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

17 Vashishtha 2012. The distributions of firm characteristics in my sample appear similar overall to those in Ljungqvist et al. 2017, with any differences likely attributable to differences in sample composition. For example, the firms in my sample appear slightly older mean of 22.4 years versus 19.6 years in Ljungqvist et al. 2017, somewhat larger total assets of 5.40 billion versus 1.75 billion, and carry more cash 9.1 of assets versus 3.5 of assets. These differences likely arise from the fact that my sample is drawn from firms included in the SP 1500 index, whereas the sample in Ljungqvist et al. 2017 includes all Compustat firms. Table 1, Panel B reports the Pearson and Spearman correlations, which provide preliminary evidence on the relation between CEOs’ tax burdens and corporate risk. Consistent with the conjecture that higher tax burdens reduce risk-taking, the Pearson and Spearman correlations between CEO Tax Burden and the three primary measures of risk-taking are significantly negative. Specifically, the Pearson Spearman correlations between CEO Tax Burden and LogTotal Vol, LogIdio Vol , and LogROA Vol are -0.09, -0.08, and -0.19 -0.11, -0.09, -0.22, respectively. Panel C of Table 1 displays the mean values of key CEO and firm characteristics after partitioning the sample on CEO Tax Burden into quartiles within each year. As CEO Tax Burden increases, Delta and Total Wealth increase monotonically. At the same time, however, CEOs with higher tax burdens appear to be significantly less diversified than CEOs with low tax burdens Pct. of Wealth in Equity is 0.79 for CEOs in the top quartile of tax burdens and 0.68 for CEOs in the bottom quartile. This descriptive evidence provides preliminary support for the claim that CEOs with higher tax burdens are those most in need of improved diversification.

