-years whose ratio of depreciable real property to property, plant, and equipment is in the lower two quintiles of the sample. Mean deviation from

avetage

marginal tax rate

Median deviation from average marginal tax rate

Panel A: Entire sample Pre-1982 income-increasing estimate revisions Post-1981 income-increasing estimate revisions Pre-1982 income-decreasing estimate revisions Pre-1982 income-increasing method changes

-0.209 -0.114 -0.138 -0.017

-0.231 -0.117*** 0.203 0.242

Panel B: Real property-intensive sub-sample Pre-1982 income-increasing estimate revisions Post-1981 income-increasing estimate revisions Pre-1982 income-decreasing estimate revisions Pre-1982 income-increasing method changes

-0.091 0.015 -0.226 -0.049

0.242 0.384 -0.228 0.281

Panel c: Equipment-intensivesub-sample Pre-1982 income-increasing estimate revisions Post-1981 income-increasing estimate revisions Pre-1982 income-decreasing estimate revisions Pre-1982 income-increasing method changes

-0.129 -0.069 -0.049 -0.073

0.251 -0.090* 0.241 0.251

N

41 111 42 25

8 14 2 24

6 11 24

*Significantly different from median marginal tax rate of pre-1982 income-increasing estimate revisers at 10% level or better. **Significantly different from median marginal tax rate of pre..1982income-increasing estimate revisers at 5% level or better. ***Significantly different from median marginal tax rate of pre-1982 income-increasing estimate revisers at 1% level or better.

marginal tax rates equal to one-half. Hence,a firm-year's marginal tax rate can have a value of zero, one-half, or one. To control for other factors affecting marginal tax rates over time, we calculate a firm-year's marginal tax rate as the deviation from the mean marginal tax rate for all Compustat-listed firms in that year. Panel A of Table 4 reports the results of tests of differences in mean and median marginal tax rates. The mean (median) marginal tax rate for pre-1982

A.S. Keating, J:L. Zimmerman / Journal of Accounting and Economics 28 (2000) 359-389

375

income-increasing estimate revisions is -0.209 ( -0.231), whereas the mean (median) marginal tax rate for post-1981 income-increasing estimate revisions is -0.114 ( -0.117). This difference of 0.095 (0.114) is in a direction consistent with our predictions. Moreover, the difference in medians is significantly different from zero at the 1% level, although the difference in means is not. Also consistent with our predictions, the mean and median marginal tax rates for firms making pre-1982 income-decreasing estimate revisions ( -0.138 and 0.203 respectively) are higher than the comparable statistics for the pre-1982 incomeincreasing estimate revision sample, although neither difference is significantly different from zero. Finally, the mean (median) marginal tax rate of pre-1982 firm-years with income-increasing methodchangesis -0.017 (0.242),higher, but insignificantly so, than the mean (median) marginal tax rate for pre-1982 income-increasing estimate revisions. Overall, these tax rate-based tests are directionally consistent with our hypothesis that the 1981 tax act affected the relative frequencies of estimate revisions and method changes. 4.2.3. The CLADR system of tax depreciation The previous tests assumed that book depreciation estimates for assets placed in service before 1981 could be used by the IRS to challenge taxpayers' tax depreciation estimates. However, a tax depreciation system known as the Class Life Asset Depreciation Range system (CLADR) was available for assets placed in service after 1970 and offered a depreciation 'safe haven' for taxpayers adopting IRS-specified life ranges. The IRS would not challenge life estimates within those ranges. For example, taxpayers could use a tax life of between 8 and 12 years for office equipment, regardless of their book life estimates for office equipment, and be assured the IRS would not challenge the tax life. For firms using the CLADR system for most of their assets,book and tax depreciation was largely uncoupled before 1981, and hence faced no higher incentive to make income-increasing estimate revisions for pre-1981 assetsthan for post-1980 assets. Indeed, if all firms adopted the CLADR system for all assets,then we would predict no difference in the pattern of depreciation policy changes pre-1982 and post-1981. However, a study by the IRS of 1975 corporate tax returns (Internal Revenue Service, 1979) revealed that among the largest corporate tax returns -sales greater than $100 million and representative of our sample- only 31% of all returns included depreciation of at least one assetusing the CLADR system. While it is possible that CLADR adoption increased between 1975 and 1980,this IRS study suggeststhat all firms did not immediately embrace the system. Indeed, our findings suggestthat the CLADR system was not widely adopted. The existence of the CLADR system compromises the power of our tests reported above. CLADR-adopting firms did not face the same change in book depreciation incentives in 1981 as did non-CI-,ADR-adopting firms. To increase

376

A.S. Keating,

J.L. Zimmerman

/ Journal

of Accounting

and Economics

28 (2000)

