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

227 NTU Management Review Vol. 29 No. 1 Apr. 2019 5. Robustness Tests 5.1 Propensity Score Matching (PSM) Examination Relatively few firms recognize asset impairment, resulting in an asymmetric sample distribution. Following Lawrence, Minutti-Meza, and Zhang (2011) and Mitra, Jaggi, and Hossain (2013), this study uses the propensity-score matching approach to obtain a matched control sample and reruns the regressions. We estimate the propensity-score matching model by including company characteristics and managerial incentives that are expected to influence a firm’s recognition of asset impairment (Riedl, 2004). The estimation model is a logit regression of the assets write-off using all firm-years (with and without asset write-offs, D_IM) with the explanatory variables, i.e., ΔGDP (the percent change of gross domestic product from year t -1 to t ), ΔROA (a firm’s industry-adjusted change of return on assets from year t -1 to t ), ΔSALES (a firm’s percent change in sales from year t -1 to t ), ΔUE (a firm’s change in pre-write-off earnings from year t -1 to t , divided by total assets at the end of t -1), ΔOCF (a firm’s change in operating cash flows from year t -1 to t , divided by total assets at the end of t -1), ΔMGT (a dummy variable equal to 1 if a firm experiences a change in CEO from year t -1 to t , and 0 otherwise), BATH (a proxy for “big bath” reporting, equal to the change in a firm’s pre-write-off earnings from year t -1 to t , divided by total assets at the end of t -1, when below the median of nonzero negative values of this variable, and 0 otherwise), SMOOTH (a proxy for “earnings smoothing” reporting, equal to the change in a firm’s pre-write-off earnings from year t -1 to t , divided by total assets at the end of t -1, when above the median of nonzero negative values of this variable, and 0 otherwise), and LEV (a firm’s leverage measured as total debts divided by total assets in year t ). The estimating regression is presented as follows: D_IM t = β 0 + β 1 ΔSALES t + β 2 ΔROA t + β 3 ΔUE t + β 4 ΔOCF t + β 5 LEV t + β 6 ΔMGT t + β 7 BATH t + β 8 SMOOTH t + β 9 ΔGDP t + ε t (3) For each firm with impaired assets, without replacement and using a caliper distance of 0.03 (Rice, Weber, and Wu, 2014; Bills, Cunningham, and Myers, 2016), 14 we choose a 14 We also match each impaired firm with the non-impaired firm most similar in terms of Reg. (3), without replacement and no limitations on the caliper distance, and reexamine the empirical regressions. The results from these examinations do not qualitatively change the empirical findings.