Acharya et al (2012).pdf

8/16/2019 Acharya et al (2012).pdf

1/36

Corporate Governance and Value Creation:Evidence from Private Equity

Viral V. AcharyaNew York University Stern School of Business, National Bureau of EconomicResearch, Centre for Economic Policy Research, and European CorporateGovernance Institute

Oliver F. GottschalgHEC School of Management, Paris

Moritz HahnLudwig Maximilian University of Munich

Conor KehoeMcKinsey & Company

Using deal-level data from transactions initiated by large private equity houses, we ndthat the abnormal performance of deals is positive on average, after controlling for leverageand sector returns. Higher abnormal performance is related to improvement in sales and

operating margin during the private phase, relative to that for quoted peers. Generalpartners who are ex-consultants or ex–industry managers are associated with outperformingdeals focused on internal value-creation programs, and ex-bankers or ex-accountants withoutperforming deals involving signicant mergers and acquisitions. The ndings suggestthe presence, on average, of positive but heterogeneous skills at the deal-partner level inlarge private equity transactions. ( JEL G31, G32, G34, G23, G24)

ViralAcharya completed thiswork in partat theLondon BusinessSchool and while visitingtheStanford GraduateSchool of Business. We are especially grateful to the PE rms that assembled and gave us access to sensitive deal

data, with a special mention to the invaluable support of Golding Capital Partners. Viral Acharya was supportedduring thisstudy by the London BusinessSchool Governance Center and PrivateEquity Institute, the LeverhulmeFoundation, INQUIRE Europe, and London Business School’s Research and Materials Development (RAMD)grant. Oliver Gottschalg was supported by the HEC Foundation and the HEC PE Observatory. McKinsey &Company also devoted signicant human resources to the eldwork and analysis, not for any client but on itsown account. We are grateful to excellent assistance and management of data collection and interviews byAmithKaran, Ricardo Martinelli, Prashanth Reddy, and David Wood of McKinsey & Co. Ramin Baghai-Wadji, AnnIveson, Hanh Le, Chao Wang, and Yili Zhang provided valuable research assistance. The study has benetedfrom comments of two anonymous referees, Laura Starks (editor), Yael Hochberg (discussant), Steve Kaplan(discussant), Tim Kelly (discussant), and from seminar participants at the European Central Bank and Centrefor Financial Studies Conference (2008) in Frankfurt, the Inaugural Symposium (2008) at the London BusinessSchool’s Coller Institute of Private Equity, the joint conference of the University of Chicago and University of Illinois at Chicago, the London Business School, McKinsey & Co., the Western Finance Association Meetings(2008), and participating PE rms. All errors remain our own. Send correspondence to Oliver Gottschalg,HEC School of Management, 78351 Jour en Josas, Paris, France; telephone: +33 (0) 670017664. E-mail:[email protected].


2/36

Corporate Governance and Value Creation: Evidence from Private Equity

1. Introduction

In a seminal piece on private equity, Jensen (1989) argued that leveragedbuyouts (LBOs) create value through high leverage and powerful incentives.He proposed that public corporations are often characterized by entrenchedmanagement that is prone to cash-ow diversion and averse to taking onefcient levels of risk. Consistent with Jensen’s view, Kaplan (1989), Smith(1990), Lichtenberg and Siegel (1990), and others provide evidence that LBOscreate value by signicantly improving the operating performance of acquiredcompanies and by distributing cash in the form of high debt payments.

By contrast, the recent literature has focused on the returns that privateequity (PE) funds—which usually initiate the LBO and own, or more precisely,manage, at least a majority of the resulting private entity—generate for theirend investors, such as pension funds. In particular, Kaplan and Schoar (2005)studied internal rates of return (IRRs) net of management fees for 746 funds in1985–2001 and found that the median fund generated only 80% of the S&P 500return, and that themean fund’s return was only slightly higher, at around 90%. 1

However, the evidence suggests that returns are better for the largest and mostmature PE houses—those operating for at least ve years. Kaplan and Schoarnote that, for funds in this subset of houses, the median performance is 150%of S&P 500 return and the mean is even higher at 170%. Furthermore, thisperformance is persistent, a characteristic that is generally associated with thepotential existence of “skill” in a fund manager. It is interesting to note thatsuch long-run persistence has rarely been found in mutual funds, and whenfound has generally been in the worst performers (Carhart 1997).

Our article is an attempt to bridge these two strands of literature concerningPE, the rst of which analyzes the operating performance of acquiredcompanies, and the second that analyzes fund IRRs. In addition, we investigatehow human capital factors are associated with value creation in PE deals. Wefocus on the following questions: (i) Are the returns to equity investments bylarge, mature PE houses simply due to nancial leverage and luck or markettiming from investing in well-performing sectors, or do these returns representthe value created at the enterprise level in the so-called portfolio companies,over and above the value created by the quoted sector peers? (ii) What is theeffect of ownership by large, mature PE houses on the operating performance of portfolio companies relative to that of quoted peers, and how does this operatingperformance relate to the nancial value created (if any) by these houses?(iii) Are there any distinguishing characteristics based on the background and

1 This evidence hasbeen replicated by studies in Europe(PhalippouandGottschalg 2009; Phalippou2007), though

they raise the issue of certain survivorship biases in data employed that might imply no median outperformancerelative to the market even for large and mature PE houses. This by itself does not necessarily refute Jensen’soriginal claim; it could simply be that PE funds keep the value they create through fees. The puzzle that the

id di t f PE f d i i th b t h th i i t (th li it d t ) h


3/36

The Review of Financial Studies / v 26 n 2 2013

experience of PE houses or partners involved in a deal that are best associatedwith value creation?

