4. DATA AND METHODOLOGY
4.5. Descriptive statistics
paring turnaround and non-turnaround firms within the approximately same time periods (e.g.
Mueller & Barker, 1997). However, as noted by Bibeault (1999), “a boom covers many sins, and a bust uncovers many weaknesses”. Bibeault refers to the fact that macroeconomic events, economic change, and business cyclic behaviour often reveal unsound corporations, which are reflected by a larger number of firms experiencing severe performance declines at the onset of economic downturns (Bibeault, 1999). Based on this conception, I introduce time dummies to capture potential fixed year effects for the firms in the sample.
4.4.4. Summary of variable definitions and data sources
Table 5 summarizes the variables used in the econometric analyses.
Table 5: Summary of explanatory and control variables
Variable Variable explanation Definitions and description Expected sign HHI Herfindahl index The sum of individual squared ownership share by all
blockholders.
+ OCR Ownership
concen-tration ratio
The variable is defined as the percentage of the total ownership share of all blockholders
+ DOMI Blockholder dominance Takes on the value 1 if the firm is dominated by a single
blockholder, otherwise 0
+ / - BI Block investment Takes on the value 1 if there is a block investment in the
given year, otherwise 0.
+ TO Takeover Takes on the value 1 if the firm experience a takeover or a
blockholder increases its share to above 50 pct., otherwise 0.
+ COSTRY Cost retrenchment Change in cost base defined as (Cost baset – Cost baset-1)/Cost
baset-1
- ASSETRy Asset retrenchment Change in asset base defined as (Asset baset – Asset base
t-1)/Asset baset-1
-
SIZE Firm size Natural logarithm of the number of employees + / -
The table summarizes the independent variables applied in this thesis except dummies to control for industry, country and time specific effects. The hypotheses and expected signs are formulated under the ceteris paribus condition. The dependent variables are given the following abbreviations:
Turnaround performance (AdjROA), turnaround performance measured by ROIC (AdjROIC), turnaround outcome depending on the definition;
TURNa and TURNb.
Table 6: Sample descriptive statistics
Variable Mean Std.dev. Min Max Median
Turnaround performance -0.1009 0.2494 -3.4390 0.3779 -0.0385
Herfindahl ownership index 1859.07 2104.22 0.0000 1000.00 1005.08
Ownership concentration ratio 0.4927 0.2559 0.0000 1.0000 0.5139
Blockholder dominance 0.2468 0.4213 0.0000 1.0000 0.0000
Takeover 0.0156 0.1238 0.0000 1.0000 0.0000
Block investment 0.3126 0.4637 0.0000 1.0000 0.0000
Cost retrenchment 0.0868 1.6170 -16.3764 32.7089 -0.0151
Asset retrenchment 0.0574 0.6257 -1.0000 8.1238 -0.0253
Size 6.9489 1.8473 1.6094 13.0470 6.8101
N=1734 (equal to 289 cases in each year). The sample is restricted to the years in the turnaround cycle period, i.e. year 3-8. This table reports descriptive statistics for all variables (both dependent and independent) and measures used in my estimations related to the main definition.
It is evident that the mean (average) of turnaround performance among the firms in the period is negative 10.19 pct., indicating that the average firm averagely has a non-viable and poor performance during the turnaround cycle period. The average firm has an HHI of 1859, which is the alternative definition to ownership concentration that also is measured by the ownership concentration ratio, taking the average of 49.27 pct. The average of the dominant blockholder variable is 24.69 pct., indicating the variable takes the value 1 in approximately one fourth of the observations. The descriptive statistics reveals that on average 1.56 pct., and in absolute values amounting to 27, of the firms experienced a takeover activity during the turnaround process.
Similar, the block investment is on average 31.26 pct., meaning that a block investment occurred in approximately one third of the observations, which confirms the perception of that acquisitions and/or increases in holdings of shares increases regularly. The mean value of size is 6.94, suggesting the average firms has approximately 1042 employees. The mean value of cost and asset retrenchment is 8.68 pct. and 5.74 respectively, while the median is -1.51 and -2.53 for the average firm. This suggests that outliers14 in the sample affect the mean value, which is sensitive to extreme observations, why the median value is also reported to describe the middle observation. The standard deviations are reported to describe the spread of the data, indicating that the values for some variables are wide spread from the mean.
14 The descriptive statistics reflect that the panel dataset is likely to be subject to (extreme) outliers, which is also confirmed by assessing the distribution of the variables. This ignites the considerations to remove some of these observations. Two options are possible: 1) restrict the sample to not include the (most extreme) outliers, or 2) keep outliers in the sample. As the first option implies removing data from the analysis, which additionally would reduce the number of firms in the sample, I do not remove any outliers for two reasons. First, I have checked the most extremes outliers for miscalculations and validated the data, which did not lead to any incorrect measures and, thus, no exclusion of outliers. Second, I follow the mindset that altering the dataset is constructing the reality as wished for. The outliers present the fact that some turnaround measures yield extreme values, and removing outlying observations may change the relation among variables. Therefore, outliers are not restricted from the sample despite the fact that outlying observations may affect the panel data estimations. To mitigate the issue with outliers, I take corrective actions in the SAS procedures when possible.
In Table 7 the descriptive statistics are classified by the year in the turnaround cycle period, which illustrates the development of the variables year-wise during this process on an aggregated level.
