臺大管理論叢 NTU Management Review VOL.30 NO.2

285 NTU Management Review Vol. 30 No. 2 Aug. 2020 In sum, our hypotheses are supported, and the results are consistent with the viewpoint from Cho et al . (2013), who suggest the importance of separately considering both positive and negative CSR performances. Generally, these results suggest that a firm with superior CSR performance would have lower liquidity risk and market risk. 4.2 Robustness Tests 4.2.1 1% Outliers Deletion Rule To reduce the potential effects of outliers, we winsorize top and bottom 1% observations for each continuous variable and repeat risk regressions from Table 3 to Table 6. In unreported analyses, all of our primary inferences remain largely unchanged. 4.2.2 A Direct Measure of the Cost of Capital In this section, we estimate the cost of capital directly to investigate the effect of CSR performance on the cost of capital instead of risk. To construct the cost of capital variables, we follow Fu, Kraft, and Zhang (2012), who define five proxies for the cost of capital based on the CAPM, Fama-French three-factor model, and earnings-to-price ratio to calculate five measures of the cost of capital. The first measure uses compounded twelve monthly returns. In the second measure, we regress the stock return on the market return (value-weighted CRSP returns) by each firm and past five years, and the estimated expected return is the cost of capital estimate. In the third measure, we use Fama-French three-factor model to obtain the estimated expected return as a proxy for the cost of capital. In the fourth measure, we use the earnings-to-price ratio as the measure of the cost of capital by using earnings per share including extraordinary items as the earnings. In the final measure, which is similar to the fourth measure, we use earnings per share excluding extraordinary items as the earnings to calculate the earnings-to-price ratio. The results for four models to test the effects of CSR performance on the cost of capital are shown in Table 7. In panel A, we find that CSR_STR is significantly negatively correlated with FRET1 (-0.185, t = -5.44; -0.032, t = -2.84) (non-standardized and standardized value), EP_EPSPI (-4.251, t = -1.47; -0.875, t = -1.37) (non-standardized and standardized value), and EP_EPSPX (-4.243, t = -1.46; -0.893, t = -1.40) (non-standardized and standardized value), with the exception of the coefficient on R_HAT_FF (1.270, t = 1.36) (standardized value) which is significantly positive at 10% significance level under one-tailed test. We also find that CSR_CON is significantly negatively correlated with FRET1 (-0.259, t = -4.23) (non-standardized value), R_HAT_MARKET (-4.716, t = -2.21) (non-standardized value), with the exception of R_HAT_FF (1.797, t = 2.53) (standardized value). In panel