The Influence of Private Equity Fund Growth on
Fund Performance
By Robert Goossens Master Thesis University of Amsterdam MSc Finance (Corporate Finance)
Statement of Originality
This document is written by student Robert Goossens 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.
The Influence of Private Equity Fund Growth on
Fund Performance
Robert Goossens
June 23, 2017
Abstract
In June 2016, the record amount of capital under management of private equity funds amounts to 2.49 trillion dollars. All private equity funds combined raise almost 350 billion dollars in 2016. Nevertheless, recent research points out decreasing returns for many funds. Also, literature finds that the gap between top and bottom performing funds increases. Although private equity, as an investment type, enjoyed increased attention over the last years there remain many questions to be answered. In this paper, I will investigate one of the drivers of fund returns. I find that fast growing funds perform worse than funds that grow at a slower pace. For my sample, this is mainly the case in the United States.
1 Table of Contents 1. Introduction ... 2 2. Literature review ... 4 3. Empirical method ... 9 4. Data ... 14 5. Results ... 19 6. Robustness checks ... 33 7. Discussion ... 52 8. Limitations ... 55
2
1. Introduction
During the last decades, the size of capital flowing to private equity funds has grown explosively. During the beginning years of private equity funds’ returns were satisfying. Despite recent lower returns of private equity funds (Phalippou & Gottschalg, 2009) (Kaplan et all. 2016), a bad name of the industry and increasing regulation on tax deductions, capital flows into private equity funds keep increasing. Also, Braun, Jenkinson and Stoff (2017) find that the previously existing fund persistence does not hold anymore, which means that investors can no longer base future performance on previous performance. However, because public markets are risky and interest rates are low, investors still prefer an investment in this asset class. Preqin (PE database) established a widening gap between top and bottom performing funds between 2008 and 2013. Therefore, it seems that it has become more important for investors to choose the right fund to invest in. What are factors that influence fund returns besides picking the right investments? Does, for instance, fast growth of a fund depress the return?
Previous literature investigated possible drivers of fund performance such as the expertise of the general partners and the moment of investment. However, there has not been a consensus on what the influence of fund size on fund performance is and, to my knowledge, there has not been any research recent done on what the influence of fund growth between two funds is on performance. As mentioned earlier, reported returns for private equity funds decrease and performance between top/bottom funds diverges. This makes it more important for investors to make the right choice in which fund they invest. But it is also interesting from a fund managers’ point of view as this paper will try to establish a relationship between performance and fund growth. Fund managers could use this paper to set out an optimal growth path.
The aim of my study is thus to further investigate what are the determinants of private equity fund performance and especially whether fund inflows negatively influence fund returns and whether initial funds buyout funds perform better than follow-up funds. Furthermore, I will assess whether difference between realized fund size and target fund size has an influence on fund returns. This research adds to literature on the drivers of private equity funds’ performance in various ways. Firstly, the effect of fund growth on performance has not been investigated so far. Also, the effects of the gap between realized fund size and target fund size on fund performance has, until now, not been investigated. I have created a variable to measure the difference between both and created 6 groups to assess whether larger gaps have larger effects.
3
Moreover, the hypothesis that first-time funds perform best is also a new idea. Lastly, by integrating a large dataset containing information on funds’ portfolio companies, I can create a unique dataset that incorporates fund level and deal level data.
Using a large set of data on private equity fund sequences and separate sets on their portfolio companies obtained from the private equity fund database Preqin, I have been able to establish a relationship between fund size and performance and fund growth and performance. In my paper, I have used multiple sets of (OLS and logit) regressions to see if the effects differ between regions and fund sizes. The rest of the thesis is organized as follows: first I will review a large body of existing literature in which I will shorty explain how private equity funds work and a quick overview on the different kind of funds that exist. Next, I will discuss previous papers that have investigated possible determinants of private equity fund returns. In part 3, the empirical methods used in this paper are discussed. Also, al the control variables used in the regressions are justified by stating previous investigations that have found relationships between these variables and fund performance. In part 4, I discuss the origin of the dataset and how the combined datasets were formed. Furthermore, some descriptive statistics are given. In part 5 and 6 the results and the robustness test are shown and discussed followed by the conclusion and some limitations of my study.
4
2. Literature review
A private equity firm typically starts a new fund every three to five years. The general partners or fund managers (GP’s hereafter) raise capital from (institutional) investors. These investors or limited partners (LP’s hereafter) commit around 80% of the fund size and the GP’s contribute the other 20% percent. This asset class is called private equity because the private funds are not listed on public exchanges although listed private equity funds are becoming more common (McCourt, 2016). The lifetime of a fund is generally ten years but this can be extended (Metrick & Yasuda, 2010). The first five years are used by the GP’s to invest the committed capital. There are many investment types for private equity firms of which buyout funds and venture fund are the most common. Therefore, I will focus on these two funds. Venture funds are further under divided in subcategories. Venture capital funds focus on small/start-ups. These investments are typically riskier than the investments made by buyout funds but can also have a large upside. Famous, recent examples of companies backed by venture capitalists are Uber, Snapchat and Airbnb. Buyout funds usually acquire a controlling interest in a developed or full-grown company. Buyout funds commonly use a significant amount of debt, relative to equity, to acquire these companies. Therefore, these buyouts are often called leveraged buyouts. The portion of debt commonly ranges from 60 to 90 percent of the total sum (Kaplan & Stromberg, 2009).
Liu (2017) identifies three channels through which private equity funds using LBO’s create value. The first channel is the “cherry picking” channel. This means that general partners have the skills to pinpoint undervalued companies. Moreover, private equity firms are able to enhance the operating performance of target companies. Also, general partners increase the debt level of the target company to exploit tax benefits. Lastly, the target companies’ management is adjusted to increase company value. Furthermore Jensen (1989) argues that private equity involvement in a company improves the governance and thus creates value.
Although investments in both buyout and venture capital funds have increased explosively, and private equity has caught the attention of academics, literature on the determinants of PE funds is scarce. The literature that is available focusses on three main topics (Diller & Kaserer, 2009). Firstly, does private equity contribute to economic welfare? Secondly, do investors gain informational advantages when choosing this asset class to invest in? And finally, what factors influence the performance of private equity funds? In this paper, I will focus on the determinants of private equity returns.
