FEDERAL AND STATE TAX CUTS ANALYSIS

21 To address these endogeneity concerns, I perform a series of tests exploiting federal and state tax cuts to isolate the influence of changes in the tax rate on corporate risk-taking.

V. FEDERAL AND STATE TAX CUTS ANALYSIS

Federal Tax Cuts: Research Design First I examine the federal tax cut contained in the Taxpayer Relief Act of 1997 TRA97 which reduced the top individual capital gains tax rate on long-term investments from 28 to 20, effective May 7, 1997. This tax act has several features which make it a desirable research setting. First, the change in the tax rate is economically significant – nearly a 30 relative drop in the maximum rate compared to the previous tax regime. Second, the capital gains tax reduction was the most noteworthy feature of the TRA97, which contained few other provisions. 18 Third, the tax reduction was largely unexpected. On April 30, 1997, the Congressional Budget Office made a surprise announcement that it had reduced its estimate of the 1997 deficit by 45 billion. Two days later, on May 2, President Clinton and Congressional leaders announced an agreement to balance the budget by 2002 and reduce the capital gains tax rate. The following week, on May 7, Senate and House leaders announced that the effective date for any capital gains tax cut would be May 7, 1997 Lang and Shackelford 2000. Given the unexpected and swift nature of the enactment of TRA97, it is unlikely that CEOs anticipated the tax cut and adjusted their behavior preemptively. Although the legislative change affected all CEOs simultaneously, I hypothesize that it had a stronger effect on CEOs with high tax burdens just prior to the tax cut. My prediction rests on two main arguments. First, CEOs with high tax burdens are more likely to be under-diversified as a result of the tax lock-in effect, and thus the most eager to take advantage of a reduction in the tax cost of selling. Second, CEOs with high tax burdens are likely to experience the greatest benefit from the tax cut, as they are limited in their other options for reducing their tax bills upon a sale of 18 Other tax legislation included in TRA97 consisted of a child tax credit, estate tax relief, education tax incentives, and an expansion of individual retirement accounts Dai, Shackelford, and Zhang 2013. 22 stock for example, they are less likely than low-tax-burden CEOs to have some stock in a loss position that they can use to offset a taxable gain. Since high-tax-burden CEOs are likely to benefit the most from incremental diversification and stand to gain the most from a reduction in tax rates, I predict a larger increase in risk-taking among these executives following the tax cut. I construct a six-year window of observations surrounding the tax cut and require all CEO- firm pairs to have at least four consecutive years of data to be included in the sample. I compute each CEO’s tax burden at the end of 1996 just prior to the tax cut, and estimate the following regression: Firm Risk i,t = α i + β 1 Post-Fed Tax Cut t + β 2 Post-Fed Tax Cut t × Pre-Tax Cut Burden i + ɤ′X i,t-1 + ϵ i,t . 3 Year t runs from 1995 through 2000. Thus Firm Risk i,t includes firm i ’s risk-taking in two years before, the year of, and three years following the tax cut in order to allow sufficient time for any change in risk-taking to manifest. 19 Post-Fed Tax Cut t is an indicator variable equal to one for years 1998 onwards, and equal to zero in the earlier years. Pre-Tax Cut Burden is equal to the CEO’s tax burden measured as of the end of 1996, just prior to the tax cut. Firm Risk is defined as previously, X represents the same control variables as in Equation 2, and α i represents firm fixed effects. 20 I predict a positive β 2 indicating that CEOs with higher tax burdens prior to the tax cut will have a larger subsequent increase in risk-taking relative to low-tax-burden CEOs. Next I perform a similar analysis for the period surrounding the capital gains tax cut contained in the Jobs and Growth Tax Relief Reconciliation Act of 2003 JGTRRA, which reduced the top individual capital gains tax rate from 20 to 15. Compared with the tax cut in TRA97, JGTRRA is a relatively noisy setting in that the act contained other significant tax 19 Refer to Figure 2 for a graphical representation of the time periods over which the variables are measured and tests are conducted. 20 Note that the main effect Pre-Tax Cut Burden i is excluded from Equation 3 because it reflects CEO tax burden values measured in just one year for each of the two federal tax cuts CEO Tax Burden is measured in 1996 for the first federal tax cut, and in 2002 for the second federal tax cut. Thus it is subsumed by firm fixed effects for the federal-level tests. 