Income Inequality, Tax Policy, and Economic Growth∗ Siddhartha Biswas

Indraneel Chakraborty

Rong Hai†

August 9, 2015 Abstract We investigate how reduction of income inequality through tax policy affects economic growth. Taxation at different points of the income distribution has heterogeneous impacts on households’ incentives to invest, work, and consume. Using U.S. state-level data and microlevel household tax returns over the last three decades, we find that reduction of income inequality between low and median income households improves economic growth. However, reduction of income inequality through taxation between median and high income households reduces economic growth. These economic growth effects are attributable both to supply-side factors (changes in small business activity and labor supply) and to consumption demand.

JEL Codes: E2, H2, H3.

∗ We

thank Scott Ashworth, Christopher Berry, Dan Black, St´ephane Bonhomme, Naijia Guo, James J. Heckman,

Dirk Krueger, Alok Kumar, Ethan Bueno de Mesquita, Anthony Fowler, Jo˜ao Gomes, Tong Li, B. Pablo Montagnes, Jos´e-V´ıctor R´ıos-Rull, James Smith, Rex Thompson, Xi Weng, Harald Uhlig, Motohiro Yogo, Li-An Zhou and seminar participants at University of Chicago, Peking University, Fourth Society of Labor Economists/EALE World Conference 2015, and Chinese University of Hong Kong for helpful comments and suggestions. We thank Daniel Feenberg at NBER for access to TAXSIM individual tax return data. † Siddhartha Biswas: Department of Economics CB 3305, University of North Carolina, Chapel Hill, NC 27599. Email: [email protected]. Indraneel Chakraborty: Department of Finance, School of Business Admin., University of Miami, Coral Gables, FL 33124. Email: [email protected]. Tel: (312) 208-1283. Rong Hai: Center for the Economics of Human Development and Becker Friedman Institute for Research in Economics, University of Chicago. 1155 E. 60th Street, Room 223, Chicago, IL 60637. Email: [email protected]. Tel: (267) 254-8866.

Modern governments have utilized tax policy to not only raise capital for government operations, but also to reduce income inequality among citizens. Progressive taxation with negative net tax rates for the lowest income households aims to achieve two distinct objectives: (i) to provide a minimum level of consumption for the low income population, and (ii) to reduce income inequality between different groups of population.1 The underlying economic justification for this tax policy is that income inequality creates lower economic growth. Researchers using cross-country regressions of GDP growth on income inequality, find mixed evidence regarding the relationship between income inequality and economic growth.2 Our paper investigates how tax policies that reduce income inequality have affected economic growth in U.S. states in the last three decades.3 We distinguish between the impact of tax policy on households below median income level and on those above median income level. We find that poverty alleviation i.e. reduction of income inequality between low income and median income households improves economic growth. At the same time, we find that reduction of income inequality through taxation between median income households and high income households reduces economic growth. We explore three major economic components of economic growth as well. We find that reduction of income inequality between below median and median households encourages growth in labor supply on both the extensive and intensive margin, while reduction in income inequality between above median households and median households reduces labor supply. Further, reduction in income inequality through taxation between above median and median households reduces job creation by small businesses. In contrast again, below median income inequality reduction through tax policy encourages small businesses to employ more workers. Finally, we also find that reduction in income inequality between lower income households and median in1 See

Figure I, which shows how income inequality is reduced in U.S. through tax policy which effectively compresses the income distribution around the median household. 2 Alesina and Rodrik (1994), Persson and Tabellini (1994), Perotti (1996) among others find that there is a negative correlation between average growth and inequality since 1960s. Persson and Tabellini (1994) document that a similar negative relationship existed in nine developed economies since 1830s. However, Forbes (2000) finds a positive relationship between income inequality and economic growth in high and mid-income countries, and Barro (2000) finds a positive relationship between income inequality and growth in rich and negative relationship in poor countries. See B´enabou (1996), Ostry et al. (2014), and Cingano (2014) for detailed surveys of the literature. 3 Literature on endogenous growth of countries includes seminal work by Romer (1986), Lucas (1988), Barro (1990), and Barro (1991) among others. Our paper focuses on the impact of U.S. tax policy to reduce income inequality on economic growth of U.S. states.

1

come households increases per capita consumption expenditure growth rate. Our empirical strategy relies on within-state variation in tax policies that reduce income inequality to explain within-state variation in growth rates of U.S. states over time. We utilize a simple measure that calculates the changes in income distribution induced by income tax policy for each state using actual tax return data. In other words, the measure calculates the additional average income tax paid for each additional dollar earned by a person at the higher/lower income level, compared to the reference point of median income level household. Since this is analogous to a contraction function on income distribution, we refer to it as the contraction factor. As we show, this cross-sectional differential tax rate measure is able to explain economic growth even after controlling for marginal tax rate for a given individual over time (See, for example, Barro and Redlick (2011) who consider the impact of marginal tax rate). This is because we argue that an individual considers the impact of tax policy on her personal income in two dimensions. She considers the tax-induced change in her income with respect to the reference point of median income household. This is in addition to the impact of tax policy on her marginal dollar, where her reference point is herself. We calculate this contraction factor between the median income and bottom income group, and the median income and top income group for each U.S. state and year from 1979 to 2008. To address potential endogeneity of tax policy to economic conditions in the state, we use a set of exogenous instrumental variables (IV). We estimate our model using a generalized method of moments (GMM) approach developed by Blundell and Bond (1998) which refines the approach of Arellano and Bond (1991) for panel data that is persistent. The instrumental variables are the exogenous tax shocks identified by Romer and Romer (2009, 2010) and later refined by Mertens and Ravn (2013) at the national level, and their interactions with state-specific initial income inequality and initial propensity towards charity. The intuition is that state level conditions, specifically initial income inequality and initial propensity towards charity, affect tax policies to reduce inequality, and different states respond differently to tax shocks based on these initial conditions. Since we include state fixed effects to control for any time-invariant state level heterogeneity, only the exogenous tax shocks and interaction of the shocks with initial state conditions are exogenous 2

instruments. The main dataset of the project is a large sample of income tax returns, TAXSIM micro data, provided by the Internal Revenue Service (IRS) and made available to researchers by the National Bureau of Economic Research (NBER). Combining both TAXSIM micro data and simulation program, we calculate the tax rates and contraction factors for each state and each year. The state level per capita real GDP growth is created from the U.S. Bureau of Economic Analysis (BEA) data. We calculate labor market figures using data from the March Current Population Survey (CPS) Supplement. We collect data on small business activity from Business Dynamics Statistics (BDS), as made available by U.S. Census Bureau. BEA also provides us with state level consumption data. Additional datasets include state minimum wage data from Autor et al. (2014), state level expenditure data from U.S. Census Bureau’s State Government Finances, state level schooling from CPS, state level population growth from BEA, and unionization rates in each state from the State Union Coverage Density database provided by Hirsch et al. (2001). Our results are robust to a large set of controls that literature has suggested in prior work.4 The exhaustive specification, the instrumental variables approach that addresses potential concerns regarding endogeneity of tax policy to economic conditions, data on U.S. states which are relatively more homogenous than countries, and the relatively long length of the sample period, provide us confidence in our results – that tax policy’s effects of inequality reduction at different income percentiles relative to the median household have a heterogenous impact on economic growth. We find that one percentage point (pp) reduction in income gap between the median household and the 10th percentile household through taxation increases the GDP growth rate by 0.38 pp.5 The reduction of income gap between the above median households and the median household, however, has a negative effect on GDP growth. We find that if the income gap between the median household and the 90th percentile household is reduced by one pp, then GDP growth decreases by 0.47 pp. As discussed later, these results are obtained using an instrumental variables approach, 4 See

Barro (1991), Levine and Renelt (1992), Alesina and Rodrik (1994), Persson and Tabellini (1994), Perotti (1996), B´enabou (1996) and Sala-I-Martin (1997) among others. 5 The mean contraction factor in the sample is 9.1 pp and the mean state level GDP growth rate is 1.46 pp; a one pp increase at the mean to 10.1 pp leads the state level GDP growth rate to increase to 1.84 pp.

3

that controls for marginal tax rates, state fixed effects, and other important economic characteristics. Regarding components that drive this difference in state GDP growth rate, we find that one pp reduction in income inequality through tax policy between below median and median household increases the average employment rate by 0.29 pp.6 In contrast, employment rate is reduced by 0.32 pp for each pp increase in contraction factor between above median households and median households. On the intensive margin, the growth rate of average annual hours worked reduces by 0.18 pp for each pp reduction in income inequality through tax policy between above median and median household. The contraction factor between below median and median household does not have a statistically significant effect on the growth rate of hours worked. The second GDP component we investigate is business activity. We find that one pp increase in income inequality between above median households compared to median households reduces growth of the number of small businesses by 0.28 pp.7 Reduction of income inequality between below median households and median households encourages business activity as one pp reduction in below median household income inequality with respect to median households increases growth rate of the number of small business by 0.25 pp. Finally, one pp increase in contraction factor between below median income households and median income households increases per capita consumption growth by 0.43 pp. This result shows that tax policy to reduce inequality has a demand side effect as well.8 We also find that the sen6 Our

paper does not study the impact of specific programs on particular population groups. Meyer and Rosenbaum (2000) document the magnitude and timing of various social and tax policy changes in recent years that have encouraged work by single mothers, and provide causal evidence of tax effects on employment. Meyer and Rosenbaum (2001) show that a large share of the increase in work by single mothers during 1984–1996 can be attributed to the Earned Income Tax Credit (EITC) and other tax changes, with smaller shares for welfare benefit cuts, welfare waivers, training programs and child care programs. Eissa and Hoynes (2004) study the impact of EITC on labor market participation of married couples, and finds that EITC is reducing labor supply of married women, while increasing that of married men. Michalopoulos et al. (2005) shows that, in a social experiment setting, financial incentives increased full-time employment, earnings, and income. Eissa and Hoynes (2006) and Eissa et al. (2008) find that labor supply responds to EITC, specially on the extensive margin. 7 Aghion et al. (2015) find causality from innovation to top income inequality in the U.S. over the past decades. We focus only on small business establishments in this paper because we study the impact of personal income taxes, and many such businesses file taxes at the personal income tax rates using the S corporation structure. See http://www. irs.gov/Businesses/Small-Businesses-&-Self-Employed/S-Corporations for more details regarding S corporation status. 8 Meyer and Sullivan (2004) examine the consumption patterns of single mothers and their families from 1984– 2000, and find that material conditions of single mothers have improved slightly due to tax and welfare programs. We

4

sitivity of consumption of durable goods to tax policy is higher (almost three times) than that of non-durable goods and services, which is intuitive. Our results show that reduction of income inequality between low income households and median income households through tax policy supports economic growth. However, we also document that reduction in income gap for above median households with respect to median households through taxation hurts economic growth. These results support the incentives based argument as formalized by Mirrlees (1971), Okun (1975) and Becker (2011) and also support the arguments that taxation can have positive effects if it is used to provide social insurance for the low income households (see B´enabou (2000), Saint-Paul and Verdier (1993, 1997)). Becker (2011) has argued that “good inequality” motivates individuals to exert much effort, including creative effort, whereas “bad inequality” reduces efficiency, productivity, and utility. As far as we know, this asymmetric effect of redistributive tax policy along the income distribution has not been shown empirically before this paper. Our results contribute to the literature on tax policy and economic growth. Theoretical predictions regarding impact of taxes on economic growth are mixed.9 Thus, the question is primarily an empirical one. Empirical literature has investigated the effects of taxation on economic growth within U.S., across U.S. states, and across countries. Using U.S. post-WWII data, recent studies find that a positive change in taxes has a negative impact on GDP growth.10 Several studies focus on state level taxes and economic growth. Based on annual data from 1965 to 1979 for 48 states, Helms (1985) shows that state and local tax increase is bad for economic growth when the revenue is used to fund transfer payments and is good for economic growth when the revenue is used to fido not study consumption responses of specific policies for individual population groups. 9 Mirrlees (1971), Okun (1975) and Becker (2011) argue that taxes reduce economic growth by dampening incentives to work and invest. Barro (1990) shows that taxes can be beneficial for economic growth in presence of public goods, but as government size increases, the benefits are outweighed by the costs of taxation. B´enabou (2000) shows that taxation can help growth if they finance public investment. Saint-Paul and Verdier (1993, 1997) shows that higher health and education spending benefits the poor, helping to offset labor and capital market imperfections. 10 Blanchard and Perotti (2002) finds positive tax shocks have negative effects on output in the U.S. from 1947 to 1997. Romer and Romer (2010) find that a tax increase of 1 percent of GDP implies a 3 percent fall in output in the U.S. economy from 1947 to 2007. Mertens and Ravn (2013) find that a large short run output effects of tax shock and personal income tax cuts being more effective in creating jobs and stimulating consumption in the short run than cuts to corporate profit taxes from 1950 to 2006. Barro and Redlick (2011) find a large and significantly negative impact of an increase in marginal tax rates on U.S. annual economic growth over the time period 1950 to 2006.

5

nance improved public services (such as education, highways, and public health and safety). Reed (2008) finds that taxes used to fund general expenditure are associated with significant, negative effects on income growth from 1970 to 1999 over U.S. states. There are also a large number of studies using cross-country data, which generally find negative effects of tax increase on output.11 Building upon this literature, we show that redistributive policies have heterogenous effects on households. Our empirical analysis relies on a large sample of administrative record of actual tax return data in the U.S. states over three decades. Thus, our analysis suffers less from measurement error issues that can exist in cross-country data comparisons.12 Furthermore, compared to cross-country differences, the economic development and institutions are more homogenous across U.S. states. This allows us to assume that the same underlying economic relationship between GDP growth, tax policy, and inequality exists across states, which is more defensible than a similar assumption regarding countries.13 A large literature has also studied the relationship between inequality and economic growth, and has also provided mixed predictions. Inequality can reduce economic growth through the following three channels. First, political economy theory suggests that greater inequality is conducive to the adoption of distortionary redistributive tools and growth-retarding policies, which hurt economic growth (See, for example, Alesina and Rodrik (1994), Persson and Tabellini (1994), and Benhabib and Rustichini (1996)). Second, in presence of financial market imperfections, higher inequality exaggerates the adverse effects of credit constraints on human capital accumulation, reducing growth (See, for example, Galor and Zeira (1993) and Galor and Moav (2004)). A similar argument can also be applied to business growth. Third, Murphy et al. (1989) show that a more equal society with homogenous tastes helps to create large market for domestic manufactures. 11 See

Koester and Kormendi (1989), Easterly and Rebelo (1993), Mendoza et al. (1997), Miller and Russek (1997), Kneller et al. (1999), Lee and Gordon (2005), Alesina and Ardagna (2010), Gemmell et al. (2011), Arnold et al. (2011), Ferede and Dahlby (2012), Padovano and Galli (2001), among others. 12 Measurement error can cause estimation bias. For example, if more unequal society underreport their inequality statistics and also grow slower, cross-country estimates of the impact of inequality on growth may suffer from a negative bias. 13 This is because the relationship between inequality, tax policies and economic growth may depend on time-varying developmental stage or political environment of countries.

6

On the other hand, greater inequality might increase growth. Higher inequality provides the incentives to work harder, invest more and undertake risks to take advantage of high rates of return (See, for example, Mirrlees (1971), Lazear and Rosen (1981), Rebelo (1991), Heckman et al. (1998) and Guvenen et al. (2013)). Higher inequality can also foster aggregate savings and therefore capital accumulation, because the rich have a lower propensity to consume (See Kaldor (1956) and Bourguignon (1981)). Our results show that both sets of predictions above may be at work simultaneously. Our results have important implications for tax policy. Kuznets (1955) conjectured that inequality increases in the early stages of economic development for a country (due to industrialization and urbanization). As industries attract a larger fraction of the labor force, inequality starts decreasing. As observed by Aghion and Williamson (1998), up to the 1970s, the prediction of Kuznets (1955) was corroborated by data. However, in recent times, wage inequality between and within groups of workers has been increasing.14 These evidence provide support for action by policy makers to reduce income inequality.15 Our results demonstrate that reduction of income inequality between all groups of income may not have similar effects. When income inequality is reduced between above median households and median households, economic growth may decrease. Our results do not say what the optimal tax rate at various levels should be.16 We just document the asymmetric nature of redistributive tax policies on economic growth, and argue that policymakers should not assume that reduction of income inequality would translate into economic growth, as the opposite occurs in some cases. The rest of the paper is structured as follows: Section I introduces a measure that captures the reduction in income distribution through tax policy. Section II discusses the framework of the analysis. Section III describes the data. Section IV describes the results. Section V discusses 14 Katz and Murphy (1992) document increasing wage inequality in recent years,

and show that rapid secular growth in the demand for more-educated workers, “more-skilled” workers, and females appears to be the driving force behind observed changes in the wage structure. Juhn et al. (1993) find that the trend towards increased wage inequality is apparent within narrowly defined education and labor market experience groups as well. Piketty and Saez (2003) show that top shares of income and wages in the United States display a U-shaped pattern over the 20th century. 15 Blundell et al. (2008) show that such redistributive tax policies in U.S. have helped reduce consumption inequality, even in presence of income inequality. 16 See Diamond and Saez (2011) for a survey of the literature on optimal taxation. Recent papers include Conesa et al. (2009) and Holter et al. (2014) among others.

7

robustness tests, and Section VI concludes.