To answer the rst question, we account for the impact of leverage on returnsby breaking down the deal-level equity return, measured by the IRR, into two

components: the un-levered return, and an amplication of this un-leveredreturn by deal leverage. Next, we subtract from the un-levered deal return theun-levered return that the quoted peers of the deal generated over the life of thedeal, the latter representing luck or the ability to timemarkets in PEinvestments.The difference between these two un-levered returns is what we call “abnormalperformance”: a measure of a deal’s enterprise-level outperformance relative toitsquoted peers, removing theeffects of nancial leverage.We hypothesize, andlater show, that a deal’s abnormal performance captures the return associatedwith changes in operating performance of the portfolio company, as well as

human capital factors such as a deal partner’s skill.We apply this methodology to 395 deals closed during the period 1991 to2007 in Western Europe by thirty-seven large, mature PE houses (each withfunds larger than ∼ $300 mil) 2 with a mean, gross IRR of 56.1%. We ndthat, on average, about 34% (19.8% out of 56.1%) of average deal IRR comesfrom abnormal performance, another 50% (27.9% out of 56.1%) is due tohigher nancial leverage, and the remaining portion (16% out of 56.1%) is dueto exposure to the quoted sector itself. Although abnormal performance hassubstantial variation across deals, it is positive and statistically signicant on

average, even during periods of low sector returns. This is consistent with theview that large, mature PE houses generate higher (enterprise-level) returnscompared with benchmarks. 3 Regarding the second question, we nd thatownership by large, mature PE houses has a positive impact on the operatingperformance of portfolio companies, relative to that of the sector. In particular,during PE ownership the deal margin (EBITDA/Sales) increases by around0.4% p.a. above the sector median, and the deal multiple (EBITDA/EnterpriseValue) increases by around 1 (or 16%) abovethe sector median. We interpret theoperational improvements as causal PE impact, since we nd no evidence for

a violation of the strict exogeneity assumption of the PE acquisition decision.That is, we assume—and later conrm—that there is nothing inherent in thecompanies targeted by the PE rms in our sample that would have caused theiroperating performance to improve without being acquired by private equity. 4

2 We believe this time period is particularlywellsuited for studyingvalue creationthroughoperational engineering.Kaplan and Stromberg (2008) note that operational engineering became a key private equity input to portfoliocompanies primarily in the past decade.

3 In the cross-section of deals, abnormal performance has a positive, albeit imperfect, correlation to IRR,

the “public-market equivalent” (PME) measures employed by Kaplan and Schoar (2005), and measuresof performance based on the “cash-in/cash-out multiple” (ratio of outows to inows). These alternativeperformance measures are described in the Appendix and discussed in the text.


4/36


In addition, we examine the impact of major merger and acquisitions (M&A)events during the private phase on operational performance during the sameperiod, since M&A events can generate—due to their direct impact on theoperational measures—a considerable distortion of underlying operational

performance. However, we still nd margin and multiple improvements abovesector when we separately analyze deals with and without M&A events.We then provide evidence that higher abnormal performance is asso-

ciated with a stronger operating improvement in all operating measuresrelative to quoted peers: we nd that sales growth, EBITDA margin,and multiple improvement are important explanatory factors for abnormalperformance.

Overall, this evidence is consistent with the assertion that top, maturePE houses create economic value through operational improvements. Such

improvements require skill, and the return to this skill may explain the persistentreturns these funds generate for their investors (Kaplan and Schoar 2005).This brings us to our nal question, in which we study whether characteristicsof the deal partner affect deal performance. We use deal partner backgroundand experience as a human capital or skill factor that may be relevant to dealsuccess; the econometric advantage of using this factor is that it is xed, ortime invariant, and hence exogenous (except for the matching of deal partnersto specic deals).

We nd evidence that there are combinations of value creation strategies

and partner backgrounds that correlate with deal-level abnormal performance.Deal partners with a strong operational background (e.g., ex-consultantsor ex–industry managers) generate signicantly higher outperformance in“organic” deals. In other words, partners who work as managers in theindustry or as management consultants before joining a PE house appear toaccumulate skills to improve a company internally by, for example, cuttingcosts or expanding to new customers and new geographies. In contrast,it appears that partners with a background in nance (e.g., ex-bankersor ex-accountants) more successfully follow an M&A-driven, “inorganic”

strategy.One couldarguethat westudied deals only from the funds wesampled, whichmight havebeen cherry-picked by the PE fund. We show that this is not the case.While wehave a biasinour samplefor largePEfunds, this isbydesigngiven thatwe wished to understandwhat drives their persistent outperformance.However,

sample size of deals with more than two years of pre-acquisition data is small, they show no difference totheir respective sector companies in performance trends pre-acquisition: both targeted and sector companiesshow nearly the same increase in nominal sales and constant protability. Yet, PE might still be able to identify

companies that will be subject to a positive future shock. This is something we cannot rule out. However, asystematic relationship between PE-ownership and anticipated future performance shocks can induce abnormalperformance in PE deals. To nancially exploit individual shocks on a company, a PE house must have a

t ti i f ti l d t i f ti th f t i i t th ll d th biddi PE


5/36


within the funds we sampled for our deals, we nd no statistically signicantbias between the performance of deals sampled and those not sampled. 5