359-389

power, we exploit the fact that the CLADR system was rarely used for depreciable real property (see, for example, Gravelle, 1980; Scholler et al., 1973).12 Firms with a high proportion of depreciable real property were likely to have a relatively small fraction of their assetsdepreciated for tax purposes using the CLADR system and hence were likely to experience an increase in incentives after 1981 to make income-increasing estimate revisions. Firms with little depreciable real property were likely to have a relatively large fraction of their assetsdepreciated using the CLADR system and hence were unlikely to experience an increase in incentives to make income-increasing estimate revisions after 1981. We, therefore, predict a bigger increase in the frequency of incomeincreasing estimate revisions between the pre-1982 period and the post-1981 period among firms with a high proportion of depreciable real property than among firms with a low proportion of depreciable real property. We divide our sample into two groups: those with ratios of depreciable real property to total property, plant and equipment in the fourth or fifth quintile (the 'real property-intensive sub-sample') and those with ratios in the first or second quintile (the 'equipment-intensive sub-sample').13 We predict the realproperty-intensive sub-sample to exhibit a significant increase in estimate revisions after 1981 relative to before 1982, and a significant decrease in method changes over the same two periods. Among the real-property-intensive firmyears, there were nine income-increasing estimate revisions and 16 incomeincreasing method changes before 1982, and 30 income-increasing estimate revisions and only nine income-increasing method changes after 1981.Thus, the frequency of estimate revisions increased after 1981 while the frequency of method changes decreased.A chi-squared test rejects the independence of these frequencies at the 1% level. By contrast, among the equipment-intensive sample both estimate revisions and method changes increase in frequency after 1981. There were 15 income-increasing estimate revisions and seven income-increasing method changes before 1982, and 29 income-increasing estimate revisions and ten income-increasing method changes after 1981.A chi-squared test fails to reject the independence of these frequencies a1:even the 10% level. This is further evidence that the change in the relative frequency of estimate changes and method changes in the pre-1982 and post-1981 periods is due to changes in tax depreciation rules. We also repeat our marginal tax rate tests on the real-property-intensive and equipment-intensive sub-samples. Since real-property-intensive firms are least

12Our thanks to the referee for pointing out the fact that the ADR system was generally not used for depreciable real property and for suggesting a test exploiting this fact. 13We estimate a firm-year's real depreciable property as the sum of buildings (Compustat

items

155) and construction in progress (item 158). We use book values net of accumulated depreciation because gross book values are not available on Compustat for firm-years earlier than 1984.

A.S. Keating, J:L. Zimmerman / Journal of Accounting and Economics 28 (2000) 359-389

377

likely to have adopted the CLADR system for assetsplaced in service between 1971 and 1980, we would expect to obtain stronger results from our marginal tax rates tests using this sample. Conversely, we would expect to obtain weaker or even opposite results from equipment-intensive firms which are most likely to have adopted the CLADR system. These statistics and their significance levels are reported in Panels Band C of Table 4. For the real-property-intensive firms, the mean and median marginal tax rate of pre-1982 income-increasing estimate revisions is lower than the mean and median of both post-1981 incomeincreasing estimate revisions and pre-1982 income-increasing method changes. The mean and median marginal tax rates for pre-1982 income-decreasing estimate revisions are lower than pre-1982 income-increasing estimate revisions -inconsistent with our predictions -although the sample size is small (n = 2). For the equipment-intensive firms the results are inconsistent. The mean but not the median marginal tax rate of the pre-1982 income-increasing estimate revisions is smaller than the marginal tax rates of the other three samples. The results in Panels Band C of Table 4 are at least weakly consistent with our tax law change hypothesis, although small sample sizes compromise the power of these tests. 4.2.4. Changes in disclosure requirements We also investigated whether the observed trends in depreciation-policy changes reported in Table 1 resulted from changes in disclosure requirements. The history of disclosure requirements does not bear this out however. APB 20, governing disclosure of changes in methods or estimates, went into effect in 1971, before the beginning of our sample period. Statement of Auditing Standards (SAS) No.58, governing the form of an auditor's discussion of a method change, was considered by the AICPA beginning in 1985, adopted in 1988 and became effective in 1989. The frequency of method changes declined noticeably in 1984 and 1985, before SAS 58 was even under consideration. Moreover, SAS 58 simply changed the manner in which auditors discussed method changes in their opinion letter, not the disclosure or non-disclosure of method changes. Finally, the SEC requirement that auditors certify the preferability of newly adopted methods was in force over this entire period. Hence, it appears unlikely that the change in the relative frequency of method changes and estimate revisions is due to a change in reporting or disclosure requirements. 4.3.

Three dimensions

of depreciation-policy

changes

Table 5 presents summary statistics on sales, assets, and the effect of the depreciation-policy change for our sample.14 The median firm in our sample

14Note that our sample size declines from 646 firm-years in Tables 1-3 to 440 firm-years in Tables 4 and 5, reflecting the fact that not all NAARS-Iisted firms are also Compustat-listed.