5
Although there seems to be a consensus on het negative effects of mutual fund size on mutual fund returns that is documented by Chen et al. (2004) and Wermers (2000), research did not reach a decisive conclusion on the effects of private equity fund size and especially not on the effects of the growth of fund inflows on fund performance. Kaplan and Schoar (2005) find that fund performance declines when funds grow fast, however growth until 2000 was not large enough to influence fund persistence. Also, Kaplan and Schoar (2005) find that the influence of fund growth on performance is upward biased because of the fact that top performing funds grow at a slower pace. They mention the possibility that a greater rise in capital inflows in the future could influence the returns of private equity funds but for their sample this was not yet the case. They also find that larger VC funds perform better then smaller VC funds. Also, VC funds outperform buyout funds in their sample. Furthermore, they find that returns are positively related with fund size and that the relation is concave. This means that larger funds perform better but when funds become too large the performance starts to decrease. Moreover, they argue that funds that perform well will have a greater chance of performing better than the market in a follow-up fund and even the fund after that. The same holds for funds that perform worse than the market. Braun, Jenkinson and Stoff (2017) find that this “fund persistence” found by Kaplan and Schoar largely disappeared in recent years. The decline/disappearance of fund persistence is mainly due to an increase in competition in the private equity sector. Furthermore, Braun, Jenkinson and Stoff (2017) do find that, although the evidence on persistence for their whole sample is almost non-existent, there is still performance persistence among the best and the worst performing funds. Kaplan and Lerner (2010) argue that follow-up funds usually grow significantly if the previous fund performed well. One very important side note however is that they argue that funds that grow fast do not perform as well as follow-up funds of which the fund size does not increase. The performance measures that are used most often are the “internal rate of return” (IRR) and “public market equivalent” (PME). However, Harris et al. (2012) argue that the multiples of invested capital are a more reliant measure for fund performance. Kaplan and Stromberg (2009) and Harris et al. (2012) find that performance measures, such as IRR and PME, decline when there is an increase in capital inflow to PE. Robinson and Sensoy (2011) do not find a similar relationship. The findings of Kaplan and Stromberg (2009) could be caused by, for instance, the “money chasing deals” phenomenon described by Gompers and Lerner (1999). They find that an increase in capital flows to funds increases valuation of investments but does not increase the success of these investments. Diller and Kaserer (2009) extend the research done by Gompers and Lerner
6
(1999). They investigate what the drivers are for private equity returns with a sample of 200 European private equity funds. Their findings are in line with the “money chasing deals” phenomenon. They conclude that capital flowing to a buyout fund or a venture capital fund has a negative influence on the performance of that fund type. They argue that this effect is stronger for venture capital funds than for buyout firms. However, Kaplan and Schoar (2005) argue that for funds that are raised during boom times, returns are lower. But that the decrease in performance of total PE funds is not due to established firms but due to incumbents. Degeorge,
Martin and Phalippou (2014) investigate whether secondary buyouts by private equity firms can create value. They find that secondary buyouts done during the beginning of the investment period perform just as well as other investments but that SBO’s done at the end of the investment period of a fund perform worse. This could be evidence that general partners invest in these SBO’s for personal gains. I assume that the faster a fund grows, the more dry-powder is lying around when the end of the investment period is nearing and thus more value decreasing SBO’s are done. Preqin reported that the total amount of dry-powder increased from EUR 292.2 billion to EUR 1411.4 billion between December 2000 and December 2016.
Lopez-de-Silanes, Phalippou and Gottschalg (2015) find that private equity investments are not scalable, which means that if a fund generates positive returns, the returns in follow-up funds cannot just be increased by increasing the fund size. They propose three different reasons. Firstly, they propose that diseconomies of scope cause performance to be inferior. When the PE firm increases its fund size, it will also invest in industries it did not used to invest in. This could lead to diseconomies of scope. Secondly, Lopez-de-Silanes et al. (2015) hypothesize that the increased size of the investments causes the lower returns. Based on previous literature, they argue that firms with more capital at their disposal buy companies at higher prices. A third explanation could be that the increase in fund size leads to limited attention.
The limited attention hypothesis, which is also discussed by Cumming and Dai (2011), argues that PE funds do not increase their staff in line with the increase in fund size. This causes the quality of the control on each investment to decrease and eventually decreases performance of the fund. DellaVigna and Pollet (2009) also find evidence of limited attention in trading. They find that even the nearing of the weekend will distract traders from trading adequately when firms have a new earnings announcement. Another rationale why larger funds could perform worse is given by Cumming and Dai (2011). They find diseconomies of scale because larger funds tend to need more highly skilled managers to pick and manage investments, but
7
find that the talent pool is limited. This could ultimately deteriorate the quality of the investments and the return. Cumming and Dai (2011) further find that some venture capital fund managers will let the fund become inefficiently large because they pursue their own interest (salary, prestige). This relates to the agency problem theory and the free cash flow hypothesis of Jensen (1986). Humphery-jenner (2011) also suggests that large private equity funds should perform worse than small funds. Over a large sample of U.S. private equity firms, they find that large funds earn lower returns. These returns are even lower when they invest in small companies.
However, not all previous literature agrees on the relationship between funds size and performance. Metrick and Yasuda (2007) find that it is easier to increase the scale of a buyout funds than it is to increase the scale of a venture capital fund. They argue that this is the case because venture capitalists always invest in small firms, whereas buyout firms can just increase the size of their investments. Higson and Stucke (2012), for instance, find that fund size positively influences fund performance. Kaserer and Diller (2004) however, find no conclusive relationship between fund size and performance. Kaplan and Schoar (2005) write that top performing funds intentionally grow less than other funds to keep returns higher. This argument would support my first hypothesis which is based on, among others, the limited attention theory Dellavigna and Pollet (2009) and Cumming and Dai (2011), the “late SBO” theory of Degeorge, Martin and Phalippou (2014), and the theories posed by Lopez-de-Silanes, Phalippou and Gottschalg (2015). Hence my first hypothesis is that:
“That performance of a private equity follow-up fund is negatively influenced by the growth rate of the value of the follow-up fund”.
However, if the finding of Kaplan and Schoar (2005) is really the case, the outcome of my regression could be biased due to reversed causality.
Furthermore, I will investigate whether fund performance is the highest if the realized fund size is the same as the target fund size. If the realized fund size is larger than the fund size general partners aimed for, the fund grows faster than the general partners expected. Consequently, the firm does not have the right amount of human capital to manage the increase in investments, which could influence fund performance. Moreover, if realized fund size is smaller than the target fund size, investors could have a lack of confidence in the general partners because of a bad reputation or low returns in the past.
8
The second hypothesis is thus formulated as follows:
“Private equity funds, of which realized fund value is equal to target fund value have higher returns”.
Aggarwal and Jorion (2010) find that hedge funds achieve higher returns when they just get started because they have more incentives to perform better. Although hedge funds differ significantly from buyout funds I hypothesize that this effect will be similar for buyout funds. Nevertheless, Lopez-de-Silanes et al. (2015) do not find that the age of the firm or the numbers of the fund have an influence on the performance. Kaplan and Schoar (2005) argue that first funds perform worse than follow up funds because of excessive risk taking of these younger funds. Ljungqvist, Richardson and Wolfenzon (2003), and Phalippou and Zollo (2005) also argue that small and inexperienced have worse returns than larger experienced funds. However, I argue that because initial funds are often the smallest that these funds will outperform follow-up funds. Moreover, I propose that initial funds will perform better because GPs are better incentivised to perform better since they want to realize higher fund sizes in follow-up funds. The third hypothesis is therefore:
9
3. Empirical method
The main relationship that I will be testing in this paper is whether the growth rate of the realized fund size between a private equity fund and its follow-up fund has a negative influence on the performance of the follow-up fund. For the dependent variables, I will use a measure for the difference in net IRR between funds and Net IRR. Fund performance is normally measured by using the net Internal Rate of Return (IRR). Net IRR is the IRR of which fees for the fund managers are subtracted. Preqin reports the difference between net IRR (%) of the fund and the IRR of a benchmark. The IRR is the discount rate for which the current value of an investment is zero. A higher net IRR also means that the fund returns are higher. The variable “Δ Net IRR” is calculated manually by going through the entire sample and excluding the difference between the net IRR of funds that were not follow-up funds or funds from different fund families.
The main independent variable of interest is “Change in Fund Value”. “Change in Fund Value” is measured as the relative difference between the realized fund size of a fund and the realized fund size of its preceding fund. The variable “Change in Fund Value” is created by manually going through the sample and only calculating the differences between fund values if the funds are follow-up funds in the same fund family. Phallipou and Zollo (2005) find that the return of PE funds positively covaries with economic cycles and indices. According to them funds that operate during an upward economic cycle should perform better. Although the net IRR measure used already accounts for market movements, I will include time-fixed effects to control for upward and downward cycles in the economy. If the variable has the hypothesized effect on the dependent variable, “Change in Fund Value” will negatively influence the return of a fund.
