We use model (1) to empirically test hypotheses 1 and 2. In doing so, asset impairment losses, and the reversal of an impairment loss are the primary independent variable, and the dependent
variable, respectively.5 If a firm reverses a previously recognized assets impairment loss, the value of the dependent variable will be a positive number; otherwise it will be zero, giving rise to a potential of being a truncated datum. We employ a Tobit regression to explore the reversal of assets impairment as follows:
REVit: the reversal of an impairment loss by firm i in year t deflated by total assets at the end of year.
NIit-REVit-NIit-1: firm i’s pre-reversal earnings change from year t-1 to year t deflated by total assets at the end of year t.6
IMPit-1: the impairment loss recognized by firm i in year t-1 deflated by total assets at the end of year.
NIit-REVit-NIit-1*IMPit-1: interaction between variables NIit-REVit-NIit-1, and IMPit-1
DEBTit: firm i’s debt ratio, measured by the ratio of the total debt to the total assets at the end of the tth year.
MGTit-1,t-2: an indicator equal to 1 if a firm changed its president or general manager in year t-1 or t-2, and 0 otherwise.
ΔSALESit: the percentage change in firm i’s net sales from fiscal year t-1 to t.
MTBit: firm i’s market to book ratio, measured by the ratio of market value to the stockholders’
equity at the end of year.
SIZEit: the logarithm of firm i’s total assets at the end of year.
Hypothesis 1 predicts that the magnitude of reversal of an impairment loss is positively related to the amount of previously recognized impairment losses. We thus expect the coefficient of IMPit-1 to be positive. Hypothesis 2 predicts that the magnitude of reversal of an impairment loss is negatively
5 Although total amount of impairment losses consists of the impairment loss on the income statement and decreases in unrealized revaluation, this study adopts only the impairment loss on the income statement.
6 According to Tax Law in Taiwan, an impairment loss is unrealized and not deductible from taxable income. Similarly, reversal of an impairment loss is not taxable. It is not necessary to consider the tax impact when computing pre-reversal earnings.
related to the interaction between the amount of a previously recognized impairment loss and the pre-reversal earnings change. As discussed earlier, managers have incentives to reverse an impairment loss when earnings cannot meet the prior year’s earnings (Bartov 1993; Peasnell et al. 2005). We use NIit-REVit-NIit-1 to represent a firm’s pre-reversal earnings change. Positive numbers mean that the pre-reversal earnings meet the target and negative fail. If managers reverse a previously recognized impairment loss to avoid earnings decline, the coefficient of the interaction between IMPit-1 and NIit-REVit-NIit-1 will be negative.
Following prior research, we include DEBTit, MGTit-1,t-2, ΔSALESit, MTBit,and SIZEit as control variables. It has been proposed that the larger a firm’s debt ratio, the more likely its managers are to engage in greater manipulation. Prior research uses debt-equity ratio to proxy debt covenants (Fields et al. 2001). We thus expect the coefficient of debt ratio (debt/total assets) to be positive. Prior studies also show that firms experiencing recent changes in top management are more likely to recognize a higher amount of impairment loss (Loh and Tan 2002; Riedl 2004). We then expect that the recent change in top management is positively related to the reversal of an impairment loss and the coefficient of MGTit-1,t-2 to be positive. We expect ΔSALESit to be positively associated with the magnitude of reversing an impairment loss. It is because firms with sales growth whose assets will be more valuable and thus more likely to reverse an impairment loss. Similarly, the coefficient of MTBit is expected to be positive. Finally, we include SIZEit as another control variable. Elliott and Shaw (1988) provide evidence that firms disclosing large asset impairment losses are larger than other firms in their industries.
Hypothesis 3 predicts that, other things being equal, an effective corporate governance mechanism will deter earnings management. We incorporate a measure of corporate governance
mechanism in the analysis and estimate the following regression model:
CGit: the variable representing corporate governance mechanism.
NIit-REVit-NIit-1*IMPit-1*CGit: interaction among variables NIit-REVit-NIit-1, IMPit-1 and CGit. Variable CGit is a composite measure for corporate governance mechanism. It consists of (1) Board size (B_SIZE), measured as the total number of directors on the board; (2) Independent director (IND_D), an indicator variable which equals 1 if none of directors is an insider of the company and holds more than one percent of stock;7 (3) Independent supervisor (IND_S), an indicator variable which equals 1 if none of supervisors is an insider of the company and holds more than one percent of stock; (4) Institutional investors’ shareholding (%INST); (5) Foreign institutional investors’
shareholding (%FORE), and (6) The difference between control rights and cash flow rights (V-C) , computed as the percentage of voting rights minus the percentage of cash flow rights. B_SIZE, %INST, and %FORE are sorted in ascending order and V-C is sorted in descending order before computing the percentile values. We compute a composite variable (CG) which is composed of total percentile values of B_SIZE, %INST, %FORE, and V-C plus IND_D and IND_S to capture the strength of corporate governance (Bushman et al. 2004). Hence, high values of CG represent relatively strong corporate governance mechanism.
Hypothesis 2 predicts that, to meet the target, the magnitude of reversal of an impairment loss is negatively related to the interaction of the amount of a previously recognized impairment loss and pre-reversal earnings change. Hypothesis 3 further predicts that this earnings management behavior
7 The definitions of independent directors and independent supervisors follow the stipulation by Financial Supervisory Commission in Taiwan.
will be deterred by an effective corporate governance mechanism. We therefore expect β4 to be positive.