ADDITIONAL ANALYSIS Locked-in: The Effect of CEOs’ Capital Gains Taxes on Corporate Risk-Taking

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

In this section, I perform several additional analyses that serve two main purposes. First, I aim to shed light on the mechanism underlying the relation between capital gains tax changes and corporate risk-taking. Second, I hope to further reduce concerns that alternative explanations could account for my primary results. With these objectives in mind, I study CEO stock sales, the strength of firm responses to the tax cuts conditional on CEO outside wealth, changes in corporate policies as other proxies for risk-taking, and firm responses to tax increases. 31 CEO Stock Sales Tax cuts mitigate the lock-in effect by reducing the tax cost of selling shares, and I predict that CEOs respond to the cuts by selling shares with unrealized gains. In this test, I modify the specification from Equation 4 to display a separate coefficient for each individual year surrounding the tax cuts. The reason for the modification is because I expect CEOs to respond rapidly to the tax cuts in terms of selling appreciated stock, and then the implications of those stock sales for corporate risk-taking manifest over time, as shown in Tables 3 and 4. I use two proxies for CEO stock sales designed to capture two distinct features of the sales. The first proxy is the number of shares sold in each year, with which I aim to capture changes in the absolute magnitude of stock sales . The second proxy is the proportion of the CEO’s total equity portfolio sold, designed to reflect changes in the relative magnitude of stock sales. My aim with the second proxy is to estimate the degree to which the CE O’s total portfolio, or diversification, has shifted in response to the tax cuts. I use data from Thomson Financial Insiders Data Feed to 31 For the sake of brevity, I present the results for all additional analysis using the state tax changes setting. In untabulated analysis, I perform all tests using the two federal tax cuts and my inferences are generally similar. 32 estimate CEO stock sales in each year. Appendix A contains a detailed description of the construction of the measures. Table 5 Panel A presents the results from estimating the impact of state tax cuts on CEO stock sales. Column 1 shows that prior to the tax cuts i.e., in years t=-3 to t=-1, CEOs with higher tax burdens tend to sell fewer shares than CEOs with lower tax burdens, consistent with the tax lock-in effect. But stock sales spike in the year the tax cuts take effect year t=0, displaying a positive and significant coefficient coef.=1.232; t-stat.=1.85. Column 2 shows a similar pattern when measuring sales as the percentage of CEOs’ total equity holdings. Economically, a one percent drop in the long-term gains tax rate leads CEOs at the 75 th percentile of CEO Tax Burden to sell incremental stock worth 2.5 of their total equity portfolio. Interestingly, the bulk of the incremental stock sales appear to take place in the year in which the tax cut takes effect year t=0, and to a lesser degree in the year afterwards year t=+1. Thus, the evidence suggests that locked- in CEOs respond quickly to tax cuts by selling shares, reducing their exposure to the firm and improving diversification, and begin to make riskier corporate decisions that manifest in the firm over the following years. To directly connect CEO stock sales to increased corporate risk-taking, I estimate a two- stage-least-squares specification to first obtain a predicted value of stock sales in year t=0 based on the magnitude of the tax cut and the CEO’s tax burden prior to the tax cut, and then use that predicted value to examine the impact on subsequent risk-taking. Panel B of Table 5 presents these findings, using both Num Shares Sold and Pct. Equity Sold to measure stock sales, and LogTotal Vol and LogROA Vol measured over three years to proxy for risk-taking. In all cases, I find that larger predicted stock sales lead to statistically significant increases in risk-taking in the three years following the tax cuts. 33 Cross-Sectional Variation: CEO Diversification In addition to the federal and state tax cut analyses, another approach I consider to address endogeneity concerns is to investigate cross-sectional variation in the effect of CEOs ’ tax burdens on corporate risk-taking, when this effect should vary predictably. Table 4 showed that CEOs with high tax burdens experience larger increases in risk-taking following state-level tax cuts. Next I consider the extent to which the strength of this effect varies with CEOs ’ outside wealth. Specifically, I conjecture that when CEOs hold more personal wealth outside of the firm, they have greater opportunities to diversify despite the tax-induced lock-in effect on their equity holdings in the firm. As a result, I predict that high tax burdens have a weaker risk-deterring effect on CEOs with significant outside wealth, and tax cuts induce a milder increase in risk-taking. To obtain an estimate of the percentage of the CEO’s total wealth held in the firm’s equity, I compute the ratio of the value of each CEO’s stock and option portfolio held in the firm compared to the CEO’s total wealth Pct. of Wealth in Equity. 