An analysis of the determinants of secondary buy-Out persistence.

(1)

An Analysis of the Determinants of

Secondary Buy-Out Persistence.

By

Kuchulain O’Flynn

Supervised by

Dr. J. K. Martin

September 2014

University of Amsterdam

Amsterdam Business School

Master in International Finance

(2)

Abstract:

This paper examines a selection of potential determinants of secondary buyout persistence. Data was collected from 1342 deals, involving 318 companies, over a period of 34 years, along with data (such as capital market data). The method in which the data was collected is unique to other studies on secondary buyouts and private equity, in the sense that the data was collected at target company level, and collected for the life of the target company. Logit models were constructed to predict the likelihood of secondary buyout persistence. The models were separated into the following categories: Capital Providers, Capital Markets, Industry & Region Specifics, History of Alternative Exit Failure, Deal Specifics, and Acquisition Behaviour of Target Companies. The statistical significant variables from each model were then combined into a global model. While there was a selection of statistically significant determinants of secondary buyout persistence in each model, they all seem to be driven by the amount paid for the target firm. The only exception to this is the number of agents on the sell side of a deal. Therefore, we have concluded that that the deal size, amount paid for the target company, and number of sellers, number of agents involved in the sell side of the deal, are statistically significant determinants of secondary buyout persistence. For every additional seller involved in a deal the likelihood that the deal is a Secondary Buyout Persistent Deal increases by 3.5%. Likewise, for every $1million increase in the amount paid for the target firm the likelihood that the deal is a Secondary Buyout Persistent Deal increases by 0.403%

(3)

Table of Tables:

Table 1: HHI Index Description ... 16

Table 2: Country Code Rank Description ... 24

Table 3: Descriptive Statistics - Club Deal Investors ... 59

Table 4: Descriptive Statistics - Club Deal Sellers ... 60

Table 5: Descriptive Statistics - Club Deal Lenders ... 61

Table 6: Descriptive Statistics - Number of Sellers ... 62

Table 7: Descriptive Statistics - Fed Tightening Index ... 63

Table 8: Descriptive Statistics - High-Yield Spread ... 64

Table 9: Descriptive Statistics - Number of IPOs ... 65

Table 10: Descriptive Statistics - Herfindahl-Hirschman Index ... 66

Table 11: Descriptive Statistics - Cumulative M&A Failure ... 67

Table 12: Descriptive Statistics - Cumulative IPO Failure ... 68

Table 13: Descriptive Statistics - Deal Size ... 69

Table 14: Descriptive Statistics - Years Held ... 70

Table 15: Descriptive Statistics - Investment - Same Country ... 71

Table 16: Descriptive Statistics - Investment - Different Country ... 72

Table 17: Descriptive Statistics - Investment - Value ... 73

Table 18: Descriptive Statistics - SBO Persistence Deal ... 74

Table 19: Model Construction - Acquisition Behaviour of Target Company ... 75

Table 20: Model Construction - History of Failed Alternative Exit Options ... 76

Table 21: Model Construction - Providers of Capital ... 77

Table 22: Diagnostics - Providers of Capital - Multicollinearity Check ... 79

Table 23: Diagnostics - Providers of Capital - Pearson Goodness of Fit ... 79

Table 24: Model Construction - Deal Specifics ... 80

(7)

3 | P a g e

Table 26: Diagnostics - Deal Specifics - Pearson Goodness of Fit ... 82

Table 27: Model Results - Deal Specifics - Marginal Effects ... 83

Table 28: Model Construction - Industry & Region ... 84

Table 29: Diagnostics - Industry & Region - Multicollinearity Check ... 86

Table 30: Diagnostics - Industry & Region - Pearson Goodness of Fit ... 86

Table 31: Model Results - Industry & Region - Marginal Effects ... 87

Table 32: Model Construction – Capital Markets ... 88

Table 33: Diagnostics - Capital Markets - Multicollinearity Check... 90

Table 34: Diagnostics - Capital Markets - Pearson Goodness of Fit... 90

Table 35: Model Construction – Combined Model ... 91

Table 36: Diagnostics - Combined Model - Multicollinearity Check ... 93

Table 37: Diagnostics - Combined Model - Pearson Goodness of Fit ... 93

(8)

4 | P a g e

Table of Figures:

Figure 1: Diagnostics - Providers of Capital - ROC Curve ... 78

Figure 2: Diagnostics - Deal Specifics - ROC Curve ... 81

Figure 3: Diagnostics - Industry & Region - ROC Curve ... 85

Figure 4: Diagnostics - Capital Markets - ROC Curve ... 89

(9)

5 | P a g e

1. Introduction

The last three decades have seen the Private Equity Industry, in Europe and the US, grow at a rapid rate (both in terms of deal size and assets under management) (Cornelius 2011). Given the industry’s rapid growth and potential offering of exceptional returns, the Private Equity Industry has attracted a great amount of attention from investors and academics. As the Private Equity industry has grown, the Secondary Buyout method of exit has become more frequent.

A Secondary Buy-Out (SBO) is a method of exit where a private company is sold from one private equity fund to another, or a group (syndicate) of private equity funds (Cannon 2007). Technically, a SBO is only “secondary” the first time a private company is sold from one private equity fund to another. For the purpose of this proposal and the following thesis, we will use the term SBO to refer to all exits which involve the sale of a private company from one private equity fund to another, therefore including tertiary buy-outs and so on. Furthermore, it has be shown that deals are more likely to be exited through SBO if they were entered into through an SBO (F. J. Degeorge 2013).

The primary objective of this thesis is to determine which factors cause certain private companies to remain in the SBO Sphere, i.e. which factors cause the persistence of Secondary Buy-Outs. The primary focus will be on capital market factors specific to the Private Equity Industry, industry-specific factors of the private companies, company-specific factors of the private companies, deal-specific factors and acquisition strategies of the target companies.

(10)

6 | P a g e Favourable conditions in the capital market, equity and debt markets, may increase the likelihood of SBO persistence. If the capital markets as a whole are favourable, and there is more money across all asset classes, the phenomenon of “too much money, chasing too few deals” could occur (Gompers 2000) and result in Private Equity Funds recycling deals, and thus cause an increase in SBO exits. A high-yield index and the Federal Tightening Index will be used to measure conditions in the debt markets. The number of IPOs per year will be used as a proxy for the condition of the equity markets. This, along with other factors, will be investigated in the following thesis.

Factors specific to the private companies, and industries they operate in may affect the likelihood of SBO persistence of these companies. Certain companies, or industries, may possess certain characteristics that require them to be in private equity hands, in order to be successful. An example of such a characteristic is the industry concentration, i.e. competitiveness of the industry, that rules out a merger and acquisition exit due to anti-trust regulation in a concentrated industry (Justice n.d.), or IPO if the company had to comply with the reporting requirements of an exchange, would be exposed, and thus the company would lose its advantage. Another example of such a characteristic is a case where a company has large free cash flows and is more efficient when highly levered and has the oversight that comes with private equity ownership (Jensen 1989). If this category of companies has a requirement of being held by private equity, this could explain, to a large degree, the persistence of SBOs. This, along with other factors, will be investigated in the following thesis.