Braun, Jenkinson and Stoff (2017) find that persistence of fund performance has disappeared the last decade. However, I do need to control for GP skill and experience. Therefore, I will include a one year lagged fund performance as a proxy for GP skill and experience. If the effect of “Lagged Performance” is in line with the findings of Braun, Jenkinson and Stoff (2017) it will have a positive effect for the best performing funds, a negative effect for the worst performing funds and no effect for the funds in between. I will also include a second variable that acts as a proxy for trust of LP’s in the GP’s of a fund and of GP skill and experience.
The variable “Time to Closing” is constructed as the amount of days between the first and final closing of a fund. This variable could also be interpreted as a proxy for the experience/skill of GP’s and their ability to raise capital. A shorter period between the first and
10
final closing of a fund could imply that GP’s are more experienced and skilful. However, because the information on first and final closing dates is only available for a limited amount of observations, the variable “Time to Closing” will be included in a separate regression. Furthermore, because of the findings of Kaplan and Schoar (2005) and Harris, Jenkinson and Kaplan (2014) that aggregate capital inflows to the specific investment types negatively influence performance, I will control for the total inflow to buyout and venture funds per vintage year in my sample by incorporating the variable “Total Relative Inflow”. This measure is constructed by adding the total realized value per investment type and vintage year and assigning the total value to each corresponding fund. A priori, I expect the effect of “Total Relative Inflow” on performance to be negative. This means that a larger inflow of capital into a specific investment type will decrease returns of fund that operate within that particular investment type. Kaplan and Schoar (2005) find that the logarithm of the realized fund size and the logarithm of the sequence number of a fund have positive influence on the performance. Although other research finds other effects of fund size on performance, I still include this control variable to control for a possible bias that might occur when this variable is left out of the regression. The logarithm of the variable is used to make the make the variable more normally distributed.
Also, Braun, Jenkinson and Stoff (2017) find an opposite effect of the sequence number on performance for a more current dataset. I will use the sequence number as a control variable because I propose that this has an influence but that the direction of its effect has changed over the years. Furthermore, Kaplan and Schoar (2005) argue that returns are higher for venture capital funds. I will therefore include a dummy variable that takes the value of 1 if the fund is a venture capital fund. Also, in every set of regressions will control for fund vintage fixed effects to account for macroeconomic effects such as, economic growth or inflation and for general partner location fixed effects to account for regional trends and effects on performance.
11
Regression 1 and 2 are used for testing the first hypothesis:
1) Net IRRi = β0 +β1∆FundValuei+ β2FundValuei + β3lnTotalRelativeInflowit+
β4Perfomanceit-1 + β5Sequencenri + β6VentureFundi + β1PortfolioSizei + β7FirsttoFinali + ui
2) Δ Net IRRi = β0 +β1∆FundValuei+ β2FundValuei + β3lnTotalRelativeInflow it+
β4Perfomanceit-1 + β5Sequencenr i + β6VentureFund i + β7PortfolioSizei + β8TimetoClosingi + ui
The second set of regressions will test my hypothesis that a firm that has the same realized fund size as the original target size will perform better. The sample that I will use to test the second hypothesis slightly differs from the dataset used in the first regression. Because it is not necessary to calculate the difference between a follow-up and its predecessor, the first observations of a fund family do not have to be deleted. This causes the dataset to be larger than in the first regression. The dependent variables are again “Net IRR” and “ΔNet IRR”, which is the difference between the net IRR of the fund and the net IRR of the benchmark.
To assess whether the performance of a fund is influenced if the realized fund value is lower or higher than the target fund value I constructed two sets of independent variables. In the first set of (OLS and logit) regressions set of independent variables of interest “Under Target “and “Precisely Target” are constructed by dividing the realized fund size by the target size. If the outcome is lower than 1.0 the realized fund size is smaller than the target fund size. If the outcome is between 1 and 1.1, the realized fund value is considered to be the same as target fund value. The 10% buffer is included to ensure that minor differences between realized- and target fund value are not considered as a difference.
Furthermore, I will run a second set (OLS and logit) regression, in which the main independent variable is divided into 6 bins to see whether the effect is stronger for larger differences between realized- and target fund value. The values in “Realized / Target 1” range from 0.1 to 0.49, in “Realized / Target 2” between 0.5 and 1, in “Realized / Target 3” from 1 to 1.1, in “Realized / Target 4” from 1.11 to 1.4, in “Realized / Target 5” from 1.41 to 1.79 and in “Realized / Target 6” from 1.8 to 3.5. The other independent/control variables are similar to the variables used in the first regression. These second set of regression could be biased due
12
to endogeneity since it might be the case that GPs partly decide whether the actual fund size will be larger than the target size. I will conduct research as to what the determinants are in this process and take them in to account when assessing the results. However, I assume that they are not perfectly able to influence fund size. This will also be an important assumption for the quasi-experiment that I will be also using to test the second hypothesis.
I will also use a regression discontinuity design to establish whether the Net IRR “jumps” around the threshold (in this case 1). RDD takes advantage of the fact that the funds, of which the realized fund value is just lower than the target fund and funds that have the same realized fund value as target fund value are similar to each other. As said before, in order to construct a valid RDD it is important that fund managers are not able to exactly influence the realized fund value. I assume that they are not able to do this since investors make the decision whether they invest or not. Furthermore, investors are also not able to precisely know how much capital has been committed to the fund so far.
3) Net IRRi = β0 + β1UnderTargeti + β2∆PreciselyTargeti + β3∆FundValuei+
β4FundValuei + β5lnTotalRelativeInflow it+ β6Perfomanceit-1 + β7FundSequence i +
β8VentureFund i + β9PortfolioSizei + ui
4) P[Yi=1] = β0 + β1UnderTargeti + β2∆PreciselyTargeti + β3∆FundValuei+ β4FundValuei + β5lnTotalRelativeInflow it+ β6Perfomanceit-1 + β7FundSequence i + β8VentureFund i + β9PortfolioSizei + ui
5) Net IRRi = β0 + β1Realized/Target1-6i + β2∆FundValuei+ β3FundValuei +
β4lnTotalRelativeInflow it+ β5Perfomanceit-1 + β6FundSequence i + β7VentureFund i + β8PortfolioSizei + ui
6) P[Yi=1] = β0 + β1Realized/Target1-6i + β2∆FundValuei+ β3FundValuei +
β4lnTotalRelativeInflow it+ β5Perfomanceit-1 + β6FundSequence i + β7VentureFund i + β8PortfolioSizei + ui
*(yi=1 if the fund has a Net IRR of above the mean Net IRR of the total sample, and yi= 0
13
The final set of (logit) regressions assess if initial funds have a better return than follow-up funds. To create a dataset that includes both performance variables and information on initial (first-time) funds two datasets are merged together. The first sample that is used to create the merged set is the complete base sample that I also use for the second regression, the second sample is a sample that only contains first-time funds. The dependent variable is the same as the dependent variable used in the other logit regression, namely “High IRR”. Again, two sets of independent variables are created. The independent variable of interest in the first set is a dummy variable for whether the fund is an initial fund. If the fund is an initial fund the variable “First Fund” takes on the value 1 and zero if the fund is not a first-time fund. The independent variables of interest in the second set are dummy variables for the follow-up funds. In this set the initial fund is excluded and used as a reference point. The other independent/control variables are similar to the variables used in the first regression.
