CHAPTER VII - RESULTS OF STATISTICAL DATA ANALYSIS
7.3 Results of segmented data analysis
To our knowledge, no previous studies have investigated relative performance differences between industries in DK within the private equity domain of study.
In chapter 2, we documented that from 2007-2015, some industries of the Danish buyout market have been invested and divested more than others and hence it was concluded that PE-firms change their investment and divestment focus from year to year.
With the above in mind, we find it interesting to investigate whether the different investment and divestment focus by industry is due to differences in operational performance of those industries.
This is the reason for testing hypothesis #4 of this thesis by applying our data in the Kruskal-Wallis H-test i.e. whether industry differences with regard to operational performance can be observed in our data.
7.3.1 Growth metrics
The segmentation of our data is based on six different industries with which the 43 portfolio companies have been divided into. The segmented data is tested on the same three areas of operational performance measures as in the full data analysis.
Table 18 presents the empirical results from the statistical test of hypothesis #4 of this thesis.
83
Table 18: Growth test statistics for differences between industries
From the test statistics in table 18 it is evidenced that the operational performance of the six industries measured in terms of growth in assets is significantly different between the industries at a 95% significance level. This result seems to support hypothesis #4 that industry differences with respect to asset growth is observed in data. The median change in the Kruskal-Wallis H-test cannot be used to compare e.g. the largest increase with the largest decrease and comment on this. It solely states, if significant, that one or more of the groups, deviate significantly from the entire sample (Newbold, Carlson, & Thorne, 2013). Even though it would be logical to conclude that this must be the largest decrease vs. largest increase, it cannot be interpreted like this, because of the correlation between all six groups.
If the test is significant, then to isolate the group or groups that deviate from the sample, you Growth test statistics [entry -1; exit +1] Observations
Revenue
Median change (Healthcare) 29.75% 7
Median change (TMT) 52.00% 6
Median change (Industrials) -0.04% 18
Median change (Service) -3.41% 5
Median change (Consumer goods) -153.52% 2
Median change (Other) 26.14% 5
P-value (significance) 0.1398
Assets
Median change (Healthcare) 56.53% 7
Median change (TMT) 29.52% 6
Median change (Industrials) -12.00% 18
Median change (Service) -17.25% 5
Median change (Consumer goods) -158.67% 2
Median change (Other) 55.69% 5
P-value (significance) 0.0471**
Employees (FTEs)
Median change (Healthcare) -8.10% 7
Median change (TMT) 42.49% 6
Median change (Industrials) 23.24% 18
Median change (Service) 27.26% 5
Median change (Consumer goods) -134.18% 2
Median change (Other) 37.10% 5
P-value (significance) 0.1029
Note: Test statistics from Kruskal-Wallis H-test. Median changes are measured as percentage changes.
***, **, and * imply statistical significance at confidence levels 99%, 95%, and 90%, respectively.
84
must test each group bilaterally towards each other in another statistical test called the Mann-Whitney U-test.
With respect to the growth measures of revenue and employees, the median changes between the industries are different from each other in both increasing and decreasing direction.
However, the median differences measured in terms of growth in revenue and employees are not statistically significant, but very close to be at a 90% significance level.
In chapter 2 we documented that industries on average having received most attention in terms of buyout investments in the period 2007-2015 are ‘Computer and Consumer Electronics’
(21.2%), ‘Business and Industrial Products’ (16.1%), ‘Life Sciences’ (13.3%), and ‘Consumer Goods and Retail’ (11.8%). Note that the dividing of buyout investments by industry in chapter 2 was done on 15 industries due to the available data from Invest Europe (2017), whereas the dividing of portfolio companies from our sample by industry for the segmented data analysis cf. Kruskal-Wallis H-test is done on six industries due to the size of our sample.
The industry on average performing best relatively in terms of growth in revenue, assets, and employees in the event window cf. median changes in above table 18 is the ‘TMT’ industry.
This supports the finding from chapter 2 that the industry of ‘Computer and Consumer Electronics’, which is the closest industry to compare with TMT, is the most invested industry in the Danish buyout market in the period 2007-2015 with 21.2% of total investments.