(11)

7 | P a g e In order to test the above hypotheses data were collect on private equity target companies. As the name suggests, private equity data is difficult to retrieve given the private nature of the investment. Dr. Jens Martin kindly provided me with a list of private equity deals, from which we were able to construct a list of private equity target companies.

Data on private equity deals and acquisitions of the target companies of these deals was collect from Pitch Book (PitchBook n.d.), and when not available in Pitch Book, from Zephyr (Dijk 2014). Capital market data were collected from the Federal Reserve Board Database (Fed Tightening Index), (Board of Governors of the Federal Reserve System n.d.), University of Florida (Number of IPOs per Year) (Ritter 2014), and Federal Reserve Bank of St. Louis (High-Yield Index) (Lynch n.d.). Finally, data related to the industries which the target firms operate in was collected from Wharton Research Data Services (The Wharton School, University of Pennsylvania n.d.).

Data on 1342 deals, involving 318 companies from North America, Europe (Northern, Southern, and Western), and Australia (Rest of World) was individually collected, and merged to form a single sample. Additionally, data on the acquisitions of the target companies during the life of the deal was collected. Data on 533 acquisitions made by 150 target companies, was collected and merged.

The statistical model selected to build the explanatory models in this was the logit model. The logit model was selected due to the fact that the independent variable (SBO Persistent Deal) is a “dummy” variable (0 or 1), and if a traditional OLS regression was run, the outputs would not be restricted to the [0;1] interval.

The remainder of this thesis is structured as follows: Section 2 discusses the relevant literature. Section 3 provides details on the primary objectives (hypotheses) and methodology of the analysis. Section 4 provides a description of each variable used, along with pertinent descriptive statistics of these variables. Section 5 reports the findings of the empirical research. Section 6 discusses the statistical fit of the empirical results. Lastly, Section 7 concludes.

(12)

8 | P a g e

2. Literature Review

While there exists an extensive literature on Private Equity, Secondary Buyout returns, and factors that lead to changes in the prevalence of secondary buyouts, there remains little academic analysis on the persistence of Secondary Buyouts. Therefore, the focus of the literature research is on the private equity as a whole, with specific attention to determinants of secondary buyouts.

Degeorge, Martin, and Phalippou in their paper “Is the rise of secondary buyouts good news for investors” show the existence of the persistence of Secondary Buyouts. They show that SBOs are more likely to be exited through a secondary buyout, than a Primary Buyout. In fact, the difference in the likelihood of a secondary buyout exiting through a secondary buyout, versus a Primary Buyout exiting through a Secondary Buyout, is substantial, 43% to 20% (F. J. Degeorge 2013). Furthermore, they show that secondary buyouts are less likely than Primary Buyouts to be exited through an IPO. This leads to the potential point of entry for an investigation into the determinants of secondary buyouts. Persistence, as it may suggest that public equity market conditions affect the likelihood that a Secondary Buyout is exited through a secondary buyout.

(13)

9 | P a g e Han T.J. Smit and Vadym Volosovych developed a model that showed that SBO waves emerge as financial sponsors react to past returns and flock into the private equity market, through funding private equity funds. They also show that investment volumes and returns increase with higher leverage and lower market interest rates (Smit 2013). This suggests that when debt markets are favourable, and private equity has shown good returns, the number of Secondary Buyouts increase. This does not, however, indicate the SBOs become more prevalent during these waves. It could, in fact, be the case that all private equity deals are increasing and that the proportion of Secondary Buyouts to other deals remain close to constant. Additionally, this does not answer the question of whether or not SBO persistence increases during these waves. This does, however, provide a good entry point for this thesis to build on.

Robinson, and Berk in their paper “Cyclicality, performance measurement, and cash flow liquidity in private equity.” show valuation waves in public capital markets co-move with waves in private equity markets. The traditional hot-market underperformance states that private equity raised during booms tend to perform significantly worse than those raised during bust years. They find that in a relative performance assessment, using plausible estimates of beta, this hot-market underperformance no longer exists. They conclude that this is a consequence of the interconnectedness of private and public capital markets (Robinson 2011). When the public market is illiquid, the exit opportunities in the private market disappear, this is coupled with low performance in the public market. This research provides a point at which to test if market conditions affect the likelihood of SBO persistence. It may be the SBO persistence occurs due to a lack of alternative options, such as IPO.

(14)

10 | P a g e Kitzmann and Schiereck in their paper “A Note on Secondary Buyouts-Creating Value or Recycling Capital” analyse whether SBOs are a suitable alternative exit strategies to IPOs or trade sales. They argue that SBOs should not be viewed as a second best alternative for recycling the private equity investor’s capital in cases where alternative exit channels are not available. Furthermore, they show that profitability of SBOs are not significantly different from that of trade sales. Additionally, they argue that SBOs add value from different sources, such as the reduction in agency costs, and increased liquidity to the market (Kitzmann 2009). These findings are not directly applicable to this thesis, because the finding of liquidity to the market, as an advantage, is not necessarily a decision that would be made at fund level and thus may not determine SBO persistence. However, there may be value in the point of agency cost reduction, for this thesis. This is a point that is reiterated across many articles, and thus will be investigated in this thesis.

Cressy, Munari, and Malipiero. Found the that private equity funds that specialise have a competitive advantage in unlocking value of private companies in the industry they specialise in, or that are at the stage in their development that the fund specialises in (Cressy 2007). This has interesting implications for a study on SBO persistence as it allows for the possibility that the first private equity company that owned the portfolio company, may not have unlocked all the value, and that SBOs can still make financial sense if the purchaser is either more specialised, or is specialised in the phase the portfolio company has grown to. This will be analysed from both perspectives, industry and stage specialisation, in the thesis.

(15)

11 | P a g e Tim Jenkinson and Miguel Sousa found that exit choice is heavily influenced by the capital market conditions (debt and equity markets). Furthermore, they find that the exit route that maximises value varies with capital market conditions, and that private equity funds take advantage of these “windows of opportunity” (Sousa 2010). If capital market conditions influence the exit choice, i.e. is a determinant of Secondary Buyout occurrence, then it is plausible that it could affect the persistence of Secondary Buyout persistence. This provides a potential point of analysis of Secondary Buyout persistence, which will be analysed in this thesis.