378

A.S. Keating,

J.L. Zimmerman

/ Journal

of Accountj'ng

and Economics

28 (2000)

359-389

Table 5 Descriptive statistics on 440 firm-years of changes in depreciation methods and/or revisions of depreciation estimates. (Dollar figures are price level adjusted to 1992 dollars and reported in millions.) Medians

(1)

(2)

Method (n =

changes

Estimate

(3) revisions

Entire

sample

175)

(n = 265)

(n = 440)

Sales 271 Net income 7.4 Assets 226 Impact of change on net income" 0.296* Income impact as % of net incomeb 6** Impact of change as % of assets" 0.3**

251 5.2 284 0.802* 14** 0.5**

270 5.~\ 270 o.~;00 11 o.~.

aAs reported by firms in their annual reports or 10 Ks. bAbsolute value of impact of change/absolute value of net income. cAbsolute value of Impact of change/assets. * Median of the first series is significantly or better. **Median or better.

of the first series is significantly

smaller than the median of the second series at 10% level smaller than the median of the second series at 5% level

(Column 3) experiences a $500,000 increase in earnings as a result of the depreciation-policy change, representing 11% of income and 0.5% of assets. While method changers (Column 1) do not differ significantly from estimate revisers (Column 2) in terms of sales, earnings, or assets, the dollar effect of method changes, as well as the earnings and asset impact of those changes, are significantly smaller than comparable statistics for estimate revisers. As discussed above and demonstrated in Table 6, the impact of method changes is smaller than that of estimate revisions because many method changes are applied to new assets only. Table 6 divides the sample along three dimensions: the type of change (method change versus estimate revision), the direction of the change (incomeincreasing, income-neutral and income-decreasing), and the way in which the change is applied (to new assetonly or to both new and existing assets).Table 6, Panel A (Column 4) indicates that, of 175 method changes, 102 (58%) are applied to both new and existing assets,while 73 (42%) are applied only to assets placed in service on or after the adoption of the new depreciation method. The earnings impacts of these two types of method changes differ significantly. Income-increasing method change applied to both new and existing assets increases median earnings by $3.9 million, or 23% of earnings (Column 1), while

A.S. Keating, J.L. Zimmerman / Journal of Accounting and Economics 28 (2000) 359-389

379

income-increasing method change applied only to new assetsincreases median earnings by $600,000 (6%). In Table 6 Panel B (Column 4), 258 of the 265 estimate revisions are for both new and existing assets. Only ten of the 258 have an immaterial effect

Table 6 Income effects of depreciation method changes and estimate revisions broken down by (a) type of change (method change or estimate revision), (b) the directional effect of the change on net income (income-increasing, income-neutral or income-decreasing),and (c) how the change is applied (to new assetsonly or to both new and existing assets).Sample I::onsistsof 440 depreciation-policy changes made by NAARS-Iisted firms between 1972 and 1994 for which Compustat data is available. (3) Income

increasing

(2) Income neutral

changes

changes

changes

(1) Income

(4)

decreasing Total

Panel A: Method changes Change applies to both new and existing assets

Median dollar impact on earnings (millions)

$3.9

$0.0

Median percentage impact on earnings

23%

0%

Median percentage impact on assets

Change applies to new assetsonly

Total

1.3%

0%

N

66

22

Median dollar impact on earnings (millions)

$0.6

$0.0

-$14.

$0.7

56%

2.3%

14

15%

0.8% 102

$0.

-$0.2

Median percentage impact on earnings

6%

0%

7%

3%

Median percentage impact on assets

0.3%

0%

0.2%

0.1%

N

41

25

Median dollar impact on earnings (millions)

$1.6

$0.0

Median percentage impact on earnings

14%

0%

Median percentage impact on assets N

0.6% 107

7 -$6.5

$0.3

12%

0%

47

73

1%

21

6%

0.2% 175

380

A.S. Keating,

J:L. Zimmerman

/ Journal

of Accounting

and Economics

28 (2000)

359-

Table 6 (continued) (3) Income

increasing

(2) Income neutral

changes

changes

changes

(1) Income

(4)

decreasing Total

Panel B: Estimate revisions Change applies to both new and existing assets

Median dollar impact on earnings (millions)

$2.0

$0.0

Median percentage impact on earnings

16%

0%

Median percentage impact on assets N Change applies to new assetsonly

Median dollar impact on earnings (millions) Median percentage impact on earnings

Total

0.5% 190

15%

16%

$0.0

58

-$10.5

8%

0.3%

0.5%

N

3

2

2

Median dollar impact on earnings (millions)

