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

The Effect of Corporate Social Responsibility Performance on Financial Risk 294 4.2.4 Models Controlled for Reverse Causality In order to alleviate possible reverse causality concern and potential endogeneity, we follow Jiraporn, Jiraporn, Boeprasert, and Chang (2014) to re-estimate CSR performance models (Model 1 to Model 4) by using lagged CSR performance instead of contemporaneous CSR performance in Table 10 to Table 11, and we reduce the sample size to 3,689 observations. The results remain similar to those using contemporaneous CSR performance. Although some variables of interest turn to be insignificant, many variables turn to be more significant. For example, for risk regressions, the coefficients of Table 10 Risk Regression Analysis Controlled for Reverse Causality β L Non-standard β L Standard cov ( L i , L m ) Non-standard Panel A: Model 1 CSR_STR 0.014 0.005 -0.032* (0.35) (0.43) (-1.64) CSR_CON 0.043* 0.008 0.036** (1.42) (1.26) (2.11) Panel B: Model 2 TOTAL_CSR_STR 0.002 0.005 -0.005* (0.35) (0.43) (-1.64) TOTAL_CSR_CON 0.006* 0.008 0.005** (1.42) (1.26) (2.11) Panel C: Model 3 COMMUNITY -0.012 0.014* 0.017 (-0.06) (1.37) (0.22) ENVIRONMENT -0.111 -0.008* -0.121* (-0.54) (-1.46) (-1.41) DIVERSITY -0.006 -0.003 -0.080* (-0.04) (-0.22) (-1.35) EMPLOYEE 0.005 -0.005 -0.016 (0.03) (-0.62) (-0.35) PRODUCT 0.000 0.000 -0.099* (0.00) (0.02) (-1.58)