Private equity performance, performance persistence, and capital flows in emerging markets

(1)

UNIVERSITY OF AMSTERDAM

Private Equity Performance, Performance Persistence, and Capital

Flows in Emerging Markets

Walter Sarin (10272992) Supervisor: Dr. Jens Martin

MSc. Finance – Asset Management

(2)

2

Statement of Originality

This document is written by Student Walter Sarin who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

3

Abstract

This study analyzes the performance, performance persistence, and capital flows of private equity in emerging and frontier markets. Growth and Venture Capital funds have been the highest performers in emerging markets, while Real Estate funds have performed the worst. Asia and Africa focused funds have performed in line with benchmarks, while other geographies have underperformed. Fund managers based on the same continent as they are investing tend to underperform foreign managers. Funds investing in emerging markets do not exhibit performance persistence, but instead exhibit mean reversion. The main driver of capital flows in private equity in emerging markets is the size of the previous fund.

(4)

4

1. Introduction

As of June 2016, the private equity industry had assets under management of USD 2.49 trillion (Preqin Global Private Equity Report, 2017). As emerging and frontier markets continue to develop, private equity investors are increasing their allocations to these countries. MSCI classifies 25 countries as emerging, and 22 countries as frontier (MSCI, 2017). The Preqin Report on Private Equity in Emerging Markets (2016) found that between 2006 and 2016 USD 332 billion was raised by private equity

managers in these countries. As more capital flows into these regions, research regarding investments in emerging markets becomes more important. Research in this field can provide valuable information to investors investing or considering investing in these markets. Research is especially relevant now, as interest is growing and information is lacking. Due to the nature of emerging markets, a lack of data has so far slowed the production of reports about private equity in these regions. This paper is one of the first to attempt to analyze performance of private equity funds in emerging and frontier markets. While the results are based on a rather small sample size, this research can provide an indication of the state of private equity in emerging and frontier markets and can serve as a foundation for further research with more complete data. As more funds become realized the availability of data will increase and the quality of research will increase along with it.

Two important papers in the field of private equity performance were written by Kaplan and Schoar (2005) and Phalippou (2010). The research in this paper will be based on the research conducted by the three aforementioned authors. Both of these papers focused on private equity in developed countries, with a large majority of the data being US-based. Nonetheless, the methodologies can still be relevant to an analysis of emerging and frontier markets.

The dataset used for the current research comes from Preqin, a private equity fund database. Preqin has performance data on a total of 417 funds investing in emerging and frontier markets. As this is one of the first papers focusing on emerging markets, this data set has not been used before. However, samples of developed market private equity funds have been used in previous research regarding private equity performance.

This research looks at various parts of performance of private equity funds. Firstly, it provides a summary of historical performance in emerging and frontier markets. In addition to this, this paper analyzes which fund characteristics drive performance, and finds that fund size and location of the GP are two important drivers of fund performance.

(6)

6 Secondly, it analyzes whether funds exhibit performance persistence. Performance persistence relates to the performance of successive funds. It is a measure that looks at whether a fund can consistently outperform or underperform relative to a benchmark. The evidence in this research points towards funds in emerging and frontier markets not exhibiting performance persistence, but instead exhibiting mean reversion.

Thirdly, it analyzes what drives the capital inflows to these funds, and finds that the only significant driver of capital inflows is the size of the previous fund. Performance of a fund does not seem to have an effect on the amount of capital a general partner is able to raise for its subsequent fund.

Throughout the research, this paper studies whether domestic or foreign fund managers tend to perform better than the other in these markets. This research shows that domestic managers who invest in the same continent on which they are based tend to underperform foreign managers. However, when accounting for the performance of the past fund, there does not appear to be a difference between domestic and foreign managers.

These results suggest that private equity funds in emerging markets do not necessarily behave like the more well studied private equity funds in developed markets. The skill of General Partners (GP) does not seem to play as important of a role as in developed markets. The finding of mean reversion implies that funds that outperform one year will likely underperform the following year, and all fund performances will trend toward this baseline over time.

2. Literature Review

2.1 The state of private equity in emerging markets

As of June 2016, the private equity industry had assets under management of USD 2.49 trillion (Preqin, 2017a). This indicates that the assets under management have more than doubled since 2010, when private equity funds managed around USD 1 trillion (Metrick, Yasuda, 2010). A sentiment analysis conducted by the Preqin report found that 95% of investors believed their private equity portfolios had met or exceeded their performance expectations of the last 12 months. It also found that 48% of investors planned to increase their allocation to private equity, compared to only 6% who planned to decrease their allocation. Along with this positive sentiment, private equity partnerships disbursed a record total of USD 257 billion in H1 2016 (Preqin, 2017a).

(7)

7 The 2016 Preqin Report on Private Equity in Emerging Markets provides information about yearly

fundraising in emerging markets. According to this data, 2011 had the most fund closes of any year with 355. Since then, however, the number of funds closing each year has steadily declined to 199 funds in 2015. In spite of this, 2014 was the year with the highest total aggregate capital raised by private equity partnerships focused on emerging markets. In general, while the number of funds has been decreasing, the aggregate capital raised has been fluctuating over the past few years. The proportion of capital raised in emerging markets has also continually shifted towards being raised by managers not based in emerging markets. As of June 2016, out of all the capital raised in emerging markets, 67% was raised by non-emerging markets-based managers. Eight of the last ten largest private equity funds to close since 2013 that invest in emerging markets have been raised by GPs based outside of these regions. Emerging markets-based managers raised a total of USD 332 billion between 2006 and 2016 (Preqin, 2016). The Preqin Report on Private Equity in Emerging Markets also states that in terms of geography, emerging Asia draws the majority of capital. In 2015 it accounted for a total of 77% of all fundraising. The region experiencing the largest growth is Sub-Saharan Africa. Capital raised in Sub-Saharan Africa has increased from USD 1.6 billion in 2014 to USD 4.5 billion in 2015. This growth is spurred by the young population and growing middle class. Since 2008, 55% of all capital raised in emerging markets has been raised by venture capital and growth funds. This is in contrast with developed markets, where buyout funds dominate Additionally, the majority (52%) of emerging markets-based fund managers have only raised one previous fund. For North America-based managers this number is 42% and for

European-based it is 38%. Developed market-based investors consists mostly of pension funds, foundations, wealth managers, and endowments, while emerging markets-based investors tend to be banks, corporate investors, and government agencies (Preqin, 2016).

Performance of private equity funds in emerging markets has in general been rising since funds of vintage 2008. There is a large gap between the median net internal rate of return (IRR) and the top quartile boundary, indicating that a few select managers are greatly outperforming the average fund. (Preqin, 2016).

2.2 Private Equity Background

Metrick & Yasuda (2006) outline the structure of private equity funds. They state that almost all private equity funds are set up as limited partnerships. Private equity firms serve as the general partners (GPs), while institutional investors and wealthy individuals provide the capital as limited partners (LPs). A typical partnership lasts for 10 years, and a successful private equity firm will stay in business by raising

(8)

8 a new fund. This is typically done every three to five years. They go on to state that if a private equity fund performs well, investors will take it as a sign of the skill of the fund manager. This is especially the case when private equity funds can perform well over a series of funds, which is known as performance persistence. This research is based around analyzing the performance persistence or GP “skill” that develops between successive funds. Private equity has historically outperformed public markets. This outperformance can be seen as compensation for holding a 10-year illiquid investment. The illiquidity of private equity stems from the fact that there is not an active secondary market, investors have little control over how the capital is invested, and the long investment horizon. (Ljungqvist & Richardson, 2003).