In the same paper, Tim Jenkinson, and Miguel Sousa state that given the requirement of private equity firms’ stakes in an IPO exit to be locked up for at least six months, and the difficulty in selling a significant stake on the market, IPO exits do not result in quick, or certain proceeds for private equity funds (Sousa 2010). This fact coupled with the fact that private equity funds have a limited life, the compensation nature of the private equity funds, may result in private equity investors choosing a secondary buy-out route to avoid the disadvantages of an IPO exit. This may be more prevalent when the fund which holds the target company is reaching the end of its “life,” resulting in the private equity investor requiring a quicker method of liquidating its investment in the target company. The probability of this effect is likely to increase as the number of private equity funds on the sell side increases due to the fact there is more chance that one of the private equity funds is close to the end of its “life”. This provides a potential point of analysis of Secondary Buyout persistence, which will be analysed in this thesis.

(16)

12 | P a g e SBO waves emerge as financial sponsors react to past returns and flock into the private equity market, through funding private equity funds. They also show that investment volumes and returns increase with higher leverage and lower market interest rates (Smit 2013). This suggests that when debt markets are favourable, and private equity has shown good returns, the number of Secondary Buyouts increase. This does not, however, indicate the SBOs become more prevalent during these waves. It could, in fact, be the case that all private equity deals are increasing and that the proportion of Secondary Buyouts to other deals remain close to constant. Additionally, this does not answer the question of whether or not SBO persistence increases during these waves. This does, however, provide a good entry point for this thesis to build on.

While there is little academic research on SBO persistence, there is sufficient research on related topics that will allow for the extension of those research findings to the analysis of SBO persistence.

(17)

13 | P a g e

3. Methodology & Hypothesis

3.1 Primary objective:

The primary objective of this thesis is to determine which factors cause certain private companies to remain in the SBO Sphere, i.e. which factors cause the persistence of Secondary Buy-Outs. The primary focus is on capital market factors specific to the Private Equity Industry, industry-specific factors of the private companies, company-specific factors of the private companies, deal-specific factors and acquisition strategies of the target companies. The following hypotheses are tested under the primary objective:

Hypothesis 1: Favourable conditions in the capital market, equity and debt markets, increase the likelihood of SBO persistence.

Hypothesis 2: Factors specific to the private companies, and industries they operate in, affect the likelihood of SBO persistence of these companies.

Hypothesis 3: Factors specific to the deal, and subsequent acquisitions of the target companies, affect the likelihood of SBO persistence in these companies.

(18)

14 | P a g e

3.2 Methodology:

The following section will outline the method and source of data collection, and subsequent manipulation.

3.2.1 Private Equity Deal Data Retrieval

The first step in the data collection process was to collect data on private equity target companies. As the name suggests, private equity data is difficult to retrieve given the private nature of the investment. Dr. Jens Martin kindly provided me with a list of private equity deals, from which we were able to construct a list of private equity target companies.

This list was used to collect data on all the deals involving all the companies on the list. This data were collected from Pitch Book (PitchBook n.d.), and Zephyr (Dijk 2014) where not available in Pitch Book. Data on 1342 deals, involving 318 companies from North America, Europe (Northern, Southern, and Western), and Australia (Rest of World) was individually collected, and merged to form a single sample. Secondary Buyout Persistence deals, defined as secondary buyout in companies that have previously been exited through at least one secondary buyout, accounted for 178 of the total deals in the sample used in the estimation.

Additionally, data on the acquisitions of the target companies during the life of the deal was collected. Data on 533 acquisitions made by 150 target companies, was collected and merged to form a single sample. As with the target company deal information, this was collected from Pitch Book (PitchBook n.d.), and Zephyr (Dijk 2014).

Data on the ranking of the countries which the target companies have headquarters in was collected from the GDP ranking data from the World Bank (The World Bank n.d.).

The majority of the data used originated from the above two data sets. These two samples were merged to form one sample from which to run regressions, this process is explained in the following sections.

(19)

15 | P a g e 3.2.2 Industry Data Retrieval

The second step in the data collection process was to collect data on the industries the target companies belonged to. Given the fact that the deal data spanned the period from 1980 to 2014; the industry data had to begin earlier in order to calculate long-term (10 years) growth rates within the industries. Therefore, the time period of 1970 to 2014 was selected.

The next sub-step was to collect data on all companies within each industry, for which data exists. This was done by taking the Standard Industry Classification (SIC) code of each target company, which was available within the private equity deal data.

Unfortunately, the databases which were used (CRSP & Compustat) (The Wharton School, University of Pennsylvania n.d.) did not contain sufficient information in either the SIC or NAICS codes to build the required industry database. Therefore, the SIC codes were then converted to North American Industry Classification System (NAICS) codes. Both lists of industries were used to obtain the ticker and company information from the merged database. In order to obtain the final list of companies within each industry over the 44-year time period, duplicates in the lists were removed.

This list of companies over time was then used to obtain stock returns, stock prices, and number of shares outstanding over the 44 year time period, measured monthly. The stock price and share data were used to calculate market capitalisation data for each company, which was then used to calculate industry capitalisation at each data. This was then used together to calculate a Herfindahl-Hirschman Index (HHI) (Justice n.d.) as follows:

𝐻𝐻𝐼𝑚𝑡= 𝑆12+ 𝑆22+ 𝑆32+ ⋯ 𝑆𝑛2

𝑊ℎ𝑒𝑟𝑒: 𝐻𝐻𝐼𝑚𝑡= 𝑡ℎ𝑒 Herfindahl − Hirschman Index of industry 𝑚 at time 𝑡 𝑎𝑛𝑑 𝑆𝑛2= 𝑡ℎ𝑒 𝑚𝑎𝑟𝑘𝑒𝑡 𝑠ℎ𝑎𝑟𝑒 𝑜𝑓 𝑡ℎ𝑒 𝑖𝑡ℎ 𝑠ℎ𝑎𝑟𝑒

(20)

16 | P a g e The HHI is a measure market concentration, and therefore competitiveness. The US Department of Justice uses the HHI to determine the competitiveness of the industry when analysing a merger case (Justice n.d.). They categories industries as unconcentrated (highly competitive), moderately concentrated (moderately competitive), and highly concentrated (low competition), as outlined in the table below:

Table 1: HHI Index Description

HHI Value Market Competitiveness

0 – 1 500 Highly Competitive – Perfect Competition

1 500 – 2 500 Moderately Competitive

2 500 – 10 000 Low Competition - Monopoly

The HHI Index values for each month end were then matched against the month in which each deal occurred. In cases where there was no HHI Index data for a given deal data, the yearly average HHI Index value for the year in which the deal occurred, was used. In the case where neither was available, the data were left blank, in order to not be included in subsequent regressions.

(21)

17 | P a g e 3.2.3 Capital Markets Data Retrieval

The last step in the data collection process was to collect data on the capital market conditions over the time period of the sample collected from Pitch Book. The sample time period ran from 1980 to 2014 (over 34 years). The data collected focused on US capital markets for two reasons.