On the other side, what is also quite interesting with regard to the test statistics in table 18 and the focus of PE-firms on industries in the Danish buyout market in terms of investments is that the fifth most invested industry in 2007-2015 i.e. ‘Consumer Goods and Retail’ referred to as ‘Consumer goods’ in table 18, have performed very bad in terms of growth in revenue, assets, and employees relative to the other industries with median changes of 153.52%, -158.67%, and -134.18%, respectively. Note that a potential sample bias can be attributed to these statistics, as the number of portfolio companies in our sample in this industry is only two firms.
Due to the evidence provided in table 18 on differences in operational performance in terms of growth between industries that only median differences measured in terms of assets is significant on a 95% significance level, the overall conclusion on the growth tests statistics is that we cannot reject the null hypothesis 𝐻04 and hence cannot document that the six industries perform significantly different in the event window with regard to growth measures.
85 7.3.2 Profitability metrics
Table 19 presents the empirical results from the statistical test of hypothesis #4 i.e. that industry differences is observed in our data in terms of profitability measurements.
Table 19: Profitability test statistics for differences between industries
By comparing the median changes of the six samples of industries cf. above table 19 it appears that for all the performance measures within profitability, industry differences is observed in both increasing and decreasing direction. However, none of the four profitability performance
Profitability test statistics [entry -1; exit +1] Observations EBITDA
Median change (Healthcare) 120.04% 7
Median change (TMT) 42.56% 6
Median change (Industrials) -25.77% 18
Median change (Service) 79.84% 5
Median change (Consumer goods) -70.63% 2
Median change (Other) 45.37% 5
P-value (significance) 0.8054
EBITDA margin
Median change (Healthcare) 4.67% 7
Median change (TMT) 1.67% 6
Median change (Industrials) -3.59% 18
Median change (Service) 1.65% 5
Median change (Consumer goods) -1.21% 2
Median change (Other) 0.22% 5
P-value (significance) 0.9174
EBITDA/Assets
Median change (Healthcare) 7.94% 7
Median change (TMT) 0.44% 6
Median change (Industrials) -7.69% 17
Median change (Service) 6.65% 4
Median change (Consumer goods) 9.43% 2
Median change (Other) -6.55% 5
P-value (significance) 0.7076
ROE
Median change (Healthcare) -40.97% 7
Median change (TMT) -5.58% 6
Median change (Industrials) -15.53% 17
Median change (Service) 38.36% 4
Median change (Consumer goods) -6.13% 2
Median change (Other) -32.19% 5
P-value (significance) 0.5670
Note: Test statistics from Kruskal-Wallis H-test. Median changes are measured as percentage changes, except for ratios, which are measured as percentage point changes.
***, **, and * imply statistical significance at confidence levels 99%, 95%, and 90%, respectively.
86
measurement differences by industry are statistically significant, and hence it is not possible to reject the null hypothesis 𝐻04.
Even though we cannot reject the null hypothesis and cannot conclude that significant industry differences with regard to performance in terms of profitability appears in our data, there are again some interesting linearity’s between some of the relative median changes in table 19 and statistics which suggest that there may be a positive correlation between the performance of an industry and the investor focus of the PE-firms on the Danish buyout market.
7.3.3 Productivity metrics
In below table 20 is presented the test statistics derived from the statistical testing of hypothesis #4 with respect to productivity effects.
Table 20: Productivity test statistics for differences between industries
From the P-values of the two productivity performance measurements observed in above table 20 we can conclude that there exist no significant differences in those performance measurements by industry, and hence we cannot reject the null hypothesis 𝐻04 in respect to productivity measurements.
Productivity test statistics [entry -1; exit +1] Observations Revenue/FTEs
Median change (Healthcare) 39.75% 7
Median change (TMT) 7.87% 6
Median change (Industrials) -23.84% 18
Median change (Service) 4.29% 5
Median change (Consumer goods) 43.91% 2
Median change (Other) -17.21% 5
P-value (significance) 0.3968
Asset turnover
Median change (Healthcare) 10.09% 7
Median change (TMT) 13.14% 6
Median change (Industrials) 12.63% 17
Median change (Service) 15.69% 4
Median change (Consumer goods) -20.26% 2
Median change (Other) 16.91% 5
P-value (significance) 0.4500
Note: Test statistics from Kruskal-Wallis H-test. Median changes are measured as percentage point changes.
***, **, and * imply statistical significance at confidence levels 99%, 95%, and 90%, respectively.
87
Similar patterns with regard to relative performance by industry and PE firm investment focus can be observed in the above test statistics as with growth and profitability test statistics.