7) P[Yi=1] = β0 + β1FirstFundi + β2∆FundValuei+ β3FundValuei +
β4lnTotalRelativeInflow it+ β5Perfomanceit-1 + β6FundSequence i + β7VentureFund i + β8PortfolioSizei + ui
8) P[Yi=1] = β0 + β1Fund 2-15i + β2∆FundValuei+ β3FundValuei +
β4lnTotalRelativeInflow it+ β5Perfomanceit-1 + β6FundSequence i + β7VentureFund i + β8PortfolioSizei + ui
*(yi=1 if the fund has a Net IRR of above the mean Net IRR of the total sample, and yi= 0 if this is not
14
4. Data
The information on funds, fund managers and fund performance is obtained from Preqin. Preqin is a comprehensive database on the private equity & venture capital industry. I will include buyout funds and venture capital funds of which the general partners. There are no restrictions for location of the fund. The final sample contains vintages between 1978 and 2007. Initially total sample contained 3064 funds. However, because I will look at the growth rate between funds in the same fund sequence I dropped all observations for funds where there was only 1 fund available manually. I also manually dropped every first fund in a sequence because there were no previous funds and therefore also no growth rate. Furthermore, after dropping all vintages before 1980 and after 2007, the total sample contains 1006 funds.
A second sample could be constructed by only including al funds between 1978 and 2014 or including funds of which more than 85% of the capital has been called. These samples could also be used in robustness checks. Using these latter samples construction my sample will contain more recent transactions. After performing the base regressions extra datasets are merged with the base dataset. A dataset containing deal level information on 55409 companies is merged in with the base sample to check whether the amount of, for instance, portfolio companies held by a fund has influence on the performance. Also, an extra data set containing information on the date of first and final closing of funds is merged in to create the variable “Time to Closing”.
Table 1 displays the amount of funds per vintage year divided between the investment types. The amount of funds per investment type has increased significantly (27 times larger) between 1980 and 2007. The table also shows that there are often more buyout funds than venture funds. Moreover, the mean value of a private equity fund increased 875.39% from 1980 and an increase of 2678.52% from 1981. Also, in the last few years in the sample (2005-2007) the amount of funds grew significantly.
Table 2 shows display the summary statistics for the main variables that I will be using in my study. The mean value of the funds in the sample is approximately 575 million euro, while the median is only 272.1 million euro. This could indicate that there are some large outliers. The value of the funds differs significantly. Table 2 shows that the smallest fund in the dataset raised 0.42 million euro, while the largest fund raised approximately 15.9 billion euro. “Change in Fund Value” is a variable that was created to assess the increase/decrease of the value of a fund between a fund and its predecessor. The largest increase of fund value between
15
two funds is 5394%. The increase of fund value between Storm Ventures Fund I and Storm Ventures Fund II of 5.95 million euro to 326.87 million euro is an enormous increase. A quick scan of the raw dataset also shows that the performance of fund II is significantly worse. This finding supports the idea that large increases in fund value depress returns. The mean of “Change in Fund Value “teaches us that on average follow-up funds more than double in size. The variable “Net IRR” displays the IRR net of fees for the funds. Table 2 shows that the lowest Net IRR in my sample is -41.9 and the highest Net IRR is 514.3. Furthermore, the lowest amount of portfolio companies held by a fund in my sample is 1, while the largest amount of portfolio companies held is 90. The largest fund sequence within a single fund family is 15. Also notable is that the largest fund in my sample collected more than EUR 15 billion. If more recent funds are taken into account the fund sizes increase even further.
Table 3 displays the mean fund value, the mean change in fund value and the mean net IRR per investment type. The change in fund value is the main independent variable in most of the regressions and displays the percentage change of fund value between two funds. There are roughly the same amount of buyout funds (517) and venture funds (489) in the final sample. However, the table does show that buyout funds on average are larger than venture funds. It also shows that the mean net IRR is the highest for Early Stage: Seed funds. Lastly, table 4 displays a summary and a short description of the variables that I will use most often in this paper.
16
Vintage
Year Buyout Venture Fund Value 1980 2 1 124.37 1981 0 2 43.66 1982 1 3 113.04 1983 2 3 66.58 1984 2 7 154.58 1985 1 3 56.06 1986 4 5 142.35 1987 4 6 590.48 1988 3 8 267.53 1989 3 14 131.06 1990 2 9 126.82 1991 2 7 69.78 1992 5 11 210.22 1993 8 9 195.45 1994 13 12 298.68 1995 15 12 380.55 1996 12 15 206.37 1997 27 28 537.10 1998 34 31 684.01 1999 26 43 551.16 2000 35 67 844.21 2001 27 34 797.71 2002 27 19 569.28 2003 24 15 820.37 2004 28 18 728.13 2005 73 24 904.43 2006 70 39 1,381.30 2007 67 44 1,213.10
Table 1: Funds and Fund Value. The table displays the amount of
funds per vintage year used in the final sample. Furthermore, the mean value of the fund value per vintage year is displayed. The Fund Value is the fund value at the closing of the fund in millions of euros
17 T able 2 : Descript iv e Sta tis tics. T h e tab le sh o w s th e d escr ip tiv e stati stic s fo r th e sa m p le o f 1 0 0 6 o b ser v atio n s. I t d is p la y s th e m in im u m , m a x im u m , m ea n , m ed ian a n d s ta n d ar d d ev iatio n f o r th e v ar iab les. T h e Fu n d V alu e is t h e fu n d v al u e at th e cl o sin g o f th e fu n d i n m illi o n s o f eu ro s T h e C h an g e in Fu n d V alu e is m ea su red as a p er ce n tag e c h a n g e b et w ee n a fu n d an d its p rec ed in g f u n d . Net IR R is t h e in ter n al rate o f retu rn o f th e fu n d n et o f fee s. P er fo rm a n ce Gr o w th m ea su res t h e in cr ea se/d ec rea se o f retu rn s b et w ee n a fu n d an d its p rec ed in g f u n d . Fu rt h er m o re, th e v ar iab le P o rtf o lio Size m ea su res t h e a m o u n t o f p o rtf o lio co m p a n ies p er f u n d , Fu n d Seq u en ce d en o te s th e seq u e n ce n u m b er o f th e fu n d an d R ea lized Valu e/T ar g et Valu e sh o w s th e relatio n sh ip b et w ee n th e p la n n ed fu n d s ize a n d th e ac tu al fu n d s ize. Va ri ab le Nu m be r M in im um Lo w er Q ua rt ile M ed ia n Up pe r Q ua rt ile M ax im um M ea n St an da rd De vi at io n Ch an ge in Fu nd V al ue 1006 -1 .0 0 0. 15 0. 67 1. 50 53 .9 4 1. 27 3. 08 Ne t I RR 1006 -4 1. 9 2. 8 10 .5 20 .5 51 4. 3 15 .1 1 29 .6 9 Po rt fo lio Si ze 1006 1. 00 1. 00 1. 00 5. 00 90 .0 0 4. 92 8. 95 Fu nd Se qu en ce 1005 1. 00 2. 00 3. 00 5. 00 15 .0 0 3. 92 2. 02 Pe rfo rm an ce G ro w th 1001 -5 10 .8 -1 1. 2 -1 .1 8. 00 32 0. 5 -3 .5 3 37 .8 7 Fu nd V al ue 1000 .4 2 11 5. 05 27 2. 1 65 9. 64 15 93 6. 28 75 7. 43 14 84 .3 2 Re al iz ed V al ue /T ar ge V al ue 423 0. 10 0. 96 1. 05 1. 23 3. 75 1. 09 0. 32 Ti m e to Cl os in g 158 2. 00 12 4. 00 28 6. 5 45 0. 00 10 21 .0 0 30 7. 62 19 9. 71
18
Table 3: Sample per Investment Type. The sample is divided in two investment types: Buyout and Venture.