In Section 2, we review the related literature. In Section 3, we providea description of the data we collected and some summary statistics. In

Section 4, we describe the methodology for calculating abnormal performance.In Section 5, we discuss operating performance. In Section 6, we link abnormalperformance and operating performance. Section 7 discusses the role of dealpartner background. Section 8 concludes.

2. Related Literature

Following the seminal work of Jensen (1989) on LBOs, the early empiricalcontributions veried the impact of PE ownership on rms (Kaplan 1989;Smith 1990; Lichtenberg and Siegel 1990). 6 However, the most recent wave of PE transactions (2001 to mid-2007) has prompted researchers to reexaminewhether buyouts are still creating value in this new era. Guo, Hotchkiss,and Song (2011) answer this question with a sample of ninety-four U.S.public-to-private transactions between 1990 and 2006. They nd that gainsin operating performance are not statistically different from those observedfor benchmark rms. Leslie and Oyer (2008) also nd weak or no evidenceof greater protability or operating efciency for 144 LBOs in the UnitedStates between 1996 and 2004, relative to public companies. However, Lerner,Sorensen, and Stromberg (2008) nd evidence in a sample of 495 buyoutsthat, in contrast to the oft-cited claim that PE has short-term incentives, buyoutdeals in fact lead to signicant increases in long-term innovation. They ndthat patents applied for by rms in PE transactions are more frequently cited (aproxy for economic importance), show no signicant shifts in the fundamentalnature of the research, and are more concentrated in the most important andprominent areas of companies’ innovative portfolios.

This literature has focused mainly on U.S. data, while our data are based onPE deals in the United Kingdom and Europe. 7 Several studies have examinedLBOs in the United Kingdom, which also experienced a tremendous increase inbuyout activity prior to the crisis of 2007–09. Nikoskelainen and Wright (2007)

5 Moreover, in contrast to the extant literature that focuses mainly on public-to-private deals, our data set alsocoversdealswhere only part of a company is acquired(e.g., carve-outdeals)and private-to-private deals,where anon-listed business is acquired. Using carve-out and private-to-private deals is important, because they compose74% of PE deals in Western Europe over the past decade, and they are different in size (enterprise value) andprotability (EBITDAmargin) frompublic-to-private deals (according to a simple, nonstatistical analysisof dataprovided by Private Equity Insight).

6 Note that Kaplan (1989), Smith (1990), and Lichtenberg and Siegel (1990) also investigate whether LBOsimproved operating performance at the expense of workers. They nd that the wealth gains from LBOs were

not a result of signicant employee layoffs or wage reductions (see Palepu 1990 for a detailed survey of thesearticles and also Kaplan and Stromberg 2008 for a comprehensive survey).

7 O di ti i hi f t f th U K d E PE d l l ti t th i th U it d St t i th t th


6/36


study 321 exited buyouts in the United Kingdom in the period 1995 to 2004. Onaverage, these deals generated a 22%return to enterprisevalue and a 71%returnto equity, after adjusting for market return. In a related article, Renneboog,Simons, and Wright (2007) examine both the magnitude and sources of the

expected shareholder gains in 177 public-to-private transactions in the UnitedKingdom in 1997–2003. They nd that pre-transaction shareholders receivea premium of 40%. They also nd that the main sources of post-transactiongains in shareholder wealth are attributable to an undervaluation of the pre-transaction target rm, increased interest tax shields, and the realignment of incentives. Harris, Siegel, and Wright (2005) study the productivity of U.K.manufacturing plants subject to management buyouts (MBO). Such plantsexperienced substantial increases in productivity, the post-MBO magnitudesof which are substantially higher than those reported in the United States by,

for example, Lichtenberg and Siegel (1990).Among the limited evidence on the impact of human capital or skill inPE investments, Kaplan et al. (2008) analyze the relationship between 316portfolio company managers (CEOs) and the success of buyouts. They nd thatexecution skills appear to be more strongly related to success than interpersonalskills. To our knowledge, with the exception of this study, there has beenno systematic analysis of the link between the nancial returns of LBOs andhuman capital factors. As Cumming, Siegel, and Wright (2007) state, “Thereis a need to understand the human capital expertise that successful PE rms

require. There appears to be a need to broaden the traditional nancial skillsbase of private equity executives to include more product and operationsexpertise.”