2.3 The Challenges of Private Equity in Emerging Markets

Leeds et al. (2002) provide an overview of private equity in emerging markets. They show that emerging markets funds have historically underperformed. They cite three possible reasons for

underperformance: low standards of corporate governance, weakness of the local legal system in enforcing contracts, and the inability of domestic equity markets to offer exit opportunities through Initial Public Offerings (IPO).

Both financial and operating information provided to GPs and investors by firms in developing countries tend to lack accuracy, timeliness, and transparency when compared to firms in developed countries. This problem is especially prevalent in family-owned firms, which are a common business structure in the developing world. This leaves the prospective investor at the mercy of the entrepreneur in order to receive access to information necessary to make critical judgements about company performance and value.

Limited legal recourse is also an issue when investing in emerging markets, and it also serves to compound the corporate governance problems. Low shareholder protection and a lack of contract enforcement can make investing in private equity in emerging markets a challenge. Local entrepreneurs and GPs tend to be more adept in working with the local legal system, leaving foreigners at a distinct disadvantage.

Finally, dysfunctional capital markets are the third reason for the historical underperformance of private equity in emerging markets. The major obstacle in this case being the lack of exit opportunities for investments. Exit opportunities for private equity in general consist of IPO, management buyouts, and

(9)

9 sales to strategic investors or later-stage financial investors. Due to the lack of well-functioning IPO markets in many developing countries, IPO exit opportunities are limited.

2.4 Private Equity Performance

Harris, Jenkinson, Kaplan & Stucke (2014) provide an in depth study of the performance of private equity funds in the United States using a data set spanning 1984 to 2004. They find that US buyout funds outperformed the S&P 500 through the 1980s, 1990s, and 2000s, while Venture Capital (VC) funds outperformed the benchmark in the 1980s and 1990s, and underperformed since. In fact, Higson & Stucke (2012) find that when looking at US buyout funds with vintage years between 1980 and 2008, these funds outperformed the S&P 500 by over 500 basis points per annum. Harris et al. (2014) also find that performance persistence for buyout funds in the 2000s fell considerably, while it remained stable for VC funds. They hypothesize that the greater performance persistence for VC funds compared to buyout funds can suggest that GP skills and network are harder to replicate for VC funds than buyout funds. Out of entire US private equity fund population around 60% of funds outperform the S&P 500, indicating that there is cross sectional variation in fund performance. The average fund performs much better than the median fund, indicating that excess returns are driven by positive outliers. In fact, excess returns are driven by top-decile rather than top-quartile funds (Higson & Stucke, 2012). Finally, Harris et al. (2014) examine whether there exists a link between fund performance and capital inflows into private equity funds. They find that both fund performance in absolute terms and fund performance relative to public markets are negatively correlated with aggregate capital commitments to private equity funds.

Robinson and Sensoy (2011) expand on the link between public and private equity markets. They found that funds that were raised during hot markets underperformed in absolute terms, however they did not underperform with respect to a public market equivalent. This conclusion is in line with Kaplan and Schoar (2005), who suggest a boom and bust cycle in private equity fundraising. Positive market-adjusted returns increase entry, which then in turn leads to negative market-market-adjusted returns, thus forming a cycle. Capital calls and distributions were both larger and more likely to occur when public equity valuations rise (Robinson & Sensoy, 2011). They found that distributions were more sensitive than calls, meaning that net cash flows are procyclical. They conclude that this causes private equity funds to be liquidity providers (sinks) when valuations are high (low). Building on these findings,

Robinson and Sensoy explain that market conditions can also affect contractual terms. GP compensation increases and shifts towards fixed components during fundraising booms.

(10)

10 Kaplan and Schoar (2005) also examine the performance of private equity funds, while placing an

emphasis on performance persistence. They find that over the entire sample period of 1980-1997, average fund returns net of fees were close to equal to those of the S&P 500. They find that buyout funds return slightly less than the public market. They also report that VC funds underperform the public market slightly when looking on an equal-weight basis, but outperform on a capital-weighted basis. Kaplan and Schoar find that returns persist strongly across funds raised by the same private equity partnership. They find that returns persist more strongly for VC funds than Buyout funds. Finally, they state that performance increases in the cross-section with fund size and GP experience. Kaplan and Schoar also study the capital inflows of private equity partnerships. They focus on the link between past fund performance and inflows into the subsequent fund, and find a positive and significant relationship. This indicates that a GP’s ability to raise funds is positively related to the GP’s track record.

Phalippou (2010) builds on the performance persistence results obtained by Kaplan and Schoar (2005). Firstly, Phalippou develops an ex-ante measure of performance persistence. This is due to the fact that private equity partnerships tend to raise the subsequent fund within three years of closing the

preceding fund. Therefore, through this ex-ante measure, one is able to gauge to what extent the past performance at the time of fundraising carries predictive value for the performance of the subsequent fund. Phalippou finds that private equity funds do no exhibit ex-ante performance persistence, and due to this Kaplan and Schoar (2004) have overstated their results.

Phalippou (2010) also discusses the effect of unsophisticated investors versus sophisticated investors on performance persistence. Banks and pension funds are examples of unsophisticated investors, while endowments are an example of sophisticated investors. Banks and pension funds are classified as unsophisticated for the following reasons: rigid decision criteria, lack of sufficient understanding of the asset class, inappropriate incentives, poor incentives, poor human resource practices, and conflicting objectives (Lerner et al., 2007). Endowments on the other hand tend to have superior investment committees and more skilled investment managers, and better networks (Lerner et al., 2008). When an Endowment re-invests in a private equity partnership, the newly raised fund tends to outperform, on the other hand if an Endowment decides to stop investing in a partnership, the newly raised fund tends to underperform. Therefore Endowments have high learning capacities that are not matched by other, unsophisticated investors (Lerner et al., 2008). Phalippou (2010) states that the performance

persistence is largely due to unsophisticated investors, and he finds that when investors are sophisticated, the performance, size, and sequence do not help predict the performance of the

(11)

11 subsequent fund. This is due to the fact that funds with higher performance are backed by more

sophisticated investors. Inferior funds, therefore, which are expected to be backed by less skilled investors, do not exhibit nearly as much performance persistence. Cavagnaro et al. (2016) confirm that higher (lower) skilled investors consistently outperform (underperform). By comparing the distributions of investors’ returns to a bootstrapped distribution which replicates the returns if funds were randomly distributed to investors, they find that the variance of actual performance is higher than the

bootstrapped distributions. They estimate the actual effect of investor skill on performance, and find that a one standard deviation increase in skill leads to a three percentage point increase in returns.

2.5 The role of this paper

The role of the current study lies in applying the methodologies above to a new subset of data. While all previous research has been focused on the United States and other developed markets, this research will analyze private equity performance and performance persistence in emerging and frontier markets. Previous research focused on developed markets due to the availability of data. Harris et al. (2014) state that the uncertainty of historical private equity returns remains uncertain due to the uneven disclosure of private equity returns, as well as due to questions about its quality. Preqin offers a database covering private equity partnerships focused on both developed and emerging markets, and the availability of data is finally enough to be able to conduct research into both performance and performance

persistence of private equity in emerging markets. This analysis therefore fills a gap in existing literature. As the current study is focused on a different market of private equity, it is likely that different factors will affect private equity performance in developed and emerging markets. Therefore, this study has included additional variables and controls which could affect performance and performance persistence. Previous research has been conducted using the following databases: Burgiss, Cambridge Associates, Venture Economics, and Preqin. Harris et al. (2014) compares the performance results they obtained between the data sets, and they find that performance in Cambridge Associates, Preqin, and Burgiss is similar, while it is lower in Venture Economics. Stucke (2011) explains the reasoning behind the lower return figures of the Venture Economics database. He states that while Venture Economics had been a staple database for judging performance of private equity, the underlying data contains several anomalies that can negatively bias results. Stucke finds that around 40 percent of the funds in Venture Economics sample stopped being updated during their active lifetime. This leads to a negative bias, especially with regards to the claim that private equity has not outperformed private equity.