Venture is further divided into Early Stage, Early Stage: Seed, Early Stage: Start-up, Expansion/Late Stage, Venture (General), and Venture Debt. The Fund Value is the fund value at the closing of the fund in millions of Euros. The Change in Fund Value is measured as a percentage change between a fund and its preceding fund. Net IRR is the internal rate of return of the fund net of fees. The same holds for the IRR of the benchmark.
Type Observations Fund Value Change in Fund Value Net IRR
Buyout 517 1,234 120% 15.51 Early Stage 122 238 146% 6.76 Early Stage: Seed 12 75 648% 20.36 Early Stage: Start-up 8 121 123% 13.55 Expansion / Late
Stage 34 273 62% 16.01
Venture (General) 304 267 117% 17.52 Venture Debt 9 263 123% 14.57
Table 4: Description of Variables
Variables Description Net IRR
Net IRR is the internal rate of return, of which fees for the fund managers are subtracted
∆ Net IRR The % pnt. difference in net IRR between two sequential funds. Change in Fund Value
The percentage change between the realized fund size of a fund and the realized fund size of its preceding fund.
Fund Value The realized fund value in millions of Euro's Fund Sequence The number of a fund in a fund sequence. Total Relative Inflow
The total capital inflow to a specific fund type (buyout or venture) for a particular vintage
Lagged Performance The net IRR of the preceding fund
Venture Fund A dummy variable that takes on a value of 1 if that fund is a venture fund Portfolio Size The amount of portfolio companies held by a fund
Time to Closing The amount of days elapsed between the first closing and the final closing of a fund Change in Fund Value*Venture Interaction variable between the change in fund value and venture funds
19
5. Results
The results for testing hypothesis 1 are displayed in table 5. For every regression, there are 888 observations. To estimate the effect of the change in fund value between two consecutive funds I ran OLS regressions on two different dependent variables. The first dependent variable used is “Net IRR”, which is the IRR of the fund net of fees. The second variable used is the “Difference in Net IRR” between two consecutive funds and is also net of fees. The second dependent variable is used to see how the net IRR changes relative to the previous fund. The main variable of interest in table 5 is the “Change in Fund Value”. The variable measures how much the value of a fund changes between two consecutive funds.
The results in table 5 show a negative relationship between the performance measure (Net IRR, Change in Net IRR) and the” Change in Fund Value”. This implies that a larger change in fund value causes a decrease of performance/return of the fund. Furthermore, the variable is significant in all six regressions, but is only significant on the 5% level if I do not include fixed effects in the regression. Since the “Change in Fund Value” is measured as the percentage change between two funds, the result in, for instance, regression 1 can be interpreted as follows: a 1 percent increase in the “Change of Fund Value” between two consecutive funds leads to a decrease of 0.701% of net IRR. The absolute fund value also has a significant negative effect on the performance measures in all six regressions. In this case, an increase in “Fund Value” leads to a decrease of returns of the fund. However, because the outcome could be driven by a couple of outliers I plotted the fund value divided in ten bins against the mean net IRR of those bins in figure 1. The figure shows that the mean net IRR is the highest for the bins with the lowest absolute fund value and decreases when funds become larger. The variable “Fund Sequence” represents the fund number of the fund in a sequence. If, for instance, the fund is the third fund in a fund family, “Fund Sequence” will be 3.
Table 5 shows that the relationship between “Fund Sequence” and the performance measures is positive. Each consecutive fund adds 5.028% to the net IRR. This indicates that more experienced funds or fund managers realize better returns. It might also present evidence against my second hypothesis that first-time funds perform better than follow-up funds. The variable “Total Relative Inflow” measures the inflow to a certain investment type (either buyout or venture funds) in a specific vintage.
20
Regressions 1-4 find a significant negative effect for the total relative inflow to an investment type from a certain vintage. This could mean that an increased capital inflow during a certain year leads to a decrease in performance for that fund. This is in line with the results found by, among others, Gompers and Lerner (1999) and Diller and Kaserer (2009). They conclude that increased capital flows to an investment type have negative effects on performance due to, for instance, the “money chasing phenomenon”. An increase of inflow of capital to an asset class in a specific vintage year of 1% leads to a decline of more than 9.137% in net IRR. The results for the regression 5 and 6 are negative, but not significant. The relationship between lagged fund performance and the performance measures seems to change from positive to negative between the first three and the last three regressions. However, for the dependent variable for the last three regressions measures the change in “Net IRR” between two consecutive funds. Thus, a negative relationship between “Lagged Performance” and the “Change in Net IRR” does not necessarily mean that the “Change in Net IRR” becomes negative but it can also mean that the “Change in Net IRR” is still large and positive but became smaller because the last return was larger.
The variable “Venture Fund” displays a negative relationship in all three regressions, which is the opposite of the result found by Kaplan and Schoar (2005). The negative relationship might be explained by the larger risk that is associated with venture capital. During a conversation with a partner at a Dutch venture capital firm, the venture capitalist explained that of the 8 portfolio companies they hold, 6 are written off and 2 companies take care of the return of the fund. Because these companies are mostly very young start-ups, it is difficult to assess whether they grow into a profitable firm. However, the variable is only significant in the first three regressions. The coefficient in regressions 1,2 and 3 is large and significant on a 1% level. This tells us that venture funds have a significant lower net IRR than non-venture funds. In the regressions for table 5 I have added two venture fund interaction variables. The variable “Portfolio Size” measures the amount of portfolio companies per fund. The results in table 5 show that an increase in portfolio companies has a positive effect on the performance measures. However, the effect of the amount of portfolio companies on the performance measures is only significant in regression 2 and 3. This means that it is not necessarily detrimental for fund performance to increase the amount of portfolio companies.
Furthermore, in figure 1 and in figure 2 the “Change in Fund Value” is plotted against the mean of the “Difference in Net IRR” and the mean “Net IRR”. In both figures, I created bins based on the “Change in Fund Value”. Figure 1 clearly shows that if a fund grows faster
21
(large change in fund value) the mean “Difference in Net IRR” is negative. According to figure 1 an increase in fund value thus leads to a lower “Net IRR” than last fund. Figure 2 shows how the Change in Fund Value affects the mean net IRR. The figure exhibits that the fastest growing funds perform the worst, but not necessarily that the slowest growing funds perform the best. The scatterplot shows that, on average, the growth of a fund between 0% and 100% leads the highest mean net IRR. In figure 3, I plotted the absolute “Fund Value” against the mean Net IRR. The observations in bin 2 achieve the highest mean net IRR. The fund sizes in bin 2 vary between EUR 50.72 million and EUR 94.83 million.
For the regressions, of which the output is displayed in table 6, I have added two interaction variables. The first interaction variable is an interaction term between “Venture Funds” and ‘Change in Fund Value”, the second is an interaction between “Venture Funds” and “Fund Sequence”. The addition of these extra variables increases the R-squared (the amount of variance in the performance measures explained by the independent variables) slightly, but also increases the statistical and economical significance of the effect of “Change in Fund Value”. The coefficient for the variable “Venture Fund” becomes even more negative after adding the interaction variables, which could mean that, ex ante, venture funds perform worse than buyout firms. However, the interaction variables both show a positive relationship with the interaction variables. The interaction variables should be interpreted as follows: if the fund is a venture fund, the extra difference in “Net IRR” in column 3 for each percentage change in fund value increases the “Net IRR “with 1.326%. Same holds for the interaction between “Fund Sequence” and “Venture Fund” in, for instance, column 1. For each sequential fund the “Net IRR” increases by 1.803%. These results indicate that it is less harmful for venture fund to grow rapidly, but it also indicates that venture funds later in the fund sequence perform better than venture funds earlier on.