I

Taxes, Incentives and Contraction Factor

A

Progressive Taxation and Tax Credits

Income inequality reduction through tax policy is obtained through two main mechanisms: progressive taxation, which leads to higher marginal taxes for higher income households, and tax credits which provide negative tax rates for the lower income households. The source of variation in taxation between states and within states comes from multiple sources. While federal taxes are the same for households in nominal terms, irrespective of where they reside in United States, state income tax schedules vary (i) by state and (ii) within each state over time.17 Further, the actual net taxes paid by a household depends on (iii) the demographics of the households which also varies within and between states, and over time. These differences in demographics affect income tax rates at federal and state level and also affect the tax credits available to each household. Our data on tax returns is the NBER’s TAXSIM micro data, prepared by the Statistics of Income Division of the Internal Revenue Service (IRS) for public use. TAXSIM data is a large stratified random sample of total 3, 544, 410 actual tax returns from the IRS administrative records for each states and for each years from 1979 to 2008 (except 1982, where total tax liability is not available). It contains detailed information on Form 1040, such as adjusted gross income, total federal and state tax liability, and tax credits.18 We use adjusted gross income (AGI) as our measure of pretax income (hereafter referred to as “income”), which is defined by the IRS as total income (line 22, Form 1040) minus statutory adjustments (lines 23 through 36, Form 1040).19 The total tax 17 For state income tax rates over time, please follow the link by The Tax Foundation: http://taxfoundation. org/article/state-individual-income-tax-rates. 18 The IRS does not identify the state for the returns with above $200, 000 adjusted gross income. NBER reassigns these records to states, such that the number of those returns by state matched figures provided by the Joint Committee on Taxation. However, since our top income bracket is the 90th percentile, which does not cross $200, 000 of income in the sample, this issue is not a concern. 19 Total income is all positive sources of income less negative amounts, including wages and salary, taxable interest and dividends, net income from a business, etc. Statutory adjustments include educator expenses, certain business expenses, health savings account deduction, moving expenses, IRA deductions, etc.

8

liability (hereafter “tax”) includes all income tax after credits, self employment tax, and any tax adjustments from previous years. A detailed description of all the components of tax according to the IRS is in Section A of the Appendix. The tax data help show that progressive taxation and tax credits “compress” the income distribution. To illustrate the “compression” effect, we compare the national before-tax income distribution and after-tax income distribution in Figure I. The figure utilizes the full sample of TAXSIM tax return data for all U.S. states for the years 1979 to 2008, in 2009 U.S. Dollars.20 The mean pre-tax income (AGI) in 2009 U.S. Dollars is $49, 548, with inequality measured by the standard deviation of log income of 1.188. The darker region shows that the after tax distribution is shifted to the left and contracted – it has a lower mean ($42, 508) and smaller standard deviation of log income (1.139). Measured in terms of reduction of variance of log income, the income inequality on average is reduced by about 8% through taxation. The progressive taxation and tax-credit policies, in effect, are moving the after-tax income of both the lower and upper income households towards the median to reduce inequality. Thus, the median income household is a natural reference point to measure the impact on inequality reduction through tax policy. In Section B, we measure the extra tax liability for each additional dollar earned by the lower income and upper income households compared to the reference household (the median income household) respectively.

B

Contraction Factor

Before we can estimate the impact of reduction in income inequality through tax policy on economic growth, we need to estimate the impact of tax policy on income distribution. In this section, we propose a simple measure that evaluates the changes in income inequality induced by the progressive tax policy. Let Incomet (i) denote the average pre-tax income for households in the i percentile in the national income distribution in year t, and Taxs,t (i) denote the associated total income tax liability 20 Throughout

this paper, all monetary figures are deflated to 2009 USD using the GDP deflator from BEA.

9

in year t and state s. The before-tax income inequality is then measured by the difference between pre-tax income percentile i and j in a given year t, Incomet (i) − Incomet ( j). Similarly, the aftertax income inequality in year t and state s is given by (Incomet (i) − Taxs,t (i)) − (Incomet ( j) − Taxs,t ( j)). Using the median income household as the reference point, we define the reduction in the after-tax income inequality as a fraction of the before-tax income inequality, referred to as the “contraction factor” Cs,t (i), as follows: (Incomet (i) − Taxs,t (i)) − (Incomet (50) − Taxs,t (50)) Incomet (i) − Incomet (50) Taxs,t (i) − Taxs,t (50) = Incomet (i) − Incomet (50)

Cs,t (i) ≡ 1 −

(1)

where i denotes national income percentile. In a progressive tax system, Taxs,t (i) − Taxs,t (50) and Incomet (i) − Incomet (50) share the same sign. Further, as long as taxes are used to reduce income inequality, Cs,t ∈ [0, 1). In the future section, we suppress the subscript s,t of Cs,t (i) for simplicity. The contraction factor measures the taxation induced reduction in the income gap between the income percentile of interest and the median income households. In particular, one percentage point increase in the contraction factor C(i) is the same as one percentage point reduction in the after-tax income gap relative to before-tax income gap between the ith and median percentiles. The contraction factor can also be interpreted as the additional average income tax paid for each additional dollar earned by a person at the ith income percentile, compared to the median income. Our variable is different from the average marginal tax rate (See Barro and Redlick (2011)) that affects an individual’s decision to earn an additional dollar given her present income and income tax rate. Barro (1990) and Barro and Redlick (2011) among others, have discussed the importance of marginal tax rates for economic growth. In contrast, contraction factor aims to capture the impact of tax policy induced reduction in income inequality, with respect to the median household as the reference household, on economic growth. Our framework for analysis considers the impact of contraction factor for above median households and below median households separately, and also controls for changes in marginal tax rate over time for the reference household. Additional discussion of the framework of the analysis, including the empirical specification, instrumental 10

variables and system GMM approach, is in Section II. Our main analysis will rely on two contraction factors: C(90), that measures the effects of taxation on after-tax income gap reduction between 90th percentile household (high income household) and the median household; and C(10) which evaluates the reduction of after-tax income gap through taxation between the median household and the 10th percentile household (low income household). The top and bottom panels of Figure C.I graphically demonstrate an increase in contraction factors C(90) and C(10) respectively. In our robustness check section (Section V), we also show that our analysis yields similar results if we use different thresholds for above median and below median contraction factors such as C(80) and C(20). We use the TAXSIM tax return data to construct contraction factors at different income percentiles for each state and for each year from 1979 to 2008.21 In a given year, using national income distribution, we create a 10 percentile income bracket centered around the national pre-tax income percentile i. For each state, we then calculate the state-level average tax liabilities associated with the national income percentile i by averaging all tax liabilities associated with incomes that fall within the specified 10 percentile national income band.22 Similarly, for each state we also calculate the average income that are within the 10 percentile national income bracket. Finally, we calculate contraction factors using Equation 1. The thought experiment here is to take a group of households that are representative at the national income distribution level, and assign them to different states, imposing the state specific tax burden. Therefore the measured contraction factor in a given year captures the effects of statelevel variation in tax policy to reduce income inequality. By fixing the income distribution across states in a given year, we also reduce the concern of geographic mobility of labor force where, for example, a 90th percentile person from a relatively poorer state moves to a neighboring relatively 21 Data

from 1982 is not included as tax liability is not available in the microdata. We exclude Washington D.C. because it is not a state. We also impose a sample size restriction, dropping any state-year point if that state does not have at least 500 observations in that year. Lastly, due to small sample size and large differences between moments of TAXSIM data and IRS full SOI sample, we also exclude Alaska from our analysis. Table B.I reports the availability of data in our selected sample from TAXSIM. 22 For example, let us consider creating the median income for each state in a given year. We first calculate the 45th percentile and 55th percentile of the national income distribution as the lower and upper bound of the 10 percentile band around the median. Then, within each state, we average the adjusted gross income and total tax liability of all returns that fall between the 45th and 55th percentile of the national income distribution.

11

richer state, moving into a lower income percentile in the process. Using a national pre-tax income distribution allows us to isolate the impact of the variation in tax burden attributed to the changes in the tax policies and demographic distribution in each state over time, from confounding effects such as labor movement and state income distribution differences. As shown in our robustness check section (Section V), the alternative approach of using state level income distribution as the starting point for contraction factor calculation also yields similar results. Our identification strategy requires variation in contraction factors over time and across states. Figure II shows the distribution of contraction factors in our analysis across states for the sample period. Figure III plots the standard deviations of contraction factors for each state.23 Table B.II in the Appendix summarizes the average one period lagged contraction factors as well as average per capita annual GDP per capita growth rates for the 49 states in our analysis for the sample period of 1980–2009. The figures and table suggest significant variation in contraction factors over time and across states. Table B.III in the Appendix reports correlation between state level Gini coefficient in a given year with the contraction factors C(10) and C(90) in the state in that year. As expected, a higher level of income inequality measured by Gini correlates with less contraction. We also note that the Gini coefficient correlates more strongly with the ratio of income of 90th and 50th percentile households, than with the ratio of 50th and 10th percentile households. This is intuitive because of the positively skewed income distribution. This observation provides additional motivation for using contraction factors which allow distinguishing the impact of above and below median households on the economy.

II

Framework of Analysis

This section discusses our empirical specification and estimation strategy. 23 Figure

C.II in the Appendix plots the levels of contraction factors for each state averaged over the same time

period.

12

A

Empirical Specification

In our paper, we focus on the impact of reduction in inequality induced by tax policy on economic growth. As discussed before, this effect where households compare themselves with other households is different from the marginal tax rate that affects an individual’s decision to earn an additional dollar given her present income and income tax rate (See, for example, Barro and Redlick (2011)). Furthermore, we distinguish the inequality reduction through taxation between the upper and lower ends of the income distribution, and allow their impact on economic growth to be different. In particular, the reduction of income inequality through taxation in lower and upper income distribution are measured by the below median and above median contraction factors measures respectively. This is because households who benefit from income distribution compression (lower than median income households) may respond differently than households that are adverse impacted by income distribution compression (upper income households). We estimate the effects of below median and above median contraction factors on state-level annual per capita economic growth, using the following specification:

log GDPs,t − log GDPs,t−1 =κ1 · logCs,t−1 (10) + κ2 · logCs,t−1 (90) + γ1 · (MT Rs,t−1 (50) − MT Rs,t−2 (50)) + γ2 · MT Rs,t−1 (50) + h1 · log

Income(90) Income(50) + h2 · log Income(10) Income(50)

+ α · log GDPs,t−1 + Xs,t−1 β + δs + εs,t−1

(2)

where Cs,t−1 (90) is previous years’ contraction factor between the 90th percentile household and the median household; we use it as a measure of income inequality reduction between high income households and median income households. Cs,t−1 (10) is previous year’s contraction factor between the 10th income percentile household and the median household; it represents the income inequality reduction between median income households and low income households. Xs,t−1 is a vector of additional controls and δs denotes state fixed effects. Banerjee and Duflo (2003) show that an empirical specification that includes both the level 13

and change of the inequality measure is consistent with both the endogenous political economy argument and imperfect financial market mechanisms. Hence, we include log Income(90) Income(50) and log Income(50) Income(10) to capture the level of income inequality between the 90th percentile income household and median household, and median household and 10th percentile household respectively. The change in income inequality is captured by the contraction factors themselves. As discussed before, previous literature has also shown that changes in marginal tax rates have important implications on economic growth because they affect households’ current choices on employment and consumption compared to the previous period.24 As marginal tax rates increase, incentives to work decline for the same household irrespective of the tax rate of other households. Therefore, following Barro and Redlick (2011), we include both levels and changes in median Marginal Tax Rates (MT R(50)) in our estimation equation. Specifically, for each state s we control for changes in marginal tax rate at the median income level between year t − 1 and t − 2 (MT Rs,t−1 (50) − MT Rs,t−2 (50)), and marginal tax rate at the median income level in year t − 1. Lastly, Mankiw et al. (1992) show that a neoclassical growth model (see Solow (1956)) augmented with accumulation of human capital as well as physical capital yields an empirical specification where GDP growth rate log GDPs,t − log GDPs,t−1 depends on level of GDP in the previous period (i.e. log GDPs,t−1 ), accumulated human capital level measured by average schooling level (log Schools,t−1 ), and population growth (ns,t−1 ). Within the framework of neoclassical growth model, (1 + α) in Equation 2 measures the rate of convergence in economic growth. When the economic growth is around the steady state, we expect α to be a small negative number (which is what we find). Following Mankiw et al. (1992), we include state specific human capital level measured by average schooling level (log Schools,t−1 ) and state population growth (ns,t−1 ) into Xs,t−1 which represents the set of controls. We also control the state government per capita expenditure gI

s,t−1 on goods and services (log GDP ), because Helms (1985) finds that state expenditure on public s,t−1

services and investment (such as highways and education) is good for economic growth. Finally, Card and DiNardo (2002) argue that the changes in non-market factors such as changes in min24 Literature

includes, but is not limited to Mirrlees (1971), Barro (1990), Barro (1991), Alesina and Rodrik (1994) and Barro and Redlick (2011).

14

imum wage and declining unionization can explain the changes in inequality of wages. Hence, we also include both these factors in the growth regression to account for their effects on income inequality. In sum, Xs,t−1 includes a constant term, state level human capital measured by average schooling levels (log Schools,t−1 ), state population growth (ns,t−1 ), changes in effective state level minimin ), and changes in the fraction of labor force covered by unions (∆Union mum wage (∆Ws,t−1 s,t−1 )

(see Table I, Panel B for details).

B

Estimation Strategy

Our key parameters of interest are the coefficients of contraction factors, κ1 and κ2 , in Equation 2. We first report the estimation results using Ordinary Least Squares (OLS) with state fixed effects to address unobserved heterogeneity among states. A potential concern with OLS is that tax policy may be an endogenous response to the economic conditions. To address this potential endogeneity concern, we use a set of exogenous instrument variables and estimate our model using a generalized method of moments approach developed by Blundell and Bond (1998). The exogenous instrument variables are the tax shocks at the national level, and their interactions with state-specific initial income inequality and initial propensity towards charity. We use the exogenous tax liability shocks narratively identified by Romer and Romer (2009, 2010) and later refined by Mertens and Ravn (2013) to form exogenous instruments for changes in our contraction factors and marginal tax rates (see the appendix in Mertens and Ravn (2013) for more details). Romer and Romer (2009) identified a series of tax liability changes that are exogenous to economic growth in the U.S. from 1945 to 2007. Using a narrative approach based on congressional reports and other government administrative data, Romer and Romer (2009) classify the tax liability changes as exogenous if the motivation for the legislative action is either arising from inherited deficit or from ideological concerns.25 Mertens and Ravn (2013) further extend the 25 Romer and Romer (2009) classify every significant tax bills into one of the four categories based on the underlying

motivation for the tax change: responding to a current or planned change in government spending, offsetting other

15

analysis by distinguishing between changes in personal and corporate income tax liability, and by distinguishing between unanticipated and anticipated tax changes on the basis of the implementation lag. These tax shocks, identified by Romer and Romer (2009) and refined by Mertens and Ravn (2013), are exogenous to the current state of the economy, because they relate to unanticipated tax liability changes that are not a response to the growth prospects of the economy.26 Figure C.III in the Appendix depicts personal income tax shocks given by Mertens and Ravn (2013) together with contraction factors averaged across states from 1979 to 2007.27 During this time period, the largest exogenous change in personal income taxes relates to the Jobs and Growth Tax Relief Reconciliation Act of 2003, which includes across-the-board reductions in marginal tax rates as well as increases in child credit. The largest exogenous increase in personal income taxes relates to Omnibus Budget Reconciliation Act of 1993, which increased income taxes, mostly for higher earners. The impact of the exogenous tax shocks at the national level varies based on conditions present at the state level. We focus on two initial conditions which are relevant to contraction factors: initial state level inequality measures (as measured by 90th percentile/50th percentile and 50th/10th percentile log income ratio in 1979), and initial attitude towards charity (measured by the share of charity to income ratio in 1979).28 The intuition is that initial inequality in a state affects tax policy, and that if households in a state are interested in charity, then that affects their attitude towards using tax policy to reduce inequality. The interaction of the tax shocks with these state specific influences on economic activity, reducing an inherited budget deficit, and attempting to increase long-run growth. Romer and Romer (2009) classify the last two types of tax changes as exogenous to the current state of the economy in the sense that they are not a response to the growth prospects of the economy. 26 Furthermore, in practice, these tax liabilities changes are often related to the changes in progressivity of the tax schedule and thus correlate with changes in contraction factors as well as marginal tax rates. In particular, Barro and Redlick (2011) use the changes in exogenous tax liabilities identified by Romer and Romer (2009) to form an instrument for changes in the marginal tax rate in the GDP growth regression analysis. 27 In particular, Mertens and Ravn (2013) create the personal income tax shock as the narratively identified exogenous personal income liability change divided by previous personal taxable income, z pi ; the corporate income tax shock is defined as narratively identified exogenous corporate income tax liability changes scaled by previous period corporate profits, zci . Mertens and Ravn (2013) provide both quarterly and annual data; we choose the annual level data for our analysis. 28 To develop a measure of attitude towards redistribution, we collect total charitable contributions, cash and assets, as recorded on tax returns from the TAXSIM data. We scale the total charitable contributions by the total income in the state.

16

initial conditions satisfies the exclusion restriction because in presence of state-fixed effects, only the time varying effect of the exogenous tax shocks interacted with initial state conditions (which are time invariant) is the instrument. Detailed discussion regarding the relevance of these exogenous instruments to our contraction factors is in Section IV.A.2 where we discuss empirical results. Table B.VI in the Appendix reports the results. In addition to endogeneity of tax policy to state level economic growth, one may be concerned about the possibility of Nickell bias in Equation 2 given a small T = 30 years and large N = 49 states. Hence, we implement our estimation using the system GMM method developed by Blundell and Bond (1998). The system GMM approach is an improvement upon the difference GMM proposed by Arellano and Bond (1991). Such IV GMM methods have become increasingly popular in the empirical literature on inequality and economic growth (See, for example, Forbes (2000), Ostry et al. (2014), Cingano (2014)). Besides exogenous instrument variables, system GMM also utilizes lagged values of control variables as internal instrumental variables in the estimation. Specifically, for our internal instruments, we use lagged dependent variable and predetermined control variables at t − 3 and t − 4.29 Our estimation results are robust to using more lags as internal instruments. In our estimation table, we also report the Arellano-Bond test for autocorrelation, as well as the test for over-identification and validity of instrument variables, all of which provide confidence in our results.