$1.9

$0.0

Median percentage impact on earnings

15%

0%

193

258

8%

0%

N

0.5%

0%

0.3%

0.5%

15%

$0.0

Median percentage impact on assets

Median percentage impact on assets

$0.9

0.7%

0%

10

$0.1

-$1.7

0% 12

-$1.7 15%

0.6% 60

$0.8 14% 0.5% 265

(Column 2).15The 190 income-increasing revisions (Column 1) have roughly the same absolute dollar and absolute percentage impact ($2.0 million and 16%) as the 58 income-decreasing revisions in Column 3 ( -$1.7 million and 16%). The earnings impact of income-increasing method changes for new and existing assets is almost twice as large ($3.9 million) as the earnings impact of incomeincreasing estimate revisions for new and existing assets($2.0 million). Although not reported in Table 6,88% of the estimate revisions are revisions of estimated lives, 7% are revisions of estimated salvage values, and 5% are revisions ofboth useful lives and salvage values. 1SFirms are not required to disclose that the estimated useful life or salvage value of new assets differs from that of assetsalready in service.As a result, we would expect the majority of disclosed estimate revisions to apply to both new and existing assets.

A.S. Keating, JL. Zimmerman / Journal of Accounting and Economics 28 (2000) 359-389

381

Overall, Table 6 shows that method changes generate smaller average dollar and percentage earnings changes than estimate revisions because managers frequently elect to apply method changes to new assetsonly, and then to only a subset of their assets.Only 80 (66 income-increasing plus 14 income-decreasing) of the 175 method changes in Panel A apply to both new and existing assets and have a material effect on income. Estimate revisions, by contrast, are almost always applied to both new and existing assetsand produce material earnings effects in 248 (190 income-increasing plus 58 income-decreasing) of the 265 estimate revisions. 4.4.

Test of responses to declining

performance

We predict that declines in accounting performance create incentives to adopt income-increasing method changes that have a material impact on earnings. To test for earnings management. by poorly performing firms, we examine two groups of income-increasing depreciation-policy changes: (a) income-increasing method changes for both new and existing assets; and (b) income-increasing method changes applied only to new assets. From Table 6 we know that group (a) experiences the greater earnings increase from the change, while group (b) experiences the lesser earnings impact. We therefore predict that the group applying the method changes to both new and existing assets is likely to exhibit worse financial performance relative to the group applying the method changes only to new assets. Notice that neither the ROA, leverage nor current ratio variable we use in our tests is adjusted for the effect of the depreciation change, and hence the tests are biased against finding the predicted relations. Table 7 presents data on ROA (income before extraordinary items divided by total assets),leverage (Iong-term debt to total assets),and current ratio (current assetsto current liabilities) for both groups of income-increasing depreciationpolicy changers. We examine the current ratio because debt covenants often contain minimum working capital provisions and depreciation-policy changes can affect working capital through inventory values. The sample sizes are smaller in Table 7 than in Table 6 because we require that each firm has Compustat data in years t -2, t -1, t, and t + 1, where t is the year of the policy change. In the year of the policy change, firms making income-increasing method changes for both new and existing assets (Column 1 in Table 7 Panel A) have a significantly lower ROA (0.019) and higher leverage (0.280) than firms changing methods only for new assets (0.054 and 0.171, respectively) (Column 2). Moreover, in Panel B sales and ROA decline more, and leverage increases more from year t- 1 to year t for firms changing methods for new and existing assets than for firms changing only for new assets. These findings are consistent with our predictions.

382

A.S. Keating,

J.L. Zimmerman

/ Journal

of Accounting

and Economics

28 (2000)

359-389

Table 7 Year-to-year levels and changes in summary statistics for firms making (a) income-increasing depreciation method changes for new and existing assets,and (b) income-increasing depreciation method changes for new assets only. (Dollar figures are price level adjusted to 1992 dollars and reported in millions.) (1) Income-increasing change for both existing assets (n = 61)

method new and

(2) Income-increasing method change for new assets only (n = 38)

Panel A: Levels in the year of the policy change (medians) Sales Net income Assets ROA a

233 1.3 259 0.019*

Leverageb Current ratio" Depreciation expense as % of net incomed as % of assets" Impact of change on net incomef Income impact as % of net incomeg Income impact as % of assetsh

0.280* 1.930 16.1 100% 4.7% 4.3* 24%* 1.3%*

442 10.3 209 0.054* 0.171* 2.148 10.7 91% 5.2% 0.72* 6%* 0.3%*

-0.003 -0.024t* -0.008

0.044 0.039* -0.007

-0.072 -0.330t -0.097

0.122 -0.011 -0.202

-0.009t -0.033t* -0.006

-0.001 -0.005* -0.011

0.011 0.009t* -0.006

-0.012 -0.013* -0.008

t-1

-0.061 t

t+

-0.043 -0.146t

0.070 -0.008 0.017

Panel B: Year-to-year changes Sales t-l t+l Earnings t-l t+ ROAa tt t+l Leverageb t-l

t+1 Current

ratio'

aROA = income before extraordinary

items (Compustat

item 18)/total assets (Compustat

item 6).