Firstly, a large portion of the sample is based in the US, and thus, the US capital markets are the most appropriate. Secondly, given the increasing interconnectedness of the global financial markets, and the ability for large financial institutions (i.e. private equity firms) to access global financial markets, the US market would act as a good proxy for global developed market conditions. This effect will be strong as time increases (given increased economic linkages over time), and thus, the proxy will be best-suited to more current time period. However, the proxy should still suffice from the start of the sample, 1980.

Three variables were collected, Fed Tightening Index (Board of Governors of the Federal Reserve System n.d.), High-Yield Index (Lynch n.d.), and the number of IPOs in each year in the US (Ritter 2014).

This data were then used by matching the month end values, in the case of High-Yield, yearly values, in the case of number of IPOs and quarterly values in the case of the Fed Tightening Index. The value of the Fed Tightening index is reported after the fact, and therefore the end of the quarter tightening index value defined the tightening index value for the month since the previous quarter to the end of the quarter.

(22)

18 | P a g e 3.2.4 Combined Database Construction

Once the individual sections of data were collected, we proceeded by combining all the data sources into one sample. The base sample is the list of deals obtained from Pitch Book. This list was then used to match with the data on acquisitions of the target companies, IPO frequencies, debt market conditions data, etc.

After the data merging process was complete, the next step was to convert the data into quantitative data, where applicable. This was a process of creating dummy variables for a selection of variables, creating ranked indicators for the target companies headquarters, etc.

Before the data could be used in the model building process, two variables needed to be winsorised in order to reduce the change that outliers in these data sets would have driven the regression results. Both Deal Value and Number of Years Held were winsorised.

3.2.5 Combined Database Analysis Methodology

In order to analyse the hypotheses separate regressions were run. All the regressions had a dummy variable, “SBO Persistence Deal,” as the dependent variable. “SBO Persistence Deal will be equal to 1 when the deal is exited through a Secondary Buyout, if the target company had been exited through a Secondary Buyout before. This allowed for the coefficients of the dependent variables to be interpreted as effects on the likelihood of SBO persistence. The specifics of explanatory variables used are outlined in the following section (4.1 Variable Description). These variables are divided into 7 models:

1. Capital Providers 2. Capital Markets

3. Industry & Region Specifics 4. History of Alternative Exit Failure 5. Deal Specifics

(23)

19 | P a g e The statistical model selected to build the explanatory models in this was the logit model. The logit model was selected due to the fact that the independent variable (SBO Persistent Deal) is a “dummy” variable (0 or 1), as if a traditional OLS regression was run, the outputs would not be restricted to the [0;1] interval (Caudill 1988).

Each model was run in an iterative process until only significant variables remained. Once the separate regressions were run, a “combined” regression of the determinants of SBO persistence was run. The model was initially comprised of the significant variables from each of the individual models. Like the other models, this was run in an iterative process until only significant variables remained.

Once the regressions were run, diagnostic tests were done to ensure there is no multicollinearity and that the model fits the data well. These were measured through a correlation table and a ROC Curve and Chi-Squared Test for Goodness of Fit respectively. Interpretations of marginal effects were only conducted if the model passed the diagnostic tests.

(24)

20 | P a g e

4. Data & Descriptive Statistics

The following section begins by describing each variable used in the models to be reported later in this paper. It concludes by discussing the significant descriptive statistics of each variable’s descriptive statistics tables. The descriptive statistics tables (Tables 1 – 18) divide the descriptive statistics by whether or not the deal was considered to be an SBO persistence deal (Secondary Buyout after previous Secondary Buyout); the time period in which the deal occurred; the region of the target company; and the country of the target company.

The variables which have been selected for analysis have been selected based on either plausibility (they could plausibly explain Secondary Buy-out persistence) or based on previous research on the determinants of Secondary Buy-outs and other related literature.

(25)

21 | P a g e

4.1 Variable Description:

“SBO Persistent Deal” is a measured as 1 if the deal type is an SBO and the target company was previously involved in an SBO. This is the dependent variable in all models and is the focus of this thesis.

4.1.1 Providers of Capital:

The variable "Club Deal Investors" is measured as 1 if there was more than 1 buyer in the deal, and 0 otherwise. This variable is included in the analysis to determine if an SBO persistent deal is more likely to occur if there is more than 1 private equity fund involved in the buy side of the deal.

The variable “Club Deal Sellers” is measured as 1 if there was more than 1 seller in the deal, and 0 otherwise. A related variable "Number Sellers” is measured as the number of different agents on the sell side of the deal. This variable was included in the model to determine if an SBO persistent deal is more likely to occur if there is more than 1 (or increasing numbers of) agents involved in the sell side of the deal. The theoretical background for this is somewhat diluted. Given the requirement of private equity firms’ stakes in an IPO exit to be locked up for at least six months and the difficulty in selling a significant stake on the market, IPO exits do not result in quick, or certain proceeds for private equity funds (Sousa 2010). This fact coupled with the fact that private equity funds have a limited life and the nature of compensation of the private equity funds may result in private equity investors choosing a secondary buy-out route to avoid the disadvantages of an IPO exit. This may be more prevalent when the fund which holds the target company is reaching the end of its “life,” resulting in the private equity investor requiring a quicker method of liquidating its investment in the target company. The probability of this effect is likely to increase as the number of private equity funds on the sell side increases due to the fact there is more chance that one of the private equity funds is close to the end of its “life”. This provides a potential point of analysis of Secondary Buyout persistence, and hence is included in the analysis.

(26)

22 | P a g e The variable "Club Deal Lender" is measured as 1 if there was more than 1 lender (syndicate) in the deal, and 0 otherwise. This variable is included in the analysis to determine if an SBO persistent deal is more likely to occur if there is more than 1 loan provider (syndicate) involved in the debt side of the deal.

4.1.2 Capital Markets:

The variable "Fed Tightening Index" is measured as the number of banks to have reported tightening (positive) the standards on loans. A related variable "High Yield Spread" is measured by the spreads between a computed Option Adjusted Spreads Index in all bonds in the High-Yield rating and a spot Treasury curve. This variable is included in the model to measure the effects debt market conditions. Private equity investments exit choice is heavily influenced by the capital market conditions (debt and equity markets). Furthermore, the exit route that maximises value varies with capital market conditions, and private equity funds take advantage of these “windows of opportunity” (Sousa 2010). If capital market conditions influence the exit choice, i.e. is a determinant of Secondary Buyout occurrence, then it is plausible that it could affect the persistence of Secondary Buyout persistence. This provides a potential point of analysis of Secondary Buyout persistence and is, therefore, included in the model.

