5. EMPIRICAL RESULTS AND ANALYSES
5.1. Evidence from panel regressions: Estimation results
Table 9: Fixed effect estimation results
Herfindahl ownership index Ownership concentration ratio Variables Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Herfindahl ownership
index
-9.56E-6 (6.87E-6)
-2.98E-6 (9.02E-6)
-3.53E-6 (9.03E-6)
-3.50E-6 (9.04E-6)
- - - -
Ownership concentra-tion ratio
- - - - -0.0458
(0.0535)
-0.0134 (0.0574)
-0.0167 (0.0574)
-0.0185 (0.0351) Dominant blockholder - -0.0470
(0.0417)
-0.0533 (0.0421)
-0.0530 (0.0422)
- -0.0530
(0.0341)
-0.0602*
(0.0347)
-0.0593*
(0.0351)
Takeover - - 0.0547
(0.0497)
0.0547 (0.0498)
- - 0.0544
(0.0497)
0.0544 (0.0497)
Block investment - - - 0.0016
(0.0139)
- - - 0.0026
(0.0141) Cost retrenchment 0.0014
(0.0038)
0.0014 (0.0038)
0.0016 (0.0038)
0.0016 (0.0038)
0.0013 (0.0038)
0.0014 (0.0038)
0.0015 (0.0038)
0.0016 (0.0038) Asset retrenchment 0.0886***
(0.0099)
0.0879***
(0.0099)
0.0878***
(0.0099)
0.0878***
(0.0099)
0.0882***
(0.0099)
0.0878***
(0.099)
0.0877***
(0.0099)
0.0877***
(0.0099)
Firm size -0.0129
(0.0164)
-0.0127 (0.0164)
-0.0126 (0.0164)
-0.0127 (0.0164)
-0.0116 (0.0164)
-0.0124 (0.0164)
-0.0122 (0.0164)
-0.0123 (0.0164)
Industry dummies Yes Yes Yes Yes Yes Yes Yes Yes
Country dummies Yes Yes Yes Yes Yes Yes Yes Yes
Time dummies Yes Yes Yes Yes Yes Yes Yes Yes
F-value <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001
R2-value 0.3607 0.3612 0.3618 0.3618 0.3601 0.3612 0.3618 0.3618
# cross-section/time 289 / 15 289 / 15 289 / 15 289 / 15 289 / 15 289 / 15 289 / 12 289 / 12 This table shows fixed effects (FE) OLS estimation results when the Herfindahl ownership index and ownership concentration ratio is the ownership concentration variable. Standard errors are presented below the parameter estimates in parentheses and are corrected for heteroscedasticity. The sample is restricted to the years in the turnaround process, i.e. year 3-8. The time and individual intercepts are not shown to save space. F-tests for no fixed effects are all rejected. Hausman tests suggest fixed effects as presented in Table 24. Although included, industry and industry effects are conditioned out and their effects absorbed by the individual intercepts. Stars indicate statistically significance at the respective levels: * p<0.10; ** p<0.05; *** p<0.01.
The coefficients for both ownership concentration definitions are negative, suggesting turnaround performance is negatively related to ownership structure, but the coefficients are all statistically insignificant. Blockholder dominance negatively influences turnaround performance, supporting the two-sided hypothesis of a negative relationship. Blockholder dominance is weakly statistically significant in Model 7 and 8, suggesting that firms having a dominant blockholder is weakly significantly experiencing a turnaround performance 6 pct.-points lower than non-dominated firms. Blockholder dominance is insignificant in the remaining models, i.e. Model 1 to 6. Takeover and block investments are having the hypothesized sign by suggesting a positive relationship to turnaround performance, but both variables are found to be statistically insignificant. The variable cost retrenchment is also found to both be insignificant related to turnaround performance and to take the incorrect sign of the expected relationship. In addition, firm size is also found to be insignificant associated with performance.