23 legislation that could confound my inferences. For example, JGTRRA reduced the top dividends tax rate from 39.6 to 15, the top ordinary income tax rate from 38 to 35, and contained bonus depreciation incentives designed to boost corporate investment Blouin, Raedy, and Shackelford 2011. While the impact of JGTRRA on corporate activity has been debated e.g., Chetty and Saez, 2005; Campbell, Chyz, Dhaliwal, and Schwartz 2013; Yagan, 2015, the range of major tax changes in the act makes it difficult to draw clean inferences about the capital gains tax cut using this setting. A potential concern related to using the difference-in-differences methodology for the federal tax cut tests is whether firms with high-tax-burden CEOs are comparable to firms with low-tax-burden CEOs. I attempt to mitigate these concerns by partitioning each sample at the median into high and low tax burden CEOs just prior to each tax cut using the end-of-year values for 1996 and 2002 for the 1997 and 2003 tax cuts, respectively, and using propensity score matching to more closely match the characteristics of the two groups of firms. 21 Panels A and C of Table 3 show the results of the matching procedure for the samples used in the 1997 and 2003 federal tax cut tests, respectively. In both cases, the matched treatment and control groups display no significant differences in any of the matching variables. Federal Tax Cuts: Results Panels B and D of Table 3 present the results from estimating Equation 3 for the two federal tax cuts. Examining the main effect on Post-Fed Tax Cut in column 1 of Panel B, there is a positive but insignificant increase in total return volatility for CEOs with a tax burden of zero in the three years following the 1997 tax cut, coef.=0.022; t-stat.=1.49. However, the coefficient on the interaction term Post-Fed Tax Cut × Pre-Tax Cut Burden shows that high-tax-burden CEOs 21 Using the sample of firm-years available just prior to each of the two federal tax cuts i.e., 1996 and 2002, I match firms with an above-the-median tax burden CEO to a corresponding firm with a below-the-median tax burden CEO. To construct propensity scores, I estimate a probit regression where the dependent variable is an indicator equal to one zero for high low burden CEOs, and the independent variables include all of the control variables in Equation 2 except for CEO Tax Burden. I use nearest neighbor matching within caliper set at 0.50. 24 experienced a substantially larger increase in corporate risk-taking following the tax cut coef.=0.214; t-stat.=2.39. Economically, I find that a CEO at the 25 th percentile of CEO Tax Burden in 1996 experiences a 2.4 increase in total return volatility, whereas a CEO at the 75 th percentile of CEO Tax Burden experiences a 4.6 increase. 22 The results in columns 2 and 3 examining idiosyncratic volatility and earnings volatility yield similar inferences. 23 Panel D displays the results for the analysis of the tax cut contained in JGTRRA of 2003. Although the negative coefficients on Post-Fed Tax Cut indicate that for CEOs with no tax burden prior to the tax cut, risk-taking decreased in the three years following the tax cut, the coefficient on the interaction term Post-Fed Tax Cut × Pre-Tax Cut Burden is consistent with Panel B. Specifically, the results indicate that firms with high-tax-burden CEOs prior to the tax cut experience increased volatility following the tax cut, relative to firms with low-tax-burden CEOs. As noted previously, JGTRRA contained other major tax legislation besides the reduction in the capital gains tax rate, and accordingly these results should be interpreted with caution. The evidence in Table 3 suggests that the federal capital gains tax cuts in 1997 and 2003 disproportionately affected CEOs with higher tax burdens prior to the cuts. I interpret the results as indicating that the tax cuts reduced the tax cost to selling shares for locked-in CEOs, allowing them to dispose of firm stock at a lower cost, diversify their personal wealth, and take on more corporate risk. Overall, the findings in Table 3 provide evidence consistent with my central prediction that higher CEO tax burdens discourage corporate risk-taking. State Tax Cuts: Research Design Next I exploit state-level changes in individual capital gains tax rates. Although smaller in magnitude than the federal tax cuts, state tax rate changes provide several advantages in that they 22 Note that the 25 th and 75 th percentiles of CEO Tax Burden in 1996 were 0.010 and 0.113. Thus the economic magnitude of the effects can be computed by adding the main effect to the respective interaction effect. For the 25 th percentile: 0.022 + 0.010 × 0.214 = 0.024; for the 75 th percentile: 0.022 + 0.113 × 0.214 = 0.046. 23 I verify the results in Table 3 are robust to interacting all control variables with Post-Fed Tax Cut. 25 change frequently, they are staggered over time, and each tax rate change only affects a subset of CEOs, leaving a plausible counterfactual of unaffected CEOs. Similar to the tests examining federal tax cuts, I predict that state tax cuts differentially affect CEOs with high tax burdens relative to CEOs with low tax burdens. I test my prediction by estimating the following model: Firm Risk i,s,t = α i + α t + β 1 |State Tax Rate Cut s,t | + β 2 Pre-Tax Cut Burden i + β 3 |State Tax Rate Cut s,t | × Pre-Tax Cut Burden i + ɤ′X i,t-1 + ϵ i,s,t . 