III A

Data Data Description

We utilize multiple data sources to construct our state-year level panel dataset. Data include information on GDP growth, income distribution, taxes, and other economic control variables. The state 29 As

shown in our specification, the controls for our primary estimation are lagged one period: inequality levels relating the 50th and 10th percentile incomes and 90th and 50th percentile incomes, log of GDP per capita, log of the average years of schooling, the change in state government expenditure, and labor market conditions, such as change in effective minimum wage and unionization rates. We do not use second order lags, i.e. Xs,t−2 , as internal instruments because we are concerned about the possible first order correlation of the error terms. However, our results are robust if we use Xs,t−2 as internal instruments.

17

level GDP per capita growth rate is constructed from the U.S. Department of Commerce Bureau of Economic Analysis (BEA). As discussed in Section I, we use TAXSIM data to obtain information on contraction factors, income percentiles, and the corresponding tax liability and marginal tax rates in each state and each year. We also utilize the TAXSIM data to gather data on initial income inequality and charitable contributions in 1979 to create variables on attitudes towards inequality and redistribution. Additional datasets include (i) U.S. Census Bureau for state government finances, (ii) Mertens and Ravn (2013) for narratively identified personal and corporate income tax shocks exogenous to economic growth, (iii) Current Population Survey (CPS) data for state level schooling and labor market variables, (iv) unionization rates from the State Union Coverage Density database (Hirsch et al. (2001)), (v) minimum wage information from Autor et al. (2014), (vi) the Business Dynamics Statistics (BDS) database for information on small businesses, and (vii) the BEA for personal consumption expenditures and population growth by state. Table I Panel A, summarizes all the dependent variables in our analysis and their data sources. Panel B of Table I lists our explanatory variables and their sources. The exogenous instruments have already been discussed in Section II along with the estimation strategy, below we discuss dependent and explanatory variables.

A.1

Dependent Variables

GDP Growth The primary dependent variable in our analysis is the state level log change in real GDP per capita (log GDPs,t − log GDPs,t−1 ). The state level real GDP is the state level real chained GDP in 2009 dollars from BEA.30 BEA also provides the state level population estimates for each year, which we use to create GDP per capita, and consequently the annual log change in GDP per capita. 30 We collect real chained GDP in 2009 dollars for 1997 to 2009, and real chained GDP in 1997 dollars from 1987 to 1997 from BEA. We use available changes in quantity indices (from BEA) to extend our state GDP series in 1997 dollars backwards for the period of 1977 to 1986. We then convert the pre-1997 real chained GDP series from 1997 dollars to 2009 chained dollars by using the ratio of 2009 dollar GDP to 1997 dollar GDP in 1997, where both series are available.

18

Labor Supply We also investigate how contraction factors impact components of GDP. In particular, we look into both the extensive and intensive margins of the labor supply. Our data on labor supply comes from March CPS for individuals aged 18 to 64. For the extensive margin, we consider an individual to be employed in year t if she works at least 26 weeks in a year and at least 20 hours per week. The employment rate (E˜s,t ) in each state is calculated as the fraction of employed individuals among the population aged 18 to 64. Furthermore, to analyze the intensive margin, we also calculate the average annual hours worked among those who are employed in each state (Hrss,t ). For each of these two variables we construct state specific change variables from the previous year (∆E˜s,t = E˜s,t − E˜s,t−1 , ∆ log Hrss,t = log Hrss,t − log Hrss,t−1 ). Small Business Activity The second component of GDP that we consider is small business activity. We use the Business Dynamics Statistics (BDS) database to collect state level data on small businesses, which are referred to as establishments. An establishment is a fixed physical location where economic activity occurs. We focus our attention on the number of establishments (BDS variable: estabs) of size 5 to 19 and of size 5 to 49. We then investigate the state level growth rate of the number of establishments in each of the two size buckets (∆ log estabss,t = log estabss,t − log estabss,t−1 ). Consumption The last GDP component we consider is personal consumption expenditures per capita (PCE). BEA provides data on nominal consumption for several categories for each state from 1997 to 2009. We deflate the nominal consumption by the GDP deflator and use the same population denominator from real GDP per capita (both available from BEA) to calculate real PCE per capita for total consumption, consumption of durable goods, and consumption of nondurable goods and services. Consumption growth rates are measured by the changes of log consumption.

19

A.2

Explanatory Variables

Our main explanatory variables in are two specifications are one period lagged log contraction factors. We focus on the effect of reducing income inequality between the 10th percentile household and the median household (logCs,t−1 (10)) and the effect of reducing income inequality between the 90th percentile household and the median household (logCs,t−1 (90)) on our various dependent variables. In addition, we add several control variables based on previous literature.

Marginal Tax Rates Using the TAXSIM micro tax return data as an input, we utilize NBER’s TAXSIM simulation program to compute overall marginal tax rates (including federal, state, and FICA tax) for each return, and then calculate the median household’s marginal tax rate MT Rs,t−1 (50).31 Furthermore, when calculating the changes in marginal tax rate, ∆MT Rs,t−1 (50), we use the t − 2 national income distribution to calculate marginal tax rates for both t − 1 and t − 2 following Barro and Redlick (2011). This strategy eliminates any shifts in income distribution that might move a household into a different tax bracket, and helps ensure that the changes in marginal tax rates are specifically due to changes in tax policies.

State Government Spending Data on state and local government spending is available from U.S. Census Bureau’s State Government Finances. Following Barro and Redlick (2011), we consider only the government spending on goods and services, excluding transfers or interest payments. Specifically, government spending on goods and services includes expenditures on education, highways, police protection, and correction. We deflate the state government spending on goods and services by the GDP deflator and create state government spending per capita (gIs,t ).32 Government spending per capita enters gI

s,t−1 ). the control variables in our estimation as a log fraction of state GDP per capita (log GDP s,t−1

31 We

utilize the TAXSIM model’s “wages” option to calculate marginal tax rates. government expenditure include the following detailed categories: higher education, elementary and secondary education, educational administrative services, libraries, highways, police protection, traffic safety, and correction. 32 The

20

State Inequality Measures To control for the previous year’s level of inequality within each state, we use the TAXSIM state income distribution in calculating two ratios measuring the 50th percentile income in comparison (50)

Income

to the 10th percentile income ( Incomes,t−1 (10) ), and the 90th percentile income in comparison to s,t−1

Income

(90)

the 50th percentile income ( Incomes,t−1 (50) ). In our specification we use the logs of these ratios s,t−1

Income

(50)

Income

(90)

(log Incomes,t−1 (10) , log Incomes,t−1 (50) ). Table B.IV summarizes the state fiscal and inequality controls s,t−1

s,t−1

for the 49 states in our analysis.

State Human Capital We also control for the level of schooling in each state as a measure of the human capital stock. Specifically, we use CPS data to calculate the average years of schooling (with the maximum value being 17, signifying more than 4 years of college) among the 18 to 64 year old population in each state. Schooling then enters our specification as a one year lagged log value (log Schools,t−1 ). State Population Growth Using the population information from BEA used to calculate per capita GDP, we create a population growth variable, ns,t = log Pops,t − log Pops,t−1 . Specifically, it enters our estimation as a one year lag.

State Labor Market Institutions Changes in labor market institutions such as minimum wages and unionization rate are also immin , is obtained from portant factors in impacting GDP growth. The state minimum wage data, Ws,t

Autor et al. (2014). If the state specified minimum wage is lower than federal minimum wage, we use the federal minimum wage as the effective state minimum wage. The effective state minimum wage is then deflated using GDP deflator, and it enters our controls as the annual lagged growth min = rate of the effective state minimum wage, from year t − 2 to t − 1 (∆Ws,t−1

21

min −W min Ws,t−1 s,t−2 ). min Ws,t−2

The unionization rates (Unions,t ) are the fractions of nonagricultural wage and salary workers covered by a collective bargaining agreement and are available from the State Union Coverage Density database (Hirsch et al. (2001)). Unionization enters our controls as the lagged change in unionization rates (∆Unions,t−1 = Unions,t−1 −Unions,t−2 ). Table B.V summarizes the remaining controls for the 49 states that enter our IV model.

B

Summary Statistics

Table II provides summary statistics for all of the variables used in this paper. All statistics shown are pooled across all states and years from 1980 to 2009 for dependent variables and 1979 to 2008 for independent variables, as we measure the effect of our lagged regressors on our variables of interest. All the nominal figures are deflated to 2009 dollars using GDP deflator. Panel A reports summary statistics for all the dependent variables in our analysis. The equal weighted average real GDP growth from 1980 to 2009 in U.S. states is 1.5% with a median growth rate of 1.7%. The equal weighted pooled standard deviation of growth rate is significant at 2.8%. Regarding small business activity variables, the number of establishments of size 5 to 19 grow at a 1.47% rate and establishments of size 5 to 49 grow at a 1.54% rate. Our calculations on labor supply show that, on average, 70.1% of the working age population are employed, with an average annual change in employment rate of −0.06%. We also note that the employed population works on average 2, 073 hours per year, with an average annual change of −0.06%. Lastly, BEA data shows that the average real total personal consumption expenditure (PCE) per capita is $30, 217 in 2009 U.S. Dollars. The growth rate of real PCE per capita from 1998 to 2009 was 1.5%. The average real PCE per capita on durable goods is $3, 871 and the annual growth rate is −0.31%. On average, real PCE on nondurable goods and services per capita is $26, 346 in 2009 U.S. Dollars, and the average annual growth rate is 1.7%. Panel B reports summary statistics pooled across all states and years from 1980 to 2009 for explanatory variables, controls, and instruments. These variables enter our estimation as lagged measures, and the subscript denotes the applied lag. The contraction factor between median and

22

10th percentile household is 9%, and that between 90th and median household is almost double of that at 17%. Thus, the marginal tax rate almost doubles on the additional income between the two groups. The other key explanatory variable we utilize is the lagged annual change in median marginal tax rate, which has an average of 0.11%, and ranges from a minimum of −5% to a maximum of 4.9% across states. We also include a large number of controls identified in the literature as possible determinants of GDP growth rate. This is in addition to state fixed effects, which are included in our OLS and instrumental variables specification. Along with the change in median marginal tax rates, we also control for the lagged median marginal tax rate, which is an average of 35.4%. On average, the 90th percentile income is 3.2 times the median income, and the median income is on average 5.9 times the 10th percentile income. We note that states on average spend $1, 723 per capita on goods and services such as education, highways, police protection, and correction, and the average state spending as a portion of GDP is −3.1 in logs. The average population growth is 1%. On average, the effective state minimum wage is $6.18 and grows at a 0.1% annual rate.33 Lastly the average unionization rate is 16.8%, and the annual change in unionization is on average −0.4%. Our system GMM analysis with IV will employ exogenous tax shocks to personal and corporate income as identified by Mertens and Ravn (2013), as well as interaction terms with charitable contributions and inequality levels in 1979. The average personal income tax liability shock is 0.06% of the previous period personal taxable income, with a minimum of −1.1% and a maximum of 0.44%. The average corporate income tax liability shock is −0.04% of the previous period corporate profits, with a minimum of −3.28% and a maximum of 7.38%. We interact these tax shocks with the charitable donations in 1979, which were on average 1.54% of income. Furthermore, we interact tax shocks with the inequality levels of 1979, where the 90th percentile income was 2.7 times the median income, and the median income was 5.5 times the 10th percentile income. 33 For

the states where the minimum wage is lower than federal minimum wage, the federal minimum wage applies, thus the effective state minimum wage has a minimum of the federal minimum wage.

23

IV

Empirical Results

This section first conducts an estimation of the relationship between reduction of income inequality through tax policy and economic growth. For this, we use ordinary least squares (OLS) estimation with state fixed effects, and instrumental variables approach with the system GMM estimator (IV GMM) to gain additional confidence in our results. Then, we explore three important channels through which reduction of income inequality through tax policy affects economic growth: (i) Employment, (ii) Business activity, and (iii) Consumption.

A

Income inequality, tax policy and economic growth

Table III reports the results of the impact of contraction factor on economic growth using both OLS and IV GMM approaches.

A.1

Ordinary Least Squares Approach

We first discuss OLS results, and then the IV GMM results. Column (1) does not include any controls other than the below median contraction factor, and Column (2) adds the above median contraction factor. Column (3) includes median marginal tax rate variables and initial condition controls from 1979. Column (4) removes the initial condition controls, but adds state fixed effects. Column (5) is our most exhaustive OLS specification adding dynamic controls denoted by Xs,t−1 above in Equation 2 along with state fixed effects. Column (1) shows that for one percent decrease in income inequality at the mean, between the 10th percentile household and the median household, GDP per capita increases by 0.013 pp. If we scale up the numbers to 1 pp decrease in income inequality at the mean, i.e. increase C(10) to 10.14 pp, then according to Column (1), there is a corresponding change of 0.14 pp in the growth rate of GDP per capita.34 This is a statistically and economically significant result given that the average real GDP per capita growth rate in our sample is 1.5 pp, and hence a 1 pp change in C(10) 34 Since the mean value of contraction factor C(10) is 9.14 pp, one percent at the mean is 0.0914 pp. We can calculate the GDP per capita increase in growth rate as approximately 0.013 × 1/0.0914 = 0.14.

24

increases the GDP per capita growth rate to 1.64 pp. When we add the above median contraction factor in Column (2) this effect strengthens, as a one percent decrease in income inequality between the 10th percentile household and the median household results in a 0.026 pp increase in GDP growth. At the same time, by noting the coefficient of the contraction factor C(90), we find that if we reduce income inequality between the 90th percentile household and the median household, by increasing taxes by one percent on the 90th percentile household, GDP growth rate declines by 0.034 pp. These results become stronger as we add marginal tax rate variables and either initial condition controls in Column (3) or state fixed effects in Column (4). We note that one pp increase in median marginal tax rate from year t − 2 to t − 1 (as measured on t − 2 income distribution) correlates with a 0.68 pp decrease in average real GDP per capita growth rate in the presence of state fixed effects. State fixed effects help control for heterogeneity between states, both observed and unobserved. Following literature, Column (5) includes additional lagged inequality, human capital, fiscal, and labor institution variables that may also affect GDP growth rates in addition to state fixed effects. In this exhaustive specification, our results remain statistically and economically significant and similar to previous columns. Specifically, a one percent higher C(90) is related to a 0.082 pp lower GDP growth rate, while a higher C(10) is associated with a 0.022 pp higher GDP growth rate. As before, in pp terms, Column (5) implies an increase of 1 pp C(10) is associated with 0.24 pp increase in the GDP growth rate. Column (5) also suggests that 1 pp change in C(90) at the mean of 17.21 pp in the sample to 18.21 pp is associated with 0.48 pp decrease in the GDP growth rate.35 We find that one percent above the mean GDP in the last year predicts an 0.064 pp below the mean real GDP per capita growth rate in this year. We also note that one pp increase in minimum wage correlates with 0.069 pp decline in per capita GDP growth rate. Higher population growth is also associated with a higher GDP growth rate. Higher state government spending on goods and services in the previous year, and change in unionization from year t − 2 to t − 1 are insignificantly related to GDP growth rate after controlling for state fixed effects. 35 As before, we can calculate the GDP per capita decrease in growth rate as approximately 0.082×1/0.1721 = 0.48.

25

A.2

Instrumental Variables Approach With System GMM Estimation

The results above regarding the relationship between contraction factor and GDP growth rates, though suggestive, do not lend themselves to causal interpretation. Even though we have included an exhaustive set of controls, one may be concerned that an unobservable or omitted variable which is correlated with the contraction factor and GDP growth rate is driving their correlation. This concern should be partially assuaged by the presence of lagged GDP growth rate and state fixed effects, which control for any observed/unobserved heterogeneity between states and years. However, the concern may be that the omitted or unobserved variable is dynamic and varies by state in a way not captured by lagged GDP growth rate. Specifically, an important concern maybe that tax policy driving the contraction factors could be itself endogenously chosen by policymakers in response to economic conditions that also affect GDP growth rate in the state. To address these concerns, Column (6) utilizes an IV approach with system GMM with our most exhaustive specification. In Section II, we have already discussed the estimation strategy and the instruments. The coefficients on the contraction factors in Column (6) of Table III again show that contraction factors have a significant impact on per capita economic growth. The economic magnitudes are comparable to the OLS estimates. In particular, one percent decrease in income inequality between the 10th percentile household and the median household increases the GDP per capita growth rate by 0.035 pp (0.38 pp increase in GDP growth rate for 1 pp increase in C(90)). At the same time, one percent reduction in income inequality between the 90th percentile household and the median household, by increasing taxes by one percent on the 90th percentile household, reduces GDP growth rate by 0.081 pp (0.47 pp decrease in GDP growth rate for 1 pp increase in C(90)). In all tables including Table III, we conduct two important diagnostic tests for the System GMM estimator. First, to test for overidentifying restrictions, we report the Hansen J statistic and the corresponding p-value. The p-value fails to reject the null (the p-value is in fact almost 1), providing confidence in our choice of instruments.36 Another important diagnostic is the test for 36 Given

the large degree of freedom (approximately 550) because of large number of internal instruments, the Jstatistic is also relatively quite small in all cases, compared to the critical value of χ 2 . J-statistic is approximately 44

26

autocorrelation of the residuals. If a significant autocorrelation is present, the lags of endogenous variables will not be appropriate instruments for their current values (in this case Xs,t−1 ) in the System GMM estimator. The results of the test are reported against M3 where 3 represents the third lag. The p-values of M3 are always above 0.10, comfortably rejecting the null that there is significant autocorrelation. This provides us confidence that Xs,t−3 and Xs,t−4 are valid internal instrument variables in the System GMM estimator.37 Table B.VI in the Appendix reports the relevance of our exogenous instruments with respect to contraction factors (see the discussion in Section II.B). The first three columns report the relation between below median contraction factor and our instruments, and the last three columns repeat the exercise for the above median contraction factor. Columns (1) and (4) do not include state fixed effects, to analyze the effect of cross sectional variations between states on the contraction factors. All other columns include state fixed effects. Columns (1) and (4) show that positive shocks to personal income tax rates are positively associated with contraction factors. This is because higher tax rates lead to decrease in post-tax income inequality relative to pre-tax income inequality. We also note from Columns (1) and (4) that higher corporate income tax rates correlate with higher C(10), although this result does not survive when state fixed effects are included in Column (3). Between states, higher levels of inequality as measured by ratio of income of median household to 10th percentile households (in the year 1979) are associated with higher C(10). However, care must be taken to interpret such associations given that Columns (1) and (4) do not have state fixed effects. We also note that states with higher charity have lower contraction factors, possibly suggesting a substitution between the public support vs private charity. Lack of state fixed effects remain a caveat. In the System GMM approach with IV, we include state fixed effects to eliminate the impact of initial conditions. Hence, all remaining columns report results with state fixed effects. We include interaction between exogenous tax shocks and inequality levels in 1979 and aggregate fraction of income in a state donated to charity in year 1979 as additional instruments. The argument is that in Table III, where the critical value of χ 2 even at p = 0.995 is close to 468. 37 We do not use second order lags, i.e. X s,t−2 , as internal instruments because we are concerned about the possible first order correlation of the error terms. However, our results are robust if we use Xs,t−2 as internal instruments.

