Preliminary Analyses

First I analyse whether there is evidence in favour of performance persistence in the sample by running regressions based on the methodology of Kaplan and Schoar (2005). This is important because as a consequence of performance persistence, funds often do more and/or bigger deals. Table 5 displays the regression results. The regressions control for year fixed effects. The results indicate strong persistence across subsequent buyout funds from the same private equity firm. The coefficient for the first lag of net IRR in column 1 is positive and strongly significant. This is in line with the findings from Kaplan and Schoar (2005). A coefficient of 0.160 means that an increase in fund performance of 1% is associated with an increase in performance of 16 basis points of the subsequent fund. In column 2 also the second lag is included. The results in column 2 show that the coefficient on the first lag is again positive and strongly significant, indicating that an increase in fund performance is associated with higher performance of the subsequent fund as well. However, the coefficient on the net IRR of second previous fund is not significant. This suggests that an increase in the performance of the second previous fund does not have a significant impact on the

performance of the subsequent fund. In regression 3 only the second lag is included, as suggested by Kaplan and Schoar (2005). They explain that overlapping investments between funds can induce performance persistence. When a subsequent fund contains investments that the previous fund also engaged in, this may be the cause of the performance persistence and thus bias the results. According to them the possibility of an overlap between the second previous fund and the current one is highly unlikely because there are often six years in between the two subsequent funds. The positive and significant coefficient of 0.064 indicates that an increase in fund performance of 1% of the second previous fund is associated with an increase in performance of 6.4 basis points in subsequent funds. The fact that this coefficient is again positive and significant shows that the results are not biased by overlapping

investments between funds. It is important to note however that the coefficient on the second lag in model 3 is only significant at a 5% and 10% level, whereas the coefficients on the first lag in model 1 and 2 are also significant at a 1% level. The coefficient on the third previous fund is not significant. This may be due to the fact that the sample size is much smaller for

this regression. Model 4 only includes 544 buyout funds, which is relatively small compared to the number of observations of the other three models. Overall these results can be seen as evidence in favour of performance persistence and they support hypothesis 1 that this persistence is strongest for the two subsequent funds and declines when you look at the performance persistence between a current fund and funds from further in the past. A possible explanation for the existence of performance persistence is that it is hard for new funds to compete with already existing funds (Kaplan and Schoar, 2005). The qualities and skills of general partners can differ a lot. If that is the case, established funds have several advantages compared to new funds. Firstly they often have proprietary access to specific deals. So funds that did well in the past may have access to better investments than new funds. Furthermore, returns can persist when high-quality general partners are scarce. In that case only a limited number of general partners can provide valuable management and advisory inputs. Lastly, funds that perform well now are also more likely to attract more and bigger funds in the future which can help them to perform well in subsequent funds.

Table 5: Performance Persistence

The dependent variable is Net IRR. The variables Net IRRt-1, Net IRRt-2 and Net IRRt-3 are lagged realised Net IRRs of a private equity firm’s previous fund, the fund before that and the one before that respectively. Size is the amount in millions of USD that is committed to the fund. Sequence is the sequence number of the fund. All four models include year fixed effects.

(1) (2) (3) (4)

Net IRR Net IRR Net IRR Net IRR

Net IRRt-1 0.160*** 0.158***

(0.036) (0.048)

Net IRRt-2 0.038 0.064**

(0.027) (0.030)

Net IRRt-3 0.047

(0.029)

Log(Size) -0.540 -0.055 -0.097 0.075

(0.376) (0.428) (0.376) (0.491)

Log(Sequence) 0.763 -0.159 -0.212 0.729

(0.789) (0.966) (0.896) (1.097)

_cons 15.066*** 12.646*** 15.243*** 12.825***

(2.591) (3.427) (2.798) (3.799)

Firm fixed effects No No No No

Year fixed effects Yes Yes Yes Yes

R-squared 0.133 0.130 0.104 0.109

Adjusted R-squared 0.110 0.096 0.072 0.062

Obs. 1146 770 802 554

Standard errors are in parenthesis and they are corrected for heteroskedasticity and clustered by year.

*** Significant at the 1 percent level.

** Significant at the 5 percent level.

* Significant at the 10 percent level.

Table 6 displays the results from the regressions regarding fund performance and fund characteristics. Columns 1 and 2 only include year fixed effects and show the cross-sectional relations between size, sequence and fund performance. Columns 3 and 4 estimate the same relations as the first two columns but include both year fixed effects as well as private equity firm fixed effects. Columns 5 and 6 exclude both types of fixed effects. The adjusted R²’s in model 3 and 4 are a lot higher compared to the other models. This suggest that the average performance varies a lot between private equity firms. In all 6 regression models the

coefficient on size is negative and strongly significant. This indicates that an increase in the value/size of the fund leads to a decrease in performance as measured by the net IRR. The coefficient on “Log(Size)” in column 1 indicates that for each increase in size by 10% the expected net IRR decreases by 5.2 basis points (that is -1.259 * log(1.10) = -0.052). By including the squared term of size in column 2, the coefficient on the linear term of size decreases and the squared term is positive and significant. From columns 3 and 4 we come to a similar conclusion. The coefficient on the linear term of fund size decreases from model 3 to model 4, and the coefficient on the squared term of size is positive and only significant at the 10% level. So the effect of including the squared term of size is similar as with the cross-sectional models in columns 1 and 2. The results from table 6 regarding fund size are

evidence against hypothesis 2. Instead of a positive and concave relation between size and performance, the results suggest that there is a negative relation. This result is different from what Kaplan and Schoar (2005) found. Kaplan and Schoar (2005) concluded that larger funds also have higher returns in the cross-section. With regards to the regressions that include both year fixed effects and firm fixed effects the results are in line with those from Kaplan and Schoar (2005). Their results also indicate a significant negative coefficient for fund size when firm fixed effects were also controlled for. There are many other papers that came to a similar conclusion and offer explanations for the negative relation between size and performance.