4 Equation 4 is a two-way fixed effects model, containing firm and year fixed effects α i and α t , respectively, with subscript s representing the state in which the firm is headquartered. |State Tax Rate Cut s,t | equals the magnitude of any reduction in the state’s top capital gains tax rate. For each tax cut, |State Tax Rate Cut s,t | is set equal to this value for a three-year period following the tax cut to allow time for any changes in firm risk-taking to manifest see Figure 2 for a graphical illustration of the time period over which the variables are measured. If no tax cut has taken place in state s within the previous three years, then |State Tax Rate Cut s,t | is set equal to zero. Pre-Tax Cut Burden i represents the tax burden of firm i ’s CEO in the year prior to the tax cut. Firm Risk i,s,t represents the three risk proxies measured in year t. I predict a positive β 3 indicating that CEOs with higher tax burdens prior to each state tax cut will experience a larger subsequent increase in risk-taking than low-tax-burden CEOs. 24 State Tax Cuts: Main Results Table 4 presents the results of estimating Equation 4. Consistent with the results from the panel tests in Table 2, the coefficients on Pre-Tax Cut Burden are significantly negative indicating that CEOs with higher tax burdens tend to have lower risk. The coefficient on |State Tax Rate Cut| can be interpreted as the effect of a state tax cut on corporate risk- taking when the firm’s CEO has 24 It is important to note that although state tax changes are likely not exogenous with respect to macro or local economic conditions, they are still useful in identifying the effect of CEO tax burdens on corporate risk-taking for two reasons. First, t he tax cuts allow me to isolate the effect driven by the tax rate component of the CEO’s tax burden, thereby distinguishing it from the effect due to the stock’s historical price appreciation and the CEO’s past portfolio choices. Second, I am able to mitigate concerns that changing economic conditions confound my analysis by comparing the effects of the same tax cuts for CEOs with high versus low tax burdens. 26 a tax burden equal to zero. Although the main coefficients on |State Tax Rate Cut| are insignificant, the coefficients on the interaction term |State Tax Rate Cut| × Pre-Tax Cut Burden are significantly positive across all three measures of risk-taking. The results suggest that tax cuts have the largest effects on CEOs with substantial embedded gains in their equity holdings i.e., locked-in CEOs, consistent with my central hypothesis. 25 Economically, a one percent drop in the long-term gains tax rate leads to a modest 0.5 increase in total return volatility for CEOs at the 25 th percentile of CEO Tax Burden, relative to CEOs in unaffected states. But the same tax cut leads to a statistically significant 2.1 volatility increase for CEOs at the 75 th percentile of CEO Tax Burden. The inferences from the estimations using idiosyncratic volatility and earnings volatility in columns 2 and 3 are similar. For the sake of comparison, Ljungqvist et al. 2017 find that a one percent increase in state corporate tax rates reduces earnings volatility by between 2.4 and 3.2 over the subsequent three years. 26,27 Figure 3 provides a visual depiction of the Table 4 results in a univariate difference-in- differences style analysis. The three lines in the figure represent three groups of firms. The first second group of firms consists of firms with high low tax burdens headquartered in states that experienced tax cuts in year t=0, whereas the third group consists of firms headquartered in states with no tax cut. The figure shows that high-tax-burden CEOs had lower volatility prior to the tax cuts, but experience a substantially greater increase in volatility than unaffected CEOs and affected low-tax-burden CEOs following the tax cuts. 25 I verify that the results in Table 4 are robust to interacting all control variables with |State Tax Rate Cut|. 26 To explore whether my results are driven primarily by the larger tax cuts in the sample, I partition the tax cuts into through groups: small from 0.25 to 0.50, medium from 0.50 to 1.25, and large above 1.25. Replacing |State Tax Rate Cut| with indicators for each group of tax cuts, I find that the largest tax cuts spur the largest increases in risk-taking, followed by medium-size tax cuts which generate marginally significant increases in risk, and finally small tax cuts which generate positive but statistically insignificant increases in risk-taking. 