臺大管理論叢 NTU Management Review VOL.29 NO.1

219 NTU Management Review Vol. 29 No. 1 Apr. 2019 Based on Hypothesis H1, the coefficient of IM t *X t ( β 7 ) is expected to be negative to reflect the decreased informativeness of current earnings for firms with asset impairment. Meanwhile, according to Hypothesis H2, the coefficient of IM t *X t3 ( β 8 ) is expected to be positive to reflect better approximation of the intrinsic economic value of the underlying impaired assets and enhanced earnings informativeness. To examine the third hypothesis, we expand Reg. (1) by incorporating the dummy variable of the reversed asset impairments (REV t ) and the relatively interactive variables. The regression is presented as follows: R t = β 0 + β 1 X t-1 + β 2 X t + β 3 X t3 + β 4 R t3 + β 5 IM t + β 6 IM t *X t-1 + β 7 IM t *X t + β 8 IM t *X t3 + β 9 IM t *R t3 + β 10 REV t + β 11 REV t *IM t *X t-1 + β 12 REV t *IM t *X t + β 13 REV t *IM t *X t3 + β 14 REV t *IM t *R t3 + β 15 LEV t + β 16 MB t + β 17 SIZE t + ε t (2) where: REV t : a dummy variable for firms with asset impairment reversals; REV t is set as one if the firm recognized asset impairment in year t and reversed some or all of the impairment in the following year, and otherwise 0. The definitions of the remaining variables are the same as for Reg. (1). According to Hypothesis H3, the coefficient of REV t *IM t *X t3 ( β 14 ) is expected to be negative to reflect the decreased informativeness of future earnings for firms with asset impairment reversals. 4. Empirical Results 4.1 Descriptive Statistics Table 3 presents the descriptive statistics for the related variables in this study. For the following analysis, this study winsorizes the top and the bottom 1.5% of outliers 11 for all continuous variables, except for the assets impairment variable (IM t ) due to its truncated characteristic. The mean (median) of annual stock returns (R t ), earnings per 11 The number of winsorized 1.5% samples is approximately the same as the number of outliers that is outside three standard deviations in the empirical samples.