In table 7, I ran the same OLS regression as for table 5. However, the observations are divided into four subcategories based on their fund size. The bins are constructed as follows: small funds range between < EUR 301mln, midsized funds between EUR 301mln- EUR 750mln, large funds between EUR 751mln- EUR 2000mln and finally mega funds, which have a value of > EUR 2000mln. The main variable of interest “Change in Fund Value” still has a negative effect on both performance measures but is not statistically significant for small and midsized funds. However, the “Change in Fund Value” is statistically significant and becomes more negative and more economically significant for large and mega funds. This could indicate that the results from table 5 are driven by these two categories of fund sizes. Moreover, the
22
effect is far stronger for large and mega funds than the effect of the coefficient of “Change in Fund Value” in the set of regression from table 5. For mega funds, an 1% increase in fund value leads to a decrease of -2.818% in the Net IRR. The mean Net IRR of the entire sample is 15.11%, so a decrease of 2.818% would mean a decrease of 18.6% of the average Net IRR.
23
Table 5: Effect of Fund Characteristics on Fund Performance. The table displays the results of an OLS
regression. The dependent variable of the first three columns in the internal rate of return net of fees (Net IRR). The dependent variable for the last three columns, Performance Growth, is measured as the difference between the Net IRR minus the benchmark IRR of a fund and the Net IRR minus the benchmark of the preceding fund. The independent variable of interest is Change in Fund Value, which is measured as de percentage difference between a fund and its preceding fund. For the variables: Fund Value, Fund Sequence, and Total Relative Inflow the natural logarithm is used. Lagged Performance is the Nett IRR minus the benchmark of the previous year. Venture Fund is a dummy variable that takes on the value 1 if the fund is a Venture Fund and 0 otherwise. The final independent variable is Portfolio Size, which measures the amount of portfolio companies in a fund.
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
(1) (2) (3) (4) 5) (6)
’80-‘07 ’80-‘07 ’80-‘07 ’80-‘07 ’80-‘07 ’80-‘07
Net IRR Net IRR Net IRR ∆ Net IRR ∆ Net IRR ∆ Net IRR
Change in Fund Value -0.707** -0.626* -0.653* -0.650** -0.560* -0.567*
(0.324) (0.330) (0.339) (0.307) (0.317) (0.326)
Fund Value -2.238** -2.041** -2.259** -1.715** -1.517* -1.753*
(0.917) (0.910) (0.999) (0.868) (0.875) (0.961)
Fund Sequence 5.020** 4.686** 5.028** 4.654** 4.457** 4.754**
(2.348) (2.315) (2.359) (2.223) (2.226) (2.271)
Total Relative Inflow -7.585*** -8.452*** -9.137*** -3.019*** -1.197 -1.774
(0.930) (3.052) (3.126) (0.881) (2.935) (3.009) Lagged Performance 0.210*** 0.216*** 0.215*** -0.786*** -0.790*** -0.792*** (0.0287) (0.0283) (0.0286) (0.0271) (0.0272) (0.0276) Venture Fund -18.37*** -18.94*** -19.00*** -3.527 0.116 -0.465 (2.751) (5.512) (5.628) (2.604) (5.300) (5.418) Portfolio Size 0.185 0.217* 0.249** 0.0785 0.0946 0.121 (0.122) (0.123) (0.127) (0.116) (0.118) (0.122) Constant 104.2*** 84.38*** 87.89*** 38.21*** 14.39 18.36 (9.646) (29.51) (33.54) (9.132) (28.38) (32.29)
Vintage Fixed Effects No Yes Yes No Yes Yes
Location Fixed Effects No No Yes No No Yes
Observations 888 888 888 888 888 888
24
Table 6: Effect of Fund Characteristics on Fund Performance with interaction variables. The table displays the results of
an OLS regression. The dependent variable of the first three columns in the internal rate of return net of fees (Net IRR). The dependent variable for the last three columns, Performance Growth, is measured as the difference between the Net IRR minus the benchmark IRR of a fund and the Net IRR minus the benchmark of the preceding fund. The independent variable of interest is Change in Fund Value, which is measured as de percentage difference between a fund and its preceding fund. For the variables: Fund Value, Fund Sequence, and Total Relative Inflow the natural logarithm is used. Lagged Performance is the Nett IRR minus the benchmark of the previous year. Venture Fund is a dummy variable that takes on the value 1 if the fund is a Venture Fund and 0 otherwise. The final independent variable is Portfolio Size, which measures the amount of portfolio companies in a fund.
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
(1) (2) (3) (4) (5) (6)
’80-‘07 ’80-‘07 ’80-‘07 ’80-‘07 ’80-‘07 ’80-‘07 Net IRR Net IRR Net IRR ∆ Net IRR ∆ Net IRR ∆ Net IRR
Change in Fund Value -1.522** -1.569** -1.647** -1.328** -1.306** -1.338** (0.638) (0.653) (0.664) (0.604) (0.628) (0.640)
Fund Value -2.081** -1.872** -2.006** -1.585* -1.385 -1.553
(0.921) (0.915) -1,007 (0.872) (0.880) (0.970)
Fund Sequence 1.219 2.139 2.663 0.831 1.569 2.075
(3.145) (3.105) (3.171) (2.977) (2.986) (3.053) Total Relative Inflow -7.987*** -9.408*** -10.16*** -3.384*** -1.982 -2.588
(0.949) (3.096) (3.173) (0.898) (2.977) (3.055) Lagged Performance 0.216*** 0.222*** 0.222*** -0.782*** -0.785*** -0.787***
(0.0291) (0.0287) (0.0290) (0.0275) (0.0276) (0.0279)
Venture Fund -26.81*** -26.17*** -25.94*** -11.81** -7.270 -7.436
(5.011) (7.063) (7.184) (4.744) (6.792) (6.918)
Change in Fund Value*Venture 1,100 1.251* 1.326* 0.919 0.988 1,025
(0.737) (0.750) (0.765) (0.697) (0.721) (0.737) Fund Sequence*Venture 1.803* 1.159 1.054 1.829* 1.353 1.237 (1.039) (1.027) (1.044) (0.984) (0.987) (1.005) Portfolio Size 0.206* 0.228* 0.257** 0.101 0.110 0.133 (0.123) (0.124) (0.127) (0.116) (0.119) (0.123) Constant 112.7*** 93.34*** 96.31*** 46.35*** 23.01 26.25 (10.41) (29.88) (33.84) (9.851) (28.74) (32.58)
Vintage Fixed Effects No Yes Yes No Yes Yes
Location Fixed Effects No No Yes No No Yes
Observations 888 888 888 888 888 888
25
Table 7: Effect of Fund Characteristics on Fund Performance per Fund Size. The table displays the results of an OLS regression. The dependent variables of the first two columns are the internal rate of return net of fees (Net IRR) and Performance Growth, which is measured as the difference between the Net IRR minus the benchmark IRR of a fund and the Net IRR minus the benchmark of the preceding fund for small funds (<€301mln). The next two columns are for midsized funds (€301mln-€750mln), then large funds (€751mln-€2000mln) and the last two for mega funds (>€2000mln). The independent variable of interest is Change in Fund Value, which is measured as de percentage difference between a fund and its preceding fund. For the variables: Fund Value, Fund Sequence, and Total Relative Inflow the natural logarithm is used. Lagged Performance is the Nett IRR minus the benchmark of the previous year. Venture Fund is a dummy variable that takes on the value 1 if the fund is a Venture Fund and 0 otherwise. The final independent variable is Portfolio Size, which measures the amount of portfolio companies in a fund.