The variable "Number of IPOs" is measured the number of IPOs during a given year in the US. This variable is included in the model to measure the effects of pro-cyclicality of the private equity market and the public capital markets. When the public market is illiquid, the exit opportunities in the private market disappear, this is coupled with low performance in the public market. (Robinson 2011). This provides a point at which to test if market conditions affect the likelihood of SBO persistence. It may be the SBO persistence occurs due to a lack of alternative options, such as IPO.

(27)

23 | P a g e 4.1.3 Industry & Region Specifics:

The variable “Herfindahl-Hirschman Index” is measured as the index value during the month (or yearly average where monthly data is not available). The HHI index is included in the model to determine if industry concentration (competitiveness) determines Secondary Buyout persistence. The logic behind its inclusion is twofold.

Firstly, if the industry is highly concentrated, there is little competition and thus little need to invest in new projects. This could lead to increased agency cost of public ownership as management has excessive free cash flow to shield their poor decisions (Jensen 1989). This would reduce the chances of an IPO exit.

Secondly, if the industry is highly concentrated, there is little competition and thus a merger will be unlikely to be allowed by the lawmakers, given anti-trust laws (Justice n.d.). This would reduce the chances of an M&A exit.

By process of elimination, a reduction in the chances of IPO exit and M&A exit, increase the changes of an SBO exit, and thus potentially secondary buyout persistence.

(28)

24 | P a g e The variable “Country Code Rank” is measured from 0 to 15 as a ranking (ascending) of the country by GDP:

Table 2: Country Code Rank Description

It is included in the model to determine if target companies from certain countries are more likely to remain in the SBO sphere. The additional measurement of this variable is if the GDP (The World Bank n.d.) (Size of the economy) rank determines Secondary Buyout persistence. “Region Rank” is a relative variable that follows the same reasoning for inclusion, and construction.

Company Country Value GDP Ranking

Luxembourg 0 73 Finland 1 42 Denmark 2 34 Belgium 3 26 Norway 4 25 Sweden 5 22 Switzerland 6 20 Netherlands 7 18 Spain 8 13 Australia 9 12 Canada 10 11 Italy 11 9 United Kingdom 12 6 France 13 5 Germany 14 4 United States 15 1

(29)

25 | P a g e 4.1.4 History of Alternative Exit Failure:

The variable "M&A Failure" is measured as the cumulative number of times the target company has had a failed M&A prior to the current deal. A related variable "IPO Failure" is measured as the cumulative number of times the target company has had a failed IPO prior to the current deal. The reason for inclusion is similar to that of the HHI Index. Failure in either may be due to factors specific to the company that makes it unable to exit through either exit routes. Additionally, reputational loss incurred during a failed IPO or M&A may result in the likelihood of a company trying to exit through an IPO or a M&A to decrease. By process of elimination, a reduction in the chances of IPO exit and M&A exit, increase the changes of an SBO exit, and thus potentially secondary buyout persistence.

4.1.5 Deal Specifics:

The variable “Deal Size” - measured in millions of $ paid for the target company. This variable was included in the model to determine if the size of the deal has an impact on whether or not the deal is an SBO persistence deal. This could be the case by default as the earliest an SBO persistent deal can occur, is in the 3rd_{deal of the target company. The 3}rd_{deal has a very high chance of having a higher}

deal value than the first 2 deals, and would, therefore, show a relationship between deal size and whether or not a deal was a SBO persistence deal or not. However, the sample includes many types of exits at all stages of a target company’s life. Therefore, this default relationship is unlikely to occur in the sample given its size and representativeness.

The variable "Years Held" is measured as the number of years from the date of the transaction, till the date of the next deal involving the target company. This variable is included in the model to determine if the number of years the target company is held, determines whether or not the deal is an SBO persistent deal or not. It is included to determine if the deal life of an SBO persistent deal or not persistent, differ from one another.

(30)

26 | P a g e 4.1.6 Acquisition Behaviour of Target Companies:

The variable "Investment - Same Country" is measured as the number of acquisitions in which the target company and the acquired firm are from the same country, during the time the target company was held in a given deal. This variable will be used as a proxy for industry consolidation or vertical integration strategies. This will allow for the testing of specialisation differences between SBO Persistent deals and not, with respect to vertical integration and industry consolidation specialisations.

The variable "Investment - Not Same Country” is measured as the number of acquisitions in which the target company and the acquired firm are from a different country than the target company, during the time the target company was held in a given deal. This variable will be used as a proxy for global expansion / new markets strategies. This will allow for the testing of specialisation differences between SBO Persistent deals and not, with respect to global expansion / New markets specialisations.

The variable "Investment - Value" is measured as the total value of acquisitions of the target company, during the time the target company was held in a given deal. This variable is included to act as a proxy for a buy-and-build specialisation, in which the private equity investor purchases the target company and acquires one or more add-on companies through the target company. This will allow for the testing of specialisation differences between SBO Persistent deals and not, with respect to buy and build specialisation.

(31)

27 | P a g e

4.2 Descriptive Statistics:

Table 18 in the Appendix contains descriptive statistics for "SBO Persistence Deal" - measured 1 if the deal type is a SBO and the target company had previously been sold through an SBO. The table (along with the other descriptive statistics tables) divides the descriptive statistics by whether or not the deal was considered to be an SBO persistence deal (SBO after previous SBO); the time period in which the deal occurred; the region of the target company; and the country of the target company.

All statistics seem to be relatively similar, apart from a clear trend of an increase in the proportion of SBO Persistent Deals, mean of 0 in the early 80s to a mean of 0.29 between 2010 and 2014. This could be due to the maturation of the Private Equity Industry as a whole. Unfortunately, this is not tested in this thesis, however, it presents a good opportunity for further research.

(32)

28 | P a g e 4.2.1 Providers of Capital:

Table 3 in the Appendix contains descriptive statistics for the variable "Club Deal Investors" which is measured as 1 if there was more than 1 buyer in the deal and 0 otherwise.. All statistics seem to be relatively similar.

Table 4 in the Appendix contains descriptive statistics for the variable "Club Deal Sellers" is measured as 1 if there was more than 1 seller in the deal and 0 otherwise. All statistics seem to be relatively similar.

Table 5 in the Appendix contains descriptive statistics for the variable "Club Deal Lender" which is measured as 1 if there was more than 1 lender (syndicate) in the deal, and 0 otherwise. SBO Persistent deals are much more like to have multiple lenders on the deal than none SBO Persistent deals. This could be due to deal size effects (the larger the deal the more likely it is that one lender will not be willing to shoulder the risk of the loan by itself). This is accounted for later in the model building section of this thesis, and it is, in fact, the size of the deal that is driving this.

Table 6 in the Appendix contains descriptive statistics for the variable "Number Sellers" which is measured as the number of different PE firms on the sell side of the deal. All statistics seem to be relatively similar. This is extremely surprising, as there is a certain relationship between SBO Persistence and number of Sellers. It may be that a small difference in mean (SBO Persistence – 1.55 V.s 1.52 – none SBO Persistence) and Std. Deviation (SBO Persistence – 0.95 V.s 0.87– none SBO Persistence), result in a confidence interval that is significantly different enough to show a relationship. However, this is not required as the relationship is confirmed in the model building later in this thesis.