Our evidence regarding the relevance of human capital factors and, inparticular, the importance of task-specic deal partner skills (operationalor nancial) lls this important gap in the literature for PE investments.We posit that task-specic skills attributable to deal partner background areone signicant part of the persistent abnormal nancial return generated bylarge, mature PE houses for their investors (Kaplan and Schoar 2005). 8 PE

partners seem to add value to portfolio companies by applying skills they haveaccumulated over time.In contrast, for venture capital (VC) investments, it has been established

that VC funds have an impact on the development of new companies, andthat fund expertise matters for performance. Hellmann and Puri (2002), forexample, show thatVC funds add professional structure and rigor to the startupsin which they invest: startups backed by venture capital (VC) funds replacetheir CEOs more frequently; introduce stock-option plans; and hire a vicepresident of sales and marketing. Dimov and Shepherd (2005) nd a positive

8 W t th t ti l i il t b t th i t f i t t b t i t it h d th i i t


7/36


relationship between successful, VC-backed initial public offerings (IPOs) andthe education of senior managers in the VC fund in science or humanities.Moreover, management consulting experience in a fund reduces the fund’sshare of bankruptcies. Bottazzi, Rin, and Hellmann (2008) go one step further

and provide evidence that experience in consulting or industry on the part of VC fund managers predicts investor activism—for instance, the frequency of interaction between the VC fund and the startup—and seems to correlate withperformance.

Most recently, Zarutskie (2010) argues that, in addition to prior businessexperience, longevity in the VC industry is relevant to the performance of VCfund investments. According to the author, task-specic human capital factorsseemto be most important and depend on the type of investment: fund managerswith science backgrounds excel only in high-tech investments, while those with

nance backgrounds excel in later-stage investments. Again, it is important tonote that such returns to task-specic skills gathered over time are not foundfor mutual fund managers, which seldom hold majority stakes and thereforeengage less actively in the management of the portfolio company (Chevalierand Ellison 1999).

3. Data and Sample Selection

Our sample combines two data sets. First, we use proprietary deal-leveldata collected together with McKinsey, a consulting company that serveslarge PE houses. This “McKinsey sample” is relatively small ( n =110), but itgives detailed deal-level information—for example, operational performancemeasures prior to PE ownership. The second data set was provided by a majorinvestor in PEfunds that hastracked thedetailedperformanceofPE investmentsover the past two decades. This “LP sample” is larger ( n = 285), but it includesless information.

The McKinsey sample represents relatively large deals, with enterprise

values greater than roughly E 50 million, acquired by fourteen large, maturePE houses from 1995 to 2005. To collect the data, we approached forty large,well-established European multi-fund houses active in either “mid-cap” or“large-cap” markets, of which fourteen (35%) agreed to provide data subjectto a condentiality agreement. (As such, all data herein are aggregated and notattributable to any single deal or PE house.) Together, these rms provideddata for 110 deals, with each group of deals pertaining to a specic fundchecked by us to ensure that the performance, or average IRR, of each groupwas representative of the fund from which it was drawn. Where groups of

deals failed this test, we removed or replaced a small number of deals to meetthis condition. For individual deals, we rigorously checked data, looking for


8/36


The LPsample has the additional advantage of being audited rather than self-reported. It is drawn from hand-collected track records of private equity rms asreported in fund-raising documents (PrivatePlacement Memorandums[PPMs])created by the PE funds. In the PPMs, we observe the list of investments made

by each PE rm. From the pool of PPMs collected by the LP, we selectedthirty-four from large, well-established European multi-fund houses active ineither “mid-cap” or “large-cap” markets. From these thirty-four PPMs, wecollected all 285 realized transactions for which both cash ows and operatingperformance were available. We checked each group of deals pertaining to aspecic fund to ensure the performance, or average IRR, of each group wasrepresentative of the fund from which it was drawn. Importantly, buyout rms’PPMs generally disclose all investments, including ones that did not performwell. As a consequence, this sample contains numerous poorly performing

investments.The nal, combined data set comprises 395 deals from forty-eight fundscovering nearly two decades. For each deal, we have the exact structure of cashows, from and to the parent fund, and detailed data on nancial and operatingmeasures.We do not have all enterprise-level cash ows, which would include,for example, the taxes or interest and principal paid on debt. 9

To compare our sample to a set of publicly listed peers, we collected data,supplied by Datastream, for approximately 7,000 publicly listed corporations(PLCs) in Europe, from which we constructed sector indices based on ICB

(Industry Classication Benchmark) sector codes. These codes group publiclylisted peers into ten industries and thirty-nine sectors. We also collected dataon TRS (Total Returns to Shareholders): enterprise value, net debt, equity,sales, and EBITDA(E). The latter removes the effects of exceptional items,which are not included in EBITDA gures provided by the rms in oursample.

Table 1, panel A, shows that, within our sample period (1991–2008), ourdeals are well distributed over time, although there is some concentration in1998–2003 in terms of deal entry or acquisition. Table 1, panel B, provides

additional summary statistics for the deals. Deals in our sample have a highmean, gross IRR (56.1%) and cash multiple (4.4), with large values on the righttail, even after winsorizing our sample (we replace the lowest 5% of values withthe fth percentile value, and the highest 5% with the ninety-fth percentilevalue). However, a high value for average IRR is to be expected from a sampleof deals from large, mature PE houses (Kaplan and Schoar 2005). We alsoreport an average duration of 3.9 years.