bLeverage = long-term debt (Compustat item 9)/total assets (Compustat item 6). cCurrent ratio = current assets (Compustat item 4)/current liabilities (Compustat item 5). dDepreciation expense/absolute value of net income. eDepreciation expense/assets. f As reported by firms in their annual reports or 10 Ks. gAbsolute value of impact of change/absolute hAbsolute value of impact of change/assets. tSignificantly *Significantly

value of net income.

different from zero at 5% level or better. different from the other sub-sample at 5% level or better.

A.S. Keating, J:L. Zimmerman I Journal of Accounting and Economics 28 (2000) 359-389

383

As an additional test of these associations, we estimate logit and probit regressions (untabulated). The dependent variable is one if the firm makes an income-increasing method change for new and existing assets (Column 1 in Table 7) and zero if the income-increasing method change is for new assetsonly (Column 2). The independent variables include ROA, leverage, and the current ratio, and changes in these variables, in the year of the depreciation-policy change. Results are consistent with those reported in Table 7. The chi-squared statistic for this regression is significant at the O.Ollevel; ROA, leverage, and the change in ROA have significant coefficients of the predicted signs. The evidence from Table 7 is consistent with firms adopting income-increasing methods for new and existing assets to offset poor performance. Firms adopting income-increasing methods only for new assets appear to do so for reasons other than poor performance. 4.5.

Tests of changes in investment

opportunities

This section tests whether changes in firms' depreciation policies are associated with changes in their investment opportunities. In response to changes in their environment, firms change their investment strategies through entering or exiting markets, introducing or deleting products, or building or disposing of plants. These new strategies often require a realignment offirms' decision rights assignments, performance measurement systems, and compensation schemes {Brickley et. al., 1997). Realigning performance measures often require changing accounting policies. Thus, accounting policies likely change when overall organization design changes. We conjecture that, and test whether, non-income-increasing depreciation-policy changes are associated with changes in firms' investment opportunities. We conduct two sets of tests: a directional test and non-directional test. In the directional test, we have no theory as to the direction of changes made in the investment opportunity set of firms adopting non-income-increasing policy changes. However, we assume that firms making these policy changes all experience increases or decreasesin their investment opportunities in the same direction. That is, our maintained hypothesis is that all non-income-increasing policy change firms adjust their investment activity in the same direction; lacking a more complete theory, however, we have no prediction as to the direction of the change. Hence, we conduct two-tailed tests. In the non-directional tests we drop the maintained hypothesis that the investment opportunity changes precipitating non-income-increasing depreciation-policy changes are in the same direction. We examine whether the variance of the change in investment opportunities increases surrounding depreciationpolicy changes. Suppose that in response to some exogenous shock one firm increases its ratio of property, plant, and equipment {PPE) to assets-a measure of investment opportunities -from 30% in year t- 1 to 35% in year t, while

384

A.S. Keating,

JL.

Zimmerman

/ Journal

of Accounting

and Economics

28 (2000)

359-389

another firm decreasesits ratio ofPPE to assetsfrom 35% in year t -1 to 30% in year t. The median level of PPE to assets has not changed. Moreover, the median change in PPE to assetsis zero. However, the variance of the change in PPE to assetsis non-zero and varies with the magnitude of the changes.We thus examine the standard deviations of the changesin variables as an additional test of an association between depreciation-policy changes and investment opportunity changes. In both the directional and non-directional tests we compare a treatment group consisting of firms making income-decreasing or income-neutral ('non-incomeincreasing') method changesor estimate revisions only for new assetsto a matched control group. The treatment firms are unlikely changing depreciation policies due to poor financial performance becausethe changes do not increase income. The fact that these changes are applied to new assetsonly suggeststhat they are not done for earnings management reasons, since the impact of the changes on earnings is relatively small. We conjecture these firms are likely changing for non-earnings management reasons, such as to re-align internal and external contracting incentives due to new investment opportunities facing the firm. We construct our control sample as follows: For each treatment firm (i.e. a firm making a non-income-increasing depreciation.policy change), we identify all Compustat-listed firms (other than the treatment firm being matched) with PPE.to-assets, capital expenditures.to-sales, and market-to-book ratios within 0.1 standard deviation of the comparable ratio of the treatment firm two years prior to the treatment firm's depreciation-policy change (Holthausen and Larcker, 1996; Barber and Lyon, 1996). For example, consider a treatment firm making a depreciation-policy change in 1988 whose PPE.to-assets, capital expenditures-to-sales and market-to-book ratios in 1986 (2 years prior to the 1988 policy change) are 0.4, 0.05, and 1.2, respectively. Assume the standard deviations of PPE-to-assets, capital expenditures-to.sales, and market-to-book for all Compustat firms in 1986 are 1.0,0.04, and 3.0, respectively. We include in our control sample all Compustat firms with available data for 1988 whose 1986 PPE.to-assets ratio is in the range 0.3-0.5, capital expenditures-to-sales ratio is in the range 0.046-0.054, and market-to-book ratio is in the range 0.9-1.5. Ideally, we would like to exclude from our control group those firms making non-income-increasing depreciation-policy changes. However, as noted above, the Compustat footnote codes are at best imperfect indicators of depreciationpolicy changes. Moreover, these footnote codes do not indicate the incomeimpact of the change. We exclude from our control group any firm that has either a footnote code # 15 indicating a depreciation.policy change, or a change in footnote code # 5 between years t -1 and t indicating a change in methods from straight.line to accelerated (a non-income-increasing method change). While this selection procedure reduces the likelihood that the control sample contains firms making depreciation-policy changes, it does not eliminate that possibility altogether.

