To provide descriptive evidence on between the earnings management and equity liquidity, we begin with an ordinary least squares estimation OLS regression.
1 ,
i i i
PSP = +α β ADCA +ε
The coefficient on (ADCA) β1 is 0.007 with a t statistic of 1.69, which is statistically significant at less than 0.1 level. The result simple shows earnings management affects liquidity in equity.
Further empirical tests include the following control variables which are return volatility (SDRET), dollar value of volume (DOLVOL) and stock’s closing price (CLP). Table 3 reports that the coefficients on dollar value of volume (DOLVOL) and stock’s closing price (CLP) (β3 and β4) are reliably negative, and the coefficient on return volatility(SDRET) (β2 ) is reliably positive, as predicted. However, the coefficient on the level of earnings management (ADCA) (β1) isn’t reliably positive.
Besides, percentage bid-ask spread can be strong explained by these control variables.
It implies that there is not much space for other variables to explain the percentage bid-ask spread. The other problem is the result maybe suffers from endogeneity bias.
So these reason may result in the coefficient on earnings management (ADCA) (β1) isn’t significant.
Table 3
1 2 3 4
lnPSPi = +α β ADCAi +β lnDOLVOLi +β lnCLPi +β lnSDRETi+εi, Variable Prediction Estimated
Coefficient
t Value Pr > |t|
Intercept 0.45912 4.58 <.0001
lnCLP i - -0.37491 -23.46 <.0001
lnDOLVOL i - -0.23286 -45.51 <.0001
ln SDRET + 0.46634 18.25 <.0001
ADCA i + 0.02077 0.14 0.8863
Adjusted R2 0.83 Variable definitions:
lnCLP = the natural log of the mean stock’s close price for firm i during our i sampling period
lnDOLVOL = the natural log of the mean dollar value of volume for firm i during i our sampling period
ln SDRET = the natural log of the mean for firm i of daily returns during our sampling period
ADCA = the mean for firm i of absolute value of discretionary current i
accruals during our sampling period
In order to solve omitted variables problem, I use two-stage least squares (2SLS) regression. Two-stage least square estimation involves two steps. In the first stage, the endogenous variable (ADCA ) is regressed on all exogenous variables included in the i model ( lnCLP , lni DOLVOL ,i ln SDRET and MBRT ). The predicted values of i
ADCA which should exhibit less correlation with the error term. In the second stage, i
the predicted values of ADCA be used as he explanatory variables in the regressions i that estimate equation (6). Table 4 presents the results of estimating equation (6) using two-stage least squares (2SLS) estimation. Dollar value of volume and stock’s close price are significant explanatory variables in the predicted direction. The coefficient on return volatility is also significant positive, as predicted direction. The coefficient on earnings management variable (ADCA ) is positive, as predicted, but it isn’t i statistically significant. The results are similar to previous multiple regression analysis.
The endogeneity bias usually arises in one of three ways: omitted variables, measurement error and simultaneity. Simultaneity arises when at least one of the explanatory variables is determined simultaneously along with dependent variables.
Because we consider the percentage bid-ask spread (PSP) and earnings management (ADCA) are simultaneously determined. Therefore, the two-stage least squares (2SLS) estimation is inefficiency in here.
Table 4
2SLS regression results showing the effects of the earnings management (ADCA ), i the natural log of return volatility, dollar value of volume and price on percentage bid-ask spreads.
1 2 3 4
lnPSPi = +α β ADCAi +β lnDOLVOLi +β lnCLPi +β lnSDRETi+εi, Variable Prediction Estimate t Value Pr > |t|
Intercept 0.271908 0.43 0.6663
lnCLP i - -0.38193 -13.4 <.0001
lnDOLVOL i - -0.23222 -40.97 <.0001
lnSDRET i + 0.424637 3.01 0.0027
ADCA i + 0.856009 0.31 0.7581
Adjusted R2 0.56
Variable definitions:
lnCLP = the natural log of the mean close price for firm i during our sampling i period
lnDOLVOL = the natural log of the mean dollar value of volume for firm i during i our sampling period
lnSDRET = the natural log of the mean for firm i of daily returns during our i sampling period
ADCA = the mean for firm i of Absolute value of discretionary current i
accruals during our sampling period
The simultaneous equation set can solve the endogeneity problem among factors. The result of 3SLS shows in Table 5. Hence, the endogenous problem does exist form the cross section regression and 2SLS. After taking the endogenous problem into account, I find that the relation between the earnings management and equity liquidity is significantly positively. On the one hand, the coefficient on
lnADCA (i β1) is 0.67 with a t statistic of 1.81, which is statistically significant at less than 0.1 level.