32 I then partition the sample of CEOs above and below the median annually based on the percentage of the CEO’s wealth held in the firm’s equity. I classify executives below the median as CEOs with high diversification High Diverse, and those above the median as CEOs with low diversification Low Diverse. Table 6 presents the results of estimating Equation 4 for the Low Diverse and High Diverse groups. Examining the results for LogTotal Vol, I find that the coefficient on the interaction term |State Tax Rate Cut| × Pre-Tax Cut Burden for the Low Diverse group is larger than for the High Diverse group, although the difference is marginally significant coef.=0.484 versus coef.=0.226; p-value=0.09 for the difference. I find similar results using LogROA Vol. 33 Overall, the cross- 32 Estimates of CEO total wealth are obtained by adding together the total value of CEO equity held in the firm and CEO outside wealth. CEO outside wealth is obtained by cumulating past compensation as described in Dittmann and Maug 2007, and missing values are imputed following Armstrong et al. 2015. Details on variable construction are contained in Appendix A. 33 Due to space constraints, I tabulate only the estimates for LogTotal Vol and LogROA Vol, but in untabulated analysis I find that my inference for LogIdio Vol is similar. Specifically, the coefficient magnitude is larger for the Low Diverse group than for the High Diverse group, and the difference in the coefficients has a p-value of 0.08. 34 sectional results indicate that the tax cuts induce a weaker response with respect to corporate risk- taking when the executives have greater outside wealth, and thereby have the potential to diversify their personal holdings in spite of the tax lock-in. Channels of Risk-Taking: Corporate Policies Next I examine the effect of the state tax cuts on firm policies commonly used as proxies for corporate risk-taking: RD expenditures, leverage, and working capital. 34 Prior literature finds that RD investment and leverage are increasing in firm risk, whereas working capital is decreasing in firm risk Coles et al. 2006; Cassell et al. 2012. Thus I estimate Equation 4 using these three proxies as the dependent variables, and I predict positive coefficients on the interaction term |State Tax Rate Cut| × Pre-Tax Cut Burden for RD and leverage, and a negative coefficient on the interaction term for working capital. The results in Table 7 are consistent with the tax cuts leading to increased risk-taking as measured by corporate policies, but only for previously locked-in CEOs. Economically, a one percent tax cut induces no significant change in RD investment for a CEO at the 25 th percentile of CEO Tax Burden, but a 3.7 relative increase in RD investment for a CEO at the 75 th percentile. Similarly, a CEO at the 75 th percentile of CEO Tax Burden experiences a 1.3 relative increase in leverage, and a 1.4 relative decrease in working capital. In addition to shedding light on specific corporate policy responses to individual-level tax cuts, these findings help further mitigate concerns that the equity volatility results in Table 4 are due to the effect of capital gains taxes on outside shareholders. The Table 7 results provide some reassurance that the capital gains tax cuts are changing incentives for decision-makers within the firm, not solely for external investors. Still, these tests cannot completely rule out the possibility 34 In untabulated analysis, I examine changes in RD expenditures, leverage, and working capital surrounding the two federal tax cuts, and my inferences are largely similar to those drawn from the state tax cuts. 35 that firm policies are changing in response to a changing cost of capital or that investors who are now facing different tax incentives indirectly influence the operations of the firm. State Tax Increases Next I perform an analysis of state tax increases. The prediction regarding the effect of tax increases on corporate risk-taking and whether this effect will vary depending on CEOs ’ pre-tax cut burdens is less clear than the prediction regarding tax cuts. Tax increases may exacerbate the lock-in effect, but presumably it takes time for this effect to result in a build-up of stock holdings that subsequently reduces the executive’s risk-taking behavior. Furthermore, to the extent that CEOs with large unrealized gains anticipate tax increases, they may preemptively sell stock to take advantage of the lower tax rates before the increase. 