A.S. Keating, J:L. Zimmerman I Journal of Accounting and Economics 28 (2000) 359-389

385

Table 8 Year-to-year levels and changes in summary statistics for (a) firms choosing income-decreasing or income-neutral ('non-income-decreasing') depreciation method changes or estimate revisions for new assets only ('treatment' firms) and (b) a set of 'control' firms matched to treatment firms based on PPE/total assets, CapEx/sales, and market-to-book two years prior to the depreciation-policy change (t -2). (Dollar figures are price level adjusted to 1992 dollars and reported in millions.) , Treatment firms (n on-income- increasing changes for new assets only) (n = 37)

Control firms (n = 1108)

Panel A: Levels (medians) Sales Net income Assets ROAa

240 9.3 161 0.061* 0.147 0.301 0.047* 2.280 1.16 5.4 53% 3.5% 0.0 0% 0%

Leverageb PPE/total assets CapEx/Sales Current ratioC Market-to-book ratio Depreciation expense As % of net incomed As % of assets" Impact of change on net incomer Income impact as % of net incomeg Income impact as % of assetsb

229 4.5 140 0.043* 0.179 0.264 0.033* 2.156 1.01 5.3 67% 3.7%

Panel B: Year-to-year changes

PPE/total assets t-l t+1 CapEx/Sales t-l

t+l

Median

Standard deviation

Median

change

of change

change

0.004 0.014t* -0.002 0.001 0.006t* -0.003

0.033°

0.001 0.000* -0.004t

0.041 0.110°

0.001 -0.001 *

0.018 0.083°

0.000

0.106

Standard deviation of change

0.050 0.050 0.063° 2.040 1.991 0.098°

&ROA = income before extraordinary items (Compustat item 1&)/total assets (Compustat item 6). bLeverage = long-term debt (Compustat item 9)/total assets (Cfmpustat item 6). .Current ratio = current assets (Compustat item 4)/current liabilities (Compustat item 5). dDepreciation expense/absolute value of net income. .Depreciation expense/assets. f As reported by firms in their annual reports or 10 Ks. gAbsolute value of impact of change/absolute hAbsolute value of impact of change/assets.

i

value of net incotne.

*Significantly different from the other sub-sample at 5% level or better (2-tailed test). tSignificantly different from zero at 5% level or better (2-tailed test). "Standard deviation is significantly different from the previous year's standard deviation at the 5% level or better (2-tailed test).

386

A.S. Keating,

J.L. Zimmerman

/ Journal

of Accounting

and Economics

28 (2000)