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
(1) (2) (3) (4) (5) (6) (7) (8)
<€301mln <€301mln
€301mln-€750mln €301mln-€750ml €751mln-€2bn €751mln-€2bn >€2bn >€2bn
Net IRR ∆ Net IRR Net IRR ∆ Net IRR Net IRR ∆ Net IRR Net IRR ∆ Net IRR
Change in Fund Value -0.695 -0.670 -0.141 -0.195 -0.938* -1.236** -2.818*** -2.426*** (0.690) (0.660) (0.262) (0.231) (0.561) (0.542) (0.846) (0.673) Fund Value -1.759 -1.284 -2.853 -4.060 -0.545 -1.424 1.755 1.112 (2.000) (1.914) (4.540) (4.008) (4.196) (4.056) (2.394) (1.905) Fund Sequence 12.52*** 11.65*** 1.523 -0.132 -2.041 -1.775 0.467 0.899 (4.178) (3.998) (2.638) (2.329) (2.659) (2.570) (2.473) (1.968) Total Relative Inflow -8.364*** -3.324** -4.763*** -1.445 -3.478** -1.678 -8.202*** -3.792** (1.447) (1.385) (1.278) (1.129) (1.402) (1.355) (1.890) (1.504) Lagged Performance 0.397*** -0.613*** 0.0361 -0.956*** 0.113 -0.824*** 0.241*** -0.790*** (0.0513) (0.0491) (0.0238) (0.0210) (0.0785) (0.0758) (0.0844) (0.0671) Venture Fund -19.14*** -4.444 -16.43*** -1.025 -14.62*** -0.254 -12.94 2,080 (4.777) (4.572) (2.912) (2.571) (3.272) (3.163) (7.825) -6,226 Portfolio Size 0.233 0.0225 0.0845 0.0616 0.0378 0.0261 0.0636 0.00661 (0.519) (0.497) (0.129) (0.114) (0.127) (0.122) (0.0708) (0.0563) Constant 101.0*** 31.51** 80.62*** 40.48 58.59* 30.49 91.59*** 33.61* (15.42) (14.76) (29.07) (25.66) (35.07) (33.90) (25.20) (20.05) Observations 476 476 204 204 116 116 92 92 R-squared 0.192 0.280 0.163 0.919 0.231 0.846 0.262 0.688
26
Figure 1: Two-way plot of the Change in Fund Value against Mean Change of Net IRR. In the figure the
percentage change in Fund Value in millions of Euro’s between two funds is plotted against the Mean Change in Net IRR between two funds. The funds are divided in ten bins according the amount of change (in percentages) value between a fund and its preceding fund. The Net IRR is the IRR net of fees and minus the IRR of its benchmark
Figure 2: Two-way plot of the Change in Fund Value against Mean Net IRR. In the figure, the percentage
change in Fund Value in millions of Euro’s between two funds is plotted against the Mean Net IRR. The funds are divided in ten bins according the amount of change of value between a fund and its preceding fund. The Mean Net IRR is the IRR net of fees
Figure 3: Two-way plot of Fund Value against Mean Net IRR. In the figure, the Fund Value in millions of
Euro’s is plotted against the Mean Net IRR. The funds are divided in ten bins according to their fund value. The Mean Net IRR is the IRR net of fees.
27
To test my second hypothesis if the performance of a fund decreases when the target fund value does not correspond to the realised fund value I first set up a regression discontinuity design to graphically show whether or not the gap between target and realised fund value has an influence on performance. For the regression discontinuity design (RDD) the threshold lies at 1, which means that realized fund value is the same as target fund value (realized fund value / target fund value = 1). A very important condition of a RDD is that that the fund manager cannot exactly determine the realised fund value. In this case, I assume that fund managers can to some extend influence the amount of capital that they collect since they usually adapt a hard-cap, a maximum of funds they collect. However, this hard-cap is always larger than the target fund value and thus does not affect my threshold of 1. The total amount of capital dedicated to a fund is ultimately determined by investors. A RDD evades endogeneity because funds just above and just below are assumed to be similar funds.
Figure 4 and figure 5 both show the outcome of a RDD but in figure 5 both “Realized Fund Value/ Target Fund Value” and “Net IRR” are winsorized at 10% and 90% to exclude any outliers that may drive the outcome. In both figures a distinct “jump” is displayed at the threshold of 1 which could indicate that if realized fund value is lower than target fund value, funds perform worse. A possible explanation would be that investors do not have confidence in the general managers of the fund because of, for instance, their past performance. In that case investors seem to be right in their assumption. However, above the threshold the Net IRR stays higher until a ratio of around 1.6, after which it sharply declines. The same jump in Net IRR at the threshold of 1 is portrayed in figure 5.
To double check this relationship, I ran an OLS regression (column 1 and 2) of Net IRR on the main independent variable of interest: “Realized Fund Value / Target Fund Value” and a logit regression (column 3 and 4) on “High IRR”, which takes on the value of 1 if the IRR is above the mean IRR of the total sample on the same independent variables. The results are shown in table 8. To assess whether larger gaps between actual fund value and target value have a different effect on the performance, I divided the independent variable in 6 different bins and in 3 different bins (other ranges). None of the variables “Realized/Target 1-6” are statistically significant. However, it should be noted that the sign of the coefficient changes from negative to positive for “Realized/Target 3”, which is the bin that ranges from 1 until 1.1. The coefficient for bin 4 and 5 are also positive, but the coefficient turns negative again for bin 6. This is similar
28
to what is displayed in figure 4 and 5. In column 2 an OLS regression of Net IRR on “Under Target” and “Precisely Target” and some additional control variables is shown. The third dummy “Over Target” is left out to avoid the dummy variable trap. Again, the coefficient for “Under Target” is negative but not significantly significant.
In column 3 and 4 a logit regression is performed. The dependent variable “High IRR” takes on the value 1 if the IRR is above the mean IRR of the total sample. The main independent variables of interest are again the “Realized/ Target 1-6” and “Under Target” and “Precisely Target”. The coefficients are similar to the regressions in showed in column 1 and 2. In column 4 however, the coefficient for “Under Target” shows a statistically and economically significant effect on the probability of having an IRR above the mean IRR of the total sample. This means that if realized fund value is lower than the target fund value the probability of having a high IRR is 49.1% lower.
29
Figure 4. Change in Net IRR around the threshold. The figure shows the difference in net IRR when realized
fund value is equal to target fund value. The sample consists of 405 funds.
Figure 5. Change in Net IRR around the threshold (winsorized). The figure shows the difference in net IRR
30
Table 8: The Influence of the Realized Fund Size on Performance. The table displays the results of both an OLS and logit
regression. The dependent variables of the first two columns is the internal rate of return net of fees (Net IRR). The dependent variable for the last two columns is the binomial variable “High IRR”. “High IRR” takes on the value 1 if the net IRR is amongst the top half performing funds. The independent (dummy) variables of interest are “Realize / Target 1-6”, “Under Target” and “Precisely Target”. Furthermore, Change in Fund Value, which is measured as de percentage difference between a fund and its preceding fund. For the variables: Fund Value, Fund Sequence, and Total Relative Inflow the natural logarithm is used. Lagged Performance is the Nett IRR minus the benchmark of the previous year. Venture Fund is a dummy variable that takes on the value 1 if the fund is a Venture Fund and 0 otherwise. The final independent variable is Portfolio Size, which measures the amount of portfolio companies in a fund.