35 I estimate a model similar to Equation 4 in which I replace state tax cuts with state tax increases, but in light of the above considerations, I make no prediction with regard to the sign of the coefficients. In Panel A of Table 8, the first column shows that the main effect of |State Tax Rate Inc| for CEOs with no tax burden is negative and statistically significant coef.=-0.031; t- stat.=-3.69. But interestingly, there appears to be no disparity in the effect for CEOs with high tax burdens compared to CEOs with low tax burdens coef.=0.116; t-stat.=0.98. In fact, if anything there appears to be a weaker effect from the tax increases on CEOs with higher tax burdens prior to the changes although the difference is not significant. The results in columns 2 and 3 lead to similar inferences. To explore the possibility that high tax burden CEOs preemptively sell appreciated shares prior to the tax increases, I perform a year-by-year analysis of stock sales, analogous to that in Table 5 with respect to tax cuts. Panel B of Table 8 displays evidence of increased stock sales in 35 Saez 2016 provides evidence of such preemptive selling among the top one percent of income earners in the year prior to the federal tax increases effective beginning in 2013. He finds that following the re-election of President Barack Obama in November 2012, top earners realized it was almost certain the top tax rates would increase significantly in January 2013, and in response accelerated the realization of capital gains prior to the year-end of 2012. 36 the year prior to the tax increases. Both Columns 1 and 2 show a temporary spike in equity sold in year t=-1, just prior to when the tax increases take effect. Economically, executives at the 75 th percentile of CEO Tax Burden sell an incremental 1.0 of their equity portfolio in year t=-1. The pattern of preemptive stock sales by high-tax-burden CEOs prior to tax increases may provide an explanation for the results in Panel A of Table 8. Although the tax lock-in effect is exacerbated following a tax increase, CEOs with high ex ante tax burdens hurry to sell stock while the lower rate is still in effect, which reduces their exposure to the firm and improves their overall diversification. Alternative Measures of Tax Burden The tax burden measure is calculated as the tax based on selling all vested stock scaled by the CEO’s total equity holdings. Although I believe this construction has a solid theoretical and institutional basis, there are a number of reasonable alternative design choices. In this section, I construct several alternative tax burden measures and use each one to re-estimate Equation 4. The first alternative I consider is to use an unscaled tax burden measure which is simply the dollar amount of taxes owed on the sale of vested stock. For the second alternative measure, I replace the denominator the CEO’s total equity holdings in the firm, with an estimate of the CEO’s total wealth, using the outside wealth estimates as computed following Dittmann and Maug 2007 and used in Table 6. For the third alternative measure, I attempt to mitigate concerns about outlier values affecting the results by creating an indicator variable based on above- and below- median values of CEO Tax Burden. And for the fourth alternative measure, I construct a “marginal” tax burden measure assuming the CEO sells just the first 10 of her equity portfolio assuming she chooses to sell the 10 of the portfolio with the highest tax basis to minimize any potential gain on the sale. The results from estimating Equation 4 using these alternative tax burden measures are displayed in Table 9. Regardless of which measure of the tax burden is used, the inferences are the 37 same. That is, higher tax burdens are associated with lower risk-taking a negative coefficient on Pre-Tax Cut Burden , and risk-taking increases following tax cuts for CEOs with high tax burdens a positive coefficient on |State Tax Rate Cut| × Pre-Tax Cut Burden. These results provide some reassurance that the results are robust to various constructions of CEO Tax Burden.

VII. CONCLUSION