IV. PANEL ANALYSIS

Panel Analysis: Research Design To investigate whether CEOs ’ tax burdens are associated with reduced risk, I test for the potential relation between risk measures Firm Risk i , t and CEO tax burden for the previous year CEO Tax Burden i,t-1 and control variables, using the following regression model: 18 Firm Risk i,t = α i + α t + β 1 CEO Tax Burden i,t-1 + ɤ′X i,t-1 + ϵ i,t . 2 In the equation above, i and t index firms and years, respectively; � � and � � are firm and year fixed effects. The dependent variable, Firm Risk, represents the three measures of corporate risk. CEO Tax Burden represents the CEO’s tax burden on her stock holdings in the firm, and the remaining control variables are drawn from prior research on compensation and corporate risk, and are described in detail in Appendix A. To control for firm-specific factors and general macroeconomic time trends, I include both firm and year fixed effects. 15,16 I estimate the model using ordinary least squares with standard errors clustered by executive and year. The variable of interest in Equation 2 is CEO Tax Burden and my prediction is that the coefficient on this variable, β 1 , will be negative, indicating that an increase in CEO Tax Burden in year t-1 is associated with reduced risk-taking in year t. 17 Panel Analysis: Results Table 2 presents the results from estimating Equation 2. The first column of Panel A shows that the coefficient on CEO Tax Burden is negative and statistically significant coef.=-0.323; t- stat.=-4.22. This result indicates that increases in CEOs’ tax burdens in year t-1 lead to lower stock return volatility in year t, consistent with my central hypothesis. Similarly, columns 2 and 3 show negative associations between CEOs’ tax burdens and idiosyncratic volatility, as well as earnings volatility. In economic terms, moving from the 25 th to the 75 th percentile of CEO Tax Burden is associated with a 2.0 relative decrease in LogTotal Vol. 15 In untabulated analysis, I estimate Equation 2 while excluding year fixed effects in order to retain the variation in the federal capital gains tax rates over time. In the absence of year fixed effects, the results are slightly stronger but are similar. I include year fixed effects in the tabulated results so as to avoid overstating the magnitudes of the effects. 16 Note also that firm fixed effects largely subsume industry fixed effects, except in the rare instances when a firm changes industry. In untabulated analyses, I perform all of the main tests replacing firm fixed effects with Fama- French 48 industry fixed effects, and the results again become slightly stronger but my inferences are unchanged. 17 I construct alternative measures of total stock return volatility and idiosyncratic volatility, measured over a three- year window from t to t+2. My inferences are unchanged when using these alternative measures. Earnings volatility is already constructed over a three-year window, in order to obtain a standard deviation of the difference between quarterly ROA and ROA for the same quarter of the previous year. 19 Turning to the control variables, I find that the coefficients on LogVega are either negative or insignificantly different from zero. Although early work in the area found a positive relatio n between the convexity of the manager’s wealth-performance relation and risk-taking e.g., Guay 1999; Coles et al. 2006, some studies examining more recent periods show an insignificant or negative relation between vega and volatility e.g., Hayes, Lemmon, and Qiu 2012; Anderson and Core 2015. The coefficient on LogDelta is insignificantly positive in column 1, whereas the coefficient on LogCash Comp is insignificantly negative, generally consistent with Armstrong and Vashishtha 2012. The coefficient on LogCEO Tenure is insignificant in column 1, consistent with the result in Hayes et al. 2012. The coefficients on firm characteristics appear overall consistent with results in Bova et al. 2015 and Ljungqvist et al. 2017. These results include negative coefficients on variables for firm age, size, and cash surplus, and positive coefficients on the presence of loss carryforwards, sales growth, and recent stock returns. Bova et al. 2015 find a positive coefficient on the book-to-market ratio, corresponding to the negative coefficient I find on the market-to-book ratio. And Ljungqvist et al. 2017 find a negative relation between long-term debt and earnings volatility, similar to my finding for long-term debt in column 3. In addition, I control for managerial overconfidence with Holder67, which displays a significantly positive coefficient in column 1, consistent with overconfident CEOs making riskier corporate decisions. Further, I control for the portion of firm shares held by taxable individual investors with SH Tax Sensitivity, which also displays a positive and significant coefficient. In Panel B of Table 2 I re-estimate Equation 2 after replacing Firm Risk with corporate policies frequently used to proxy for risk-taking, including the levels of RD expenditures, leverage, and working capital. Economically, I find that moving from the 25 th to the 75 th percentile of CEO Tax Burden leads to 3.7 lower RD expenditures, 1.8 lower leverage, and 2.1 higher working capital, all consistent with a greater aversion to risk. Overall, the results in Table 2 provide evidence of a robust negative relation between CEOs ’ tax burdens and corporate risk-taking. 20 Panel Analysis: Potential Confounding Factors The panel analysis above, while informative, does not rule out the possibility of other factors confounding my inferences. One potential concern is that some unobserved feature besides executives’ tax burdens is driving CEOs’ equity holdings. An example is target ownership plans, which commonly require CEOs to hold significant equity in the firm. While I acknowledge that target ownership plans may affect CEOs’ equity holdings, I note that the inclusion of firm fixed effects in my estimation model helps to mitigate any confounding influence of firm-specific contract features. In addition, the evidence in Armstrong et al. 2015 regarding explicit ownership guidelines indicates that the amount of equity required to be held by CEOs is relatively low an average of 1.2 million in their sample, further alleviating the concern that target ownership plans have a substantial impact on my inferences. A more significant challenge to my inferences from the panel analysis derives from the construction of the tax burden measure. As noted previously, the CEO’s tax burden is determined by several factors: the capital gains tax rate federal plus state facing the CEO, the stock’s historical pr ice appreciation during the executive’s holding period, the number of shares held that were obtained during each year , and the value of the CEO’s equity holdings in the firm. These features raise potential endogeneity concerns with respect to interpreting the effect of the tax burden on risk-taking. One concern is that the negative relation documented in the panel analysis may be due largely to the firm’s past performance, rather than by the tax rate. Christie 1982 suggests that firms which have performed well in the past are likely to experience lower volatility, as the market value of debt relative to equity declines. Known as the “leverage effect” Aït-Sahalia, Fan, and Li 2016, this mechanical relation between firm performance and volatility may confound my inferences, as firms with positive past performance are likely to have CEOs with substantial tax burdens. Another concern is that the measure is partly determined by executives’ past portfolio choices and option exercise behavior, which are also likely associated with corporate risk-taking. 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