First, a univariate test is employed, the test compares the DLLP and EBLLP of credit unions that witness earnings declines to those that witnessed earnings increases. Similar to Beatty et al. (2002), to avoid reporting earnings declines, I expect credit union managers to underestimate the discretionary part of loan loss provision. Therefore, DLLP is expected to be negative if ΔEBLLP is negative and vice versa.
Table 7 presents the results of the univariate tests. The table shows summary statistics on one subsample of credit unions that have a negative change in EBLLP and another subsample of
26
credit unions that have a positive change in EBLLP. Credit unions with a negative earnings change engage in negative DLLP (income increasing), the mean is -0.0000389. Credit unions with a positive earnings change engage in positive DLLP (income decreasing), the mean is 0.0000364. The difference of the means of the two subsamples is significant at the 1 percent level for the two-tailed t-test. Then the sample is divided according to the size categories. Large credit unions with a negative earnings change engage in negative DLLP, the mean is -0.0000386; and large credit unions with a positive earnings change engage in positive DLLP, the mean is 0.0000351. The difference of the means of the two subsamples is significant at the 1 percent level for the two-tailed t-test. Small credit unions with a negative earnings change engage in negative DLLP, the mean is -0.0000379; and small credit unions with a positive earnings change engage in positive DLLP, the mean is 0.0000361. The difference of the means of the two subsamples is significant at the 1 percent level for the two-tailed t-test.
[INSERT TABLE 7 ABOUT HERE]
For the multivariate analysis, an OLS regression is used for model (2) using the error term of model (1) as the dependent variable. Depending on the magnitude and sign of the coefficient of EBLLP, this indicates whether credit union managers use their discretion in estimating loan loss provision to manage earnings; and the magnitude and sign of the coefficient of ΔEBLLP indicates whether credit union managers manage earnings to avoid reporting earnings declines. The results of model (2) are reported in Table 8. The positive and significant coefficient of EBLLP (0.19154, p-value<0.01) in table 8 indicates that credit union managers are engaging in income smoothing, which is a form of earnings management. The coefficient of the ΔEBLLP (0.00865, p-value<0.01) is positive and significant at the 1 percent level. Therefore, credit union managers engage in income smoothing to avoid reporting earnings declines. The explanatory power of the model is high (adjR2 = 36.8%). This is consistent with hypothesis H1. Economically, the results are significant, an increase of one standard deviation in each of EBLLP and ΔEBLLP increase DLLP by 84%.
[INSERT TABLE 8 ABOUT HERE]
For the large category, the coefficient of EBLLP (0.17389, p-value<0.01) is positive and significant at the 1 percent level, and the coefficient of the ΔEBLLP (0.01394, p-value<0.01) is
27
also positive and significant at the 1 percent level. This implies that large credit unions smooth earnings to avoid reporting earnings declines. For the small category, the coefficient of EBLLP (0.19881, p-value<0.01) is positive and significant at the 1 percent level, and the coefficient of the ΔEBLLP (0.00685, p-value<0.01) is also positive and significant at the 1 percent level. This also implies that small credit unions smooth earnings to avoid reporting earnings declines. A test for the significance of the difference in coefficients of large and small credit unions is conducted. The coefficients of EBLLP and ΔEBLLP for the large category are not statistically significant different from the coefficients of EBLLP and ΔEBLLP for the small category (p>0.05), respectively.
2.5.4 Credit union characteristics and earnings management
Table 8 also presents the variables that represent the characteristics of the credit unions engaging in earnings management. The positive and significant coefficient of lnTA (β3 =
0.00008, p-value<0.01) is consistent with hypothesis H2a suggesting that earnings management increases as size increases. The negative and significant coefficient of ROAA (β5 = -0.00094, p-
value<0.01) is consistent with hypothesis H2b suggesting that earnings management increases as profitability decreases. The positive and significant coefficient of NW (β6 = 0.00001, p-
value<0.01) is consistent with hypothesis H2c suggesting that earnings management increases as net worth increases.
For the control variables that are indirect proxies for oversight, state-chartered credit unions engage more in earnings management than federal-chartered credit unions, the coefficient is negative and significant (β7 = -0.00006, p-value<0.01); consistent with hypothesis H2d.
Moreover, the coefficients of Pop_Density (β9 = -0.0000, p-value<0.01) and Educ_Level (β10 =
0.00068, p-value<0.01) are significant but the sign for Educ_Level is contrary to the expected sign. As a sensitivity check, the regression controls for the economic variables at the Metropolitan Statistical Area, and the County levels; results are similar. Results are not tabulated for brevity.
The findings for the two subcategories, the large and small credit unions, are similar to the results of the whole sample. A test for the significance of the difference in coefficients of large and small credit unions is conducted. Only the coefficients of Stat, and Educ_level for the
28
large credit unions are statistically significant different from the coefficients of Stat, and Educ_level for the small category (p<0.05), respectively.