Most importantly, I find asset retrenchment to be highly significant with a negative (positive sign) influence in the relationship with turnaround performance. The effect of asset
retrenchment is generally very robust across the different model specifications. In terms of effect, an increase in the asset base, which is equal to a decrease in asset retrenchment, by 1 pct.-point is related with an increase in turnaround performance by approximately 8.8 pct.-pct.-points.
Asset retrenchment is measured by the change in asset base, where a negative coefficient indicates that firms decreasing its asset base experience improved performance. That is asset retrenchment results in better performance. Somewhat different, my results suggest a positive relationship between increases in the asset base and turnaround performance, meaning asset retrenchment is negatively related to turnaround performance.
In terms of stability, the ownership concentration variables were sensitive to the different model specifications, which mainly stem from the inclusion of the dominant blockholder variable. None of the other variables, as reported above, were sensitive to different specifications. This was expected as they are strongly and highly inter-correlated. The adjusted R2 is reported to indicate model performance and it is approximately 36 pct. in all models.
Table 22 and Table 23 (Appendix 9) replicate the model specifications in Table 9 above and report results for fixed effect estimations when not controlling for fixed time effects and pooled estimations respectively. The fixed effect models in Table 22 do not differ substantially except the variable dominant blockholder is more significant in Model 7 and 8. Asset retrenchment is highly significant, while firm size is weakly significant when not considering time effects. The remaining estimation of parameters is not altered in terms of sign or significance. Table 23 reports pooled estimation results. The results differ compared to the fixed effects method estimations, which is evident from the parameters changing signs, parameters becoming significant, and the adjusted R2 decreasing considerably. This behaviour emphasizes the importance of using the fixed effect approach by including firm and time fixed effects. This is important since the variations between the two methods indicate that the independent variables are correlated with the error term in the pooled regression models, causing the estimates to be biased and inconsistent (Borsch-Supan & Koke, 2000)
Table 25 (Appendix 10) shows estimation results when using return on invested capital (ROIC) as the dependent variable. Surprisingly, estimating random effects is the most appropriate methods in this case, which is confirmed by the insignificant Hausman test for all models. However, the adjusted R2 is significantly low, making the estimations uninteresting.
ROIC is not considered in the forthcoming models. Similar, block investment is not reported as
the variable has no explanatory effect in the forthcoming estimations. The inclusion of the block investment variable did not affect the estimates of the other variables.
5.1.2. Dynamic panel estimation results
The key purpose of the dynamic models in Table 11Table 10 is to evaluate the influence of past turnaround performance on current turnaround performance through inclusion of the lagged dependent variable in the model specifications as shown in Equation (7). In the second column, Table 10 shows dynamic fixed effect models results, while first column present the results obtained by the two-step GMM estimations using the Arellano and Bond methodology build into the SAS EG procedure.
Table 10: Results of dynamic panel regression with GMM and FE estimation
Dynamic panel models GMM Fixed effects
Variables Model 5 Model 6 Model 7 Model 5 Model 6 Model 7
Lagged turnaround performance
0.3420***
(0.0014)
0.3827***
(0.1001)
0.4130***
(0.1038)
0.0627**
(0.0302)
0.0647**
(0.0302)
0.0651**
(0.0302) Ownership concentration ratio -0.3346
(0.3007)
-0.4649 (0.3295)
-0.4956 (0.3515)
-0.0480 (0.0534)
-0.0135 (0.0573)
-0.0168 (0.0574)
Dominant blockholder - 0.1106
(0.1773)
0.2711 (0.2134)
- -0.0559
(0.0341)
-0.0633*
(0.0347)
Takeover - - -0.4352*
(0.2452)
- - 0.0559
(0.0497)
Cost retrenchment 0.0517**
(0.0215)
0.0564**
(0.0243)
0.0569**
(0.0248)
0.0004 (0.0038)
0.0005 (0.0038)
0.0006 (0.0038) Asset retrenchment 0.2300***
(0.0597)
0.2193***
(0.0693)
0.2318***
(0.0708)
0.0886***
(0.0099)
0.0882***
(0.0099)
0.0881***
(0.0099)
Firm size -0.1425
(0.1123)
-0.0829 (0.1228)