27 In untabulated analysis, I use propensity score matching to more closely match affected CEOs with below-median tax burdens to those with above-median tax burdens within the same state and year. As in the federal tests, I construct propensity scores by estimating a probit regression where the dependent variable is an indicator equal to one zero for high- low- tax burden CEOs, the independent variables are the controls from Equation 2 excluding CEO Tax Burden , and the caliper is set at 0.01. My inferences are unchanged when using this matched sample of firms in the state tax cut tests. 27 State Tax Cuts: SP 500 Firms One potential concern with the analysis so far is that the increases in volatility following federal and state tax cuts may be driven by the tax incentives of outside shareholders rather than those of inside managers. For example, prior research focusing primarily on outside shareholders in the firm has shown that equity volatility increases following federal capital gains tax cuts Dai et al. 2013. Thus to the extent that the tax cuts affect outside shareholders in the firm as well as the CEO, they have the potential to cloud my inferences. Although the potential for outside shareholders to drive changes in firm risk in response to tax cuts is certainly a concern in the federal tax cut setting, it is somewhat less of a concern in the state tax changes setting because investors are taxed based on where they reside, not where the firm is headquartered. For outside shareholders to drive the effects documented in Panel A of Table 4, there would have to exist a very strong within-state home bias among investors. Prior work documents the existence of a local e.g., within-state home bias in investor portfolios, but the effect is driven primarily by smaller firms with less publicly available information Coval and Moskowitz 1999, 2001; Ivkovic and Weisbenner 2005; Hong, Kubik, and Stein 2008. The rationale is that investors have more familiarity with and access to private, value-relevant information for local firms. However, the local information advantage is largely absent with regard to SP 500 companies that are followed by many stock analysts, covered closely by the financial press, and researched and owned by numerous institutional investors, and thus the investor home bias is weaker for these larger, more visible firms. 28 Motivated by the relative lack of investor home bias for SP 500 firms, I re-estimate Equation 4 after restricting the sample to only the firms included in this index. If the increased volatility following tax cuts shown in Panel A of Table 4 is driven by outside shareholders who 28 Specifically, Ivkovic and Weisbenner 2005 find that individual investors place 18 of their actual portfolios into local SP 500 stocks where local is defined as firms headquartered within 250 miles of the investor’s household, whereas these stocks would make up 10 of the portfolio if the households invested in the market. 28 are residents of the state affected by the tax cuts, then this effect is likely to be weaker for firms less susceptible to home bias e.g., SP 500 firms. However, the results for this estimation in Panel B of Table 4 show that in fact the magnitude of the increase in volatility following the tax cuts is somewhat larger for SP 500 firms than it is for the whole sample of firms. This finding helps to mitigate but does not rule out entirely concerns that the equity volatility response to the state tax cuts is driven by outside shareholders who are also residents of the state. 29 State Tax Cuts: Neighboring States Falsification Test In this section I construct a falsification test for the results using the state tax cut setting. Specifically, I re-estimate Equation 4 using the original sample of firms, but I assign “pseudo-tax cuts” to the states that neighbor the “true” tax cut states. For example, in 2010 New Jersey reduced its top long-term gain tax rate from 10.75 to 8.97. In this test, I assign the tax cut to CEOs of firms headquartered in states that share a border with New Jersey i.e., New York, Pennsylvania, and Delaware. This falsification test helps to address two potential concerns. The first is that states may change capital gains tax rates in response to changes in local business conditions. That is, a general economic trend in the region could lead simultaneously to state-level tax cuts and increased corporate risk-taking. While this potential confounding factor is less likely to explain the disparity in the effects of the tax cut on CEOs with high versus low tax burdens, it could explain the general relation between tax cuts and corporate risk-taking. Since presumably economic conditions carry across state borders Heider and Ljungqvist, 2015, failing to find any significant response by firms in neighboring states helps to mitigate concerns that my results are driven by local economic trends. 29 Note that the inferences are similar when using other proxies for firm visibility, such as analyst coverage and firm size as measured by total sales. In all cases, the effects of the tax cuts for the more visible firms are no weaker than, and in some cases are stronger than the effects for the less visible firms. 