(33)

29 | P a g e 4.2.2 Capital Markets:

Table 7 in the Appendix contains descriptive statistics for the variable "Fed Tightening Index" which is measured as the number of banks to have reported tightening (positive) the standards on loans. All statistics seem to be relatively similar across Region and Country criterion. There is a significant difference between the average for SBO Persistence Deals (-1.42%) and the average for none SBO Persistence Deals (3.89%). This makes intuitive sense and agrees with the literature. The number of SBO persistence deals will increase when standards on loans are loosened (favourable debt market), do to cheaper debt, and thus cheaper leverage. The increase in none SBO Persistence Deals during times of tightening is a result of private equity markets being “tougher” and thus an equity-based exit would be more attractive. This fits with the “windows of opportunities hypothesis” (Sousa 2010)

Table 8 in the Appendix contains descriptive statistics for the variable "High Yield Spread" which is measured by the spreads between a computed Option Adjusted Spreads Index in all bonds in the High-Yield rating and a spot Treasury curve. All statistics seem to be relatively similar across SBO Persistence, Region, and Country criterion. However, there appears to be a trend of increasing average High-Yield Spreads from 3.48% in the first half of the 1980s to 5.2% between 2010 and 2014.

Table 9 in the Appendix contains descriptive statistics for the variable "Number of IPOs" which is measured as the number of IPOs during a given year in the US. All statistics seem to be relatively similar.

(34)

30 | P a g e 4.2.3 Industry & Region Specifics:

Table 10 in the Appendix contains descriptive statistics for the variable “Herfindahl-Hirschman Index” which is measured as the index value during the month (or yearly average where monthly data is not available). All statistics seem to be relatively similar. This is somewhat surprising as one would expect there to be a difference. Specifically, one would expect the SBO persistence deals to have a higher HHI Index value given the advantage the private equity ownership model has in such industries.

This, unfortunately, leads to two possible conclusions, either there is an issue with the theory or there is an issue with the data. It is most certainly not the former. The data were collected using best practices data collection methods, the collection process is faultless. The collection strategy, however may be at fault. Using SIC and NAICs to define industries might be too drilled down, which results in thinner data, and thus less representative. A better method to have built the HHI Index value would have been to start with Farm-French Industries as there are fewer of them and are listed in most databases.

4.2.4 History of Alternative Exit Failure:

Table 11 in the Appendix contains descriptive statistics for the variable "M&A Failure" which is measured as the cumulative number of times the target company has had a failed M&A. All statistics seem to be relatively similar.

Table 12 in the Appendix contains descriptive statistics for the variable "IPO Failure" which is measured as the cumulative number of times the target company has had a failed IPO. All statistics seem to be relatively similar. The only noticeable statistic that comes from this table is the spike in IPO average cumulative failure during the 2010-2014 time period 0.08 from 0.03.

(35)

31 | P a g e 4.2.7 Deal Specifics:

Table 13 in the Appendix contains descriptive statistics for the variable “Deal Size” which is measured in millions of $ paid for the target company. All the possible sample statistics (number, mean, median, standard deviation, etc.) are larger for SBO persistent deals. This implies that SBO persistent deals are larger or just priced higher, both of which are plausible. The increase in price (relative) could be a result of reduced agency costs (Kitzmann 2009). Alternatively, an increase in deal size could be simply a result of the fact that in order for a deal to be considered a SBO persistent deal the minimum number of previous deals it 2, where none SBO persistence deals do not have this restriction. This prove the difference between the two groups, simply by way of selection (assuming most targets increase in price as they are traded hands).

Table 14 in the Appendix contains descriptive statistics for the variable "Years Held" which is measured as the number of years from the date of the transaction, till the date of the next deal involving the target company. All the possible sample statistics (number, mean, median, standard deviation, etc.) are larger for none SBO persistent deals. The key statistic between the two is the mean, with SBO persistent deals lasting on average 1.94 years, and none SBO persistent deals lasting 2.54 years. There is a clear decreasing trend in recent years from and the average number of years held of 4.9 between 1980 and 84; to 1.79 between 2010 and 2014.

(36)

32 | P a g e 4.2.8 Acquisition Behaviour of Target Companies:

Table 15 in the Appendix contains descriptive statistics for the variable "Investment - Same Country" which is measured as the number of acquisitions in which the target company and the acquired firm are from the same country, during the time the target company was held in a given deal. The number of Non-SBO Persistence Deals that went on to acquire other firms in their same domicile (142) far outnumbers the number acquired by SBO Persistence Deals (37). Additionally, the target firms in U.S seem to make acquisitions (103) far more than any other country, with the United Kingdom in 2nd

place at 45 acquisitions.

Table 16 in the Appendix contains descriptive statistics for the variable "Investment - Not Same Country" which is measured as the number of acquisitions in which the target company and the acquired firm are not from the same country, during the time the target company was held in a given deal. There does not appear to be any significant/ noteworthy statistics in this table, apart from the extremely small sample size (20).

Table 17 in the Appendix contains descriptive statistics for the variable "Investment - Value" which measured as the total value of acquisitions of the target company, during the time the target company was held in a given deal. All the possible sample statistics (number, mean, median, standard deviation, etc.) are larger for none SBO persistent deals. There appears to be no pattern of change over the time periods, region, or country (outside of the normal differences in the sample).

(37)

33 | P a g e

4.3 Variable Categories Assigned to Hypotheses

The following section will provide a summary of the allocation of variable with respect to the 4 original hypotheses.

Hypothesis 1: Favourable conditions in the capital market, equity and debt markets, increase the likelihood of SBO persistence.

1. Capital Markets

a. “Fed Tightening Index" b. "High Yield Spread" c. "Number of IPOs"

Hypothesis 2: Factors specific to the private companies, and industries they operate in, affect the likelihood of SBO persistence of these companies.

1. Industry & Region Specifics

a. “Herfindahl-Hirschman Index” b. “Country Code Rank”

c. “Region Rank”

2. History of Alternative Exit Failure a. “Cum. Failed IPO” b. “Cum. Failed M&A”

(38)

34 | P a g e

Hypothesis 3: Factors specific to the deal and subsequent acquisitions of the target companies, affect the likelihood of SBO persistence in these companies.

