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

151 NTU Management Review Vol. 30 No. 2 Aug. 2020 5. Primary Results 5.1 Asymmetric OCI Adjustments We estimate Eq. (3) using Ordinary Least Squares (OLS). Following Anderson et al. (2003), we perform a series of tests for our specifications and data. White’s (1980) test indicates that heteroskedasticity is not a problem for the log linear model used in this study. We also applied the diagnostic proposed by Belsley, Kuh, and Welsch (1980) to test for multicollinearity in the pooled estimation. We find that none of the condition indexes exceed 5, which is well below the suggested cutoffs. Moreover, we evaluate serial correlation in the data on a firm-by-firm basis. To do so, we estimate Eq. (3) for the 9,610 firms in our sample and retain the errors to calculate the Durbin and Watson (1951) statistics. 10 Using the 5% significance level as a benchmark, we find that less than 3% of the 9,610 firms have a positive autocorrelation. This result indicates that it is unnecessary to correct for serial correlation in the data. In Table 2, we present the results for testing H1 using the pooled sample. The results are similar when we estimate a fixed-effects model. 11 In Model (1), the estimated value of = 0.129 ( t -statistic = 52.63), indicating a 0.129% increase in net assets for a 1% increase in the market value of equity, a proxy for an improvement in market-based information inputs. The estimated value of = -0.127 ( t -statistic = -37.53) provides strong support for the hypothesis of asymmetric OCI adjustments. The combined value of + is 0.002 (i.e., 0.129 – 0.127), and statistically insignificant ( F -value = 0.77). These results suggest that firms either do not adjust their book value of net assets or adjust the net assets downwards only by 0.002% per 1% decrease in the market value of equity. In Model (2) of Table 2, we use changes in sales revenue as accounting-based information inputs to reveal variations in economic conditions. Consistent with the market-based information inputs in Model (1), the estimated value of is 0.283 ( t -statistic = 97.12), indicating a positive association between increased sales revenue and OCI adjustments. Also consistent with in Model (1), the estimated value of is -0.125 ( t -statistic = -34.17), suggesting that OCI adjustments are asymmetric: for every 1% change in sales revenue, the corresponding OCI adjustment is 0.283% upwards but only 0.158% (i.e., 0.283 – 0.125) downwards. 10 In SAS code, we request the DWPROB as a regression option. 11 When estimating the fixed-effects model, we add a dummy variable for each firm (omitting one firm to avoid multicollinearity). This is numerically equivalent to the fixed effect model (Garcia, 1983).