Beyond the Quartiles. Understanding the How of Private Equity Value Creation to Spot Likely Future Outperformers. Oliver Gottschalg HEC Paris

Transcription

1 Beyond the Quartiles Understanding the How of Private Equity Value Creation to Spot Likely Future Outperformers Oliver Gottschalg HEC Paris July 2016 This Paper was prepared for a Practitioner Audience Please address all correspondence to: Oliver GOTTSCHALG, HEC School of Management, Paris; Jouy-en-Josas, France; gottschalg@hec.fr

2 Introduction Private equity (PE) performance measurement is by design initially about the how much of value creation of a given fund manager. It is implicitly backward-looking as today s investors cannot buy a fund s past returns but only commit to a manager s future fund with the hope of obtaining high returns from this investment. For measures based on traditional performance measures (IRR and Multiple of Invested Capital), a growing number of studies show that historic performance is not a particularly useful indicator of future returns 1. Whenever detailed cash flows are available, there are more advanced measures of past performance that provide a meaningful level of guidance to likely future outperformers, but for many funds, such data is unavailable to many investors. The present paper proposes a novel methodology to measure four distinct components of a PE fund s performance, three of which are proxy indicators for the underlying capabilities of the fund manager. This approach compares data on tens of thousands of underlying investments made by thousands of PE funds to identify funds that are similar according to three different factors: similar in the time period in which the funds were raised (three-year vintage cycle) (Group A); similar in terms of when investments were made based on acquisition years of the underlying deals (Group B); and similar in terms of the type of target companies that were chosen (i.e. similarity in terms of their industry sector, size category and geographic location) to be acquired at those times (Group C). Starting from the funds raised in the same three-year vintage year cycle, the other two peer groups correspond to sequential filtering for peer funds that are similar to a given fund in terms of two key investment decisions of a PE firm: The Timing or The When, i.e. the decision of how investments were timed over the investment period of the fund; and the Strategy or The What, i.e. the decision of which type of target companies were chosen (industry sector, size category, geography) to acquire at those times. We can therefore assess the performance impact of the Timing and Strategy decisions by comparing the average performance of Groups A, B & C. Finally, comparing the actual performance (TVPI) of a fund with that of similar funds (in terms of what they acquired and when they acquired it i.e. Group C), reflects the portion of a fund s overall performance that is attributable to The Who and How decisions, which are really the "key ingredients" of implementation of the fund, i.e. the performance of the specific investment approaches into specific target companies above the returns of other, similarly invested funds. Fig. 1 shows Venn diagrams comparing the vintage year approach with the nested samples of similar, competitor funds. Fig. 2 illustrates these different performance components based on hypothetical data. 1 See, for example Harris, Robert S. and Jenkinson, Tim and Kaplan, Steven N. and Stucke, Rüdiger, Has Persistence Persisted in Private Equity? Evidence from Buyout and Venture Capital Funds (February 28, 2014). Darden Business School Working Paper No ; Fama-Miller Working Paper. Available at SSRN: or

3 Fig. 1: Comparing a fund to sets of competitors with different levels of similarity Vintage Year Peers Fund to be benchmarked Funds with similar timing and similar strategy Funds with similar timing All funds raised in same period (Vintage Year +/-2 years) Fig. 2: Illustrative example of performance components MOIC x 0.25 x x 0.70 x 2.00 x Avg performance of all funds raised in the same VY period Impact of Timing Impact of Strategy Impact of Implementation TVPI Based on the breakdown of overall performance into these components, we can not only quantify the magnitude of the performance impact of these choices, but also provide benchmark statistics with annual quartile cut-offs for each of these components, making it possible for investors to determine to what extent a given fund manager has been a top-quartile value generator within each of these categories. Finally, we conduct a series of analyses to assess the degree to which a fund manager s performance persists in aggregate, and within each of the value creation components. Importantly, this provides the basis for the identification of factors that historically would have enabled investors to identify subsequent outperformers with a reliability that would have led to a statistically significant and economically relevant level of performance improvement in the portfolio.