7. Capital Providers

a. "Club Deal Investors" . b. “Club Deal Sellers” c. "Number Sellers” d. "Club Deal Lender" 8. Deal Specifics.

a. “Deal Size” b. “Years Held”

9. Acquisition Behaviour of Target Companies a. "Investment - Same Country" b. "Investment - Not Same Country” c. "Investment - Value"

Hypothesis 4: A combination of Hypothesis’ 1 to 3 affect the likelihood of SBO persistence. 1. Capital Markets

a. “Fed Tightening Index” 2. Industry & Region Specifics

a. “Country Code Rank” 3. Capital Providers

a. "Number Sellers” b. "Club Deal Lender" 4. Deal Specifics.

a. “Deal Size”

The following section will provide the results of the logit model construction of the above

models.

(39)

35 | P a g e

5. Results

Each variable subsection was tested in isolation, with the statistically significant variables from each final model, being promoted to be included in the final combined model.

5.1 Acquisition Behaviour of Target Companies:

Table 19 in the Appendix contains results from logit regressions run on the sample data with explanatory variables related to acquisition behaviour of the target companies. The dependent variable in all cases is SBO Persistence Deal, i.e. 1 if the deal is an SBO and the company had been through an SBO before.

The explanatory variables are Value of Investments (the value of the acquisitions the target company made during the life of the deal – years held), Company Investment Same Country (the number of acquisitions the target company made of other companies with headquarters in the same country, during the life of the deal), Company Investment Different Country (the number of acquisitions the target company made of other companies with headquarters in a different country, during the life of the deal).

The model was adjusted until all remaining variables were statistically significant. The result is that none of the variables showed a statistically significant impact on whether or not the deal will be an SBO persistent deal or not.

A no result such as this can be as important as a statistically significant positive result. This result implies that target companies investments values, and where these investments are located, do not impact SBO persistence. With a further dilution, one could infer that industry consolidation, global expansions, buy-and build and/or vertical integration strategies of private equity funds do not differ significantly between SBO persistent deals and other deals.

(40)

36 | P a g e

5.2 History of Alternative Exit Failure:

Table 20 in the Appendix contains results from logit regressions run on the sample data with explanatory variables related to the history of failed other (M&A & IPO) exit options.

The dependent variable in all cases is SBO Persistence Deal, i.e. 1 if the deal is an SBO and the company had been through an SBO before.

M&A failure was omitted from the model due to the fact that it is a perfect predictor when SBO persistence = 0. Since this would require the statistical software (Stata 13) to be able to deal with infinite values (which it is not able to do) it was omitted.

The result is that none of the variables showed a statistically significant impact on whether or not the deal will be an SBO persistent deal or not.

The explanatory variables are Cumulative IPO Failure (the number of times the target company had a failed IPO, before the current deal), and Cumulative M&A Failure (the number of times the target company was a failed merger or acquisition target, before the current deal).

Again, this no result can be as important as a statistically significant positive result. This result implies that target history of failure in either IPO or M&A does not predict if the deal will be an SBO persistent deal or not. With a further dilution, one could infer that reputational loss of a failed IPO or M&A does not SBO Persistence.

(41)

37 | P a g e

5.3 Capital Providers:

Table 21 in the Appendix contains results from logit regressions run on the sample data with explanatory variables relating to the providers of capital of the deal.

The dependent variable in all cases is SBO Persistence Deal.

The explanatory variables are Club Deal Investors (1 if there was more than 1 buyer on the deal), Club Deal Sellers (1 if there was more than 1 seller on the deal), Club Deal Lenders (1 if there was more than 1 lender on the deal), and Number of Sellers (number of agents involved on the sell side of the deal.

The models were adjusted until all remaining variables were statistically significant. The result is a model indicating the number of sellers and the whether or not there was more than 1 debt provider on the deal, have a statistically significant impact on whether or not the deal will be an SBO persistent deal or not.

The Pseudo R2 is relatively low, at 0.0482, however this should be interpreted with caution as it does not imply that the model only explains 4.82% of the variation in SBO persistence, as a traditional R2 would.

The tables in the following section assess the goodness of fit of the final model. The result is that of a poor fit, thus, the marginal effects will not be interpreted. However, the significant variables will be included in the final model as the only failure was goodness of fit.

It is no surprise that “Number of Sellers” and “Club Deal Sellers” were not both in the final model as they are a different measurement of the same underlying cause – the need for quicker liquidation than IPO offers. The “Club Deal Lenders” may only be significant due to the fact that deal size is not included in this model, and thus it is acting as a proxy for the size of the deal. This is intuitive as the larger the deal the more likely it is that one lender will not be willing to shoulder the risk of the loan by itself.

(42)

38 | P a g e

5.4 Deal Specifics:

Table 24 in the Appendix contains results from logit regressions run on the sample data with explanatory variables related to the deal specifics.

The dependent variable in all cases is SBO Persistence Deal.

The explanatory variables are Years Held (number of years the deal was held till sale), Deal Size (measured in millions of $).

Both explanatory variables were winsorised to remove the effects of outliers on the regression results (as indicated by (W)).

The result is a model indicating the Deal Size has a statistically significant impact on whether or not the deal will be an SBO persistent deal or not.

The Pseudo R2 is relatively low, at 0.0152, however this should be interpreted with caution as it does not imply that the model only explains 1.52% of the variation in SBO persistence, as a traditional R2 would. The tables in the following section assess the goodness of fit of the final model. The result is that of a good fit, thus, the marginal effects will be interpreted.

This could be the case by default as the earliest an SBO persistent deal can occur, is in the 3rd_{deal of}

the target company. The 3rd_{deal has a very high chance of having a higher deal value than the first 2}

deals, and would, therefore, show a relationship between deal size and whether or not a deal was a SBO persistence deal or not. However, the sample includes many types of exits at all stages of a target company’s life. Therefore, this default relationship is unlikely to occur in the sample given its size and representativeness.

Table 27 in the Appendix provide the average, and conditional marginal effects of the final model. The result indicates that for every $1mil increase in deal size, the probability of the deal being an SBO persistent deal increases by 0.0054%. The marginal effects at the mean (conditional marginal effects) and the average marginal effects are close to identical.

(43)

39 | P a g e

5.5 Industry & Region Specifics:

Table 28 in the Appendix contains results from logit regressions run on the sample data with combined explanatory variables related to the industry and region of the target companies.

The explanatory variables are Herfindahl-Hirschman Index (an index measuring the competitiveness of an industry, the higher the less competitive), Country Code Rank (an indicator variable of the country which the target company has a Head Quarters, ordered by the GDP of the country in ascending order), Regional Rank (an indicator variable of the region (North America, Europe, Rest of Word) which the target company has a Head Quarters).

The model was adjusted until all remaining variables were statistically significant. The result is a model indicating the Country Code Rank has a statistically significant impact on whether or not the deal will be an SBO persistent deal or not.

Table 31 in the Appendix provide the average, and conditional marginal effects of the final model. The result indicates that for every unit increase in the country indicator variable, the probability of the deal being an SBO persistent deal increases by 0.7193%. Given the fact that the country code ranks are such that they are increasing with respect to GDP of the country that the target company has a headquarters, this implies that the higher the relative GDP of the country (GDP rank), the higher the probability that a deal with be an SBO persistent deal.