27

tax rate shocks affect states differently based on level of initial inequality and charity level. After including state fixed effects, we find that variation in below median contraction factor is explained by personal income taxes and interactions; and variation in C(90) is explained by corporate income tax shocks and interactions. Column (2) shows that in presence of state fixed effects, personal income tax shocks remain positively associated with C(10). Interaction between initial below median inequality and personal tax shocks are negatively related to C(10). This suggests that states with higher initial income inequality may be partially undoing the redistributive effects of federal tax shocks through state level policies. Similarly, Column (6) suggests that the corporate income taxes and interaction with initial income inequality remain relevant for above median income households i.e. C(90). This is intuitive given that above median households own equity in firms more often. However, Column (3) shows that in presence of state fixed effects, corporate tax shocks and related interactions are unable to explain variation in C(10). This is also reasonable given less ownership of equity in firms by below median income households. Similarly, Column (5) shows that personal tax shocks and interactions are unable to explain variation in C(90). The reported test statistics also confirm that the instruments are jointly relevant to contraction factors, and satisfy instrument validity concerns regarding under-identification and weak instruments problem.

B

Components of GDP

Next, we focus on three main components through which reduction of income inequality through tax policy affect economic growth: (i) Labor supply by households, (ii) Activities of small businesses, and (iii) Consumption. The intuition is as follows: (i) If the below median contraction factor is higher, it is generally achieved by providing tax credits to the lower income households. Earned income tax credit, and other such programs should encourage households at the lower income levels to supply more labor relative to similar income levels in other states. This should support economic growth. At the same time, households at higher income level note that their taxes are higher compared to the reference of

28

median households, and this reduces their incentives to supply labor. (ii) Regarding small business activity, if the tax rate at the higher percentiles of income in a state is higher compared to other states, then small businesses who file as households in that state will be less encouraged to invest and grow than in other states, hampering economic growth.38 (iii) Finally, the changes in tax rate also affect households’ consumption through both the substitution effects and wealth effects.

B.1

Labor Supply

Table IV reports the impact of above and below median contraction factors on our first component, i.e. labor supply. Columns (1) through (3) investigate the extensive margin, and Columns (4) through (6) investigate the intensive margin of labor supply. Columns (1) and (4) include the contraction factors, the lagged change in median marginal tax rates, and state fixed effects, while Columns (2) and (5) show the full OLS specification, and Columns (3) and (6) show the most exhaustive IV GMM specification. The extensive margins (Columns (1) through (3)) look at change in annual employment rate for all workers aged 18 to 64 in a state, considering an individual employed if she has worked at least 26 weeks and at least 20 hours per week in a year in CPS data. The intensive margins (Columns (4) through (6)) consider the growth rate of average annual hours worked by the employed population in each state (See Panel A of Table I for more details). Column (1) shows that the number of individuals providing labor supply increases by 0.03 pp for each percent reduction in below median income inequality through tax policy C(10). On the other hand, an equal increase in above median income inequality reduction through tax policy C(90), reduces labor supply by 0.034 pp. Column (2), under the full OLS specification, shows that these results remain robust to introduction of full set of controls. One percent reduction in below median income inequality increases labor supply by 0.021 pp, while an equal reduction in above median income inequality reduces labor supply by 0.041 pp. Column (3), where we conduct an instrumental variables approach with system GMM estimator as in Section A, shows similar economic magnitudes to that obtained from OLS approach. Column (3) shows that 1 pp reduction 38 We

focus on small businesses because identification of impact of tax policy on larger businesses can be complex, due to multiple locations, transfer pricing and additional concerns.

29

in income inequality between 90th percentile household and median income household achieved through taxes reduces GDP growth rate by 0.32 pp. Reducing income inequality between the 10th percentile household and median household, however, causes GDP growth rate to increase by 0.30 pp. The next three columns of Table IV reports the results for the intensive margin of labor supply, i.e. the growth rate of average annual hours worked for the employed individuals. Column (4) shows that annual hours worked growth rate for all workers are reduced by 0.023 pp for each percent increase in above median contraction factor. The column also shows that if C(10) increases by one percent, then average hours worked growth rate for all employed workers increases by 0.014 pp. Column (5) adds the full set of controls and shows that annual hours worked growth rate decrease by 0.033 pp for one percent reduction in above median income inequality through taxation, and annual hours worked growth rate increases by 0.006 pp for an equal reduction in below median income inequality. Column (6) conducts the IV GMM analysis, and finds effects of reducing above and below median inequality factors that are similar in magnitude to effects found in Columns (4) and (5), providing more confidence in the results.

B.2

Small Business Activity

Table V reports the impact of above and below median contraction factors on our second mechanism – small business activity. We focus on small businesses which are affected by personal income taxes due to many small businesses filing as S Corporations.39 All columns focus on the growth rate of the number of establishments as the dependent variable. Columns (1) through (3) focus on businesses of size 5 to 19, and Columns (4) through (6) consider businesses of size 5 to 49. Columns (1) and (4) include the contraction factors, marginal tax rate variables, and state fixed effects, Columns (2) and (5) have the full OLS specification with state fixed effects, and Columns (3) and (6) report the most exhaustive IV specification with system GMM estimator. 39 Through

the Technical Amendments Act of 1958, small businesses were allowed to file as Subchapter S corporations. The benefit of such a tax structure is that firms can operate as limited liability corporations, without suffering double taxation on business earnings.

30

We note in Column (1), for establishments with 5 to 19 employees, the above median contraction factor has a negative impact on establishment growth, while the below median contraction factor has a positive impact on establishment growth. Column (2) reports similar results, which are corroborated by the system GMM estimator in Column (3). One percent reduction in above median income inequality results in a 0.049 pp decrease in establishment growth, and an equal reduction in below median income inequality results in a 0.023 pp increase in establishment growth. In other words, when income inequality between 90th and 50th percentile is reduced through taxes by 1 pp, growth rate of small businesses declines by 0.28 pp. At the same time, reducing inequality between 10th and 50th percentile by 1 pp causes growth rate of small businesses to increase by 0.25 pp. When we consider establishments with 5 to 49 employees, the growth rates remain similar in sign and magnitude: the above median contraction factor has a negative impact on establishment growth, while the below median contraction factor has a positive effect of establishment growth. These results suggest that the incentives of small businesses to grow upto median income increase, and beyond median income decrease due to tax policy that seeks to bring everyone towards the median.

B.3

Consumption

Table VI reports the impact of above and below median contraction factors on our final mechanism – personal consumption growth rates in states. The first two columns consider the impact of inequality reduction through tax policy on total personal consumption. The next two columns focus on durable goods, and the final two columns focus on personal consumption growth rates of nondurable goods and services. All columns include the full set of controls as before with Columns (1), (3), and (5) showing OLS results and Columns (2), (4), and (6) showing IV results with system GMM estimator. Although Column (1) does not show a significant relationship between the contraction factors and consumption growth, the system GMM estimator in Column (2) shows a stronger relationship. We see that decreasing inequality between 10th percentile and median household by one percent 31

through tax policy results in consumption growth rate increasing by 0.04 pp (or 1 pp increase in C(10) results in 0.44 pp increase). We do not find that consumption is increased by reducing inequality between above median households and median households. This may be because budget constraints bind more often on households below median income, and reduction in inequality in that income band improves the ability of households to consume more. Columns (3) and (4) focus on durable goods, and find higher sensitivity to changes in contraction factor compared to overall consumption discussed earlier. This is also logical, because durable goods may relatively be more discretionary than non-durable goods and services. Finally, Column (6) with an IV approach and system GMM estimator highlights, that consumption growth rate of non-durable goods and services increase by 0.032 pp for each percent reduction in income inequality through tax policy for below median income households. Again, we do not see a statistically significant impact of reducing above median inequality. In sum, this section shows that inequality reduction through tax policy for below median households encourages labor supply, business activity, and consumption, and ultimately economic growth. At the same time, inequality reduction between above median and median households through tax policy deters labor supply and business activity, hampering economic growth. Additional mechanisms may also be at work, but in this section, we only focused on labor supply, business activity and personal consumption.

V

Robustness

We have tested that our results are robust to many reasonable alternative specifications. We report our main results (as shown in Table III) with three alternative specifications next. The first alternative specification utilizes contraction factors calculated for the 20th and 80th percentile households (i.e. C(20) and C(80)) in place of C(10) and C(90). This helps address possible concerns that our results are extremely sensitive to the chosen points on the income distribution. The second alternative specification substitutes the log income ratio controls with the Gini coefficient of the given state in the previous year, as an alternative measure of state level inequality. The third specification 32

removes the charitable contributions exogenous instrument from our IV GMM model, to test the robustness of our approach. Table VII reports the consolidated results from these robustness checks. Columns (1) and (2) report the first robustness check, while Columns (3) and (4) report the second test. Columns (1) and (3) report OLS results with the full specification, and Columns (2) and (4) show IV results with system GMM estimator under the most exhaustive specification. Columns (5) and (6) utilize our last robustness check and show IV GMM results with log income ratios and the Gini coefficient, respectively, as controls for income inequality. The table shows that our main findings hold under each robustness check: a reduction in income inequality between median income households and low income households improves economic growth, and a reduction in income inequality between median income households and high income households decreases economic growth. In particular, changing the contraction factors to different income percentiles under the exhaustive IV GMM model in Column (2) shows that one percent decrease in income inequality between the 20th income percentile and the median household increases the state level economic growth rate by 0.026 pp, while an equal decrease in income inequality between the 80th income percentile and the median household reduces state level growth by 0.064 pp. In Column (4), where we switch our income inequality control to the Gini coefficient, we find that a one percent reduction in below median income inequality results in a 0.028 pp increase in economic growth, while a similar reduction in above median income inequality through taxation results in a 0.084 pp reduction in economic growth. Our results are also robust to removing the charitable contributions from our instrumental variables set, as the resulting coefficients remain similar in sign and magnitude. To further test the robustness of our results, we conduct our analysis on two different income distributions. As shown in Table VIII, we consider using the state income distribution and the national market income distribution from TAXSIM. Columns (1) through (3) use the state income distribution; specifically, we use the 10th, 50th, and 90th income percentiles within each state to calculate contraction factors, rather than at the national level as previously conducted. We find similar results, and as seen in Column (3), a one percent reduction in below median income 33

inequality results in a 0.032 pp increase in economic growth, while a similar reduction in above median inequality results in a 0.07 pp decrease in economic growth. Columns (4) to (6) utilize the national level market income. Market income is our original income measure, less the taxable portions of social security and unemployment insurance. We find, in Column (6), that a one percent reduction in above median income inequality results in a 0.071 pp decrease in economic growth, while a similar reduction in below median income inequality results in positive economic growth, albeit not significant. Obtaining similar results using different income distributions gives us further confidence in our main findings.

VI

Conclusion

Modern democracies have accepted the role of income taxation in addressing income inequality. This paper distinguishes the economic growth effect of income taxation in reducing inequality between below median and median income households, from the economic growth effect of taxation in reducing inequality between median household and above median households. We find that the tax policy that provides poverty alleviation improves economic growth. At the same time, we find that reduction of incentives that is caused by lower after-tax income gap between median and rich households, reduces economic growth. This paper does not address optimal taxation and general equilibrium effects of tax policy. We also do not investigate the impact of specific tax policies and welfare programs on economic growth. Literature has been addressing these questions. More research needs to be done on the impact of tax policy on economic growth. Hopefully, our research will help policymakers make more informed decisions regarding tax policy by carefully balancing social insurance with incentive preservation.

References Aghion, P., Akcigit, U., Bergeaud, A., Blundell, R., and Hemous, D. (2015). Innovation and Top Income Inequality. Working paper, Harvard University.

34

Aghion, P. and Williamson, J. (1998). Growth, Inequality, and Globalization: Theory, History, and Policy. Raffaele Mattioli Lectures. Cambridge University Press. Alesina, A. and Ardagna, S. (2010). Large Changes in Fiscal Policy: Taxes versus Spending, pages 35–68. The University of Chicago Press. Alesina, A. and Rodrik, D. (1994). Distributive Politics and Economic Growth. The Quarterly Journal of Economics, 109(2):pp. 465–490. Arellano, M. and Bond, S. (1991). Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations. The Review of Economic Studies, 58(2):277. ˚ Johansson, Schwellnus, C., and Vartia, L. (2011). Tax Arnold, J. M., Brys, B., Heady, C., Asa Policy for Economic Recovery and Growth. Economic Journal, 121(550):F59–F80. Autor, D., Manning, A., and Smith, C. L. (2014). The Contribution of the Minimum Wage to U.S. Wage Inequality over Three Decades: A Reassessment. Working paper, Massachusetts Institute of Technology. Banerjee, A. V. and Duflo, E. (2003). Inequality and Growth: What Can the Data Say? Journal of Economic Growth, 8(3):pp. 267–299. Barro, R. J. (1990). Government Spending in a Simple Model of Endogeneous Growth. Journal of Political Economy, 98(5):pp. S103–S125. Barro, R. J. (1991). Economic Growth in a Cross Section of Countries. The Quarterly Journal of Economics, 106(2):pp. 407–443. Barro, R. J. (2000). Inequality and Growth in a Panel of Countries. Journal of Economic Growth, 5(1):5–32. Barro, R. J. and Redlick, C. J. (2011). Macroeconomic Effects From Government Purchases and Taxes. The Quarterly Journal of Economics, 126(1):51–102. Becker, G. S. (2011). Bad and Good Inequality. Blog post, The Becker-Posner Blog. B´enabou, R. (1996). Inequality and Growth. In NBER Macroeconomics Annual 1996, Volume 11, NBER Chapters, pages 11–92. National Bureau of Economic Research. B´enabou, R. (2000). Unequal Societies: Income Distribution and the Social Contract. The American Economic Review, 90(1):pp. 96–129. Benhabib, J. and Rustichini, A. (1996). Social Conflict and Growth. Journal of Economic Growth, 1(1):pp. 125–142. Blanchard, O. and Perotti, R. (2002). An Empirical Characterization of the Dynamic Effects of Changes in Government Spending and Taxes on Output. The Quarterly Journal of Economics, 117(4):1329–1368. Blundell, R. and Bond, S. (1998). Initial Conditions and Moment Restrictions in Dynamic Panel Data Models. Journal of Econometrics, 87(1):115 – 143. Blundell, R., Pistaferri, L., and Preston, I. (2008). Consumption Inequality and Partial Insurance. The American Economic Review, 98(5):pp. 1887–1921.