The results are consistent with our pervious hypotheses show that the equity liquidity would be small as the insiders large manipulate reported earnings. It also implies the outsiders detect manager manipulate reported earnings, they would unwilling to trade this firm’s stock in equity market. At the same time, market will punish the firms which manipulate reported earnings through market mechanism, the equity liquidity is small. On the other hand, the coefficient on (ln PSP) (λ1) is 0.19 with a t statistic of 2.71, which is statistically significant at less than 0.01 level. The results reported that the level of earnings management increase as the level of information asymmetry increase.
Therefore, it’s same with Richardson (2000) empirical work. About the others control variables, the return volatility (SDRET), dollar value of volume (DOLVOL) and stock’s closing price (CLP). Panel A of Table 5 reports that the coefficients on dollar value of volume (DOLVOL) and stock’s closing price (CLP) (β3 and β4) are reliably negative, and the coefficient on return volatility(SDRET) (β2) is reliably positive, as predicted. The coefficient on financial leverage (LEV) (λ2) is reliably negative and growth opportunities (MBRT) (λ4) is reliably positive in the panel B of Table 5, as predicted. The coefficient on firm size (SIZE) (λ3) is reliably positive.
Even I don’t predict the sign of firm size (SIZE). Some previous empirical studies show that the sign of firm size (SIZE) is negative.
Why our result different form previous empirical studies? It’s a puzzle.
In summary, the regression results are consistent with the hypothesis that the insiders manipulated the reported earnings largely would result in the liquidity in equity market small. On the other aspect, the high level of information asymmetric between insider and outsider will cause the insider can easy manipulate reported earnings. This result is similar with Richardson (2000) empirical work.
Table 5
0 1 2 3 4
lnPSPi =β +β ln ADCAi +β lnSDRETi+β lnDOLVOLi+β lnCLPi+εi, (7)
0 1 2 3 4
lnADCAi =λ λ+ lnPSPi+λ lnLEVi+λ lnSIZEi+λ lnMBRTi+ωi, (8) Three-stage least squares (3SLS) regression for Simultaneous equation model
Panel A: Equation (7)—3SLS Panel B: Equation (8)—3SLS Variable Estimate t Value Pr > |t| Variable Estimate t Value Pr > |t|
Intercept 2.769518 5.08 <.0001 Intercept -2.85131 -10.43 <.0001 lnCLP i -0.29394 -5.51 <.0001 lnMBRTi 0.046725 1.83 0.0678
lnDOLVOLi -0.25362 -21.55 <.0001 ln LEV -0.021 -1.76 0.0784
lnSDRET i 0.512668 2.06 0.0402 lnSIZEi 0.083452 3.32 0.001 lnADCA i 0.669188 1.81 0.0715 ln PSP 0.192239 2.71 0.0069
Variable definitions:
lnCLP = the natural log of the mean close price for firm i during our sampling i period
lnDOLVOL = the natural log of the mean dollar value of volume for firm i during i our sampling period
lnSDRET = the natural log of the mean for firm i of daily returns during our i sampling period
ADCA = the mean for firm i of Absolute value of discretionary current i
accruals during our sampling period
lnMBRT = the natural log of the mean market-to-book ratio for firm i during our i sampling period
ln LEV = the natural log of the mean the ratio of long-term debt and short-term debt to total assets for firm i during our sampling period
lnSIZE = the natural log of the mean total assets for firm i during our i sampling period
ln PSP = the natural log of the mean for firm I of the percentage bid-ask s period during our sampling period