(12)

12 The aim of this study is to analyze whether the performance of private equity funds in emerging and frontier markets is affected by the same characteristics which the aforementioned authors have found for funds in developed markets.

2.6 Hypotheses

The hypotheses of this paper are based on the results obtained by Kaplan & Schoar (2005) and Phalippou (2010). Both papers confirm the existence of performance persistence in private equity in developed markets. Kaplan & Schoar state that one cause for funds exhibiting performance persistence is heterogeneity in manager skill. This means that highly skilled GPs are able to invest more successfully. The first hypothesis of this paper is that GPs operating in emerging markets will exhibit performance persistence. The large gap between the median Net IRR and top quartile boundary in emerging markets private equity funds indicates that a few select managers are outperforming the average fund (Preqin, 2016). As Kaplan and Schoar (2005) state, this effect can be due to fund manager skill. As outlined by Leeds et al. (2003), GPs investing in emerging markets face additional difficulties. The low standards of corporate governance, weakness of the local legal system in enforcing contracts, and the inability of domestic equity markets to offer exit opportunities through the IPO market can mean that GP skill can be an important factor driving good results in private equity in emerging markets. Above average managers should be able to cope with these additional difficulties better than below average managers. The second hypothesis follows from the first, and states that GPs domiciled on the same continent as they invest will outperform the benchmark to a higher degree than managers who are domiciled on a different continent than the focus of their investments. Domestic managers can outperform foreign managers due to the fact that they are less affected by the difficulties mentioned by Leeds et al. (2003). Domestic managers have a deeper understanding of the culture of the region they are investing in, as well as better knowledge about the business environment and legal systems. This knowledge and cultural advantage leads to increased performance and performance persistence.

The final hypothesis of this research concerns capital flows of private equity funds in emerging markets. In developed markets, better past fund performance leads to larger capital inflows for subsequent funds (Kaplan & Schoar, 2005). It is expected that this phenomenon also occurs for funds investing in emerging markets, and therefore the final hypothesis states that past performance and capital inflows into a focal fund are positively related in emerging and frontier markets.

(13)

13

3. Methodology

3.1 Data

Data on historical fund performance is be taken from Preqin, a private equity fund database containing both fund characteristics as well as performance data. Preqin contains data on 16,923 funds investing globally (Preqin, 2017b). Since this study is only focused on emerging markets, the sample size will be smaller. In total Preqin offers performance data on 417 funds that fit the scope of this study. In order to be able to properly evaluate performance persistence, data from several consecutive funds from the same GP is needed. Therefore, the entire emerging markets private equity data set spanning 1981-2014 is used to ensure the largest possible sample and data on consecutive funds. The first available data on funds investing in emerging markets with performance is from 1981, thus it provides a starting point for this research. Only data up until 2014 is used, as funds of a later vintage do not yet have reliable

performance data and do not reliably reflect the final net IRR returned to investors. Funds of vintage 2014 are able to report an indicative performance. In order to have a sufficiently large sample size this research also includes funds that are not completely realized. The funds of interest in the sample are called “focal funds.” These funds have at least one predecessor fund in the same fund family.

Data on both fund performance and fund characteristics is collected from Preqin. Fund performance is reported through the following variables: percentage of capital called, DPI, RVPI, Net Multiple, Net IRR, Benchmark Net IRR, and Net IRR difference from benchmark. The following fund characteristics will be collected from Preqin: firm name, fund name, vintage, status, fund size, target size, type of fund, region focus, GP location, and industry focus. This data will be compared to relevant benchmarks which are provided by Preqin. These benchmarks are generated by looking at the overall performance of a sample of funds from the same vintage year, region, and strategy.

The 417 funds that are analyzed in this study invest in global emerging markets, and contains managers who are based on all continents. The funds have been separated into the following geographic regions: Africa, America, Asia, Europe, and Global. The only specific exclusion in this study is South Korea. Funds investing solely in South Korea were excluded from this study as South Korea exhibits traits of both developed and emerging markets (Taranukha, 2016).

(14)

14

3.2 Method

The methodology of this research is largely based around the methodology of Kaplan & Schoar (2005) and Phalippou (2010). The analysis will be split into three different sections: fund performance, performance persistence, and the relation between fund performance, capital inflows, and fund size. In order to analyze fund performance in emerging markets the following methodology will be used. Firstly, an overview of fund performance separated by fund strategy and fund region focus is provided. The second part of the analysis consists of regressing fund characteristics on different metrics of performance. In this case, the metrics consist of: Net IRR, Net Multiple, and Net IRR difference from benchmark (Outperformance). As part of this analysis, a correlation matrix is also provided to establish the relationships between the variables.

The first step in analyzing performance of private equity funds in emerging and frontier markets is to generate an overview of the performance of these funds separated by fund strategy and f und

geography. These tables report performance in three ways, Net IRR, Net Multiple, and Outperformance. Since fund size can vary greatly, both the strategy and geography analyses will be conducted normally-weighted and size-normally-weighted.

The first regression regresses performance on fund characteristics. The basic regression specification is as follows: 𝑃𝑒𝑟𝑓𝑜𝑟𝑚𝑎𝑛𝑐𝑒_𝑖𝑡= 𝛼_𝑡+ 𝛽₁(𝐹𝑢𝑛𝑑𝑆𝑖𝑧𝑒_𝑖𝑡) + 𝛽₂(𝑆𝑒𝑞𝑢𝑒𝑛𝑐𝑒_𝑖𝑡) + 𝛽₃(𝐹𝑢𝑛𝑑𝑆𝑖𝑧𝑒_𝑖𝑡)2+

𝛽₄(𝑆𝑒𝑞𝑢𝑒𝑛𝑐𝑒_𝑖𝑡)2_{+ 𝛾}

𝐹𝑇𝐹 + 𝜑𝑀𝐷 + 𝜀𝑖𝑡. The dependent variable is the Net IRR, Net Multiple, or

Outperformance of the fund, and the independent variables are fund size, fund sequence number, first time fund dummy, and a manager domicile dummy. The manager domicile dummy indicates whether the manager is domiciled in the same geography as they invests in, and the first time fund dummy indicates whether it is the first fund in a fund family. The above regressions is run three times, once for each dependent variable. Net IRR, and Net Multiple are both provided by Preqin. Preqin also provides a variable called NetIRRDiffpts, which states the percentage difference between fund performance and the performance of its respective benchmark. In order to make this variable easier to work with in the regressions, NetIRRDiffpts has been modified into a variable called Outperformance. The

Outperformance variable is a variable that indicates to what extent a fund was able to outperform or underperform compared to the relevant benchmarks generated by Preqin. The variable is equal to the percentage outperformance/underperformance divided by 100, plus one. Thus, an Outperformance value greater than one indicates that a fund outperformed its benchmark, and a value less than one indicates that a fund underperformed relative to its benchmark. All specifications of these regressions

(15)

15 include year-fixed effects to control for inter year variation in return. Firm-fixed effects may also be used. The second and third regressions are similar to the first, only differing in that the Net IRR is replaced by the Net Multiple and the Outperformance. These three regressions show to what extent fund characteristics affect the performance of funds.