35

Bourguignon, F. (1981). Pareto Superiority of Unegalitarian Equilibria in Stiglitz’ Model of Wealth Distribution with Convex Saving Function. Econometrica, 49(6):pp. 1469–1475. Card, D. and DiNardo, J. E. (2002). Skill-Biased Technological Change and Rising Wage Inequality: Some Problems and Puzzles. Journal of Labor Economics, 20(4):733–783. Cingano, F. (2014). Trends in Income Inequality and its Impact on Economic Growth. Working Paper 163, Organisation for Economic Co-operation and Development. Conesa, J. C., Kitao, S., and Krueger, D. (2009). Taxing Capital? Not a Bad Idea after All! American Economic Review, 99(1):25–48. Diamond, P. and Saez, E. (2011). The Case for a Progressive Tax: From Basic Research to Policy Recommendations. Journal of Economic Perspectives, 25(4):165–90. Easterly, W. and Rebelo, S. (1993). Fiscal Policy and Economic Growth. Journal of Monetary Economics, 32(3):417 – 458. Eissa, N. and Hoynes, H. W. (2004). Taxes and the Labor Market Participation of Married Couples: The Earned Income Tax Credit. Journal of Public Economics, 88(9–10):1931 – 1958. Eissa, N. and Hoynes, H. W. (2006). Behavioral Responses to Taxes: Lessons from the EITC and Labor Supply. In Tax Policy and the Economy, Volume 20, NBER Chapters, pages 73–110. National Bureau of Economic Research, Inc. Eissa, N., Kleven, H. J., and Kreiner, C. T. (2008). Evaluation of Four Tax Reforms in the United States: Labor Supply and Welfare Effects for Single Mothers. Journal of Public Economics, 92(3-4):795–816. Ferede, E. and Dahlby, B. (2012). The Impact Of Tax Cuts On Economic Growth: Evidence From The Canadian Provinces. National Tax Journal, 65(3):563–94. Forbes, K. J. (2000). A Reassessment of the Relationship between Inequality and Growth. The American Economic Review, 90(4):pp. 869–887. Galor, O. and Moav, O. (2004). From Physical to Human Capital Accumulation: Inequality and the Process of Development. The Review of Economic Studies, 71(4):1001–1026. Galor, O. and Zeira, J. (1993). Income Distribution and Macroeconomics. The Review of Economic Studies, 60(1):pp. 35–52. Gemmell, N., Kneller, R., and Sanz, I. (2011). The Timing and Persistence of Fiscal Policy Impacts on Growth: Evidence from OECD Countries. Economic Journal, 121(550):F33–F58. Guvenen, F., Kuruscu, B., and Ozkan, S. (2013). Taxation of Human Capital and Wage Inequality: A Cross-Country Analysis. The Review of Economic Studies. Heckman, J. J., Lochner, L., and Taber, C. (1998). Tax Policy and Human-Capital Formation. The American Economic Review, 88(2):pp. 293–297. Helms, L. J. (1985). The Effect of State and Local Taxes on Economic Growth: A Time Series– Cross Section Approach. The Review of Economics and Statistics, 67(4):pp. 574–582. Hirsch, B. T., Macpherson, D. A., and Vroman, W. G. (2001). Estimates of Union Density by State.

36

Holter, H. A., Krueger, D., and Stepanchuk, S. (2014). How Does Tax Progressivity and Household Heterogeneity Affect Laffer Curves? Working Paper 20688, National Bureau of Economic Research. Juhn, C., Murphy, K. M., and Pierce, B. (1993). Wage Inequality and the Rise in Returns to Skill. Journal of Political Economy, 101(3):410–42. Kaldor, N. (1956). Alternative Theories of Distribution. The Review of Economic Studies, 23(2):pp. 83–100. Katz, L. F. and Murphy, K. M. (1992). Changes in Relative Wages, 1963-1987: Supply and Demand Factors. The Quarterly Journal of Economics, 107(1):35–78. Kneller, R., Bleaney, M. F., and Gemmell, N. (1999). Fiscal Policy and Growth: Evidence from OECD Countries. Journal of Public Economics, 74(2):171–190. Koester, R. B. and Kormendi, R. C. (1989). Taxation, Aggregate Activity and Economic Growth: Cross-Country Evidence on Some Supply-Side Hypotheses. Economic Inquiry, 27(3):367–386. Kuznets, S. (1955). Economic Growth and Income Inequality. The American Economic Review, 45(1):pp. 1–28. Lazear, E. P. and Rosen, S. (1981). Rank-Order Tournaments as Optimum Labor Contracts. Journal of Political Economy, 89(5):pp. 841–864. Lee, Y. and Gordon, R. H. (2005). Tax Structure and Economic Growth. Journal of Public Economics, 89(5-6):1027–1043. Levine, R. and Renelt, D. (1992). A Sensitivity Analysis of Cross-Country Growth Regressions. The American Economic Review, 82(4):pp. 942–963. Lucas, R. E. (1988). On the Mechanics of Economic Development. Journal of Monetary Economics, 22(1):3 – 42. Mankiw, N. G., Romer, D., and Weil, D. N. (1992). A Contribution to the Empirics of Economic Growth. The Quarterly Journal of Economics, 107(2):407–37. Mendoza, E. G., Milesi-Ferretti, G. M., and Asea, P. (1997). On the Ineffectiveness of Tax Policy in Altering Long-Run Growth: Harberger’s Superneutrality Conjecture. Journal of Public Economics, 66(1):99–126. Mertens, K. and Ravn, M. O. (2013). The Dynamic Effects of Personal and Corporate Income Tax Changes in the United States. American Economic Review, 103(4):1212–47. Meyer, B. D. and Rosenbaum, D. T. (2000). Making Single Mothers Work: Recent Tax and Welfare Policy and its Effects. National Tax Journal, 53(n. 4):1027–62. Meyer, B. D. and Rosenbaum, D. T. (2001). Welfare, the Earned Income Tax Credit, and the Labor Supply of Single Mothers. The Quarterly Journal of Economics, 116(3):1063–1114. Meyer, B. D. and Sullivan, J. X. (2004). The Effects of Welfare and Tax Reform: The Material Well-Being of Single Mothers in the 1980s and 1990s. Journal of Public Economics, 88(78):1387–1420. Michalopoulos, C., Robins, P. K., and Card, D. (2005). When Financial Work Incentives Pay for Themselves: Evidence from a Randomized Social Experiment for Welfare Recipients. Journal of Public Economics, 89(1):5 – 29. Tax and Transfer Programs for Low-Income People. 37

Miller, S. M. and Russek, F. S. (1997). Fiscal Structures and Economic Growth: International Evidence. Economic Inquiry, 35(3):603–613. Mirrlees, J. A. (1971). An Exploration in the Theory of Optimum Income Taxation. Review of Economic Studies, 38(114):175–208. Murphy, K. M., Shleifer, A., and Vishny, R. W. (1989). Income Distribution, Market Size, and Industrialization. The Quarterly Journal of Economics, 104(3):537–64. Okun, A. (1975). Equality and Efficiency, the Big Tradeoff. The Brookings Institution. Ostry, J. D., Berg, A., and Tsangarides, C. G. (2014). Redistribution, Inequality, and Growth. Staff discussion note, International Monetary Fund. Padovano, F. and Galli, E. (2001). Tax Rates and Economic Growth in the OECD Countries (1950–1990). Economic Inquiry, 39(1):44–57. Perotti, R. (1996). Growth, Income Distribution, and Democracy: What the Data Say. Journal of Economic Growth, 1(2):149–187. Persson, T. and Tabellini, G. (1994). Is Inequality Harmful for Growth? The American Economic Review, 84(3):pp. 600–621. Piketty, T. and Saez, E. (2003). Income Inequality in the United States, 1913-1998. The Quarterly Journal of Economics, 118(1):1–41. Rebelo, S. (1991). Long-Run Policy Analysis and Long-Run Growth. Journal of Political Economy, 99(3):500–521. Reed, W. R. (2008). The Robust Relationship between Taxes and U.S. State Income Growth. National Tax Journal, 61(1):pp. 57–80. Romer, C. D. and Romer, D. H. (2009). A Narrative Analysis of Postwar Tax Changes. Unpublished, University of California, Berkeley. Romer, C. D. and Romer, D. H. (2010). The Macroeconomic Effects of Tax Changes: Estimates Based on a New Measure of Fiscal Shocks. American Economic Review, 100(3):763–801. Romer, P. M. (1986). Increasing Returns and Long-Run Growth. Journal of Political Economy, 94(5):pp. 1002–1037. Saint-Paul, G. and Verdier, T. (1993). Education, Democracy and Growth. Journal of Development Economics, 42(2):399 – 407. Saint-Paul, G. and Verdier, T. (1997). Power, Distributive Conflicts, and Multiple Growth Paths. Journal of Economic Growth, 2(2):pp. 155–168. Sala-I-Martin, X. X. (1997). I Just Ran Two Million Regressions. The American Economic Review, 87(2):pp. 178–183. Solow, R. M. (1956). A Contribution to the Theory of Economic Growth. The Quarterly Journal of Economics, 70(1):pp. 65–94.

38

Table I: Variables List Panel A provides descriptions and sources for the dependent variables in our ordinary least squares and instrumental variables approaches. Panel B provides descriptions and sources for all of the independent variables and instruments.

Panel A: Dependent Variables Variable

Description

Source

Annual state level log change in chained real GDP per capita

BEA

Annual log change in number of establishments of size i to j

BDS

Annual change in state employed population (worked at least 26 weeks AND 20 hours per week) as fraction of 18-64 aged population Annual log change in state average annual hours worked if employed

CPS

Total personal consumption expenditures per capita annual log change Durable goods personal consumption expenditures per capita annual log change Nondurable goods and services personal consumption expenditures per capita annual log change

BEA BEA BEA

Economic Growth ∆ log GDPs,t Small Business Activity ∆ log Estabs(i)to( j),s,t Labor Supply ∆E˜s,t ∆ log Hrss,t

CPS

Consumption* ∆ log PCEs,t ∆ log PCEDG,s,t ∆ log PCENDGS,s,t

All changes in monetary figures are deflated to 2009 US Dollars using GDP deflator. *Consumption change variables are available from 1998 to 2008.

39

Panel B: Independent and Instrumental Variables Independent Variables Contraction factor, i.e. reduction of income inequality, between ith percentile household and median household Log of the contraction factor, i.e. reduction of income inequality, between ith percentile household and median household Change in overall marginal tax rate at ith percentile from previous year Overall marginal tax rate at ith percentile Income ratio of the ith percentile to the jth percentile, where i > j

Cs,t (i) logCs,t (i) ∆MT Rs,t (i) MT Rs,t (i) Income(i) Income( j) Income(i) log Income( j) log GDPs,t gIs,t log GDP s,t

Log of the income ratio of the ith percentile to the jth percentile, where i > j Log chained real GDP per capita Log of per capita state government spending on goods and services (education, highways, police protection, and correction) as a fraction of state GDP per capita Log change in total population Log of average years of schooling for population of age 18 to 64 Change in real state level effective minimum wage from previous year Change in fraction of state labor force covered by unions from previous year

ns,t log Schools,t min ∆Ws,t ∆Unions,t

NBER Taxsim NBER Taxsim NBER Taxsim NBER Taxsim NBER Taxsim NBER Taxsim BEA US Census State Finances, BEA BEA CPS Autor et al. (2014) Hirsch et al. (2001)

Exogenous Instruments (Zs,t ) z pi,t zci,t Income

(i)

z pi,t × Income s,1979( j) s,1979 Incomes,1979 (i)

zci,t × Income

s,1979 ( j)

z pi,t ×CCs,79 zci,t ×CCs,79

Narratively identified personal income tax shock Narratively identified corporate income tax shock Narratively identified personal income tax shock interacted with state inequality ratio in 1979 Narratively identified corporate income tax shock interacted with state inequality ratio in 1979 Narratively identified personal income tax shock interacted with state share of total income contributed to charity in 1979 Narratively identified corporate income tax shock interacted with state share of total income contributed to charity in 1979

Fixed Effects δs

State fixed effects

All changes in monetary figures deflated to 2009 US Dollars using GDP deflator. *Mertens and Ravn (2013)

40

MR* MR* NBER Taxsim, MR* NBER Taxsim, MR* NBER Taxsim, MR* NBER Taxsim, MR*

Table II: Summary Statistics Panel A shows summary statistics (number of observations, mean, standard deviation, minimum, maximum, and median) for all the dependent variables that enter our specification, as well as the corresponding levels to each growth and change variable. Panel B shows the summary statistics for the main explanatory variables, controls, and instruments that enter our model, and the corresponding levels to each growth and change variable. All characteristics are at the state level, and all summary statistics shown are pooled across states from 1980 to 2009. Panel C shows average adjusted gross income and tax rates for select income percentiles, pooled across states from 1980 to 2009. Observations are state (s) by year (t). Panel A: Summary Statistics for Dependent Variables in Model Variable

Observations

Mean

SD

Min

Max

Median

1076 1076

37703.91 .0146

9019.30 .0282

18624.96 -.1090

69946.16 .1323

37450.88 .0167

1076 1076 1076 1076

53015.30 66369.73 .0147 .0154

45412.78 57450.33 .0204 .0206

4044 4904 -.0948 -.0873

275491 353980 .1310 .1314

38313.5 47674.5 .0142 .0148

1076 1076 1076 1076

.7009 -.0006 2073.22 -.0006

.0468 .0189 42.48 .0129

.5142 -.0606 1937.42 -.0408

.8226 .0679 2204.47 .0380

.7030 -.0002 2075.49 -.0007

484 448 484 448 484 448

30216.51 .0150 3870.95 -.0031 26345.57 .0174

4378.86 .0243 448.66 .0535 4121.36 .0222

19320.49 -.0769 2674.85 -.1721 15981.81 -.0649

43628.41 .0770 5081.66 .1663 39472.44 .0915

29433.85 .0161 3855.95 .0045 25641.63 .0186

Economic Growth GDPs,t ∆ log GDPs,t Small Business Activity Estabs5to19,s,t Estabs5to49,s,t ∆ log Estabs5to19,s,t ∆ log Estabs5to49,s,t Labor Supply E˜s,t ∆E˜s,t Hrss,t ∆ log Hrss,t Consumption* PCEs,t ∆ log PCEs,t PCEDG,s,t ∆ log PCEDG,s,t PCENDGS,s,t ∆ log PCENDGS,s,t

All monetary figures deflated to 2009 US Dollars using GDP deflator. * The consumption series is available from BEA for 1997 to 2009; the growth variables are from 1998 to 2009. The unit of observation is state-year. For more information on sample selection, see Section I.

41

Panel B: Summary Statistics for Key Explanatory Variables, Controls, and Instruments in Model Variable

Observations

Mean

SD

Min

Max

Median

1076 1076 1076 1076 1076 1076 1076 1076 983

.0914 .1721 .1058 .1457 -2.4192 -1.7679 -2.2733 -1.9386 .0011

.0205 .0221 .0236 .0231 .2387 .1269 .2363 .1596 .0109

.0362 .1134 .0433 .0819 -3.3197 -2.1767 -3.1392 -2.5029 -.0498

.1466 .2514 .1700 .2179 -1.9199 -1.3807 -1.7719 -1.5239 .0487

.0918 .1713 .1062 .1444 -2.3886 -1.7646 -2.2425 -1.9350 .0014

1076

5.4836

.5057

4.1431

7.7806

5.4630

1076

2.7051

.1349

2.4368

3.1011

2.7096

1076

1.6974

.0936

1.4214

2.0516

1.6980

1076

.9939

.0502

.8907

1.1318

.9968

1076 1076 1076

10.2282 2.4933 .0154

.1626 .0389 .0039

9.8687 2.4038 .0076

10.5983 2.5716 .0397

10.2106 2.5004 .0150

Key Explanatory Variables Cs,t−1 (10) Cs,t−1 (90) Cs,t−1 (20) Cs,t−1 (80) logCs,t−1 (10) logCs,t−1 (90) logCs,t−1 (20) logCs,t−1 (80) ∆MT Rs,t−1 (50) Initial Condition Control Variables Incomes,1979 (50) Incomes,1979 (10) Incomes,1979 (90) Incomes,1979 (50) Income (50) log Incomes,1979 (10) s,1979 Incomes,1979 (90) log Income s,1979 (50)

log GDPs,1979 log Schools,1979 CCs,1979 Dynamic Control Variables MT Rs,t−1 (50)

1076

.3537

.0322

.2633

.4481

.3557

1076

5.9233

.9033

3.2776

10.2265

5.8610

1076

3.1711

.4402

2.3927

8.8194

3.1239

1076

1.7674

.1519

1.1871

2.3250

1.7683

1076

1.1459

.1236

.8724

2.1770

1.1391

log GDPs,t−1 log Schools,t−1 gIs,t−1

1076 1076 1076

10.4938 2.5464 1722.78

.2448 .0391 460.71

9.843 2.4008 787.81

11.1555 2.6348 2965.33

10.5112 2.5500 1667.41

s,t−1 log GDP s,t−1 ns,t−1 min Ws,t min ∆Ws,t−1 Unions,t ∆Unions,t−1

1076

-3.0787

.2329

-3.7269

-2.4055

-3.0926

1027 1076 1027 1076 1027

.0104 6.18 .0010 .1681 -.0038

.0101 .63 .0528 .0744 .0136

-.0617 5.20 -.0380 .0330 -.0650

.0707 8.15 .3334 .3990 .0480

.0082 6.06 -.0204 .1625 -.0030

1000 1000 1000 1000

-.0642 -.0389 -.0010 -.0006

.2920 1.6763 .0047 .0263

-1.0796 -3.2839 -.0428 -.1302

.4350 7.3821 .0100 .1693

0 0 0 0

1000

-.0637

.2900

-1.1534

.4621

0

1000

-.0366

1.6730

-3.5081

8.0508

0

1000

-.1099

.4977

-1.9774

.7888

0

1000

-.0701

2.8565

-6.0146

14.0119

0

Incomes,t−1 (50) Incomes,t−1 (10) Incomes,t−1 (90) Incomes,t−1 (50) Income (50) log Incomes,t−1 (10) s,t−1 Income (90) log Incomes,t−1 (50) s,t−1

gI

Exogenous Instruments z pi,t−1 zci,t−1 z pi,t−1 ×CCs,1979 zci,t−1 ×CCs,1979 Income

(90)

z pi,t−1 × log Incomes,1979 (50) s,1979

Income (90) zci,t−1 × log Incomes,1979 (50) s,1979 Incomes,1979 (50) z pi,t−1 × log Income s,1979 (10) Income (50) zci,t−1 × log Incomes,1979 (10) s,1979

Panel C: Mean Income and Tax by Income Percentile

Incomes,t−1 (i) Taxs,t−1 (i) Incomes,t−1 (i) − Taxs,t−1 (i) Taxs,t−1 (i) Incomes,t−1 (i)

i = 10th Pct.

i = 20th Pct.

i = 50th Pct.

i = 80th Pct.

i = 90th Pct.