However, this could be due to GDP acting as a proxy for company size and deal size, as the larger the GDP of a country, will result in a higher average deal value (ceteris paribus). Thus, this result should be interpreted with caution until it is included in the main model, at which time the relationship to deal size will become clearer.

(44)

40 | P a g e

5.6 Capital Markets:

Table 32 in the Appendix contains results from logit regressions run on the sample data with combined explanatory variables related to the capital markets.

The explanatory variables are Number of IPOs (the number of IPOs in the US during the year the deal occurred), High Yield Index (the spreads between a computed Option Adjusted Spreads Index in all bonds in the High-Yield rating, and a spot Treasury curve, during the month the deal occurred), FED Tightening Index (the proportion of banks to have reported tightening (positive) the standards on loans. The model was adjusted until all remaining variables were statistically significant.

The result is a model indicating the Fed Tightening Index has a statistically significant impact on whether or not the deal will be an SBO persistent deal or not. The Pseudo R2 is relatively low, at 0.0091, however this should be interpreted with caution as it does not imply that the model only explains 0.91% of the variation in SBO persistence, as a traditional R2 would.

The interpretation of this is relatively straight forward and echoes the findings of the findings of the Fed Tightening Index discussion in the descriptive statistics section. SBO Persistent deals are more likely to occur when the tightening index is low (negative) and less likely when it is higher (positive). This implies the debt market conditions affect the likelihood of SBO persistence.

The tables in the following section assess the goodness of fit of the final model. The result is that of a poor fit, thus, the marginal effects will not be interpreted. However, the significant variables will be retained for the final model.

(45)

41 | P a g e

5.7 Combined:

Table 35 in the Appendix contains results from logit regressions run on the sample data with combined explanatory variables from the above models.

The explanatory variables are Last Deal Number (how many deals the target company was involved in), Club Deal Lenders (1 if there was more than 1 lender on the deal), Number of Sellers (number of agents involved on the sell side of the deal), Country Code Rank (an ordered indicator variable to distinguish country by increasing GDP, FED Tightening Index (the proportion of banks to have reported tightening (positive) the standards on loans), and Deal Size (measured in millions of $).

The models were adjusted until all remaining variables were statistically significant. The result is a model indicating the number of sellers and the deal size have a statistically significant impact on whether or not the deal will be an SBO persistent deal or not.

Given the fact that “Club Deal Lenders” became highly insignificant when combined with deal size, along with some theory and common sense, we can conclude that club deal lenders was, in fact, acting as a proxy for deal size, not the other way around. The same applied for Country Code Rank as there should intuitively be a link between Country Code Rank and Deal Size, based on how Country Code Rank was constructed.

(46)

42 | P a g e An intriguing result is that the Fed Tightening Index was dropped from the model. It would be somewhat of a stretch to find a relation between the Fed Tightening Index and Deal Value, but one does, potentially exist. This potential arises from the fact that during times of favourable debt market conditions, which the Fed Tightening Index is a measure of, private equity investors have better access to capital (debt). This causes the “too much money, chasing too few deals” phenomenon (Gompers 2000), resulting in increased deal valuations and a resultant rise in the value of deals completed during the given time period. Therefore, it is reasonable to assume that the Fed Tightening Index was dropped from the model due to its relation to deal size.

The fact that “Number of Sellers” is significant in the final model provides a conclusion that while the theoretical background for this is somewhat diluted. Given the requirement of private equity firms’ stakes in an IPO exit to be locked up for at least six months and the difficulty in selling a significant stake on the market, IPO exits do not result in quick, or certain proceeds for private equity funds (Sousa 2010). This fact coupled with the fact that private equity funds have a limited life and the nature of compensation of the private equity funds may result in private equity investors choosing a secondary buy-out route to avoid the disadvantages of an IPO exit. This may be more prevalent when the fund which holds the target company is reaching the end of its “life,” resulting in the private equity investor requiring a quicker method of liquidating its investment in the target company. The probability of this effect is likely to increase as the number of private equity funds on the sell side increases due to the fact there is more chance that one of the private equity funds is close to the end of its “life”. Therefore, the statistical significance of this variable is theoretically sound.

(47)

43 | P a g e The fact that “Deal Size” is significant in the final model does not provide an easy interpretation. Based on a simple risk-reward framework, an increase in price (relative) (Deal Size) could be a result of reduced agency costs (Kitzmann 2009) that SBOs provide. Alternatively, the relationship could exist based on the interval which SBO Persistence can exist in the lifetime of a target company (from the 3rd_{deal onwards). Due the fact that the sample used is of a large size and should possess good}

representativeness, this alternative explanation should not hold.

The first interpretation of risk-reward, is an uncomfortable one to make, as it could easily interpreted from the other side, whether or not a deal is an SBO persistent deal, and therefore an SBO deal, affects the riskiness of the deal (Kitzmann 2009), and therefore the deal value. Therefore, Deal Size should be interpreted with caution.

Table 38 in the Appendix provide the average, and conditional marginal effects of the final model. The result indicates that for every additional seller in a deal, the probability of the deal being an SBO persistent deal increases by 3.49766%. Additionally, the results show that for every $million increase in deal size, the probability of the deal being an SBO persistent deal increases by 0.00403%. The marginal effects at the mean (conditional marginal effects) and the average marginal effects are close to identical.

The following section will assess the goodness of fit of the final model. The result is a positive outcome in all the diagnostic tests.

(48)

44 | P a g e

6. Robustness Checks

6.1 Providers of Capital

Table 22 & 23, and Figure 1 in the Appendix provide output for the diagnostic tests run on the final model of the Providers of Capital variables.

ROC curve in Figure 1 is a measure of the goodness of fit of the model. The area under the ROC curve is 0.6633, the closer to 1 the AUC value, the better the model fit. The Diagonal line represents the area under the curve of 0.5, i.e. even probability of SBO Persistence or Not. The model seems to fit well, at the least, better than no model (diagonal line).

Table 22 provides the correlation between each variable considered in the model. All correlations are low, with the highest at 0.1178. This suggests that multicollinearity is not an issue in this model.

Table 23 provides the results of a Pearson Chi-Squared test for goodness of fit of the model. The p-value (0.11%) is below a 5% significance p-value, and thus we are forced to reject the null hypothesis that the model fits the data well. Therefore, we conclude that the model does not fit the data well.

The strongest test (Pearson Chi-Squared test for goodness of fit) indicates that the model does not fit the data well. Therefore, we did not analyse the marginal effects (in the results section above). Additionally, we will not include the variables in the final model of the Deal Specifics section, in the final model.

An analysis of the determinants of secondary buy-Out persistence.