9We also do not have all cash ows for six un-exited deals in the McKinsey data set because there is no exit cashow from sale, nor can it be deemed to be zero, as in the case of bankruptcies. Therefore, theend enterprise-valuecash ow was simulated using the EV/EBITDA multiple at the start of the deal and applying that number to2006 2007 d EBITDA O lt b t t lt ti ti i l di ti


9/36


Table 1Data overview

Panel A: Distribution of deals by entry and exit years

Years 1991 92 93 94 95 96 97 98 99

Entry 3 2 2 1 3 23 34 42 44Exit 2 2 3 8 12

2000 1 2 3 4 5 6 7 2008 SUM

Entry 38 46 54 45 28 20 8 2 395Exit 14 21 19 36 54 71 64 70 a 19 395

The table shows the years in which the PE houses bought (entry) or sold (exit) the portfolio companies (deals)in the McKinsey and LP data set (our sample). aa Including six deals for which exit is simulated.

Panel B: Summary statistics

Variable n mean (median) std. dev. min max t -stat of diff.exitand entry

deal IRR (gross) % a 395 56.1 (43.2) 46.6 0.0 179.0cash in/cash out multiple 4.4 (3) 9.5 0.0 130.0duration (years) b 385 3.9 (3.4) 2 0.4 10.8deal size (entry) b 385 430.2 (141.2) 756.8 0.0 6485 .6 8.48

∗∗∗

deal size (exit) b 749.6 (244) 1311 .9 0.0 10290 .0debt/equity (entry) a,b 381 2.2 (1.9) 1.5 0.3 6.2

− 15.18∗∗∗

debt/equity (exit) a,b 0.8 (0.5) 0.9 0.1 3.8mult (entry) c 354 6.8 (6.5) 11.1

− 173 .3 42.9 2.91∗∗

mult (exit) c 8.6 (7.9) 7.5 − 67.4 54.5debt/EBITDA (entry) c 354 4.1 (4.1) 6.9

− 104 .7 35.9 − 3.37∗∗∗

debt/EBITDA (exit) c 3.0 (2.5) 3.1 − 13.2 21.7

The table shows various nancial measures for the deals in the McKinsey and LP data set. The rst part reportsthe nancial performance and the duration. We calculate the deal IRRs (internal rate of return) using the entiretime pattern of cash inows and outows for each deal (portfolio company), as experienced by the PE house(before fees). The cash-in/cash-out multiple measures the absolute value of all positive cash ows divided byall negative cash ows minus 1. The duration captures the length of the deals in years, using the entry and exitmonths and years as reported by the PE house.The second part of the table compares the enterprise value (deal size) and several nancial ratios between entryand exit date. The number of observations is smaller than in the rst part, since we include only deals that thePE funds sold, and have to exclude deals with an equity value of zero. In addition, information on EBITDA atentry and exit is not available for all deals.In the last column we test for differences between exit and entry values.Note: In Mio EUR; signicance level ∗p < 0.1, ∗∗p < 0.05, ∗∗∗p < 0.01.a Due to a few but large outliers, we winsorize the distribution of the variable by replacing the 5% highest and5% lowest values by the next value counting inward from the extremes.b Only exited deals.c Only exited deals and if EBITDA for exit and entry date is available.

In the second half of Table 1, panel B, we compare nancial ratios atthe entry and exit dates. The median entry EV/EBITDA multiple is 6.5,whereas the corresponding exit multiple is 7.9, which indicates that, on averageand assuming stable or rising EBITDA(E), our deals improved their marketvaluations (consistent with the ndings of Kaplan 1989). The median debt-to-equity (D/E) ratio at entry is 1.9, which is in line with the usual LBO capital

structure, believed to be 70% debt and 30% equity (Axelson et al. 2008).However, the median D/E ratio at exit is much smaller (0.5). In percentage


10/36


Groh and Gottschalg (forthcoming), it appears that thedebt-to-equity ratio fallsfor PE deals during their life only partly due to improvements in coverage ratio(Debt/EBITDA), and mainly due to improvements in equity value over deallife.

Next, we come to the important sample-selection issues. Table 2, in panelsA–C, provides several relevant comparisons between our sample and the PEuniverse. Overall, we conclude that our sample covers mainly large funds, butseems to be representative in terms of performance, and includes all differentvendor types; that is, not just public-to-private deals but also the frequentprivate-to-private deals.

First,Table2,panelA,shows that the sampled fundsare a good representationof funds in Western Europe, especially when we take into account the fact thatwe focus on funds whose sizes are above $300 million.All 146funds in Western

Europewith a vintage year in1991–2005 had a simpleaverage net IRR of23.5%(based on 146 funds for which Preqin reports IRRs), which is not different tothe net IRR of our funds ( t -statistic = –0.41 for the difference). Also, largefunds in the PE universe (again, for which Preqin reports IRRs) had returnssimilar to our sample funds ( t = − 0.44 for the difference).