To analyze performance persistence, this research investigates if GPs who outperform/underperform the benchmark in one fund are likely to do it again in subsequent funds. The analysis will look not only at consecutive funds, as later funds are also taken into account. The basic regression specification is as follows,

𝑂𝑢𝑡𝑝𝑒𝑟𝑓𝑜𝑟𝑚𝑎𝑛𝑐𝑒_𝑖𝑡 = 𝛼_𝑡+ 𝛽₁(𝑂𝑢𝑡𝑝𝑒𝑟𝑓𝑜𝑟𝑚𝑎𝑛𝑐𝑒_{𝑖𝑡−𝑛}) + 𝛽₂(𝐹𝑢𝑛𝑑𝑆𝑖𝑧𝑒_𝑖𝑡) + 𝛽₃(𝐹𝑢𝑛𝑑𝑆𝑖𝑧𝑒_𝑖𝑡−1) + 𝛽₄(𝑆𝑒𝑞𝑢𝑒𝑛𝑐𝑒_𝑖𝑡) + 𝛾_𝐹𝑇𝐹 + 𝜑_𝑀𝐷 + 𝜀_𝑖𝑡.

The regression contains lagged outperformance/underperformance, fund size (current or lagged, depending on the regression specification), the sequence number of the fund, a first time fund dummy, and a manager domicile dummy. Year fixed effects are also included in these regressions. Including a lag of the dependent variable makes it a dynamic panel. For this reason firm-fixed effects cannot be

included, as it would lead to biased results as the lagged dependent variable would be correlated with the error term in the fixed effect specification (Nickell, 1981). Finally, when analyzing performance persistence the sample also is split according to high and low performance to test the effect of unsophisticated investors on performance persistence, as outlined by Phalippou (2010). In order to further analyze what drives the results the performance persistence regressions is run with different bins sorted by geography, performance, and size.

Finally, the study analyzes the relation between fund performance to capital inflows and fund size. This is done through two separate regressions. The first regression is an Ordinary Least Squares (OLS) regression, regressing fund characteristics on the size difference between consecutive funds. The basic regression specification is as follows:

𝐷𝑖𝑓𝑓𝑒𝑟𝑒𝑛𝑐𝑒 𝑖𝑛 𝐹𝑢𝑛𝑑 𝑆𝑖𝑧𝑒𝑖𝑡= 𝛼𝑡+ 𝛽1(𝑂𝑢𝑡𝑝𝑒𝑟𝑓𝑜𝑟𝑚𝑎𝑛𝑐𝑒𝑖𝑡 −𝑛) + 𝛽2(𝐹𝑢𝑛𝑑𝑆𝑖𝑧𝑒𝑖𝑡−1) + 𝛽3𝑆𝑒𝑞𝑢𝑒𝑛𝑐𝑒 + 𝜑_𝑀𝐷 + 𝜀_𝑖𝑡

In this regression, the lagged outperformance variable, lagged fund size, fund sequence number, and manager domicile dummy are regressed on the size difference between consecutive funds.

(16)

16 An OLS estimation of this regression can be biased since if a fund performs very poorly they may not be able to raise another fund. In order to deal with this bias, a Tobit regression is also performed that controls for the left censoring in the Fund Size variable, as the size variable is censored at 0. The Tobit regression specification is as follows:

𝑦_𝑖 = { 𝑦𝑖 ∗_{𝑖𝑓 𝑦}

𝑖∗> 0 0 𝑖𝑓 𝑦_𝑖∗= 0

𝐹𝑢𝑛𝑑 𝑆𝑖𝑧𝑒_𝑖𝑡= 𝛼_𝑡+ 𝛽₁(𝑂𝑢𝑡𝑝𝑒𝑟𝑓𝑜𝑟𝑚𝑎𝑛𝑐𝑒_{𝑖𝑡 −𝑛}) + 𝛽₂(𝐹𝑢𝑛𝑑𝑆𝑖𝑧𝑒_{𝑖𝑡 −1}) + 𝛽₃𝑆𝑒𝑞𝑢𝑒𝑛𝑐𝑒 + 𝜑_𝑀𝐷 + 𝜀_𝑖𝑡 This regression regresses lagged outperformance, lagged fund size, fund sequence number, and manager domicile dummy on focal fund size.

3.3 Summary Statistics

Table 1 shows the descriptive statistics of the sample used for the analysis. A total of 417 funds are part of the analysis, which is centered around 251 focal funds. Due to an increase in private equity activity and the availability of data around the year 2000, the sample is heavily weighted towards more recent years. Nonetheless, the data from the 1980s and 1990s is used in order to be all-encompassing and reflective of all the data available. Average fund size has also increased post-2000 as can be seen in the column “Fund Value USD.” The average fund size is USD 724 million, however in both 2007 and 2012 the average fund size was larger than USD 1 billion. While the average fund size has grown over time, it still fluctuates from year to year, as can be seen from 2012 when the average fund was only USD 381 million. The drop in fund size post 2007 can be attributed to the recession beginning in 2008. Average net IRR post-2000 tends to fluctuates between 8% and 16%, while the Net Multiple is more steady around the average of 1.63. On average, a focal fund only has 1.45 predecessor funds, indicating that a focal fund usually has only one or two preceding funds. As can be seen from the study done by Phalippou in 2010, in developed markets focal funds tend to have closer to 3 predecessor funds. This means there is a much shorter track record to take into account when performing the analysis on emerging and frontier

markets. This is due to the fact that private equity in emerging markets is not nearly as popular as private equity in developed markets, and most GPs have not been active in these markets for nearly as long. The final two columns provide the Net Multiple of all preceding funds of a focal fund, as well as the Net Multiple of the fund preceding the focal fund.

Table 2 contains descriptive statistics of Net IRR sorted by year. The majority of the sample is

(17)

17 not being driven by outliers. The table also provides the maximum and minimum Net IRR for each year, as well as the standard deviation. Due to the recession there was a significant drop in funds raised from 2008 to 2009, where it drops from 47 to 24. Since then, the number of funds has steadily been rising. Around the time of the recession funds exhibited slightly lower performance as well as a lower standard deviation in performance.

(18)

18

Table 1

Descriptive Statistics

This table gives descriptive statistics for the sample of 417 funds. A focal fund is a fund with at least one predecessor fun d. This includes 251 focal funds. Fund Value, NetIRR, Net Multiple are obtained for Preqin. Number of Predecessor Funds indicate s the average amount of predecessor funds for a fund in each year. Ex -Ante Net Multiple - Track Record shows the average net Multiple of a focal funds entire track record. Ex-Ante Net Multiple - Preceding Fund shows the Net Multiple of only the previous fund with respect to a focal fund.