-0.0741 (0.1210)
-0.0260 (0.0165)
-0.0164 (0.0165)
-0.0163 (0.0165)
Industry dummies Yes Yes Yes Yes Yes Yes
Country dummies Yes Yes Yes Yes Yes Yes
Time dummies Yes Yes Yes Yes Yes Yes
Sargan Test (Chi2-statistic) 18.42 17.44 13.77
1st-order autocorrelation AR(1) - - -
2th-order autocorrelation AR(2) - - -
R2-value 0.3621 0.3633 0.3638
# cross-section/time length 289 / 15 289 / 15 289 / 15 289 / 15 289 / 15 289 / 15 This table reports GMM and fixed effects (FE) estimation results with ownership concentration ratio as the ownership concentration variable and the lagged dependent variable. Stars indicate statistically significance at the respective levels: * p<0.10; ** p<0.05; *** p<0.01. Standard errors are presented below the parameter coefficients in parentheses and the FE model are corrected for heteroscedasticity. The sample is restricted to the years in the turnaround process, i.e. year 3-8. The time and individual intercepts are not shown to save space. The FE F-statistics for no fixed time effects are all rejected. Joint significant tests are all significant. The Sargan statistics related to the GMM estimations all verify over-restriction of restrictions in all models. First and second order autocorrelation tests fail to report statistics, suggesting autocorrelation in the first and/or second order regression residuals. Five lags of the dependent variable are introduced and used as instruments.
First, the estimates from the dynamic fixed effect models (Model 5-7) reflect that past turnaround performance has a statistically significant explanatory effect on current turnaround performance, suggesting good turnaround performance is positive related to turnaround
performance in the current period. An increase in the level of past turnaround performance by 1 pct.-point is generally associated with a 6.5 pct.-point increase in performance. Again, asset retrenchment is highly significant with the incorrect predicted sign, while blockholder dominance is weakly significant at the more generous significance level in Model 8. Model performance in terms of adjusted R2 does not change by the inclusion of the lagged dependent variable. The lagged turnaround performance has not a considerable effect on the magnitude of the estimates in Model 5-7.
The assumption of exogenous variables is violated by the inclusion of the lagged turnaround performance in the dynamic fixed effect models and the estimations do not take the endogeneity of this variable into account. It is noteworthy that the significant estimates remain significant, while all estimates maintain their sign and magnitude. However, the underlying assumptions are violated due to the endogeneity of the lagged variable (Baltagi, 2005), undermining any causal inference based on the results. Luckily, the GMM method should be able to address these shortcomings.
Based on the econometrics issues arising when estimating the dynamic fixed effect models, the dynamic GMM models are estimated. The estimation results are presented in the first column in Table 10. The dynamic GMM panel model uses lags as instruments. Hence, the Sargan test of over-identification is reported. The reported Sargan statistics does not reject the validity of the used instrument.19 The results provided by the GMM estimation suggest an extremely large and highly significant relationship lagged turnaround performance and turnaround performance. Again, asset retrenchment is highly significant and presenting a very large effect, while cost retrenchment also are found to be significant and takeover is found to be weakly significant. The results depict no significant relationship between ownership concentration and firm turnaround performance. However, the first and second order autocorrelation tests fail to be producing any statistics, which leaves me to question the estimation results and the choice of instruments46.
The extreme dynamic GMM results are likely to be caused by inappropriate instrument, wrong instrument and the ignorance of the potential endogeneity of ownership concentration.
Assuming exogeneity of ownership concentration in the GMM estimation greatly alters the magnitude of the estimates. Table 28 shows estimation results when holding ownership concentration exogenous, confirming the lack of proper instruments in the GMM estimation
19 Using lower lags induce the Sargan statistic to discard the instrument.
(Appendix 11).20 These econometric issues are addressed in the discussion section (Section 6.1.), when discussing endogeneity issues arising from ownership concentration.