CFO gender and earnings management: evidence from China

Women can hold up half of the sky.

-Mao Tse-tung.

Abstract

We study the effect of chief financial officer (CFO) gender on earnings management (EM) in China’s listed firms from 1999 to 2011. In the cross-sectional analysis, we find that female CFO firm-years exhibit significantly lower discretionary accruals, lower total accruals, lower abnormal production costs, and higher abnormal discretionary expenditures, than the male CFO firm-years. We further examine the relation between CFO gender and EM surrounding CFO transitions. We find that the departing male CFOs are more aggressive than the departing female CFOs in managing up earnings during their last year with the firm and the newly appointed male CFOs are more aggressive than the new female CFOs in managing down earnings during their first year on the job. The evidence surrounding CFO transitions suggests that male CFOs are more aggressive than female CFOs in manipulating earnings, either in the last attempt to save their jobs or to take bigger credit for any future performance gains. Overall, our empirical evidence suggests that female CFOs engage in less EM and are more conservative in financial reporting than their male counterparts.

Appendices

Appendix 1: Earnings management measures

1.1 Accruals-based measures

We employ DCAC as our primary accruals-based measure for earnings management. We estimate the following modified Jones model as in Dechow et al. (1995) to obtain the discretionary current accruals:

$$\frac{{CAC_{i,t} }}{{TA_{i,t - 1} }} = \frac{{\alpha_{1} }}{{TA_{i,t - 1} }} + \beta_{1,i} \frac{{(\varDelta S_{i,t} - \varDelta REC_{i,t} )}}{{TA_{i,t - 1} }} + \varepsilon_{i}$$

(2)

where CAC is current accruals, measured as the change in non-cash current assets less the change in current liabilities excluding the change in short-term debt. TA is total book assets. ΔREC is the change in accounts receivable. ΔS is the change in sales between current year and the year before.

The parameters in model (2) are estimated each year cross-sectionally by industry. The estimated coefficients are then used to calculate the normal or predicted current accruals. The DCAC is the difference between the actual current accruals (CAC) scaled by lagged TA and the predicted current accruals.

We also use total accruals (TAC) as an alternative measure, which is the difference between operating income before taxes and cash flows from operating activities, scaled by lagged TA.

1.2 Real activities-based measures

We follow Roychowdhurry (2006) and use abnormal production costs and abnormal discretionary expenditures as our real activities-based earning management measures, modeled in Eqs. (3) and (4) below, respectively:

$$\frac{{PCOST_{i,t} }}{{TA_{i,t - 1} }} = \alpha_{1} \left( {\frac{1}{{TA_{i,t - 1} }}} \right) + \beta_{1} \left( {\frac{{S_{i,t} }}{{TA_{i,t - 1} }}} \right) + \beta_{2} \left( {\frac{{\varDelta S_{i,t} }}{{TA_{i,t - 1} }}} \right) + \beta_{3} \left( {\frac{{\varDelta S_{i,t - 1} }}{{TA_{i,t - 1} }}} \right) + \varepsilon_{i}$$

(3)

$$\frac{{DXPN_{i,t} }}{{TA_{i,t - 1} }} = \alpha_{1} \left( {\frac{1}{{TA_{i,t - 1} }}} \right) + \beta \left( {\frac{{S_{i,t - 1} }}{{TA_{i,t - 1} }}} \right) + \varepsilon_{i}$$

(4)

where PCOST is production cost, defined as costs of goods sold (COGS) plus change in inventory. DXPN is discretionary expenditures, defined as the sum of research and development (R&D) and selling, general and administrative (SG&A) expenses.

Models (3) and (4) are also estimated each year cross-sectionally by industry. The estimated coefficients are then used to calculate the normal production cost and normal discretionary expenditures. The abnormal production costs (ABCOST) are the difference between the actual production costs scaled by lagged TA and the normal production costs estimated in model (3). The abnormal discretionary expenditures (ABDXPN) are the difference between the actual discretional expenditures scaled by lagged TA and the normal discretionary expenditures estimated in model (4).

Appendix 2

See Table 10

Table 10 Variable definitions