4 Summary of Key Findings We assessed data from Preqin on 1,119 funds for which performance data and information on investment year, size category, industry sector and geography for at least five underlying investments was available. From this data, we then identified 257 pairs of buyout funds raised by the same GP and with the same strategy and regional focus, immediately following each other in the sequence of fundraises 2. By always comparing the first fund to the corresponding subsequent fund in these 257 pairs, we found that statistically significant positive persistence occurs only in the implementation component of returns. While past TVPI does not provide insight into subsequent IRR, other components of performance have no significant (timing, When) or a significant negative (strategic choice, What) level of persistence. This finding is not only statistically relevant but also meaningful in terms of its magnitude: had an investor picked 25 of the 257 funds using only those with the highest previous implementation-based returns, the portfolio would have yielded 2.03x, which is significantly higher than if they had picked top performing funds by TVPI, which returned only 1.73x. Even if the sample was extended to include the best 50 of the 257 funds by implementation returns, there would still be a meaningful performance improvement of 1.86x. Data and Analysis We obtained data from Preqin on the performance (TVPI) of 7,524 funds from vintages 1977 to 2014, in addition to data on 41,065 underlying investments of buyout funds. Linking deal-level activity to fund-level performance, we obtained a sample of 1,119 buyout funds for which performance data and data on at least five underlying investments is available. Descriptive statistics for this sample can be found in Table 1. Table 1: Descriptive Statistics Peers Timing Strat Implementation Focal N Minimum Maximum Mean Std. Deviation (0.76) (3.82) 1.17 (0.01) (2.50) 5.17 (0.00) Valid N (listwise) 1119 We note that the average performance contribution of each of the three value components is close to zero across the sample, which makes intuitive sense. However, we can see that the greatest standard deviation and coefficient of variation is on the implementation component. This suggests PE fund managers differ most in their ability to add value relative to the performance that other fund managers generate based on investments in similar assets at similar points in time. We document the differential ability of PE fund managers to add value based on benchmark statistics with annual quartile cut-offs for each of these components (Table 2). This makes it possible for investors to 2 Vintages of last fund range from 1997 to 2007

5 determine to what extent a given fund manager has been a top-quartile value generator within each of these categories. Table 2: Quartile Cut-Offs by Performance Component Bottom Q Cut-Off Median Top Q Cut- Bottom Q Cut-Off Median Top Q Cut- Bottom Q Cut- Off Median Top Q Cut-Off Vintage Timing Timing Off Timing Strat Strat Off Strat Implementation Implementation Implementation We proceed to explore to what extent we observe performance persistence in this data, in other words whether any or all of the performance components of one fund by a given fund manager indicate above- or below-average performance (overall or in any of the components) for the subsequent fund managed by the same fund manager with the same strategic focus. To this end, we looked for such pairs of funds in the sample, i.e. pairs of buyout funds raised by the same GP and with the same strategy and regional focus, immediately following each other in the sequence of fundraises and with the successor fund raised between 1997 and This window of vintages was chosen to exclude recent and immature funds for which performance cannot be assessed with accuracy due to the J-curve effect, as well as fund pairs from the early to mid-1990s which operated in a substantially different competitive environment. We identified 257 such fund pairs. As a first step, we performed a simple bivariate correlation analysis of the overall performance and performance components of the predecessor funds and the overall performance and performance components of the successor fund in each pair. Table 3 shows the results of this analysis, which point to a number of important implications.

6 Table 3: Bivariate Pearson Correlation between performance of predecessor and successor fund We first see that in line with recent research, there is no significant persistence in performance with respect to the overall performance, as the TVPI of the predecessor fund is not significantly related to the TVPI of the successor fund. Importantly, however, we observe significant persistence-type relationships with respect to individual performance components. Notably the implementationrelated performance component (the How) seems to persist, as the implementation-related performance component of the predecessor fund is positively and significantly related to that of the successor fund (p<0.01). Interestingly, the strategic choice-related performance component (the What ) has the opposite effect, as the strategic choice-related performance component of the predecessor fund is negatively and significantly related to that of the successor fund (p<0.01). The timing-related performance component (the When ) of the predecessor fund is not significantly related to that of the successor fund. Arguably, the most relevant question for practitioners is which performance component of the predecessor fund is more indicative of the overall performance of the successor fund, as this information would be valuable for fund selection processes. Our analysis suggests that predecessor funds with a high level of implementation-related performance are followed by funds with greater overall performance (p<0.05), and that predecessor funds with a high level of strategic choice-related performance are followed by funds with lower overall performance (p<0.05). Given the strong interdependencies between the explanatory variables in Table 3, we proceed with a multivariate assessment of performance persistence.