359-389

Using these procedures we identify 1108 control nrms, or 29.9 control firms per treatment firm. We do not require each treatment firm-year to have at least one control sample firm-year (i.e. for some treatment firms this matching process will not yield any 'matched' control firms) and we do not preclude a control firm-year from being included more than once in the control sample. Table 8 presents data on the investment opportunities. Descriptive data about levels of variables such as sales and ROA are provided in Panel A. In Panel B, using changes in investment opportunity proxies, we test whether firms adopting non-income-increasing depreciation changes for new assets only are likely to increase or decreaseinvesting activity. In Table 8 Panel A, firms making non-income-increasing changes only for new assetsare significantly healthier in the year of the change than control firms. While they do not differ significantly in terms of sales, earnings, or assets, the non-income-increasing sample has significantly higher ROA (0.061) and capital expenditures to sales (0.047) than the control sample (0.043 and 0.033). In Panel B of Table 8, we present evidence regarding changes in investment opportunity variables. The median-based tests in Panel B assume that all firms adopting non-income-increasing changes for new assetsonly will change PPE or capital expenditures in the same direction. The non-income-increasing group has a significant increase in PPE to total assets(0.014) and capital expenditures to sales(0.006) in the year of the depreciation-policy change (year t). By contrast, control firms do not experience a significant change in plant and equipment (0.000) and capital expenditures (0.001) in year t. Panel B of Table 8 also reports the standard deviation of changes in PPE to total assets and capital expenditures to sales for the treatment and control sample. We use these standard deviations as the basis for the non-directional tests described above. For the non-income-increasing group the standard deviation of the changes in plant and equipment as a fraction of total assetsincreases more than four-fold (from 0.018 to 0.083), and the standard deviation of the changes in capital expenditures more than doubles (from 0.041 to 0.110), from year t -1 to year t. These results contrast with the control firms not making a depreciation-policy change, where the standard deviation of the change in PPE/assets and CapEx/sales variables does not change significantly from year t -1 to year t. The differences in standard deviations are statistically significant between the treatment and control groups in year t. Overall, these results suggestthat non-income-increasing depreciation-policy changes adopted by firms for new assets only are associated with changes in their investment opportunities. Moreover, the directional tests showing an association between increases in investment opportunities and non-incomeincreasing depreciation-policy changes suggestthat firms experiencing increases in investment opportunities shorten estimated asset lives, decrease estimated salvage values, or adopt more accelerated depreciation schedules.

A.S. Keating, J:L. Zimmerman / Journal of Accounting and Economics 28 (2000) 359-389

387

To test for any confounding tax effects in the pr~-1981 period, we redo our analysis restricting the samples used in Tables 7 and 8 to depreciation-policy changes after 1983, i.e. two years after the adoption of federal tax laws making financial reporting depreciation choices largely independent of tax considerations. Our inferences do not change.

5. Conclusions The extant accounting literature (e.g.,Hagerman and Zmijewski, 1979, Skinner, 1993, Bowen et al., 1995) documents statistically significant cross-sectional associations between depreciation method and firm size, leverage, risk, investment opportunity set, and bonus plans, among other variables. However, these tests' explanatory power is low. Moreover, attempts to explain changes in depreciation methods have not proven very fruitful (see, for example, Holthausen, 1981; Sweeney, 1994). Thus, we do not have compelling theories to explain why firms choose the depreciation methods they do or why managers change one of the largest accruals under their control: depreciation. Unlike the previous literature, we examine whether two depreciation accounting policy changes -one being changes in depreciation methods the other being revisions of useful lives and salvage values of depreciable assets-are responsesto changes in the tax code, measures taken to offset poor performance, or reactions to changes in the firm's investment opportunities. We find that these policy changes are relatively infrequent events among NAARS-listed firms, occurring in less than 1% of firm-years. Changing depreciation policies does not appear to be a frequently used earnings management tool. While the frequency of depreciation-policy changesin aggregate has remained constant over time, the frequency of method changes has declined and estimate revisions have increased. These shifts in frequencies of policy changes over time are consistent with a 1981 tax law that uncoupled useful life and salvage value estimates for financial reporting purposes from tax liabilities. In particular, following the 1981 tax revision there is a significant increase in the frequency of estimate revisions, while there is a significant decline in the frequency of method changes. This finding suggests that prior to 1981, while firms were allowed to keep separate books for tax and financial reporting, financial reporting estimates of useful lives and salvage values could affect tax liability. The median absolute value of depreciation method changes on income is small, only 6% of the absolute value of earnings. By contrast, the median absolute value of the impact of estimate revisions is 13% of the absolute value of earnings. The smaller impact of method changes is due to managers electing 42% of the time to change methods only for new assets,while retaining extant methods for assetsalready in service. Moreover, when managers elect to apply