Mean

5192.76 76.19 5116.57

10838.74 306.81 10531.93

30645.82 2375.68 28270.14

69065.13 7880.4 61184.73

98179.19 13816.57 84362.62

48068.45 6865.26 41203.19

.0147

.0283

.0775

.1141

.1407

.1428

All monetary figures in 2009 US Dollars (using GDP deflator). The unit of observation is state-year. For more information on sample selection, see Section I.

42

Table III: The Effects of Log Contraction Factors on State Level GDP Growth (∆ log GDPs,t ) This table shows our regression results for our main specification with state level real GDP per capita growth as our dependent variable. We look at the log change in GDP per capita, ∆ log GDPs,t = log GDPs,t − log GDPs,t−1 , where GDP is the GDP per capita. Column (1) shows OLS estimates with logC(10) as the only regressor, while Column (2) adds logC(90) as a regressor. Column (3) includes ∆MT Rs,t−1 (50), MT Rs,t−1 (50), and several initial condition controls at t = 1979. Column (4) shows OLS estimates for the log contraction factors, marginal tax rate variables, and state fixed effects. To this specification, Column (5) adds dynamic controls from the previous year, Xs,t−1 . Lastly, Column (6) shows the IV GMM estimation results with log contraction factors, marginal tax rate variables, and dynamic controls. The lower panel of the table shows regressions statistics: the use of state fixed effects, R2 , number of observations and states, the mean of the dependent variable and log contraction factors, and IV GMM statistics such as autocorrelation tests, Hansen statistic, and number of instruments. Specification: Income (90) Income (50) ∆ log GDPs,t = α0 + α1 logCs,t−1 (10) + α2 logCs,t−1 (90) + α3 ∆MT Rs,t−1 (50) + h1 log Incomes,t−1 (10) + h2 log Incomes,t−1 (50) + Xs,t−1 β + δs + εs,t s,t−1

Model Regressors logCs,t−1 (10)

OLS

OLS

OLS + X1979

OLS + δs

OLS + Xt−1 + δs

IV GMM

(1)

(2)

(3)

(4)

(5)

(6)

0.013*** (0.004)

0.026*** (0.005) -0.034*** (0.010)

0.034*** (0.005) -0.033*** (0.011) -0.640*** (0.060) 0.044* (0.024)

0.048*** (0.009) -0.056*** (0.017) -0.678*** (0.063) 0.138*** (0.048)

0.022* (0.011) -0.082*** (0.020) -0.697*** (0.051) 0.191*** (0.056)

0.035** (0.015) -0.081** (0.032) -0.681*** (0.076) 0.143 (0.101)

logCs,t−1 (90) ∆MT Rs,t−1 (50) MT Rs,t−1 (50) (50)

Income

s,t−1

log Incomes,1979 (10) s,1979

-0.006 (0.011)

(90)

Income

log Incomes,1979 (50) s,1979

0.019 (0.015) -0.010 (0.008) -0.008 (0.026) -0.121 (0.166)

log GDPs,1979 log Schools,1979 CCs,1979 Income

(50)

log Incomes,t−1 (10) s,t−1 log

Incomes,t−1 (90) Incomes,t−1 (50)

log GDPs,t−1 log Schools,t−1 gI

s,t−1 log GDP s,t−1

ns,t−1 min ∆Ws,t−1

∆Unions,t−1

δs

N

N

N

Y

-0.021**

-0.005

(0.009)

(0.020)

0.048**

0.036

(0.021) -0.064*** (0.014) 0.001 (0.088)

(0.048) -0.037 (0.027) 0.048 (0.244)

-0.018

-0.029

(0.012) 0.663*** (0.173) -0.069*** (0.015) 0.033 (0.084)

(0.023) 0.156 (0.508) -0.087*** (0.026) -0.091 (0.160)

Y

M3 (p-val) Hansen J statistic Hansen (p-val) R2 # Observations # States Means: Dep. Variable logC(10) logC(90) C(10) C(90)

Y 0.150 44.610 1.000

0.013 1076 49

0.025 1076 49

0.108 983 49

0.092 983 49

0.080 983 49

983 49

0.0146 -2.4192

0.0146 -2.4192 -1.7679 0.0914 0.1721

0.0154 -2.4338 -1.7757 0.0902 0.1707

0.0154 -2.4338 -1.7757 0.0902 0.1707

0.0154 -2.4338 -1.7757 0.0902 0.1707

0.0154 -2.4338 -1.7757 0.0902 0.1707

0.0914

Significance Levels: * 10% ** 5% *** 1% The exogenous instruments for the IV GMM estimation are the narratively identified exogenous tax shocks to personal and corporate income from Mertens and Ravn (2013), as well as these shocks interacted with initial conditions on state level charitable donations and inequality levels. Internal instruments include the predetermined control variables, such as inequality level, GDP per capita, schooling, state investment, population growth, and labor market institutions at t − 3 and t − 4. OLS standard errors are clustered at the state level. IV GMM model uses robust, two step System GMM estimator with Windmeijer-corrected standard errors.

43

Table IV: The Effects of Log Contraction Factors on Changes in Employment Outcomes This table shows our regression results for the change in annual employment outcomes for each state: employment rate (∆E˜s,t ) and the log change in average annual hours worked if employed (∆ log Hrss,t ). We define employed as having worked at least 26 weeks and at least 20 hours per week in a year and our base population is individuals aged 18 through 64. Columns (1) through (3) report regression results using ∆E˜s,t as the outcome variable, and Columns (4) through (6) report estimates using ∆ log Hrss,t as the outcome variable. Column (1) and Column (4) show OLS estimates for regressing the outcome variable for all workers on the log contraction factors, marginal tax rate variables, and state fixed effects. Column (2) and Column (5) show OLS estimates for the complete specification, that is using log contraction factors, marginal tax rate variables, all dynamic controls, and state fixed effects as regressors. Column (3) and Column (6) show IV GMM estimation results with log contraction factors, marginal tax rate variables, and dynamic controls. The lower panel of the table shows regressions statistics: the use of state fixed effects, R2 , number of observations and states, the mean of the dependent variable and log contraction factors, and IV GMM statistics such as autocorrelation tests, Hansen statistic, and number of instruments. Specification: Income (90) Income (50) ∆Ys,t = α0 + α1 logCs,t−1 (10) + α2 logCs,t−1 (90) + α3 ∆MT Rs,t−1 (50) + h1 log Incomes,t−1 (10) + h2 log Incomes,t−1 (50) + Xs,t−1 β + δs + εs,t s,t−1

s,t−1

Employment Outcomes for All Workers (employed if work ≥ 26 weeks/year and ≥ 20 hours/week) Ys,t

∆Employment Rate

Model Regressors logCs,t−1 (10) logCs,t−1 (90) ∆MT Rs,t−1 (50) MT Rs,t−1 (50) Income

OLS + δs

OLS + Xt−1 + δs

IV GMM

OLS + δs

OLS + Xt−1 + δs

IV GMM

(1)

(2)

(3)

(4)

(5)

(6)

0.030*** (0.004) -0.034*** (0.009) -0.121* (0.064) -0.006 (0.025)

0.021*** (0.005) -0.041*** (0.009) -0.125* (0.063) 0.035 (0.027)

0.027** (0.011) -0.055*** (0.016) -0.127 (0.100) 0.047 (0.068)

0.014*** (0.003) -0.023*** (0.006) -0.037 (0.044) 0.004 (0.018)

0.006* (0.003) -0.033*** (0.006) -0.038 (0.043) 0.038* (0.020)

0.006 (0.005) -0.031*** (0.010) -0.049 (0.052) 0.039 (0.043)

(50)

log Incomes,t−1 (10) s,t−1 log

Incomes,t−1 (90) Incomes,t−1 (50)

log GDPs,t−1 log Schools,t−1 gI

s,t−1 log GDP s,t−1

ns,t−1 min ∆Ws,t−1

∆Unions,t−1

δs

Log Change in Average Annual Hours Worked (if employed)

Y

0.003

0.003

-0.003

-0.004

(0.005)

(0.021)

(0.003)

(0.009)

0.022**

0.031

0.008

-0.016

(0.008) -0.022*** (0.006) 0.014 (0.040)

(0.028) -0.018 (0.022) -0.040 (0.169)

(0.006) -0.013*** (0.005) -0.035 (0.031)

(0.015) -0.006 (0.005) 0.000 (.)

-0.024***

-0.023

-0.005

-0.012

(0.006) -0.085 (0.135) -0.050*** (0.010) 0.016 (0.053)

(0.022) -0.400 (0.304) -0.052*** (0.016) 0.064 (0.096)

(0.004) 0.042 (0.076) -0.021** (0.009) -0.021 (0.034)

(0.009) 0.074 (0.150) -0.012 (0.010) 0.030 (0.102)

Y

M3 (p-val) Hansen J statistic Hansen (p-val)

Y

Y

Y

0.214 36.305 1.000

Y 0.125 46.951 1.000

R2 # Observations # States

0.042 983 49

0.067 983 49

983 49

0.015 983 49

0.022 983 49

983 49

Means: Dep. Variable logC(10) logC(90) C(10) C(90)

-0.0003 -2.4338 -1.7757 0.0902 0.1707

-0.0003 -2.4338 -1.7757 0.0902 0.1707

-0.0003 -2.4338 -1.7757 0.0902 0.1707

-0.0004 -2.4338 -1.7757 0.0902 0.1707

-0.0004 -2.4338 -1.7757 0.0902 0.1707

-0.0004 -2.4338 -1.7757 0.0902 0.1707

Significance Levels: * 10% ** 5% *** 1% In this regression change in employment rate is Es,t − Es,t−1 , and log change in hours worked is log Hrss,t − log Hrss,t−1 . The exogenous instruments for the IV GMM estimation are the narratively identified exogenous tax shocks to personal and corporate income from Mertens and Ravn (2013), as well as these shocks interacted with initial conditions on state level charitable donations and inequality levels. Internal instruments include the predetermined control variables, such as inequality level, GDP per capita, schooling, state investment, population growth, and labor market institutions at t − 3 and t − 4. OLS standard errors are clustered at the state level. IV GMM model uses robust, two step System GMM estimator with Windmeijer-corrected standard errors.

44

Table V: The Effects of Log Contraction Factors on Small Business Growth This table summarizes our regression results for small business growth. Our dependent variable is the growth in the number of establishments of size 5 to 19, and size 5 to 49. Specifically, we look at the log change in number of establishments for each size bucket, ∆ log Estabss,t = log Estabss,t − log Estabss,t−1 . Columns (1) through (3) report regression results using growth of establishments of size 5 to 19 as the dependent variable, and Columns (4) through (6) report estimates using growth of establishments of size 5 to 49 as the dependent variable. Column (1) and Column (4) show OLS estimates for regressing the outcome variable for all workers on the log contraction factors, marginal tax rate variables, and state fixed effects. Column (2) and Column (5) show OLS estimates for the complete specification, that is using log contraction factors, marginal tax rate variables, all dynamic controls, and state fixed effects as regressors. Column (3) and Column (6) show IV GMM estimation results with log contraction factors, marginal tax rate variables, and dynamic controls. The lower panel of the table shows regressions statistics: the use of state fixed effects, R2 , number of observations and states, the mean of the dependent variable and log contraction factors, and IV GMM statistics such as autocorrelation tests, Hansen statistic, and number of instruments. Specification: Income (90) Income (50) ∆Ys,t = α0 + α1 logCs,t−1 (10) + α2 logCs,t−1 (90) + α3 ∆MT Rs,t−1 (50) + h1 log Incomes,t−1 (10) + h2 log Incomes,t−1 (50) + Xs,t−1 β + δs + εs,t s,t−1

s,t−1

Establishment Growth Ys,t

Establishment Size: 5 to 19

Model Regressors logCs,t−1 (10) logCs,t−1 (90) ∆MT Rs,t−1 (50) MT Rs,t−1 (50) Income

OLS + δs

OLS + Xt−1 + δs

IV GMM

OLS + δs

OLS + Xt−1 + δs

IV GMM

(1)

(2)

(3)

(4)

(5)

(6)

0.034*** (0.005) -0.024*** (0.009) -0.182** (0.069) -0.060 (0.039)

0.020*** (0.005) -0.023** (0.009) -0.169** (0.067) -0.081* (0.045)

0.023* (0.012) -0.049*** (0.017) -0.163** (0.077) -0.028 (0.047)

0.035*** (0.005) -0.025*** (0.009) -0.163** (0.070) -0.036 (0.040)

0.022*** (0.005) -0.020** (0.009) -0.148** (0.067) -0.074 (0.046)

0.026*** (0.009) -0.051*** (0.012) -0.145* (0.083) 0.000 (0.052)

(50)

log Incomes,t−1 (10) s,t−1 log

Incomes,t−1 (90) Incomes,t−1 (50)

log GDPs,t−1 log Schools,t−1 gI

s,t−1 log GDP s,t−1

ns,t−1 min ∆Ws,t−1

∆Unions,t−1

δs

Establishment Size: 5 to 49

Y

-0.005

-0.016

-0.005

-0.015

(0.006)

(0.012)

(0.007)

(0.010)

0.016

0.009

0.021*

0.016

(0.011) -0.005 (0.010) 0.020 (0.065)

(0.021) -0.019 (0.018) -0.013 (0.127)

(0.011) -0.002 (0.010) 0.043 (0.067)

(0.022) -0.022 (0.016) 0.014 (0.119)

-0.060***

-0.021*

-0.065***

-0.022*

(0.010) 0.492*** (0.178) -0.033*** (0.011) -0.040 (0.052)

(0.011) 0.422 (0.326) -0.042** (0.018) -0.157 (0.119)

(0.010) 0.520*** (0.175) -0.035*** (0.012) -0.047 (0.054)

(0.012) 0.444 (0.272) -0.050** (0.019) -0.174 (0.111)

Y

M3 (p-val) Hansen J statistic Hansen (p-val)

Y

Y

Y

0.691 46.909 1.000

Y 0.344 47.007 1.000

R2 # Observations # States

0.083 983 49

0.098 983 49

983 49

0.070 983 49

0.086 983 49

983 49

Means: Dep. Variable logC(10) logC(90) C(10) C(90)

0.0146 -2.4338 -1.7757 0.0902 0.1707

0.0146 -2.4338 -1.7757 0.0902 0.1707

0.0146 -2.4338 -1.7757 0.0902 0.1707

0.0153 -2.4338 -1.7757 0.0902 0.1707

0.0153 -2.4338 -1.7757 0.0902 0.1707

0.0153 -2.4338 -1.7757 0.0902 0.1707

Significance Levels: * 10% ** 5% *** 1% In this regression, establishment growth is ∆ log Estabss,t = log Estabss,t − log Estabss,t−1 , where Estabs is the number of establishments in the specific size bucket. The exogenous instruments for the IV GMM estimation are the narratively identified exogenous tax shocks to personal and corporate income from Mertens and Ravn (2013), as well as these shocks interacted with initial conditions on state level charitable donations and inequality levels. Internal instruments include the predetermined control variables, such as inequality level, GDP per capita, schooling, state investment, population growth, and labor market institutions at t − 3 and t − 4. OLS standard errors are clustered at the state level. IV GMM model uses robust, two step System GMM estimator with Windmeijer-corrected standard errors.

45

Table VI: The Effects of Log Contraction Factors on Personal Consumption Growth This table shows our regression results for the log change in per capita personal consumption expenditure in each state, ∆ log PCEs,t = log PCEs,t − log PCEs,t−1 . We use three separate categories of these consumption log change variables: total, durable goods, and nondurable goods and services. Column (1) and Column (2) show results for total consumption growth; Column (3) and Column (4) show results for durable goods consumption growth; and Column (5) and Column (6) show results for nondurable goods and services consumption growth. Column (1), Column (3), and Column (5) show OLS estimates for the complete specification, that is using log contraction factors, marginal tax rate variables, all dynamic controls, and state fixed effects as regressors. Column (2), Column (4), and Column (6) show IV GMM estimation results with log contraction factors, marginal tax rate variables, and dynamic controls. The lower panel of the table shows regressions statistics: the use of state fixed effects, R2 , number of observations and states, the mean of the dependent variable and log contraction factors, and IV GMM statistics such as autocorrelation tests, Hansen statistic, and number of instruments. Specification: Income (90) Income (50) ∆Ys,t = α0 + α1 logCs,t−1 (10) + α2 logCs,t−1 (90) + α3 ∆MT Rs,t−1 (50) + h1 log Incomes,t−1 (10) + h2 log Incomes,t−1 (50) + Xs,t−1 β + δs + εs,t s,t−1

s,t−1

∆Per Capita Personal Consumption Ys,t

∆Personal Consumption

Model Regressors logCs,t−1 (10) logCs,t−1 (90) ∆MT Rs,t−1 (50) MT Rs,t−1 (50) Income

(50)

log Incomes,t−1 (10) s,t−1 Income

(90)

log Incomes,t−1 (50) s,t−1 log GDPs,t−1 log Schools,t−1 gI

s,t−1 log GDP s,t−1

ns,t−1 min ∆Ws,t−1

∆Unions,t−1

δs

∆Durable Goods

∆Nondurable Goods & Services

OLS + Xt−1 + δs

IV GMM

OLS + Xt−1 + δs

IV GMM

OLS + Xt−1 + δs

IV GMM

(1)

(2)

(3)

(4)

(5)

(6)

0.006 (0.010) -0.021 (0.017) -0.260*** (0.092) -0.080 (0.094)

0.040** (0.016) -0.023 (0.028) -0.290*** (0.100) 0.154 (0.147)

-0.007 (0.020) -0.031 (0.034) -0.698*** (0.177) -0.274 (0.180)

0.090** (0.042) -0.000 (0.065) -1.056*** (0.225) -0.037 (0.397)

0.008 (0.010) -0.020 (0.016) -0.193** (0.091) -0.058 (0.091)