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
(1) (3) (2) (4)
Net IRR Net IRR High IRR High IRR
Change in Fund Value -0.666* -0.667** -0.155*** -0.155***
(0.340) (0.339) (0.0525) (0.0525)
Fund Value -2.501** -2.383** -0.234** -0.203**
(1.026) (1.004) (0.0943) (0.0917)
Fund Sequence 5.095** 5.059** 0.178 0.143
(2.377) (2.360) (0.221) (0.219)
Total Relative Inflow -9.320*** -9.330*** -1.803*** -1.789***
(3.145) (3.134) (0.319) (0.317) Lagged Performance 0.215*** 0.214*** 0.0132*** 0.0131*** (0.0289) (0.0287) (0.00375) (0.00370) Venture Fund -19.46*** -19.34*** -4.407*** -4.352*** (5.664) (5.639) (0.606) (0.603) Portfolio Size 0.251* 0.251** 0.000691 0.00103 (0.128) (0.127) (0.0112) (0.0110) Realized/Target 1 -9.478 -0.978 (8.895) (0.926) Realized/Target 2 -1.787 -0.200 (3.702) (0.344) Realized/Target 3 2.084 0.342 (3.797) (0.339) Realized/Target 4 0.706 0.470 (3.677) (0.327) Realized/Target 5 0.909 0.354 (5.753) (0.499) Realized/Target 6 -3.978 0.606 (12.14) (1.136) Under Target -2.805 -0.491* (3.133) (0.294) Precisely Target 1.694 0.0759 (3.282) (0.286) Constant 90.06*** 89.79*** 21.68*** 21.59*** (33.72) (33.60) (3.879) (3.861) Observations 888 888 853 853 R-squared 0.232 0.231
31
To test the third hypothesis, whether the first fund in a sequence performs best, is correct I perform a logit regression with “High IRR” as the dependent variable. Table 9 shows the outcome of this set of logit regressions. The first two columns display the outcome for a set of vintages from between 1980 and 2007, the second two columns for a set of vintages from between 1978 and 2014. In the first regression, “High IRR” is regressed on the dummy variable “First Fund”. “First Fund” takes on the value 1 if the fund is the first fund in the sequence. The other independent (control) variables are similar to the variables used in the previous regressions. The coefficient of “First Fund” (0.375) is positive and statistically and economically significant. This indicates that the first fund of a fund sequence has an increased probability of 37.5% to perform above the mean fund performance of the total sample.
In the second regressions, the dependent variable is regressed on a dummy variable for all other fund sequences. The coefficient for the dummy variable of the second, third and sixth fund are statistically and economically significant. The probability of being in the top half performing funds for these funds decreases respectively by 46.5%, 33.6% and 83.1%. The second two regressions use a larger sample (increase of 538 observations). The effect of “First Fund” is also positive and statistically and economically significant when using a larger sample set. In this case, first funds have an increased chance of 48.6% of having a net IRR of above the mean net IRR of the total sample. In regression 4, the coefficients for variable the second, third, fifth and sixth fund are statistically and economically significant. The chances of being an above mean performer decreases respectively with 52.2%, 56.6% 52.8% and 55.9% for these funds.
32
Table 9: Do First-time Funds Perform better. The table displays the results of a set of logit regressions. The dependent
variable for all four columns is the binomial variable “High IRR”. “High IRR” takes on the value 1 if the net IRR is amongst the top half performing funds. The independent (dummy) variables of interest are “First Fund”, “Second Fund”, “Third Fund” etcetera. For the variables: Fund Value, Fund Sequence, and Total Relative Inflow the natural logarithm is used. Lagged Performance is the Nett IRR minus the benchmark of the previous year. Venture Fund is a dummy variable that takes on the value 1 if the fund is a Venture Fund and 0 otherwise. The final independent variable is Portfolio Size, which measures the amount of portfolio companies in a fund
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
(1) (2) (3) (4)
High IRR High IRR High IRR High IRR
First Fund 0.375*** 0.486*** (0.143) (0.133) Second Fund -0.465*** -0.522*** (0.170) (0.153) Third Fund -0.336* -0.566*** (0.188) (0.165) Fourth Fund -0.143 -0.256 (0.213) (0.183) Fifth Fund -0.369 -0.528** (0.252) (0.217) Sixth Fund -0.831*** -0.559** (0.309) (0.246) Seventh Fund 0.0635 -0.308 (0.345) (0.287) Eighth Fund 0.132 -0.252 (0.443) (0.349) Ninth Fund -1.666 -0.296 (1.073) (0.459) Tenth Fund - -0.977 (0.620) Eleventh Fund -0.809 -0.00997 (1.370) (0.656) Twelfth Fund 1.559 1.274 (1.533) (1.536) Fourteenth Fund - 0.869 (1.272) Fifteenth Fund - 0.960 (1.531) Fund Value -0.156*** -0.170*** -0.134*** -0.152*** (0.0568) (0.0593) (0.0465) (0.0490) Total Relative Inflow -1.121*** -1.106*** -1.149*** -1.134***
(0.200) (0.202) (0.171) (0.172) Venture Fund -2.798*** -2.800*** -2.720*** -2.732*** (0.351) (0.355) (0.324) (0.326) Portfolio Size 0.0107 0.0120 0.0127** 0.0131** (0.00859) (0.00881) (0.00634) (0.00647) Constant 8.482*** 8.934*** 13.07*** 13.43*** (1.724) (1.722) (2.056) (2.063)
Vintage Fixed Effects Yes Yes Yes Yes
Location Fixed Effects Yes Yes Yes Yes
33
6. Robustness checks
To assess whether my findings are robust, I re-run a selection of regressions using different samples and different performance measures. The different samples are constructed by, for instance, only using funds that focus on the U.S. or Europe, by extending the vintage years in the dataset, and by winsorizing the sample on a 10% / 90% level for “Change in Fund Value”. Furthermore, I will replicate a part of the regressions done by Kaplan and Schoar (2005) and extend their sample to check whether the effect changes for later time-periods. They claim that funds in their sample do not grow fast enough to have an influence on the performance, but they hypothesize that fund growth will be larger for later vintages (not included in their sample). I will investigate whether this is also the case for my sample and for vintages after 1995.
Table 10 shows the outcome of regressions similar to the base regression (table 5). However, the sample that I used in the regressions for table 10 is winsorized on “Change in Fund Value” to make sure that the results are not driven by outliers. The first two columns show the outcome for a sample that includes vintages between 1980 and 2007, the second two columns show results for vintages between 1978 and 2014. The coefficient for “Chang in Fund Value” is more negative than for the same coefficient in the base regression. The same growth in fund value should thus have a stronger effect on the performance of a fund. However, only the coefficients in column 2, 3 and 4 are statistically significant. The effect of a 1% “Change of Fund Value” on “∆ Net IRR “with vintage and location fixed effects leads to a decrease of Net IRR with 2.128 percentage points. All the other variables have similar effects on the performance measures as in the main regression. A 1% increase in “Fund value”, for instance, leads to a decrease of 2.08 percent in net IRR. For the second two columns in table 10, the effect of “Change in Fund Value” becomes even more statistically significant and more negative.
The results in table 11 are for the same OLS regression that is used to test hypothesis 1 but in this case the dataset is extended by including vintages from 1978 until 2014. For the main variable of interest “Change in Fund Value” the effects on the performance measures are more negative and are more significant. This again means that faster growth of a fund lowers the performance even more than in the main regression. The effect of absolute fund value on performance in the robustness test is more significant in regression 1,2,3,5 and 6 but the effect is less strong. Also, the variables “Fund Sequence” and “Total Relative Inflow” show somewhat less strong results, but are still statistically and economically significant.