For example, Humphery-Jenner (2012) and Lopez-de-Silanes et al. (2015) also conclude that fund performance decreases with fund size. They mention the limited attention theory as a potential explanation. This means that an increase in fund size is associated with funds reducing the number of staff per portfolio company and the attention given to them

(Humphery-Jenner, 2012). According to Lopez-de-Silanes et al. (2015) it also leads to more communication issues. In case of larger funds, it can take more time before decisions are made, information might get lost when employees report to other partners and it takes longer before people have passed on the right information to the right people. These communication issues can lead to lengthier discussions and as a consequence a slower and less efficient

decision making process. Humphery-Jenner (2012) also mentions that an increase in fund size can lead to funds investing in industries where they have less expertise and no competitive advantage, which is associated with lower performance. Furthermore larger funds are likely to experience more agency conflicts and as a consequence engage in more value-destroying buyouts (Cumming & Dai, 2011; Humphery & Jenner, 2012). Cumming and Dai (2011) also add that larger funds are more likely to overpay for buyouts and therefore earn lower returns.

The results in column 1 also indicate that funds with a higher sequence number have a higher net IRR, although the effect is not significant. When we include the squared term of sequence in column 2, it appears that the relation between sequence and fund performance is convex, although not significantly so. This is what Kaplan and Schoar (2005) found as well.

The coefficient on sequence is again positive but decreases from column 1 to column 2 and the coefficient on the squared term of sequence is also positive and not significant. When both types of fixed effects are included in columns 3 and 4, the results are slightly different.

The coefficients on sequence are now negative in both models. This could indicate that the effect of sequence on performance captured private equity firm characteristics rather than the true effect. The result is not surprising, given that the results of Kaplan and Schoar (2005) showed a similar pattern. It indicates that raising a subsequent fund by the same private equity firm is associated with a decrease in fund performance, although this effect is not significant. The effect of including the squared term of sequence is similar as with the cross-sectional models. The coefficient on sequence increases and the squared term of sequence is again positive but not significant. So the results suggest that in the cross-section, higher sequence numbers are associated with higher performance, whereas the relation between sequence and performance is negative when we control for firm fixed effects. The fact that the coefficients on sequence are never significant is contrary to what I expected. The signs of the coefficients however are in line with my hypothesis. A possible explanation for the fact that raising another fund by the same private equity firm has a negative effect on performance is that the private equity market became more mature and competitive (Braun et al., 2017).

Because of this the number of general partners that can provide valuable management and advisory input is not so limited anymore and established funds do not necessarily have better access to good investments. In that way the maturing market and increased competition can take away the advantages that established funds otherwise have over new ones. Furthermore it may be the case that older funds, so funds with a high sequence number, feel less pressure

to establish a track record/reputation in the industry and will therefore be less focussed and put in less effort, which leads to lower returns.

Regressions 5 and 6 (without fixed effects) are included to assess whether the variation in performance comes from fixed unit characteristics. From these two columns a number of things can be concluded. Firstly, the adjusted R-squared is a lot lower compared to those of the first four models. Secondly, the conclusions regarding the variables of interest remain the same: size has a negative impact on performance and sequence does not have a significant influence. In addition, the magnitude of the coefficients differ relatively little from that of the variables in the other models. This all indicates that fixed unit characteristics do not account for much of the variation in the dependent variable and it strengthens the reliability of the results.

Table 6: The Effect of Fund Characteristics on Performance

The dependent variable in all six models is net IRR: the internal rate of return, net of fees. Fund size is the amount in millions of USD that is committed to the fund. Sequence represents the sequence number of the fund. Columns 1 and 2 include only year fixed effects. Columns 3 and 4 include both firm fixed effects and year fixed effects. Columns 5 and 6 do not include any type of fixed effects.

(1) (2) (3) (4) (5) (6)

Net IRR Net IRR Net IRR Net IRR Net IRR Net IRR

Log(Size) -1.259*** -7.690*** -4.344*** -11.688*** -1.556*** -7.810***

(0.274) (1.735) (0.684) (3.600) (0.299) (1.781)

Log(Size)² 0.517*** 0.571* 0.503***

(0.129) (0.288) (0.134)

Log(Sequence) 0.563 0.013 -2.989 -2.908 0.406 -0.601

(0.532) (0.887) (2.370) (2.547) (0.503) (0.833)

Log(Sequence)² 0.126 0.226 0.300

(0.238) (0.590) (0.237)

_cons 23.230*** 42.704*** 48.072*** 70.107*** 25.389*** 44.498***

(1.576) (5.549) (4.492) (10.410) (1.850) (5.786)

Firm fixed effects No No Yes Yes No No

Year fixed effects Yes Yes Yes Yes No No

R-squared 0.115 0.122 0.453 0.457 0.016 0.023

Adjusted R-squared 0.098 0.103 0.231 0.235 0.015 0.021

Obs. 1694 1694 1475 1475 1720 1720

Standard errors are in parenthesis and they are corrected for heteroskedasticity and clustered by year.

*** Significant at the 1 percent level.

** Significant at the 5 percent level.

* Significant at the 10 percent level.