In Table 2, panels B and C provide evidence that we have not cherry-pickedthe deals from the sampled funds. Table 2, panel B, compares the averageperformance of the deals in the McKinsey sample, per fund (in terms of netIRR), to the performance of the same thirty-two funds from which our deals

were drawn; average fund IRRs are based on Preqin gures. We show that thefunds in the McKinsey sample do not appear to have cherry-picked the dealsthat they reported: the difference between the average, publicly reported netfund IRR of 28.1% and the average net IRR of our deals, per fund, of 26.3% isnot statistically signicant ( t = 0.43). In terms of deal performance, therefore,we have an unbiased representation of deals within the funds we sampled. Forthe comparison, because publicly available data on fund performance are basedon net IRRs, we had to convert our gross deal-level IRRs, prior to fees and carrypaid to the PE funds by investors, to net IRRs—or IRRs from the viewpoint

of fund investors.10

We deduct from the gross IRR a 2% annual fee and 20%carry for IRR above the typical benchmark market return of 8%. 11

10 To perform this conversion, we also construct an articial fund of our sample deals and calculate its IRR. Thepseudo-fund starts in 1995 and lasts for thirteen years, until 2007. Investments or cash inows take place in years1–9 (with small investments in years 10 and 11 as well). The bulk of the investments occur in years 3–9. Cashpayouts start in year 5; in the last three years, the fund has only cash payouts. Using this pattern of cash inowsand outows, we calculate the gross IRR of the pseudo-fund.

11More specically, if (i) gross IRR

≤10%, then LPs keep all return except 2% fees, so that net IRR = gross IRR– 2% fees; (ii) 10% < gross IRR < 12.5%, then LPs keep all return up to 10% except for 2% fees and GPs keep

all return from 10% to 12.5%, so that net IRR = gross IRR – 2% fees – (Gross IRR – 10%) = 8%; and (iii) grossIRR 12 5% th LP d GP h i 80 20 ti th t di 12 5% th t t IRR IRR


11/36


12/36


Table 2Continued

Panel C: Benchmarking of deals used from LP data set

IRR, in % deal size, in Mio EUR

n mean (median) t -stat of diff. a mean (median) t -stat of diff. a

(1) Deals from LP data set used 285 91.6 (47.75) % –0. 42 372.49 (106.505) 0.35(2) Deals from LP data set not used b 892 104.2 (40.99) % 350.74 (85.0)

This table compares net IRRs and deal size of (1) the deals from the LP data set used in the analyses with(2) the deals from the LP data set not used in the analysis (but which are from the same PE houses and forwhich information on IRR and size was available). We exclude deals in (2) from the analyses, due to missinginformation, e.g., on leverage or deal partner names.In the rst column we test if (1) and (2) are different by IRR and in the last column if (1) and (2) are different bydeal size.Note: Signicance level ∗p < 0.1, ∗∗p < 0.05, ∗∗∗p < 0.01.a Assuming unequal variance for (1) and (2).b Deals with information on IRR ( n = 671) or deal value available ( n = 342).

Finally, as previously mentioned, our sample includes all types of deals,which is important to get a full perspective on the effects of PE ownership. Theextant literature mainly focuses on public-to-private transactions, which is onlya part of the total buyout activity in Western Europe. By contrast, the majorityof deals are carve-outs, in which only part of a company is acquired, or private-to-private deals, in which PE rms acquire a non-listed business. For example,carve-out and private-to-private deals comprised more than two-thirds of all PEtransactions in WesternEuropebetween 1995 and2005—and they were smaller

in size and different in protability (EBITDA margin) from public-to-privatedeals in the Western European universe (according to a simple, non-statisticalanalysis of data provided by Private Equity Insight).

4. A Measure of Abnormal Financial Performance

4.1 MethodologyOne of the key questions we want to answer in this study is, relative toquoted peers, how much of the excess returns generated by PE rms comes

from pure nancial leverage, and how much comes from genuine operationalimprovements. To disentangle the effect of leverage from that of operationalimprovements, we rst calculate the IRR of the deal i—its levered return(R L,i )—using the entire time series of gross cash ows for the deal (i.e., beforefees), both from and to the fund, as recorded by the PE house. Then we un-leverthis IRR and benchmark the un-levered return ( R U,i ) to returns for the quotedpeers of the deal, unlevered in the same way ( R SU,i ). The resulting difference inun-levered returns is what we call the “abnormal performance” of the deal (andwhich we referred to in earlier drafts of this article as “alpha performance”).

To arrive at the unlevered return R U,I , we use

R + R (1 t )(D/E )


13/36


The un-levered IRR, RU,i , corresponds to the return generated at theenterprise level. Since the PE houses in our sample did not report RD,i , theaverage cost of debt D, we use the base rate and interest margin spread reportedfor each deal by Dealogic. 12 The leverage ratio D/E i is the average of the entry

and exit debt-to-equity ratio of the deal; that is, since the starting D/E is higherthan the exit D/E for most deals, we employ the average of the two to reectthe pattern of declining leverage for most deals. Finally, for tax rate t , we usetheaverage corporate tax rate during the holdingperiod for the country in whichthe portfolio company’s headquarters is located.

