Robustness checks You blinked: The role and incentives of managers in increasing corporate risks following the inception of credit default swap trade

26

5. Robustness checks

In this section, we examine whether our main findings are robust to alternative test specifications. 5.1. Two-Stage Least Squares 2SLS Specification We use a 2SLS specification. In the first stage, we estimate a regression of a binary variable, CDS_Trade , on all variables of the CDS determinant model specified in Eq. 1 and on control variables in Eq. 2 and Eqs. 4 –7. We also include two instrumental variables, Industry Peers’ Bond Trading Volume and Investment GradeSpeculative Grade . These instrumental variables predict the initiation of CDS trading but are likely to be unrelated to the residuals in the second- stage regression. The first proxies for the degree to which investors can hedge and speculate in the bond market in the absence of the CDS market. 8 Following prior studies, we compute this variable by the average of the industry peers ’ bond trading volume Boehmer, Chava, and Tookes, 2015; Kim, Shroff, Vyas and Wittenberg-Moerman, 2016. Bond trading volume should reduce the likelihood of CDS inception. Higher bond market liquidity alleviates trading friction, thereby reducing the demand for CDS trades. We extract the bond trading volume for industry peers from the TRACE database. We also collect the face value of the traded bonds on the issue date from the Mergent database. We divide the dollar volume of a traded bond by its face value to estimate its trading volume. We then compute the average bond trading volume of industry peers per year. We convert this measure into a decile rank Industry Peers’ Bond Trading Volume. Our second instrumental variable, Investment GradeSpeculative Grade , is also a proxy for the demand for CDS trade. Qiu and Yu 2012 demonstrate an inverse U-shaped relation between 8 Oehmke and Zawadowski 2013 illustrate that credit investors choose the CDS market as the trading venue for their credit hedging and speculative objectives when they face trading frictions in the underlying bond market. 27 CDS liquidity and credit rating. That is, b ond investors’ hedging demand is the lowest for bonds at the extreme values of investment and speculative grades. Bonds with very high credit quality have little hedging demand because of their high credit quality. For below-investment grade bonds, credit protection is too costly. Our second instrumental variable is an indicator variable takes a value of one if the credit rating of a firm’s bonds are close to the crossover from investment to speculative grades; that is, the bonds have an average credit rating of BBB –, BBB, or BBB+ Investment GradeSpeculative Grade Frontier . We obtain corporate long-term bond credit ratings from Compustat. We present the results of the Probit model in Panel A of Table 11. We use CDS_Trade as the dependent variable and Industry Peers’ Bond Trading Volume and Investment GradeSpeculative Grade Frontier as inverse and direct proxies, respectively, for bond investors’ trading demand. As expected, the coefficients are significantly negative and positive, respectively. [Insert Table 11 near here] In the second stage, we use the predicted value of CDS_Trade from the first stage and estimate a regression of Total_Comp and Vega on the fitted value of CDS_Trade along with all the control variables specified in Eq. 2. The results are presented in Panels B and C of Table 11. Our interest is the coefficient on CDS_Trade , which is positive and significant, indicating that the value and the vega of management compensation increase following CDS inception, consistent with H2A. We then estimate a regression of operating and financing policy RDEXP , CAPEX , MA_Freq , and EXCESS_DIV and bankruptcy risk using the fitted value of CDS_Trade and its interaction with Vega . Results for those tests are presented in Panels D –F of Table 12. The coefficient on the interaction term is positive for RDEXP , MA_Freq , EXCESS_DIV, and bankruptcy risk and negative for CAPEX . Thus, our main results remain qualitatively unchanged using the 2SLS model. To 28 validate our choice of instrumental variables, we follow Larcker and Rusticus 2010 and implement weak instrument identification tests. 9 We evaluate the incremental explanatory power of the instruments with a Partial F -Statistic test. The Partial F -Statistic test statistic is highly significant in all panels e.g., 718.95 with a p -value of 0.001 when we use Total_Comp as the dependent variable and 341.80 with a p -value of 0.001 when we use Vega as the dependent variable. These results suggest that the instrument passes the weak-instrument tests and that it explains a significant amount of the variation in corporate risk taking behavior. The partial pseudo- R 2 of 15.38 in Panel A indicates a well-specified first-stage model. 5.2. Identifying lender banks that most likely used CDS contracts CDS contracts are traded over the counter. Hence, banks that purchase credit insurance cannot be easily identified. Martin and Roychowdhury 2015, however, propose that a bank likely purchased CDS protection against a borrower’s default risk if the bank was able to increase its risk- based capital ratio increased in the same year in which the borrower’s CDS trading was initiated. This argument is based on the proposition that a bank can increases its risk-based capital ratio upon hedging its credit risk. We expect banks that hedge their risks using the CDS market to become more lax in their monitoring following CDS inception. Hence, we expect that the phenomena we show in this paper are stronger for firms whose lender banks likely purchased CDS contracts. 9 For Panel B, the partial F is 718.95 p -value 0.0001, and the under-identification test chi-squared is 681.36 p - value 0.0001. For Panel C, the partial F is 341.80 p -value 0.0001, and the under-identification test chi-squared is 333.07 p -value 0.0001. For Panel D, the partial F is 352.25 and 99.41 p -value 0.0001, and the under- identification test chi-squared is 95.67 and 100.19 p -value 0.0001, when the dependent variables are RDEXP and CAPEX , respectively. For Panel E, the partial F is 127.05 and 123.00 p -value 0.0001, and the under-identification test chi-squared is 126.57and 122.54 p -value 0.0001 with MA_Freq and EXCESS_DIV as dependent variables, respectively. For Panel F, the partial F is 104.91 p -value 0.0001, and the under-identification test chi-squared is 104.07 p -value 0.0001. Finally, for Panel G, the partial F is 474.04 p -value 0.0001, and the under-identification test chi-squared is 456.14 p -value 0.0001. These results suggest that the instrument passes the under- identification tests and explains a significant amount of the variation in management compensation. The weak instrument test yields a Cragg-Donald Wald F ranging from 99.06 p -value 0.001 for Panel D to 716.52 p -value 0.001 for Panel B, compared with the Stock-Yogo critical value. Stock and Yogo 2005 provide a critical value table for a 5 Max IV size 24.09, 10 Max IV size 16.38, and 15 Max IV size 8.96. 29 We identify lenders to CDS and non-CDS firms in our sample using the Dealscan database, and we obtain banks’ risk-based capital ratio from the Federal Reserve Y-9C reports. We divide the borrower sample into two groups: those with increases in their banks’ risk-based capital ratio and those with decreases in the year of CDS inception. We then reestimate Eqs. 4 –7 examining the joint effect of CDS inception and managerial incentives on operating and financing policies and on bankruptcy risks. Table 12 presents results of this analysis. The coefficients on the interaction term CDS_Trade × Vega are significantly stronger for firms experiencing an increase in banks’ risk-based capital ratio for all dependent variables except CAPEX . For bankruptcy risk, the coefficient on the interaction term is significant only for firms with an increase in banks’ risk-based capital ratio. Thus, the shifts in lender, shareholder, and managerial forces we find following the onset of CDS trading appear to be stronger when borrowers hedge their client risks with traded CDSs and, hence, likely reduce their client vigilance. [Insert Table 12 here]

6. Conclusion