0.032* (0.016) -0.026 (0.028) -0.177 (0.115) 0.215 (0.150)

-0.019

-0.004

-0.031

0.009

-0.018

-0.008

(0.013)

(0.016)

(0.019)

(0.040)

(0.013)

(0.015)

-0.054**

-0.048

-0.146***

-0.118

-0.041**

-0.024

(0.020) -0.123*** (0.035) -0.481*** (0.139)

(0.033) -0.053 (0.050) -0.083 (0.218)

(0.036) -0.487*** (0.075) -0.894*** (0.269)

(0.094) -0.154* (0.078) -0.259 (0.580)

(0.019) -0.072** (0.031) -0.435*** (0.134)

(0.029) -0.066 (0.064) 0.161 (0.342)

-0.076***

-0.049*

-0.183***

-0.095

-0.062**

-0.041*

(0.024) 0.466 (0.414) -0.038 (0.024) -0.210* (0.120)

(0.028) 0.459 (0.715) -0.078* (0.041) -0.047 (0.214)

(0.042) 0.735 (0.705) -0.119*** (0.040) -0.279 (0.200)

(0.075) 1.138 (1.130) -0.111 (0.075) -0.227 (0.382)

(0.024) 0.428 (0.382) -0.030 (0.024) -0.202* (0.113)

(0.021) 0.591 (0.767) -0.084** (0.039) -0.141 (0.165)

Y

Y

Y

Y

Y

Y

M3 (p-val) Hansen J statistic Hansen (p-val)

0.167 33.238 1.000

0.485 33.624 1.000

0.475 33.358 1.000

R2 # Observations # States

0.092 434 39

434 39

0.139 434 39

434 39

0.070 434 39

434 39

Means: Dep. Variable logC(10) logC(90) C(10) C(90)

0.0144 -2.6213 -1.8442 0.0742 0.1591

0.0144 -2.6213 -1.8442 0.0742 0.1591

-0.0045 -2.6213 -1.8442 0.0742 0.1591

-0.0045 -2.6213 -1.8442 0.0742 0.1591

0.0170 -2.6213 -1.8442 0.0742 0.1591

0.0170 -2.6213 -1.8442 0.0742 0.1591

Significance Levels: * 10% ** 5% *** 1% The consumption growth rate data series is available from 1998 to 2008. Change in consumption variables are log changes of the form log PCEs,t − log PCEs,t−1 , where PCE is the personal consumption expenditures per capita. The exogenous instruments for the IV GMM estimation are the narratively identified exogenous tax shocks to personal and corporate income from Mertens and Ravn (2013), as well as these shocks interacted with initial conditions on state level charitable donations and inequality levels. Internal instruments include the predetermined control variables, such as inequality level, GDP per capita, schooling, state investment, population growth, and labor market institutions at t − 3 and t − 4. OLS standard errors are clustered at the state level. IV GMM model uses robust, two step System GMM estimator with Windmeijer-corrected standard errors.

46

Table VII: Robustness 1 - The Effects of Log Contraction Factors on State Level GDP Growth This table presents three robustness checks for our estimation with the state level real GDP per capita growth (∆ log GDPs,t ) as the dependent variable. The first robustness check uses our main specification, keeps the change in median marginal tax rates, and replaces our main contraction factors with Cs,t−1 (80) and Cs,t−1 (20). Column (1) shows OLS results using the full specification and Column (2) shows IV GMM estimates for the full specification. The second robustness check uses our original specification and substitutes the log income ratio controls with the Gini coefficient. Column (3) shows OLS results using the full specification and Column (4) shows IV GMM estimates for the full specification. The last robustness check removes charitable contributions from our exogenous instrument list. Column (5) shows IV GMM estimates for the original specification with log income ratios as a control for income inequality, while Column (6) shows IV GMM estimates for the version using the Gini coefficient as a control for income inequality. The lower panel of the table shows regressions statistics: the use of state fixed effects, R2 , number of observations and states, the mean of the dependent variable and log contraction factors, and IV GMM statistics such as autocorrelation tests, Hansen statistic, and number of instruments. Specification: ∆ log GDPs,t = α0 + α1 logCs,t−1 (i) + α2 logCs,t−1 ( j) + α3 ∆MT Rs,t−1 (50) + h Income Inequality + Xs,t−1 β + δs + εs,t Using logC(20) & logC(80) Model Regressors

IV GMM

OLS + Xt−1 + δs

IV GMM

IV GMM

IV GMM

(1)

(2)

(3)

(4)

(5)

(6)

0.019 (0.012) -0.078*** (0.021)

0.028** (0.014) -0.084*** (0.025)

0.035** (0.015) -0.080** (0.032)

0.028** (0.014) -0.083*** (0.025)

-0.719*** (0.055) 0.226*** (0.056)

-0.726*** (0.076) 0.194** (0.095)

-0.682*** (0.076) 0.144 (0.101)

-0.726*** (0.076) 0.195** (0.095)

logCs,t−1 (90)

logCs,t−1 (80) ∆MT Rs,t−1 (50) MT Rs,t−1 (50) Income

(50)

log Incomes,t−1 (10) s,t−1 Income

(90)

log Incomes,t−1 (50) s,t−1

0.007 (0.009) -0.066*** (0.013) -0.682*** (0.050) 0.176*** (0.048)

0.026* (0.014) -0.064** (0.030) -0.688*** (0.073) 0.146 (0.110)

-0.020**

-0.006

-0.005

(0.009)

(0.020)

(0.020)

0.045**

0.022

0.036

(0.022)

(0.049)

(0.048)

-0.079*** (0.016) 0.001 (0.089)

-0.052* (0.028) 0.131 (0.191)

0.054* (0.031) -0.061*** (0.014) 0.022 (0.089)

-0.024**

-0.039**

-0.023**

-0.030

-0.029

-0.030

(0.011) 0.734*** (0.192) -0.065*** (0.015) 0.037 (0.085)

(0.017) 0.129 (0.454) -0.087*** (0.024) -0.054 (0.140)

(0.011) 0.651*** (0.169) -0.069*** (0.014) 0.029 (0.086)

(0.027) 0.219 (0.397) -0.088*** (0.025) -0.053 (0.158)

(0.023) 0.156 (0.508) -0.086*** (0.026) -0.094 (0.160)

(0.027) 0.219 (0.397) -0.088*** (0.025) -0.055 (0.158)

Y

Y

Y

Y

Y

Y

0.123 44.142 1.000

0.149 44.621 1.000

0.123 44.147 1.000

Ginis,t−1 log GDPs,t−1 log Schools,t−1 gI

s,t−1 log GDP s,t−1

ns,t−1 min ∆Ws,t−1

∆Unions,t−1

δs M3 (p-val) Hansen J statistic Hansen (p-val) R2 # Observations # States Means: Dep. Variable logC(10), logC(20) logC(90), logC(80) C(10),C(20) C(90),C(80)

No Charitable Contributions Instrument

OLS + Xt−1 + δs

logCs,t−1 (10)

logCs,t−1 (20)

Gini as Income Inequality Control

0.170 43.977 1.000

0.034 (0.065) -0.037* (0.022) 0.034 (0.165)

-0.037 (0.027) 0.048 (0.244)

0.034 (0.065) -0.037 (0.022) 0.033 (0.165)

0.065 983 49

983 49

0.099 983 49

983 49

983 49

983 49

0.0154 -2.2880 -1.9467 0.1042 0.1445

0.0154 -2.2880 -1.9467 0.1042 0.1445

0.0154 -2.4338 -1.7757 0.0902 0.1707

0.0154 -2.4338 -1.7757 0.0902 0.1707

0.0154 -2.4338 -1.7757 0.0902 0.1707

0.0154 -2.4338 -1.7757 0.0902 0.1707

Significance Levels: * 10% ** 5% *** 1% The exogenous instruments for the IV GMM estimation are the narratively identified exogenous tax shocks to personal and corporate income from Mertens and Ravn (2013), as well as these shocks interacted with initial conditions on state level charitable donations and inequality levels. Internal instruments include the predetermined control variables, such as inequality level, GDP per capita, schooling, state investment, population growth, and labor market institutions at t − 3 and t − 4. OLS standard errors are clustered at the state level. IV GMM model uses robust, two step System GMM estimator with Windmeijer-corrected standard errors.

47

Table VIII: Robustness 2 - The Effects of Log Contraction Factors on State Level GDP Growth This table shows our regression results for our main specification with state level real GDP per capita growth as our dependent variable. As a further robustness check, we consider two different income distributions to calculate our income percentiles of interest: 10th, 50th, and 90th. Columns (1) through (3) use the state level income distribution instead of the national distribution to calculate contraction factors and inequality ratio controls. Columns (4) through (6) use the national level market income distribution to calculate analogous contraction factors and inequality ratios. Market income is created by subtracting the taxable portions of transfer income, such as social security benefits and unemployment compensation, from our original income distribution. Columns (1) and (3) show the basic OLS model with log contraction factors, marginal tax rate variables, and state fixed effects. Columns (2) and (4) use the complete OLS model by adding the full set of dynamic controls from the previous year. Lastly, Columns (3) and (6) show the complete IV GMM estimation results. The lower panel of the table shows regressions statistics: the use of state fixed effects, R2 , number of observations and states, the mean of the dependent variable and log contraction factors, and IV GMM statistics such as autocorrelation tests, Hansen statistic, and number of instruments. Specification: Income (90) Income (50) ∆ log GDPs,t = α0 + α1 logCs,t−1 (10) + α2 logCs,t−1 (90) + α3 ∆MT Rs,t−1 (50) + h1 log Incomes,t−1 (10) + h2 log Incomes,t−1 (50) + Xs,t−1 β + δs + εs,t s,t−1

State Income Distribution Model Regressors logCs,t−1 (10) logCs,t−1 (90) ∆MT Rs,t−1 (50) MT Rs,t−1 (50) Income

OLS + Xt−1 + δs

IV GMM

OLS + δs

OLS + Xt−1 + δs

IV GMM

(1)

(2)

(3)

(4)

(5)

(6)

0.023*** (0.008) -0.012 (0.019) -0.560*** (0.057) 0.084* (0.042)

0.005 (0.010) -0.050** (0.019) -0.565*** (0.052) 0.156*** (0.045)

0.032* (0.019) -0.070** (0.032) -0.709*** (0.083) 0.244** (0.118)

0.040*** (0.010) -0.038** (0.017) -0.656*** (0.062) 0.101** (0.048)

0.009 (0.012) -0.067*** (0.021) -0.671*** (0.050) 0.179*** (0.054)

0.024 (0.018) -0.071** (0.032) -0.679*** (0.077) 0.144 (0.109)

(50)

(90)

log Incomes,t−1 (50) s,t−1 log GDPs,t−1 log Schools,t−1 gI

s,t−1 log GDP s,t−1

ns,t−1 min ∆Ws,t−1

∆Unions,t−1

δs

National Market Income Distribution

OLS + δs

log Incomes,t−1 (10) s,t−1 Income

s,t−1

Y

-0.022**

-0.009

-0.020**

0.001

(0.008)

(0.019)

(0.009)

(0.022)

0.058***

0.047

0.038*

0.020

(0.016) -0.065*** (0.014) -0.017 (0.097)

(0.043) -0.063*** (0.023) 0.211 (0.219)

(0.021) -0.068*** (0.015) 0.017 (0.090)

(0.042) -0.045 (0.033) 0.100 (0.212)

-0.019

-0.034

-0.022*

-0.035*

(0.012) 0.638*** (0.175) -0.076*** (0.014) 0.019 (0.086)

(0.025) -0.043 (0.506) -0.104*** (0.025) -0.100 (0.153)

(0.012) 0.692*** (0.173) -0.069*** (0.015) 0.028 (0.084)

(0.019) 0.171 (0.475) -0.088*** (0.027) -0.078 (0.163)

Y

M3 (p-val) Hansen J statistic Hansen (p-val)

Y

Y

Y

0.243 43.452 1.000

Y 0.160 43.868 1.000

R2 # Observations # States

0.062 983 49

0.051 983 49

983 49

0.086 983 49

0.074 983 49

983 49

Means: Dep. Variable logC(10) logC(90) C(10) C(90)

0.0154 -2.4546 -1.7915 0.0893 0.1684

0.0154 -2.4546 -1.7915 0.0893 0.1684

0.0154 -2.4546 -1.7915 0.0893 0.1684

0.0154 -2.4173 -1.7698 0.0914 0.1717

0.0154 -2.4173 -1.7698 0.0914 0.1717

0.0154 -2.4173 -1.7698 0.0914 0.1717

Significance Levels: * 10% ** 5% *** 1% Contraction factors and income inequality ratios are calculated using the state income distribution instead of the national income distribution in Columns (1) through (3). Columns (4) through (6) use the national level market income distribution, which is the national level income distribution less any taxable portions of social security benefits and unemployment compensation. The instruments used in the IV GMM estimation are the same as in previous tables. OLS standard errors are clustered at the state level. IV GMM model uses robust, two step System GMM estimator with Windmeijer-corrected standard errors.

48

Figure I: Pre-Tax and Post-Tax Income Distribution

Before-Tax

After-Tax

Mean(Income) 49,548 Mean(LogIncome) 10.198 SD(Income) 298,123 SD(LogIncome) 1.188

42,508 10.115 236,306 1.139

0

.1

Density .2

.3

.4

This figure shows the distribution, mean, and standard deviations of the logs of pre-tax and post-tax income across income groups in all states from 1979 to 2008. The figure is shown from the 5th to 95th percentile of the pre-tax log income distribution to better see the compression in the distribution caused by progressive taxation. The means and standard deviations shown are for the entire national income distribution.

8.5

9

9.5

10 Log(Income)

10.5

11

11.5

Before-Tax Income

After-Tax Income

Before-Tax Normal Density

After-Tax Normal Density

Figure II: Contraction Factor Distribution This figure shows the distribution of contraction factors that cover all states from 1979 to 2008.

15 0

0

5

10

10

Density

Density

20

20

25

90th Percentile Contraction Factor

30

10th Percentile Contraction Factor

.04

.06

.08

.1

.12

.14

.12

Cs,t-1(10)

.14

.16

.18

.2

Cs,t-1(90)

49

.22

.24

Figure III: Contraction Factors across States: Standard Deviation This figure illustrates the state standard deviations for both contraction factors, C(90) and C(10). The darker states represent a higher variation in contraction factor.

(a) Contraction factor C(90)

(b) Contraction factor C(10)

50

Appendix: For Review and Online Publication Only A

Tax Definitions

The IRS provides detailed definitions for total tax liability for every year it publishes a report. Below are the definitions from 1979 and 2008, the first and last years in our current panel of data. In 1979, total tax liability was the sum of income tax after credits, additional tax for tax preferences, self-employment tax, social security tax on tips, tax from recomputing prior-year investment credit, taxes from individual retirement arrangements, and other taxes, reduced by the earned income credit used to offset all other taxes. In 2008, total tax liability was the sum of income tax after credits, self-employment tax, social security and Medicare tax on tips, additional tax on HSA and MSA distributions, tax from recapturing prior-year investment credits, low income housing credit, qualified electric vehicle credit, Indian employment credit, new market credit, employer-provided child care facilities credit, alternative motor vehicle credit, alternative fuel vehicle refueling property credit, tax from recapture of federal mortgage subsidy, taxes from qualified plans (including individual retirement accounts) and other tax favored accounts, Section 72 penalty taxes, household employment taxes, tax on golden parachute payments, Form 4970 tax, excise tax on insider stock compensation from an expatriated corporation, and interest on tax due on installment income from sale of residential lot and timeshares. These taxes were then reduced by the earned income credit used to offset all other taxes, first-time homebuyer credit, recovery rebate credit and the refundable prior year minimum tax credit, limited to zero. For the statistics, unlike the Form 1040, total tax liability does not include any advance earned income credit payments.

B

Additional Tables

The following tables contain further statistics such as data availability, averages by states for instruments and controls, cross correlation tables, and first stage regressions.

51

Table B.I: Data Availability in Sample This table shows the availability of data in our sample constructed from Taxsim. We drop any state that has less than 500 observations in any given year. Furthermore, Alaska is excluded in all years from our specification due to sample size. The table shows which years for each state are dropped due to low sample size. The table also shows the minimum, maximum, and average observations for each states in a year. Data is not available for any state in 1982.