Vintage Number of Funds Number of Focal Funds Fund Value USD NetIRR (%) Net Multiple Number Predecessor Funds Net Multiple - Track Record Net Multiple - Preceding Fund 1981 1 0 4.40 11.00 1.50 0.00 0.00 0.00 1982 1 0 50.00 64.30 3.11 0.00 0.00 0.00 1984 1 1 76.50 45.70 6.03 1.00 3.11 3.11 1988 1 1 65.00 54.50 3.00 2.00 4.57 6.03 1989 1 1 6.70 52.30 5.52 3.00 4.05 3.00 1990 1 1 5.00 2.70 1.58 1.00 1.50 1.50 1992 2 0 44.28 39.50 3.25 0.00 0.00 0.00 1994 1 0 100.00 16.20 1.78 0.00 0.00 0.00 1995 3 1 280.83 12.87 1.52 1.00 3.61 3.61 1996 2 1 194.05 32.65 2.42 1.00 2.89 2.54 1997 3 0 124.63 3.87 1.31 0.00 0.00 0.00 1998 9 3 466.00 -3.01 2.12 1.33 1.51 1.69 1999 8 3 168.39 16.50 3.20 1.33 3.86 1.30 2000 7 3 161.36 14.11 1.80 1.00 4.81 1.51 2001 5 1 149.16 20.70 3.52 1.00 2.83 2.83 2002 10 7 471.44 25.44 2.17 1.43 1.66 1.36 2003 4 0 438.76 8.43 1.35 0.00 0.00 0.00 2004 12 5 263.92 15.79 1.65 1.40 2.04 1.97 2005 28 13 793.40 9.62 1.59 1.31 3.98 3.92 2006 29 6 794.56 11.53 2.36 1.83 1.68 1.52 2007 55 22 1258.01 8.33 1.51 1.36 3.27 2.28 2008 47 28 652.05 9.70 1.52 1.21 1.98 1.50 2009 24 14 531.97 8.07 1.43 1.50 3.54 1.66 2010 28 19 381.17 11.79 1.42 1.42 2.73 2.21 2011 35 25 620.52 11.46 1.36 1.68 1.95 1.55 2012 35 33 1138.08 15.05 1.47 1.45 2.16 1.60 2013 38 37 871.14 16.89 1.25 1.59 1.81 1.48 2014 26 26 749.68 8.73 1.10 1.42 1.59 1.48 Mean 724.76 12.15 1.63 1.45 2.37 1.82 Total 417 251

(19)

19

Table 2

Descriptive Statistics

Absolute Performance of the Entire Sample (Net IRR). This table reports descriptive statistics of the entire sample separated by vintage year. Mean, median, max, min, and standard deviation are reported as a percentage . Vintage

Number of

Funds Mean Median Max Min

Standard Deviation 1981 1 11.0 11.0 11.0 11.0 . 1982 1 64.3 64.3 64.3 64.3 . 1984 1 45.7 45.7 45.7 45.7 . 1988 1 54.5 54.5 54.5 54.5 . 1989 1 52.3 52.3 52.3 52.3 . 1990 1 2.7 2.7 2.7 2.7 . 1992 2 39.5 39.5 50.3 28.7 15.3 1994 1 16.2 16.2 16.2 16.2 . 1995 3 12.9 8.6 25.4 4.6 11.0 1996 2 32.7 32.7 43.8 21.5 15.8 1997 3 3.9 8.1 10.4 -6.9 9.4 1998 9 -3.0 14.0 32.0 -100.0 42.6 1999 8 16.5 15.5 35.0 2.0 11.1 2000 7 14.1 19.9 34.0 -9.3 16.1 2001 5 20.7 12.4 48.9 5.6 18.4 2002 11 25.4 27.0 93.0 -10.8 28.5 2003 4 8.4 11.5 20.0 -9.2 12.4 2004 12 15.8 6.9 89.2 -4.4 25.3 2005 30 9.6 6.6 52.6 -11.2 13.4 2006 29 11.5 11.8 61.7 -15.9 15.2 2007 57 8.3 8.0 20.3 -12.7 6.4 2008 47 9.7 10.0 28.0 -6.9 7.4 2009 24 8.1 9.9 23.0 -6.7 7.6 2010 28 11.8 11.1 26.0 -3.9 7.6 2011 35 11.5 12.1 31.0 -15.5 10.5 2012 35 15.0 15.0 57.7 -24.4 15.2 2013 38 16.9 12.3 68.6 -6.9 15.5 2014 26 8.7 6.0 63.6 -17.5 17.7 1981-89 5 52.9 52.9 52.9 52.9 . 1990-99 29 23.7 25.3 36.3 5.2 15.0 2000-09 226 15.1 15.3 51.5 -8.4 17.0 2010-14 162 13.4 12.1 48.7 -13.9 13.0 Total 422 12.1 10.7 93.0 -100.0 15.4

(20)

20

4. Results

4.1 Private Equity Performance

The first part of this research focuses on aggregating and analyzing private equity fund performance in emerging and frontier markets. Table 3 outlines the fund performance of each fund strategy. The table provides the number of funds of each strategy that is included in the research, as well as the Net IRR, Net Multiple, and Outperformance. In total there are 15 different fund strategies included in the sample. The large majority is concentrated in five categories: Buyout (81), Fund of Funds (83), Growth (90), Real Estate (68), and Venture (61). Based on fund Net IRR, Direct lending provides the lowest Net IRR at 4.42%, and Co-investments provide the largest Net IRR at 19.10%. Of the 5 main types of funds, Growth equity has the largest Net IRR at 16.23%, while Real Estate is the lowest at 8.73%. These values are reflected in the Net Multiple, with Growth providing the highest multiple at 2.07, while Real Estate is 1.29. Finally, when looking at the outperformance metric, both Growth and Venture tend to outperform the relevant benchmark, as can be seen from the outperformance coefficient of 1.05. On the other hand, Buyout has, in general, slightly underperformed its benchmark with a coefficient of 0.99. Figure 1 in Appendix 1 presents the performance of the five largest types of funds over the period 2004 – 2014. In general, the performance of Venture, Real Estate, and Fund of Funds has remained more stable than the performance of Growth and Buyout funds.

The right side of Table 3 offers the same information, but the coefficients on Net IRR, Net Multiple, and outperformance are instead weighted by fund size. Based on Net IRR and Net Multiple, Growth is still the best performing strategy, while Real Estate is the lowest performing fund strategy. The

outperformance metric, however, shows a shift, as now both Buyout and Fund of Funds tend to outperform with a coefficient of 0.99, while Venture remains the highest performing at 1.05. Table 4 consists of the same metrics as Table 3, but instead separated by geography. The five

geographies in question for this study are the following: Asia, Africa, Americas, Europe, and Global. It is important to note, that due to the available data, almost 50 percent of the sample consists of funds investing in Asia. Americas, Europe, and Global funds each offer over 50 observations, while only 13 funds in this study focus on Africa. When looking at both the normally- and size-weighted coefficients on Net IRR, it is clear that funds focusing on the Americas perform much worse than the other geographies. The other geographies exhibit a relatively stable Net IRR, however the size-weighted coefficients are clearly lower. While the discrepancy also exists when looking at the Net Multiple, funds focusing on the

(21)

21 Americas perform more in line with other funds in this category. Finally, based on outperformance, Africa and Asia have the edge in performance. The normally weighted coefficients show a small degree of outperformance, while the size-weighted coefficients show that Africa and Asia are performing in line with their relative benchmarks. The rest of the regions have all underperformed their benchmark, especially the Americas. Figure 2 in Appendix 1 presents the performance of funds in each region over the period 2004-2014. It can be seen that the Americas has the lowest average Net IRR, however no other clear trends emerge from this figure, as performance varies from year to year for each region. Table 5 provides a correlation matrix of the natural logs of fund characteristics. From the matrix it can be seen that fund size is negatively correlated with outperformance, while fund sequence and

predecessor performance are positively correlated with outperformance. Sequence and size are also positively correlated, indicating that funds tend to be larger than their predecessors.

(22)

22

Table 3

Number of Funds and Performance by Fund Strategy

This table reports both the number and performance measures of funds separated by fund strategy. The performance measures are Net IRR, Net Multiple, and Outperformance. Outperformance is the log of the Net IRR difference from its benchmark plus one. It reports these values weighted based on the number of funds, and weighted based on the amount of assets under management.