29 The second concern is that high- and low-tax-burden CEOs are not comparable, and thus risk-taking could diverge for these groups of firms regardless of whether or not a tax cut takes place. By testing for a similar response in states where no tax cut actually took effect, I can estimate the counterfactual paths the two groups of firms may have followed. Failure to find diverging trends in the absence of a tax cut should help build confidence that the previous results are in fact driven by the effects of the tax cuts. The results from the falsification test are presented in Panel C of Table 4. The coefficients on the interaction term |State Tax Rate Cut| × Pre-Tax Cut Burden are statistically and economically insignificant for all measures of corporate risk-taking. Given these results, an alternative explanation would not only have to explain why risk-taking increases significantly for high tax burden CEOs in states affected by tax cuts, but would also have to explain why there is no corresponding difference among CEOs of firms located in neighboring states. This result helps reduce concerns related to local economic conditions as well as whether CEOs with different tax burdens follow comparable trends. Tax Cuts Analysis: Potential Confounding Factors Although the tests analyzing the effects of federal and state tax cuts help to mitigate concerns regarding the potential confounding influences of the historical firm performance and CEO portfolio choices, it is impossible to rule out all potential alternative explanations. As discussed above, one potential concern is that increased volatility following tax cuts is driven by outside shareholders who are also affected by the tax cuts, rather than by changes in managerial incentives. I have attempted to address this concern by controlling for the tax-sensitivity of firm shareholders, by demonstrating strong tax cut responses among SP 500 firms which should be the least affected by within-state home bias, and in subsequent tests by directly tying CEO stock sales to increased risk-taking Table 5 and by demonstrating similar findings with respect to internal corporate decisions Table 7. 30 A second potential concern involves concurrent events, such as changes in corporate income and ordinary income tax rates taking effect at the same time in the same place. To address the concern about corporate tax changes, which have been shown to affect risk-taking Langenmayr and Lester, 2015; Ljungqvist et al. 2017, I re-estimate Equation 4 in untabulated results after excluding all of the long-term gain tax cuts in my sample that occur within three years of a corporate tax cut in the same state. Out of the 73 tax cuts in my sample, I find that 14 of them occur within three years of a corporate tax cut in the same state. 30 My inferences are unchanged when excluding the capital gains tax cuts that overlap with corporate tax cuts. Ordinary income tax rates pose a potential issue because unlike with federal taxes, many states use the same tax rates for both ordinary income and long-term gains. Thus for many states, when capital gains tax rates change, ordinary income tax rates change as well. One setting in which there is a clear drop in capital gains tax rates without a corresponding drop in ordinary tax rates is the federal tax cut contained in TRA97. Panel B of Table 3 shows results consistent with the findings in the other tax cut tests, even in the absence of any substantial change in the ordinary income tax rate. Furthermore, subsequent tests in Table 5 display increased stock sales for locked- in CEOs following tax cuts. Since gains from selling stock are taxed at capital gains tax rates and not ordinary income tax rates, this evidence suggests CEOs are responding to changes in the taxation of capital gains income. Finally, another potential concern is whether individuals and firms adjust compensation practices in response to the changes in tax incentives. In untabulated analysis, I explore this possibility by testing for changes in newly issued delta, vega, cash compensation salaries plus bonuses, and total compensation. In the years following the tax cuts, I do find statistically significant increases in new delta and vega incentives awarded as well as an increase in total 30 This finding is comparable to that in Heider and Ljungqvist 2015, who find sixteen overlapping corporate and capital gains tax cuts over a slightly longer sample period. I apply the same filter to the set of corporate tax cuts as to the capital gains tax cuts, in that I exclude any changes smaller than 0.25 in magnitude. 31 compensation. However, there appears to be no difference in the new compensation awarded for high- versus low-tax-burden CEOs, mitigating concerns that changes in newly awarded compensation are driving the differential effect of the tax cuts for high and low burden CEOs.

VI. ADDITIONAL ANALYSIS