State

Alabama Alaska Arizona Arkansas California Colorado Connecticut Delaware Florida Georgia Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina North Dakota Ohio Oklahoma Oregon Pennsylvania Rhode Island South Carolina South Dakota Tennessee Texas Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming Total

Years of Data

29 3 29 25 29 29 29 4 29 29 8 5 29 29 28 29 28 29 6 29 29 29 29 22 29 4 17 21 14 29 8 29 29 3 29 29 28 29 4 29 3 29 29 13 3 29 29 17 29 3

Years Available

79-81, 83-08 79-81 79-81, 83-08 79-81, 83, 85, 87, 89, 91-08 79-81, 83-08 79-81, 83-08 79-81, 83-08 79-81, 98 79-81, 83-08 79-81, 83-08 79-81, 83, 85, 92, 98, 00 79-81, 96-97 79-81, 83-08 79-81, 83-08 79-81, 83-85, 87-08 79-81, 83-08 79-81, 83-85, 87-08 79-81, 83-08 79-81, 83, 85, 98 79-81, 83-08 79-81, 83-08 79-81, 83-08 79-81, 83-08 79-81, 83-85, 87, 89, 91, 95-01, 03-08 79-81, 83-08 79-81, 00 79-81, 87, 93, 96-99, 01-08 79-81, 86-91, 96, 98-08 79-81, 92, 94-95, 97, 00, 03-08 79-81, 83-08 79-81, 86-90 79-81, 83-08 79-81, 83-08 79-81 79-81, 83-08 79-81, 83-08 79-81, 83-86, 87-08 79-81, 83-08 79-81, 91 79-81, 83-08 79-81 79-81, 83-08 79-81, 83-08 79-81, 90, 98, 01-08 79-81 79-81, 83-08 79-81, 83-08 79-81, 83, 85, 87-90, 01-08 79-81, 83-08 79-81

1079

52

Annual Observations in State Sample Mean Minimum Maximum 2194 921 875 511 10291 914 1123 739 4238 1214 620 962 3680 1142 629 587 727 823 565 1172 1965 2481 1236 501 1115 942 503 642 506 2806 1134 6121 1480 744 2480 651 643 3265 831 730 729 1247 4650 522 1030 1424 1299 538 776 706

4590 3238 3382 2096 20758 3261 2991 1655 9138 4197 3185 2981 8382 3241 3235 2288 2744 2467 3009 3356 4248 5894 3470 2293 3050 2783 2113 2943 2462 6462 3339 13237 3169 2033 6903 2868 2604 7218 2062 2028 2304 2810 11515 4162 1711 4033 3096 1888 2555 2672

3336.069 2299.333 1665.793 790.24 15513.14 2130.172 2003.724 1167.5 6746.414 2741.793 1498.875 1763 5641.483 2037.483 1159.893 1040.276 1417.786 1413.483 1507.667 2426.103 3066 4469.414 2086.655 943.091 1833.034 1902.25 833.353 1524.095 996.714 4462.483 1725.375 8978 2508.069 1508 4626.655 1361.379 1241.429 4862 1401 1165.793 1647.333 1980.793 8011.793 1242.923 1453.667 2818.069 2195.69 930.353 1503.207 1930.333

501

20758

3061.833

Table B.II: Average State GDP Growth Rates and Contraction Factors We pool the state level GDP, GDP growth rate, both contraction factors, and the annual change in overall marginal tax rate at the 50th percentile across 49 states from 1979 to 2008 (GDP growth rate is from 1980 to 2009). This table shows the pooled averages for each of the economic growth variables, contraction factors, and change in marginal tax rate at the median income. Real Chained GDP per Capita (2009 $)

Real Chained GDP Growth Rate (%)

Contraction Factor (90th Percentile)

Contraction Factor (10th Percentile)

Change in Median Marginal Tax Rate

Alabama Arizona Arkansas California Colorado Connecticut Delaware Florida Georgia Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina North Dakota Ohio Oklahoma Oregon Pennsylvania Rhode Island South Carolina South Dakota Tennessee Texas Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming

29,999 33,583 29,658 42,569 42,028 52,587 44,958 35,195 37,799 40,828 23,762 42,493 35,171 35,086 36,275 32,164 39,354 26,264 42,580 46,276 37,077 40,695 26,366 36,431 28,603 40,144 45,699 38,494 47,774 25,066 47,619 36,974 25,777 36,318 31,990 32,359 37,053 28,041 31,640 20,096 34,294 39,233 36,708 24,457 41,815 44,809 27,915 36,520 38,914

1.53 1.27 1.86 1.54 1.59 2.18 1.54 1.23 1.61 0.16 -0.57 1.50 1.39 1.99 1.33 1.29 1.26 2.68 1.70 2.26 0.32 1.82 0.97 1.32 -0.13 1.91 -0.29 1.55 1.74 0.99 1.77 1.63 1.22 1.30 1.62 2.37 1.59 0.78 1.45 -0.64 1.54 1.46 0.53 2.60 1.79 1.22 1.12 1.56 -2.02

0.1769 0.1676 0.1841 0.1596 0.1631 0.1751 0.2066 0.1816 0.1685 0.1672 0.1901 0.1764 0.1776 0.1707 0.1743 0.1727 0.1812 0.2106 0.1559 0.1657 0.1716 0.1548 0.1757 0.1774 0.2024 0.1717 0.1722 0.1651 0.1707 0.1870 0.1642 0.1662 0.2212 0.1727 0.1716 0.1598 0.1779 0.1917 0.1617 0.2180 0.1816 0.1859 0.1458 0.1932 0.1687 0.1741 0.1873 0.1601 0.2121

0.0805 0.0838 0.0818 0.0883 0.0932 0.0985 0.1088 0.0900 0.0863 0.1089 0.0938 0.0918 0.0941 0.0931 0.0885 0.0873 0.0866 0.1080 0.0943 0.1041 0.0914 0.0967 0.0780 0.0902 0.1125 0.0863 0.0882 0.0991 0.0931 0.0967 0.0918 0.0874 0.1123 0.0988 0.0867 0.0923 0.0963 0.1263 0.0890 0.1167 0.0881 0.0865 0.0759 0.1061 0.0903 0.0948 0.0889 0.0996 0.1250

0.00118 0.00068 -0.00001 -0.00170 0.00087 0.00029 0.00187 0.00076 0.00182 0.00256 0.01173 0.00109 0.00293 0.00212 0.00124 0.00151 0.00340 0.00330 0.00187 -0.00015 0.00034 -0.00131 -0.00078 0.00100 0.00207 0.00173 0.00165 0.00499 0.00055 0.00366 -0.00016 0.00169 -0.00390 0.00027 0.00047 0.00131 0.00095 0.00354 0.00019 0.00305 0.00138 0.00200 0.00086 -0.00006 0.00158 0.00049 0.00436 0.00341 0.00405

Total

37,704

1.46

0.1721

0.0914

0.00114

State

53

Table B.III: Contraction Factors and Inequality Cross-correlation This table shows the cross-correlations between our contraction factors, income inequality ratios, and the Gini coefficient. Panel A presents the standard correlations, while Panel B uses demeaned variables to show correlations with state fixed effects.

Panel A: Standard Cross-Correlation

Variables

Cs,t (90)

Incomes,t (90) Incomes,t (50)

Cs,t (10)

Incomes,t (50) Incomes,t (10)

Ginis,t

Cs,t (90)

1.000

Cs,t (10)

0.706

1.000

Incomes,t (90) Incomes,t (50)

-0.360

-0.586

1.000

Incomes,t (50) Incomes,t (10)

-0.202

-0.102

-0.030

1.000

Ginis,t

-0.387

-0.524

0.758

0.153

1.000

Incomes,t (50) Incomes,t (10)

Ginis,t

Panel B: State Fixed Effects Cross-Correlation

Variables

Cs,t (90)

Incomes,t (90) Incomes,t (50)

Cs,t (10)

Cs,t (90)

1.000

Cs,t (10)

0.819

1.000

Incomes,t (90) Incomes,t (50)

-0.538

-0.638

1.000

Incomes,t (50) Incomes,t (10)

-0.266

-0.277

0.166

1.000

Ginis,t

-0.584

-0.626

0.753

0.297

54

1.000

Table B.IV: Averages for State Inequality and Fiscal Controls We pool our income ratios between 50th and 10th, and 90th and 50th percentiles, log state level spending on goods and services as a ratio of GDP, and median marginal tax rates across 49 states from 1979 to 2008. This table shows the pooled averages for each of our state inequality and fiscal controls. Income Ratio: 50th to 10th Percentile

Income Ratio: 90th to 50th Percentile

Log of State Expenditure as Fraction of GDP

Median Marginal Tax Rate (%)

Alabama Arizona Arkansas California Colorado Connecticut Delaware Florida Georgia Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina North Dakota Ohio Oklahoma Oregon Pennsylvania Rhode Island South Carolina South Dakota Tennessee Texas Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming

5.519 5.018 5.485 5.470 6.288 6.712 5.318 4.910 5.400 5.793 5.468 6.589 6.703 6.302 6.222 5.581 5.615 5.106 6.299 6.297 6.854 6.599 5.257 5.831 5.443 6.509 4.500 7.052 6.538 6.357 5.806 5.366 7.062 6.130 5.610 5.768 6.613 5.202 5.062 6.720 5.806 5.798 5.607 5.476 6.073 6.043 5.772 6.219 6.632

4.002 3.200 3.111 3.343 3.058 3.257 3.133 3.383 3.280 3.317 3.807 3.190 2.893 2.895 3.199 3.144 3.407 2.695 3.070 3.072 3.175 3.006 3.366 3.176 4.254 3.102 3.547 2.940 3.213 3.332 3.271 3.177 2.763 2.952 3.184 2.954 3.092 2.729 3.135 2.898 3.124 3.421 2.916 3.043 3.145 2.781 3.106 2.983 2.436

-2.806 -2.969 -2.760 -3.127 -3.208 -3.509 -3.111 -3.250 -3.114 -3.141 -2.747 -3.408 -3.039 -2.829 -2.967 -2.814 -3.118 -3.068 -3.216 -3.448 -2.998 -2.977 -2.722 -3.230 -2.899 -3.128 -3.295 -3.321 -3.381 -2.521 -3.368 -2.978 -2.787 -3.125 -2.857 -2.883 -3.216 -3.194 -2.916 -2.856 -3.185 -3.232 -2.764 -2.752 -3.123 -3.016 -2.624 -3.029 -3.065

36.28 34.87 35.54 33.95 35.23 33.25 38.61 30.28 37.90 39.16 37.20 34.21 34.73 36.12 37.19 36.21 35.41 35.53 37.39 35.86 36.96 38.84 37.44 35.78 33.79 35.86 32.11 31.48 32.94 31.59 37.61 38.00 32.64 34.81 35.88 37.28 33.83 34.53 38.11 29.09 32.53 32.20 35.89 34.72 36.27 29.85 34.49 40.18 31.62

Total

5.923

3.171

-3.079

35.37

State

55

Table B.V: Averages for Other Controls We pool our levels of schooling, population growth, change in effective state minimum wage, and change in percent of labor force covered by unions across 49 states from 1979 to 2008. This table shows the pooled averages for each of our other controls.

Log Years of Schooling

Population Growth (%)

Change in Effective Minimum Wage (%)

Change in Labor Force Union Coverage (%)

Alabama Arizona Arkansas California Colorado Connecticut Delaware Florida Georgia Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina North Dakota Ohio Oklahoma Oregon Pennsylvania Rhode Island South Carolina South Dakota Tennessee Texas Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming

2.510 2.541 2.507 2.535 2.590 2.584 2.537 2.540 2.534 2.547 2.533 2.556 2.532 2.561 2.578 2.508 2.516 2.506 2.574 2.581 2.550 2.582 2.512 2.552 2.543 2.569 2.539 2.588 2.568 2.508 2.554 2.526 2.511 2.548 2.544 2.565 2.548 2.503 2.516 2.495 2.505 2.513 2.577 2.532 2.559 2.580 2.486 2.562 2.541

0.70 2.99 0.93 1.54 1.83 0.46 0.36 2.30 1.97 1.20 1.84 0.39 0.57 0.19 0.61 0.59 0.18 0.48 1.04 0.42 0.29 0.91 0.56 0.68 0.49 0.55 4.36 0.93 0.58 1.37 0.30 1.65 0.56 0.23 0.63 1.39 0.21 0.02 1.33 0.04 1.13 1.86 2.29 1.00 1.34 1.72 -0.22 0.67 4.22

-0.36 0.32 -0.13 0.85 0.39 0.58 1.38 0.18 -0.36 -0.57 1.24 0.61 -0.36 0.54 -0.36 -0.30 -0.36 0.02 -0.16 0.77 0.56 -0.18 -0.28 0.15 -1.77 0.01 0.58 -0.28 0.41 -0.75 0.37 -0.16 -1.55 0.38 -0.36 0.89 0.44 -0.59 -0.36 -1.55 -0.36 -0.36 0.65 2.44 -0.36 0.79 0.10 0.01 -1.55

-0.53 -0.29 -0.32 -0.35 -0.40 -0.32 0.73 -0.22 -0.46 0.14 0.40 -0.51 -0.62 -0.35 -0.42 -0.47 -0.43 -0.14 -0.36 -0.29 -0.68 -0.56 -0.21 -0.47 -0.63 0.16 -0.35 0.57 -0.50 -0.20 -0.48 -0.25 1.00 -0.68 -0.22 -0.53 -0.61 1.16 -0.11 1.05 -0.60 -0.35 0.05 -0.20 -0.40 -0.55 -0.13 -0.57 -1.05

Total

2.546

1.04

0.10

-0.38

State

56

Table B.VI: The Effects of Exogenous Instruments on Log Contraction Factors This table shows results on the relationship between our log contraction factors and exogenous instruments. Columns (1) through (3) show the effects of our exogenous instruments on the below median contraction factor, while Columns (4) through (6) show the effect of exogenous instruments on the above median contraction factors. Columns (1) and (4) show estimates for the effect of personal and corporate income tax shock on contraction factors, while controlling for initial conditions of inequality and charitable giving without state fixed effects. Columns (2) and (5) add state fixed effects, while changing the initial condition variables to become interaction terms with personal income tax shocks. Columns (3) and (6) show a similar set of estimates for corporate income tax shocks, interaction terms, and state fixed effects. The lower panel of the table shows regressions statistics: the use of state fixed effects, R2 , F statistics, Kleibergen-Paap Lagrange-Multiplier (LM) statistic and Wald F statistic, number of observations and states, and the mean of the dependent contraction factor. Specification: logCs,t−1 (i) = α0 + Zs,t−1 γ + δs + εs,t logC(10)

Model Regressors Personal Income Tax Shockt−1 × log

(1)

(2)

0.170*** (0.013)

0.845* (0.437)

Incomes,1979 (50) Incomes,1979 (10) Income

(90)

× log Incomes,1979 (50) s,1979 × CCs,1979 Corporate Income Tax Shockt−1 × log

logC(90) (3)

(4)

(5)

0.089*** (0.008)

0.343 (0.341)

-0.485***

-0.041

(0.159)

(0.102)

0.132

-0.211

(0.272) 3.331 (4.300)

(0.205) 1.502 (1.440)

0.012*** (0.002)

0.177 (0.143)

Incomes,1979 (50) Incomes,1979 (10)

-0.002 (0.002)

(6)

0.138** (0.065)

-0.074

-0.041*

(0.046)

(0.024)

× log Incomes,1979 (50) s,1979

-0.040

-0.069*

× CCs,1979

(0.092) 0.805 (1.814)

(0.035) 0.377 (0.493)

Income

Income

(90)

(50)

log Incomes,1979 (10) s,1979 Income

(90)

log Incomes,1979 (50) s,1979 CCs,1979

0.187*

-0.052

(0.096)

(0.091)

-0.162

0.088

(0.170) -12.439*** (2.293)

(0.163) -5.339*** (1.968)

N

Y

Y

N

Y

Y

R2 F statistic p value Kleibergen-Paap LM statistic p value Kleibergen-Paap Wald F statistic # Observations # States

0.115 62.452 0.0000 30.625 0.0000 122.100 907 49

0.058 168.282 0.0000 30.115 0.0000 159.327 907 49

0.034 58.081 0.0000 22.907 0.0001 54.991 907 49

0.049 32.908 0.0000 29.831 0.0000 75.166 907 49

0.030 50.018 0.0000 28.275 0.0000 47.356 907 49

0.007 11.072 0.0000 15.231 0.0042 10.483 907 49

Means: Log Contraction Factor Contraction Factor

-2.4020 0.0926

-2.4020 0.0926

-2.4020 0.0926

-1.7634 0.1727

-1.7634 0.1727

-1.7634 0.1727

δs

Significance Levels: * 10% ** 5% *** 1% Standard errors are clustered at the state level. The F statistics for the first stage using our IV GMM estimation are as follows: 578.8270 (p = 0.000) for C(10) and 196.8728 (p = 0.000) for C(90). The IV GMM model includes exogenous tax shock instruments at t − 1 and internal control instruments at t − 3 and t − 4.

57

C

Additional Figures Figure C.I: Graphical Interpretation of Contraction Factors

This figure illustrates the calculation and change in the contraction factors in relation to the median household (reference point R = (Income(50), Tax(50))). The graph plots the income distribution and taxes paid at the 10th and 90th percentile. In subfigure (a), the slope of the solid line from R to (Income(90), Tax(90)) represents the contraction factor between the 90th percentile households and the median households, C(90). As the tax rate at the 90th percentile increases, while the median taxation is held fixed, the dotted line illustrates an increase in C(90). In subfigure (b), the slope of the solid line from R to (Income(10), Tax(10)) represents the contraction factor between the 10th percentile households and the median households, C(10). As the tax rate at the 10th percentile decreases, while the median taxation is held fixed, the dotted line illustrates an increase in C(10). An increase in the contraction factor between the ith household and the median household, C(i), signifies a further reduction in income inequality between the ith percentile and the reference median household due to taxation.

Tax(90)

C(90)

C(90)

Income(10)

R

Income(90)

Tax(10)

(a) Contraction factor C(90)

Tax(90)

Income(10)

C(10)

Income(90)

R

C(10) Tax(10)

(b) Contraction factor C(10) The contraction factors are in relation to reference point R = (Income(50), Tax(50))

58

Figure C.II: Contraction Factors across States: Average This figure illustrates the state averages for both contraction factors, C(90) and C(10). The darker states represent a higher average contraction factor.

(a) Contraction factor C(90)

(b) Contraction factor C(10)

59

Figure C.III: Average Contraction Factors and Narrative Shocks Measures

-.25 -.5 -.75 -1

Cs,t(90)

Cs,t(10)

60

zpi,t

08 20

06 20

04 20

02 20

00 20

98 19

96 19

94 19

92 19

90 19

19

86 19

84 19

82 19

80 19

88

Year

Narratively Identified Personal Income Tax Shock

.25 0

.05 0 -.05 -.1 -.15 -.2 -.25

Contraction Factor

.1

.5

.15

.75

.2

1

.25

This figure shows the average contraction factors that cover all states from 1979 to 2008. The narrative shocks measures are shocks to personal income liability given by Mertens and Ravn (2013)