Normally-weighted Size-weighted Type Net IRR (%) Net Multiple

Out-performance Net IRR

Net Multiple Out-performance Balanced 6 6 6 6 6 6 8.77 1.35 0.99 6.92 1.38 0.98 Buyout 81 81 81 81 81 81 11.51 1.52 0.99 11.15 1.41 0.99 Co-investment 4 3 4 4 3 4 19.10 1.68 1.07 26.53 2.00 1.14 Direct Lending 6 6 6 6 6 6 4.42 1.13 0.95 4.58 1.15 0.96 Direct Secondaries 2 2 2 2 2 2 16.10 1.45 1.05 15.90 1.45 1.05 Fund of Funds 83 83 83 80 80 80 9.74 1.44 1.00 9.49 1.40 0.99 Growth 90 87 90 90 87 90 16.23 2.07 1.05 11.64 1.63 1.01 Infrastructure 5 5 5 5 5 5 11.22 1.43 1.01 8.35 1.34 0.98 Natural Resources 4 4 4 4 4 4 12.95 1.44 1.02 9.20 1.23 0.97 Real Estate 68 68 68 67 61 67 8.73 1.29 1.00 4.07 1.11 0.96 Secondaries 2 1 2 2 1 2 10.50 1.52 0.99 12.03 1.52 0.99 Special Situations 9 9 9 8 8 8 7.89 2.04 0.96 0.43 1.05 0.91 Timber 1 1 1 1 1 1 4.90 1.13 0.93 4.90 1.13 0.93 Venture (General) 61 59 60 61 59 60 15.38 1.81 1.05 17.40 1.77 1.05 Total 422 409 421 417 404 416 12.15 1.63 1.01 9.51 1.39 0.99

(23)

23

Table 4

Number of Funds and Performance by Region

This table reports both the number and performance measures of funds separated by region focus. The performance measures are Net IRR, Net Multiple, and Outperformance. Outperformance is the log of the Net IRR difference from its benchmark plus one. It reports these values weighted based on the number of funds, and weighted based on the amount of assets under management.

Normally-weighted Size-weighted Region

Focus Net IRR

Net Multiple

Out-performance Net IRR

Net Multiple Out-performance Africa 13 13 13 13 13 13 13.94 2.40 1.01 10.77 1.77 1.00 Americas 63 61 62 63 61 62 6.26 1.40 0.98 0.40 1.14 0.92 Asia 208 204 208 206 202 206 13.76 1.66 1.03 10.47 1.36 1.00 Europe 51 45 51 51 45 51 12.02 1.58 1.00 9.48 1.50 0.98 Global 87 86 87 84 83 84 12.36 1.64 1.02 10.52 1.44 0.99 Total 422 409 421 417 404 416 12.15 1.63 1.01 9.51 1.39 0.99 Table 5 Correlation Matrix

This table reports the correlations between variables used in this research. Outperformance is the log of the Net IRR differe nce from its benchmark plus one, Size is the size of the fund, Sequence is the sequence number of a fund in its fund family, Net Multiple - Track Record is the average Net Multiple of the entire fund family, Net Multiple - Preceding Fund is the Net Multiple of the fund preceding the focal fund. Natural logs have been applied to all variables.

Out-performance Size Size (lagged) Sequence

Net Multiple - Track Record Net Multiple - Preceding Fund Outperformance 1 Size -0.1116 1 Size (lagged) -0.0374 0.7839 1 Sequence 0.0241 0.2956 0.3442 1

Net Multiple - Track Record 0.0569 -0.1128 -0.1471 0.0974 1 Net Multiple - Preceding

(24)

24 In order to determine whether specific characteristics of a fund affect performance, various fund

characteristics are regressed on the Net IRR, Net Multiple, and Outperformance in Table 6. The fund characteristics consist of fund size, fund size squared, fund sequence, fund sequence squared, manager domicile dummy, and first time fund dummy. For each of the six specifications of the regression, both vintage and firm fixed effects are used. Standard errors are also clustered by firm to adjust for serial correlation and heteroscedasticity. In order to normalize the results each variable is a natural log, except for the two dummy variables Manager Domicile and First Time Fund.

The most notable takeaway from these regressions is the large negative coefficient on the manager domicile dummy. Managers who are based on the same continent they invest in tend to perform worse than foreign managers. A domestic manager tends to have a multiple lower by 54 percentage points, a lower Net IRR by 173.2 percentage points, and lower outperformance by 17.6 percentage points. All coefficients of Manager Domicile are significant at the one percent level. The sequence number of a fund appears to have no effect on the performance of a fund, as the coefficients are all insignificant. Size, on the other hand, can have an effect on the Net IRR of a fund, while not affecting the Net Multiple or Outperformance. An increase in fund size by one percent will increase the Net IRR by 0.566%. Size has a diminishing effect on the Net IRR as can be seen by the negative and highly significant coefficient on Size squared. The coefficient indicates that as the size gets larger, its effect on Net IRR gets smaller. This diminishing effect can also be seen with regards to the Net Multiple and Outperformance even if Size itself is not significant. The final takeaway of Table 5 is the effect on the Net Multiple through the First Time dummy. A first time fund tends to have a higher multiple than the subsequent funds.

(25)

25

Table 6

Fund Performance and Fund Characteristics

This table reports the results of an OLS regression regressing fund characteristics on various performance measures. The dependent variables are Net IRR, Net Multiple, and Outperformance. Outperformance is the log of the Net IRR difference from its benchmark plus one. Size is the size of the fund, Sequence is the sequence number of a fund in its fund family, Manager Domicile is a dummy indicating a GP is domiciled in the same continent on which it invests, First Time Fund is a dummy indicating it is a first time fund. All variables except for Manager Domicile dummy and First Time Fund dummy are natural logarithms. Standard errors clustered by partnership are reported below the coefficients.

Dep. variable: Net Multiple Dep. variable: Net IRR

Dep. variable: Outperformance (1) (2) (3) (4) (5) (6) Size -0.125*** 0.160 -0.196** 0.566** -0.0302*** 0.0350 (0.0356) (0.108) (0.0804) (0.252) (0.0104) (0.0317) Sequence 0.00325 0.224 -0.466 -0.237 -0.0594 -0.0216 (0.169) (0.189) (0.367) (0.430) (0.0479) (0.0622) -0.0289*** -0.0777*** -0.00659** (0.0108) (0.0254) (0.00296) -0.168** -0.302 -0.0309 (0.0885) (0.200) (0.0338) Manager Domicile -0.572*** -0.540*** -1.891*** -1.732*** -0.185*** -0.176*** (0.140) (0.142) (0.274) (0.352) (0.0373) (0.0353) First Time Fund 0.0661 0.219* 0.0504 0.243 -0.00392 0.0228

(0.116) (0.129) (0.257) (0.286) (0.0340) (0.0365) Constant 1.114*** 0.578** 2.513*** 1.117* 0.338*** 1.225***

(0.218) (0.259) (0.478) (0.615) (0.0995) (0.129) Firm F.E. Yes Yes Yes Yes Yes Yes Vintage F.E. Yes Yes Yes Yes Yes Yes 0.350 0.378 0.332 0.371 0.216 0.228

N-obs. 403 403 361 361 415 415

Standard errors in parentheses * p<0.1, ** p<0.05, *** p<0.01 Size2

Sequence2

(26)

26

4.2 Performance Persistence

The second part of the analysis determines whether funds investing in emerging and frontier markets exhibit performance persistence. As indicated by both Phalippou (2010) and Kaplan and Schoar (2005), private equity funds investing in developed markets do exhibit performance persistence. Performance persistence can partly be attributed to the skill level of the GP (Harris et al., 2014). Due to the additional challenges faced by investors investing in emerging and frontier markets, as outlined by Leeds et al. (2003), this paper hypothesizes that private equity funds investing in these regions also exhibit performance persistence.

