Note(s): Table 12 reports whether the effect of CEO narcissism on corporate ESG performance varies by CEO tenure, the firm’s institutional investor shareholding ratio, the industry attribute and the level of media attention. Tenure is constructed using the median of CEO tenure in the same industry and year as the boundary, and takes the value of 1 if the CEO tenure is higher than that median and 0 otherwise. Inst is constructed by taking the median of institutional investor shareholding ratio in the same industry and year as the boundary. If the ratio is higher than that median, Inst takes the value of 1, otherwise it takes the value of 0. Pollute is constructed as whether the firm is in a heavy polluting industry or not. If the firm is in a heavy polluting industry, Pollute takes the value of 1, otherwise it takes the value of 0. Media is constructed as the Media is constructed by taking the median of media attention in the same industry and year as the boundary. If the media attention is higher than this median, Media takes the value of 1, otherwise it takes the value of 0. We also use Bootstrap method to self-sample 1,000 times to conduct the test for difference in coefficients between groups, which shows that the empirical P-values are all significant at the level of 1%. Other variables are defined in Table 1. All regressions are firm-level clustered robust standard errors. *, ** and *** indicate significant at the 10%, 5% and 1% levels, respectively