388 A.S. Keating, J.L. Zimmerman / Journal of Accounting and Eeonomics 28 (2000) 359-389

the method change to both new and existing assets,in about 20% of these cases the change is only applied to one class of assets(such as buildings) yielding an immaterial effect on earnings. When managers elect to make an income-increasing method change for both new and existing assets,the earnings effect is larger than for estimate revisions. Moreover, firms adopting such changes have the worst financial performance and the highest leverage. This evidence is consistent with earnings management to avoid debt covenant violations, to increase executive compensation, and/or to reduce the likelihood of executive turnover. Firms electing non-income-increasing depreciation-policy changeshave a larger change in their investment opportunity sets than firms not making these changes. These firms are performing well and furthermore are increasing the ratio of property, plant, and equipment to assets and capital expenditures to sales in the same year as the depreciation-policy change. We conclude that depreciation-policy changes are made for a variety of reasons. Book depreciation estimate revisions prior to a 1981 tax law change appear to have been done to reduce the firm's tax liability. Income-increasing changes applied to both new and existing assets,which produce large changes in earnings, are likely adopted to boost earnings for debt covenant and compensations reasons or to reduce the likelihood of executive dismissal. Non-incomeincreasing depreciation-policy changes, meanwhile, appear to be in response to changes in firms' investment opportunity sets. Our findings have implications for research examining other accounting policy changes. The discretion afforded managers over such changes is more subtle than simply the 'change' versus 'do not change' decision. For example, a manager electing to change depreciation methods can choose (a) whether the change will be income-increasing -say by switching from the accelerated method to the straight-line method -or income-decreasing say shortening estimated lives; (b) whether to apply the change only to assetsacquired after the adoption of anew depreciation-policy, or to existing assets as well; and (c) whether to apply the new method to all types of assetsin all sub-units or to onlya sub-set of assets -say, equipment but not buildings -and a sub-set of business units. Managers' choices along each of these dimensions are associated with their incentives. We find evidence of depreciation-policy changes being correlated with tax law changes, poor performance, and changes in investment opportunities only after we identify the sub-set of policy changes most likely to be associated with each incentive. Studies of other accounting policy changes will likely benefit from similar disaggregation. References Ball, R., Keating S., Zimmerman, J., 1999.Historical cost as a conjlmitment device. Working Paper, University of Rochester. :

A.S. Keating, J.L. Zimmerman / Journal of Accounting and Economics 28 (2000) 359- 389

Barber, B., Lyon, J., 1996. Detecting abnormal operating performance: the empirical power and specification of test statistics. Journal of Financial Economic$ 41, 359-399. Bowen, R., DuCharme, L., Shores, D., 1995. Stakeholders' implidit claims and accounting method choice. Journal of Accounting and Economics 20, 255-296. : Brickley, J., Smith, C., Zimmerman, J., 1997. Managerial Economics and Organizational Architecture. IrwinjMcGraw-Hill, Burr Ridge, IL. , Choi, S., Jeter, D., 1992. The effects of qualified audit opinions on earnings response coefficients. Journal of Accounting and Economics 15,229-247. : Graham, J., 1996a. Debt and the marginal tax rate. Journal of ~nancial Economics 41,41-73. Graham, J., 1996b. Proxies for the corporate marginal tax rate. Jdurnal fFinancial Economics 42, 187-222. Gravelle, J., 1980.The capital cost recovery act: an economic analysis of 10-5-3 depreciation, Studies in Taxation, Public Finance and Related Subjects 4, 66-71. ! Hagerman, R., Zmijewski, M., 1979. Some economic determin~nts of accounting policy choice. Journal of Accounting & Economics 1,141-161. i Holthausen, R., 1981. Evidence on the effect of bond covenants]and management compensation contracts on the choice of accounting techniques: the caseof depreciation switch-back. Journal of Accounting & Economics 3,73-109. Holthausen, R., Larcker, D., 1996.The financial performance of reverseleveraged buyouts. Journal of Financial Economics 42, 293-332. Internal Revenue Service, 1979. Statistics of Income: Corporatlions Income Tax Returns 1975. Government Printing Office, Washington, DC. : Jones, S.M., 1998. Principles of Taxation. Irwin McGraw-Hill, Bbston, MA. Mutchler, J., Shane, P., 1995. A comparative analysis of firms included in and excluded from the NAARS database. Journal of Accounting Research 33, 193-202. O'Reilly, V., McDonnell, P., Winograd, B., Gerson, J., Jaenicke, H., 1998. Montgomery's Auditing, 12th Edition. Wiley, New York, NY. : Scholer, T., TraIler, J., Wagner, I., 1973. The ADR system: an ahalysis of the final regs and how practitioners should use them. Journal of Taxation July, 16-i3. Shevlin, T., 1990. Estimating corporate marginal tax rates with asymmetric tax treatment of gains and losses.Journal of the American Tax Association 12, 51-~7. Skinner, D., 1993. The investment opportunity set and accounting procedure choice: Preliminary evidence. Journal of Accounting & Economics 16, 407-445. Smith, C., Watts, R., 1992. The investment opportunity set and corporate financing, dividend and compensation policies. Journal of Financial Economics 32, 263-292. Sweeney, A., 1994. Debt-covenant violations and managers' al::counting responses. Journal of Accounting and Economics 17,281-308. Warner, J., Watts, R., Wruck, K., 1988. Stock prices and top ~nagement. Journal of Financial Economics 20,431-460. , Watts, R., Zimmerman, J., 1986. Positive Accounting Theory. Pr~ntice-Hall, Englewood Cliffs, NJ. Watts, R., Zimmerman, J., 1990.Positive accounting theory: a ten [yearperspective.The Accounting Review 65, 131-156. Weisbach, M., 1988. Outside directors and CEO turnover. Journal of Financial Economics 20, 431-460.