These regressions analyzes to what extent past performance, along with fund characteristics, affect the performance of the current fund. The main independent variable of interest is lagged outperformance. This analysis is done ex-post, as all information about past performance is now available. Due to sample size restrictions, the ex-ante methodology reflecting the information investors have at the time of investing, was not able to be performed. This methodology, outlined by Phailppou (2010), presents a more practical look at performance persistence by looking at the data available to the investors when they make their initial commitment to a fund. As such, the current study should be viewed as a review of performance persistence in private equity instead of as a guide to investing, since the information available ex-post and ex-ante can vary.

Table 7 presents the performance persistence regressions. Five separate specifications are run, the first three showing the effects of including the performance of the past fund and the performance of the past two funds. The final specifications include one or two lags of outperformance, as well as fund characteristics and the manager domicile dummy. Year fixed effects are included for all specifications. Firm fixed effects cannot be controlled for in this regression, as lagged Outperformance is included as right-hand-side variables (Nickel, 1981). As can be seen from the first two specifications, by themselves neither performance lag has any predictive power regarding outperformance. Two lags of

outperformance is included by itself in specification 2 as consecutive funds could have some

investments in common, and could therefore induce performance persistence (Kaplan & Schoar, 2005). When both lags are included together the performance of the past fund has a coefficient of -0.481 which is significant at the five percent level. This coefficient indicates that an increase in the

outperformance of the preceding fund by one percent, will decrease the outperformance of the current fund by 48.1 basis points. This effect becomes more pronounced when including the fund

(27)

27 lag of Outperformance has a coefficient of -0.516 and is significant at the five percent level. An increase in the outperformance of the past fund by one percent now coincides with a decrease in the

outperformance of the current fund by 51.6 basis points. The performance of the fund two steps earlier in the sequence does not play a role on the performance of the current fund. Neither the size of the current fund, nor the size of the previous fund have an effect on the Outperformance of the focal fund.

Table 7

Persistence of Fund Performance

This table reports the results of and OLS regression regressing lagged Outperformance and fund characteristics on Outperformance. Outperformance is the log of the Net IRR difference from its benchmark plus one, Size is the size of the fund, Sequence is the sequence number of a fund in its fund family, Manager Domicile is a dummy indicating a GP is domiciled in the same continent on which it invests. All variables except for Manager Domicile dummy are natural logarithms. Standard errors clustered by partnership are reported below the coefficients.

Dependent Variable: Outperformance

(1) (2) (3) (4) (5) -0.0392 -0.481** -0.0540 -0.516** (0.0892) (0.234) (0.0874) (0.570) 0.0278 -0.0425 -0.0642 (0.106) (0.111) (0.139) Size -0.0181** 0.00472 (0.00830) (0.0139) Size (lagged) 0.00471 -0.00407 (0.0107) (0.0109) Sequence 0.00982 -0.0364 (0.0189) (0.0440) Manager Domicile 0.0209 0.0122 (0.0398) (0.0366) Constant 0.302*** 0.114*** 0.171*** 0.361*** 0.109 (0.0392) (0.0261) (0.0374) (0.0506) (0.0937) Firm F.E. No No No No No Year F.E. Yes Yes Yes Yes Yes

0.149 0.817 0.779 0.161 0.821 N-obs. 229 67 67 225 66 Standard errors in parentheses

* p<0.1, ** p<0.05, *** p<0.01 Outperformance_i−1

Outperformance_i−2

(28)

28 There also appears to be no difference in outperformance between domestic and foreign managers. The negative coefficient on lagged Outperformance contradicts the research done in developed markets and the hypotheses of this research. The negative coefficient points to mean reversion in the

performance of private equity fund in emerging and frontier markets. If a GP outperformed the benchmark in the preceding fund, they will be less likely to outperform in the subsequent fund. Conversely, if they underperformed, they will be more likely to outperform in the subsequent fund. Fund will always tend towards the benchmark. This result confirms that private equity funds investing in these regions do not exhibit performance persistence. Appendix 3 offers a regression table containin both the results of the current research and that of Kaplan and Schoar’s (2005), allowing for comparison. Kaplan and Schoar used a public market equivalent as their dependent variable and outperformance measure. They found that both one and two lags of outperformance positively affect outperformance, indicating that funds in developed markets exhibit performance persistence. They also found positive coefficients for both size and a venture capital dummy, meaning that larger funds and venture capital funds tend to exhibit more performance persistence.

4.3 Performance, Fund Characteristics, and Capital Inflows

The final section of the analysis considers what affects the capital inflows of private equity funds in emerging and frontier markets. Kaplan and Schoar (2005) find that up to two outperformance lags, as well as lagged size, and sequence number affect the amount of capital a fund can raise for a subsequent fund.

Kaplan and Schoar (2005) use a Tobit regression with the dependent variable the logarithm of Fund Size. They sensor Fund Size at zero because of the potential of a poorly performing GP of being unable to raise a subsequent fund. In a simple OLS regression, this would lead to bias as poorly-performing first time funds would drop out of the sample. This analysis will include a Tobit regression with the natural log of fund size as the dependent variable, as well as an OLS regression with the change in fund size as the dependent variable.

Table 8 contains the results from the OLS regression with Size Difference as the dependent variable and the results from a Tobit regression censored at Size = 0 with Size as the dependent variable.

Specifications 1 and 2 of the OLS regression show that the level of outperformance of the previous fund have a positive effect on the change in the fund size. The better the outperformance of the previous fund, the larger the subsequent fund will be. When including the natural log of lagged fund size this

(29)

29 effect disappears. Lagged outperformance is not significant in specification 3, instead lagged fund size has a negative coefficient which is significant at the one percent level. The larger the previous fund, the smaller the change in fund size of the subsequent fund. This is evidence of diminishing fund growth as funds get larger. The coefficient of -0.221 indicates that a one percent increase in lagged size will lower the size difference with the next fund by 2.21%. Metrick and Yasuda (2006) state that when a fund performs well, investors take it as a sign of GP “skill,” and thus increase the demand for a follow-up fund. GPs can respond to this increase in demand by increasing their subsequent fund size. They go on to state that increasing fund size also comes with additional costs. Table 8 also confirms that the domicile of the manager does not affect its ability to attract capital, and their inflows resemble that of foreign funds.

All three specifications of the Tobit regression show that lagged outperformance does not affect the size of the subsequent fund. As in the OLS regression, adding lagged Fund Size and the Manager Domicile dummy has a large effect on the regression results, indicating the importance of lagged fund size in determining the size of the subsequent fund. The positive coefficient on lagged fund size indicates that the larger the size of the previous fund, the larger the subsequent fund will be. An increase of one percent in the size of the previous fund leads to an increase in size of the next fund by 7.79% Once again, manager domicile has no effect on fund size.

Both methods ultimately provide insignificant coefficients of lagged outperformance. Therefore the final hypothesis of this research is refuted. Past performance does not affect the ability of a GP to raise capital for a subsequent fund. The key driver of fund size is the size of the previous fund.

(30)

30

Table 8

Capital Flows and Past Performance

This table reports the results of an OLS regression of lagged Outperformance and fund characteristics on Size Difference and a Tobit regression censored at Size = 0 of lagged Outperformance and fund characteristics on Size. Size Difference is the difference between the log of fund size and the log of lagged fund size. Outperformance is the log of the Net IRR difference from its benchmark plus one, Size is the size of the fund, Sequence is the sequence number of a fund in its fund family, Manager Domicile is a dummy indicating a GP is domiciled in the same continent on which it invests. All variables except for Manager Domicile dummy are natural logarithms. The OLS regression reports R-squared within, while the Tobit regression reports R-R-squared pseudo. Standard errors clustered by partnership are reported below the coefficients.