We also apply (1) to calculate un-levered sector IRRs, RSU,I , from thebenchmark levered sector return, RS,i . In this case, a sector is dened ascontaining all quoted European “peer” companies sharing the deal’s three-digitICB (Industry Classication Benchmark) code in Datastream. To do so, we rst

calculate RS,i —the median annualized total return to shareholders (TRS) forthe sector over the life of each deal, which we un-lever using (1) and the medianD/E for the sector over a three-year average from the deal’s entry date onward.We further assumethe sametax rateand costofdebt for the sectoras for the deal.From (1), it follows that higher values of R D,i result in greater un-levered returnfor the same levered return. Since RD,i for the sector companies is potentiallylower than for the deals (due to lower leverage and hence lower risk), weoverestimate the un-levered sector returns and are therefore conservative inattributing positive abnormal performance to PE deals.

After obtaining the un-levered returns for both the deal ( R U,i ) and sector(R SU,i ), which are stripped of the effects of nancial leverage, the next keystep is to measure the amount of excess PE return that is brought about bygenuine operational improvements. For this purpose, we dene the abnormalperformance of the deal, α i , as

α i =R U,i − R SU,i . (2)

In essence, applying (1) and (2) allows us to make the following decompositionor performance attribution of each deal IRR:

(i) Deal-level abnormal performance: α i(ii) Unlevered sector performance: R SU,i

(iii) Total leverage effect: R L,i − R U,i .

12 Dealogic provides information on the base rate and the interest margin spread for only 67 deals (out of 110)in our sample. For 19 deals we can nd only the base rate (Libor vs. Euribor), and for the remaining 24 dealswe nd no information. If the margin spread is unknown, we use the median spread of all PE deals in WesternEurope in the same year. If the base rate is unknown, we use LIBOR for the U.K. deals and Euribor for all otherdeals.

We made sure that the assumption on the spread does not have a large impact on our results. First, the spreaddoes not vary much in the cross-section. In our sample period and for all deals covered in Dealogic, the standarddeviationof theweighted(by risk tranches) average spreadis 1.1%, with an average (median)spread of 2.6(2.3)%( 984) S d th iti it f th b l f f d l t diff t i t t t ti


14/36


The leverage effect measures the total effect of leverage on deal return.However, we are more interested in measuring the effect of the additionalleverage that companies take on after their acquisition by PE. To arrive at theincremental effect of increased leverage, we rst rewrite (1) in terms of RL,i

as follows:R L,i =R U,i (1+ D/E i ) − R D,i (1 − t )(D/E i ) .

Then, we substitute for R U,i based on (2):

R L,i = (α i + R SU,i )(1+ D/E i ) − R D,i (1 − t )(D/E i )

= α i (1+ D/E i )+ R SU,i (1+ D/E i ) − R D,i (1 − t )(D/E i ) .

And nally, we substitute for D/E i in terms of incremental leverage:

D/E i =D/E S,i +(D/E i − D/E S,i ) .

To arrive at the following decomposition of deal IRR:

R L,i =α + R SU,i (1+ D/E S,i ) − R D,i (1 − t )(D/E S,i )

+ (R SU,i − R D,i (1 − t ))(D/E i − D/E S,i )+ α (D/E i ) .

This equation provides an alternative decomposition of each deal IRR:

(i) Deal-level abnormal performance : α i measures the excess asset returngenerated for PE investors at the enterprise level of the portfoliocompany, purged of the effect of leverage.

(ii) Levered sector return : RSU,i (1+ D/E S,i )− R D,i (1 − t )(D/E S,i ) mea-sures the effect of contemporaneous sector returns, including the effectof leverage inherent in the sector.

(iii) Return from incremental leverage : (R SU,i − R D,i (1 − t )) (D/E i −D/E S,i ) +α i (D/E i ) captures the amplication effect that (a) theincremental deal leverage beyond the sector leverage, ( D/E i − D/E S,i ),has on sector returns, and that (b) the total leverage has on abnormalperformance. Finally, the decomposition also allows us to quantify thetax benets of the incremental leverage in private equity transactions,which is t R D,i (D/E i − D/E S,i ).

The purpose of performing such a decomposition, or return attribution, isthreefold. First, it allows us to determine if the sample deals from mature PEhouses generated a signicantly positive abnormal performance on average(or not). Second—if we believe that abnormal performance is attributable

to operating strategies and changes attempted by the PE rms—we need todetermine the cross-sectional distribution of this abnormal performance. And


15/36


operational improvement, and to the characteristics of PE houses or their dealpartners.

Before we proceed to discussing our results, it is useful to note someof the limitations of our methodology. First, it treats leverage as having a

purely nancial effect rather than as having an incentive effect. Second, ourmethodology is subject to the usual problems associated with IRRs—namely,that they are a way of describing cash ows rather than actual, realizedreturns, and that they translate into returns only under extreme assumptionsof constant and common discount rates and reinvestment rates. To addressthe second issue, another approach we adopted was to calculate a publicmarket equivalent (PME) for each deal. As a benchmark, we used the sectorreturn to discount all cash ows and then calculated the ratio of discountedcash ows to the largest cash inow for the deal (in the spirit of Kaplan

and Schoar 2005).13

Third, it might seem problematic to use realized totalreturns for the benchmark sector while using IRRs for the individual privateequity deals. To alleviate this concern, we also calculate the “cash-in/cash-outmultiple” of a deal, dened as the total cash outows of a deal divided byits total cash inows, which we annualize over the life of the deal. We thenapply the un-levering procedure of Equations (1) and (2) to this measure andsubtract from the un-levered return the annualized un-levered total return of the sector. We call the difference the “abnormal cash-in/cash-out multiple” of the deal.