Table 7: Sample descriptive data represented for each year in the turnaround cycle period Year in the turnaround process
Variable Year 3 Year 4 Year 5 Year 6 Year 7 Year 8
Turnaround performance -0.0862 (0.2080)
-0.1433 (0.2214)
-0.1476 (0.2114)
-0.1025 (0.2316)
-0.0652 (0.2368)
-0.0608 (0.3471) Herfindahl ownership index 1876.06
(2143.47)
1793.50 (2019.84)
1830.92 (2061.92)
1843.71 (2099.51)
1861.91 (2107.86)
1948.33 (2202.93) Ownership concentration ratio 0.4798
(0.2594)
0.4798 (0.2569)
0.4914 (0.2568)
0.4898 (0.2587)
0.5036 (0.2479)
0.5116 (0.2563) Dominant blockholder 0.2595
(0.4391)
0.2422 (0.4292)
0.2457 (0.4312)
0.2318 (0.4227)
0.2422 (0.4292)
0.2595 (0.4391)
Takeover 0.0104
(0.1015)
0.0069 (0.0830)
0.0138 (0.1170)
0.0104 (0.1015)
0.0277 (0.1643)
0.0242 (0.1540)
Block investment 0.2457
(0.4312)
0.3010 (0.4595)
0.3080 (0.4625)
0.3149 (0.4653)
0.3702 (0.4837)
0.3356 (0.4730)
Cost retrenchment 0.2908
(2.0744)
0.1915 (1.8087)
0.0297 (0.8451)
-0.0212 (0.9565)
-0.1244 (1.0360)
0.1546 (2.3106)
Asset retrenchment 0.3537
(1.1069)
-0.0526 (0.4612)
-0.0470 (0.5322)
-0.0437 (0.2561)
0.0334 (0.3455)
0.1004 (0.5710)
Firm size 7.0172
(1.8412)
7.0431 (1.8405)
6.9912 (1.8252)
6.9170 (1.8405)
6.8674 (1.8534)
6.8576 (1.8896) N=289 in each year. The table presents means and standard deviations in parentheses for the variables each year during the turnaround process, and is related to the main definition.
A noteworthy development is the average firm size that decreases, which indicate the average firm reduces the amount of employees during the turnaround process. The performance of the average firm decreases during the decline period, while increasing in the recovery period. Asset retrenchment follows an expected pattern by being negative in turnaround year 4 to 6, while cost retrenchment takes a pattern that is diverging to the theorized pattern.
In Table 17 and Table 18 (Appendix 6) are reported the mean and standard deviations for the two additional definitions, which are grouped by the turnaround outcome and, thus, provides a prelude of what to expect from these models. Generally, the descriptive statistics in Table 17 and Table 18 reveal that turnaround firms on average are larger than non-turnaround firms. The average turnaround firm seem less likely to be dominated by a single blockholder, while there on average are fewer cases of takeovers in turnaround firms than compared to the average non-turnaround firms. Ownership concentration does not seem to differentiate the two groups.
Table 8 reports the correlation coefficients between the dependent variable and all independent variables considered in the alternative model specifications.
Table 8: Correlations between variables considered in this thesis
Variables 1 2 3 4 5 6 7 8 9
1. Turnaround performance 1
2. Herfindahl ownership index .05** 1
3. Ownership concentration ratio .07*** .79*** 1
4. Dominant shareholder .02 .81*** .59*** 1
5. Block investment -.02 -.26*** -.11*** -.23*** 1
6. Takeover .00 .17*** .14*** .21*** -.01 1
7. Cost retrenchment -.01 -.03 .00 -.03 -.01 -.02 1
8. Asset retrenchment .18*** .01 .02 -.02 .02 .00 .17*** 1
9. Firm size .19*** -.07*** -.15*** -.07*** .01 -.01 -.10*** -.04 1
N=1734 (289 cases multiplied by six years of interest). This table reports correlations between the variables used in testing the main approach. Stars indicate statistically significance: * p<0.10; ** p<0.05; *** p<0.01, which are used to indicate the result of the null hypotheses testing for zero correlation.
If the null hypothesis is rejected, there is an indication of either positive or negative relationship between the two given variables. The sample embraces all years in the turnaround cycle period, e.g. year 3-8.
In relation to the model general model specification, the dependent variable are significantly correlated with Herfindahl index, ownership concentration, asset retrenchment and firm size, although this does not necessarily imply significance in the individual model estimations.
A high degree of correlation between the explanatory variables may suggest multicollinearity problems, which could limit the usefulness of my estimation results. For example, the correlation coefficient is 0.81 between Herfindahl ownership index and blockholder dominance, indicating a significant linear relationship between these two variables, while the latter is correlated to ownership concentration with a value of 0.59 and, thus, only moderately correlated. The pair-wise correlation between the two variables used in measuring ownership concentration is 0.79. These two variables will be substituted by each other in the model specification to test different ownership concentration measures and their inter-correlation is therefore not relevant. The high inter-correlation between Herfindahl ownership index and blockholder dominance is strongly correlated by exceeding 0.80, suggesting there may be severe multicollinareairty problems. However, high correlation is not a necessary condition for multicollinearity to exist. Therefore, I apply variance inflation factors (VIF) and condition index (CI) in order to test the presence of multicollinearity, which do not suggest any problems with multicollinearity (Appendix 7).
Table 20 and Table 21 (Appendix 8) report correlations of the alternative sample definitions used in the discrete response models. Furthermore, the tables report the mean and standard deviation for the full sample for definition 2a and 2b respectively without grouping the descriptive statistics.