7 Based on a stepwise entry function for an OLS regression with the overall performance of the successor funds as a dependent variable and the different performance components of the predecessor funds as independent variables, we confirm that the implementation-related performance component is the only variable with a statistically significant link to overall subsequent performance. The relationship is positive and significant at the p<0.05 level. The explanatory power of this model does not improve by adding any of the other predecessor fund performance components. This points to the dominant role of the implementation-related performance component as indicator of likely future outperformance. In other words, investors should give preference to GPs with a proven ability to add value based on the How of value creation in earlier funds to build outperforming portfolios. This raises the question of the economic magnitude of the performance improvements one could have obtained historically by systematically investing in funds preceded by those with high levels of the implementation-related performance component. We can also address this question based on our sample of 257 fund pairs. Selecting a portfolio based on the top 25 funds using implementationbased returns, the performance would have been 2.03x. A t-test confirms that the difference in mean performance between the 25 and the total sample is statistically significant at p< Even if the sample was extended to include the best 50 of the 257 funds by implementation-based returns, there would still be a meaningful performance improvement to 1.86x. Summary and Conclusion This paper has demonstrated that while traditional measures of the how much of value creation are limited in their ability to provide investors with reliable insight into a GP s ability to outperform in the future, a deeper understanding of the How of value creation can be a meaningful starting point for the identification of likely future outperformers. We utilized a novel methodology to measure distinct components of a PE fund s performance, leveraging data on tens of thousands of underlying investments made by thousands of PE funds. This enables us to assess the performance impact of The When, i.e. the decision of how investments were timed over the investment period of the fund and the Strategy; The What, i.e. the decision of which type of target companies were chosen (industry sector, size category, geography) to acquire at those times; and to compare this to the The Who and How decisions, i.e. the performance of the specific investment approaches into specific target companies, above the returns of other funds that made similar decisions in terms of timing and strategy. By analyzing data on 1,119 PE funds, we show that funds differ most in the implementation-related How component. We then explore the persistence of overall returns based on 257 pairs of buyout funds immediately following each other in the sequence of fundraises. By always comparing the first fund to the corresponding subsequent fund in these 257 pairs, we find that statistically significant positive persistence occurs only in the implementation component of returns. While past TVPI does not provide insight into subsequent IRR, other components of performance have no significant (timing, When) or a significant negative (strategic choice, What) level of persistence. This finding is not only statistically relevant but also meaningful in terms of its magnitude: had an investor picked 25 of the 257 funds using only those with the highest previous implementation-based returns, the portfolio would have yielded 2.03x, which is significantly higher than the sample average TVPI of 1.73x.

8 While the analyses presented in this paper constitute a major step towards the goal of better understanding the relevant and persistent drivers of PE value creation, there are inevitably limitations. While this methodology has been designed to provide directionally correct insight, despite possible isolated imperfections in the underlying data on fund performance and investment activity, the method only considers deals that actually happened. As such it cannot capture those funds that may have been in regular competition for similar deals, but eventually ended up making investments that were not very similar as per our definition of similarity. Data limitations further prohibit us from considering deal attributes other than size, industry and geography, so therefore we do not capture funds that are similar with respect to their choice of buy-and-build versus standalone buyout strategies. The aforementioned findings have important implications for PE investors, as they suggest that without a deeper understanding of the nature of a GP s value creation, the ability of investors to reliably identify likely future outperformers is substantially limited. Based on an assessment of the deeper value driver components in a GP s past funds, however, investors are likely to be able to select funds that are significantly more likely than the average to outperform.