OLS Tobit - Censored at Size = 0 Dependent Variable: Size Difference Dependent Variable: Size

(1) (2) (3) (1) (2) (3) 0.925** 0.908** 0.591 -0.702 -0.505 0.591 (0.440) (0.444) (0.453) (0.797) (0.747) (0.429) Sequence -0.0807 0.145 0.941*** 0.147 (0.123) (0.149) (0.279) (0.141) Size (lagged) -0.221*** 0.779*** (0.0496) (0.0470) Manager Domicile 0.0134 0.0134 (0.124) (0.117) Constant 0.0194 0.0827 0.929*** 4.646*** 3.907*** 0.929*** (0.193) (0.234) (0.280) (0.350) (0.373) (0.266) Firm F.E. No No No No No No

Year F.E. Yes Yes Yes Yes Yes Yes 0.172 0.164 0.400 0.0352 0.0647 0.317 N-obs. 226 226 226 226 226 226 Standard errors in parentheses

* p<0.1, ** p<0.05, *** p<0.01

4.4 Robustness Checks

In order to confirm the results obtained in the previous section, a series of robustness checks have been run. The robustness checks are used to check whether the above results hold when applied to different sub samples of the data. The first robustness check is presented in Table 9. In order to get a deeper understanding of the results from the performance persistence regression in Table 7, the sample of funds has been split into two bins. The sample has been split into half by fund size, meaning the first bin contains smaller funds while the second bin contains larger funds. This gives insight into whether performance persistence differs based on the size of the fund.

(31)

31

Table 9

Performance Persistence - Small Funds

This table reports the results of and OLS regression regressing lagged Outperformance and fund characteristics on Outperformance. The sample is split in half, and the regression is run separately for small funds and large funds. Outperformance is the log of the Net IRR difference from its benchmark plus one, Size is the size of the fund, Sequence is the sequence number of a fund in its fund family, Manager Domicile is a dummy indicating a GP is domiciled in the same continent on which it invests. All variables except for Manager Domicile dummy are natural logarithms. Standard errors clustered by partnership are reported below the coefficients.

Small Funds Large Funds

Dependent Variable: Outperformance Dependent Variable: Outperformance

(1) (2) (3) (1) (2) (3) -0.201 -0.191 -0.189 -0.0189 -0.0193 -0.0248 (0.190) (0.184) (0.190) (0.103) (0.103) (0.0926) Sequence 0.00977 0.0140 0.0152 0.0228 (0.0283) (0.0299) (0.0212) (0.0238) Size 0.00632 -0.0517** (0.0157) (0.0240) Size (lagged) -0.00151 0.0415** (0.0135) (0.0188) Manager Domicile 0.0131 0.0402* (0.0309) (0.0220) Constant 0.373*** 0.361*** 0.336*** -0.362*** -0.389*** -0.406*** (0.0832) (0.0777) (0.119) (0.0811) (0.103) (0.117) Firm F.E. No No No No No No

Year F.E. Yes Yes Yes Yes Yes Yes 0.319 0.296 0.242 0.124 0.117 0.323 N-obs. 107 107 107 118 118 118 Standard errors in parentheses

* p<0.1, ** p<0.05, *** p<0.01

As the sample has been split in half, the smaller sample size prevents two lags of outperformance from being used in these regressions. Therefore, Table 9 should be compared to specification 4 of Table 7. The results of specification 3 in both regressions show that when separated into small and large funds, neither group by itself exhibits performance persistence or mean reversion. This is in line with the results from Table 7, which shows that the entire sample does not exhibit mean reversion or performance persistence.

(32)

32 While small funds do not show any other factors to be significant in affecting fund outperformance, large funds report significant coefficients for both Size, lagged Size, and Manager Domicile. The negative size coefficient indicates that the larger the focal fund is, the less likely it is to outperform. An increase of fund size by one percent leads to a drop in outperformance by 5.17%. The positive coefficient on lagged size indicates that the larger the previous fund, the more likely the focal fund is to outperform. These two forces work against each other, as the larger the previous fund the larger the current fund will be. As could be seen in the correlation matrix in Table 5, there is a positive correlation between lagged size and size. The combined effect of both variables is, nonetheless, slightly negative, indicating that larger funds tend to be less likely to outperform. This size effect is also only visible in the large funds. The Manager Domicile dummy is included in the regressions for both the larger and smaller funds. Domestic GPs of small funds do not tend to outperform foreign GPs of small funds. Large domestic funds tend to perform better than large foreign funds. The coefficient of 0.0402 means that a large domestic funds tend to outperform foreign large funds by 4.02%. This effect does not appear over the whole sample, as the Manger Domicile dummy is not significant in Table 7.

Table 10 consists of the performance persistence regression where the entire sample is split into two bins based on fund performance. The regression is then run on funds with high performance and on funds with low performance. The regressions indicate that neither high- nor low-performance funds exhibit performance persistence or mean reversion, and therefore the performance persistence results are robust with specification 4 in Table 7. The regressions report no significant coefficients for any variables, while the overall sample reported a negative coefficient for Size. These results do not agree with those obtained by Phalippou (2010), where he finds that low performance funds exhibit

(33)

33

Table 10

Performance Persistence - High Performance Funds

This table reports the results of and OLS regression regressing lagged Outperformance and fund characteristics on Outperformance. This regression is run on half of the sample containing funds with high performance. Outperformance is the log of the Net IRR difference from its benchmark plus one, Size is the size of the fund, Sequence is the sequence number of a fund in its fund family, Manager Domicile is a dummy indicating a GP is domiciled in the same continent on which it invests. All variables except for Manager Domicile dummy are natural logarithms. Standard errors clustered by partnership are reported below the coefficients.

High-Performance Funds Low-Performance Funds Dependent Variable: Outperformance Dependent Variable: Outperformance (1) (2) (3) (1) (2) (3) -0.0835 -0.0773 -0.0815 -0.0704 -0.0586 -0.0660 (0.0839) (0.0863) (0.0867) (0.0986) (0.0944) (0.0932) Sequence 0.000286 0.00204 0.0141 0.0249 (0.0184) (0.0190) (0.0247) (0.0254) Size -0.00417 -0.0140 (0.00876) (0.0140) Size (lagged) 0.00140 -0.0110 (0.00738) (0.0127) Manager Domicile 0.00944 0.0328 (0.0192) (0.0215) Constant -0.0972*** -0.0978*** -0.0833* 0.315*** 0.300*** 0.353*** (0.00264) (0.0248) (0.0469) (0.00264) (0.0248) (0.0590) Firm F.E. No No No No No No

Year F.E. Yes Yes Yes Yes Yes Yes

0.642 0.628 0.627 0.682 0.667 0.668

N-obs. 124 124 121 105 105 104

Standard errors in parentheses * p<0.1, ** p<0.05, *** p<0.01

The third robustness check considers the effects of fund characteristics on the outperformance of a fund. In Table 11 the fund characteristics regression is run separately for each different type of fund. The original regression in Table 5 found that Fund Size squared and the Manager Domicile dummy were significant factors in determining the outperformance of a fund. When separated by fund strategy, one can see that the driving factors depend on the type of fund. Both Fund of Funds and Venture Capital funds report no significant coefficients. Buyout funds report a negative coefficient for Size and a positive coefficient for Size squared. This means that as a buyout fund grows in size, it is less likely to

outperform. This effect gets stronger the larger size is. Growth funds, on the other hand, report the

Private equity performance, performance persistence, and capital flows in emerging markets

UNIVERSITY OF AMSTERDAM