The next section discusses the relationship between our benchmark measureof abnormal performance and IRR, PME, and measures based on cash-in/cash-out. Finally, we note that because we do not have the exact cash payouts ondebt, we are unable to employ the methodology used by Kaplan (1989), whichsimulates the enterprise-level (not equity) cash ows that would be obtainedby investing these cash inows in the quoted sector and examining the cashoutows thus generated.

4.2 Average abnormal performance and its characteristicsTable 3, panel A, summarizes the results from employing the decompositionmethod of Section 4.1. It presents results for (1) for the overall sample of 395deals; for (2)–(6) for different time periods; and for (7) for the subset of 67

13 The precise way in which we calculated the public market equivalent metric (PME) was as follows: (i) Wedened the deal exit date as the date when the last CF occurred. (ii) We calculated for each CF of the deal asector “reinvestment rate” from the date when the CF occurred to the exit date, based on each sector company’stotal return series (TRS). For example, if a deal CF was in Q1/2000 and the exit date in Q1/2001, then we would

use a sector company’s Q1/2000 TRS (e.g., 100) and Q1/2001 TRS (e.g., 150), and derive a reinvestment ratefor this CF (e.g., 50%). For a deal CF in a given sector, we used the median of all companies’ reinvestment ratein its sector. (iii) We grew (effectively “reinvested”) each CF with this median sector interest rate over the time(f li ti f th CF til th d l it) d did th f th ti CF (i ) W d ll


16/36


Table 3Financial performance analyses

Panel A: IRR decomposition

Scenario n (i) deal-levelabnormal

performance

(ii) leveredsector return

(iii) returnfrom

incrementalleverage

total IRR c Abnormalperformance

per IRR

(1) A ll deals 395 19.8 (15.4) 8.5 (8.4) 27.9 (19.1) 56.1 (43.2) 33.93%

(2) 1991–93 7 32.2 (28.5) 9.8 (9.1) 29.1 (26.4) 71.1 (70.9) 45.07%(3) 1994–97 61 20.5 (13.8) 8.5 (7) 30.4 (19.7) 59.4 (41.9) 33.90%(4) 1998–2000 124 16.5 (12) 0.7 (1.5) 18 (12) 35.2 (25) 45.71%(5) 2001–02 100 17.5 (14.6) 11.2 (12.1) 21.9 (17.9) 50.6 (42.9) 34.00%(6) 2003–07 103 24.7 (21.8) 15 (16.1) 43.9 (36.6) 83.5 (73.3) 28.92%

(7) Deals withcost of debtinformation

67 11 (10.1) 11.2 (10.1) 20.7 (15.2) 42.8 (40) 26.19%

The table provides simple averages of three gross IRR components:

(i) Deal-level abnormalperformance(alpha): α i measures theexcess asset return generated at the enterpriselevel of the portfolio company for PE investors. It is purged of the effect of leverage nancing the rmtakes on, since α i =RUi − RSU i . Whereby RUi is the un-levered return of the deal i and RSU i theun-levered return of the sector i , using the standard un-levering formula. ab

(ii) Levered sector return: R SU i (1+ D/E Si )− RDi (1–t)(D/E Si ) measures the effect of contemporaneoussector returns, including the effect of sector-level leverage. For the sector, we use the median IRR ineach deal corresponding sector.

(iii) Return from incremental leverage: (RSU i − RDi (1–t))(D/E i − D/E Si )+ α i (D/E i ) captures theamplication effect that (i) the incremental deal leverage beyond the sector leverage, (D/E i − D/E Si ),has on the sector returns and (ii) the total leverage has on enterprise-level ouperformance.

We report the IRR decomposition for different scenarios: (1) We break down the returns for all deals, (2)–(6) forthe deals separately by entry year, and (7) only for deals for which the cost of debt was available.Note: All values in percent, simple averages; medians in parentheses.a We further use the average D/E ratio during deal life for the deals, a median D/E ratio over three years for thesector and β = 1. We further assume the same cost of debt and corporate tax rate for the sector as for the deal.b Out of the 110 deals in the McKinsey data set, we nd for 62 deals the cost of debt (based rate and marginspread) in Dealogic; for 15 we only nd the base rate (Libor vs. Euribor); and for 23 deals we nd no information.If the margin spread is unknown for a deal we use the median spread of PE deals in Western Europe in the sameyear. If the base rate is unknown we use Libor for U.K. deals and Euribor for all other deals. For the 285 dealsfrom the LP data set we assume an average cost of debt of 7% and 30% corporate tax rate.c Due to a few but large outliers, we winsorize the IRR distribution by replacing the 5% highest and 5% lowestvalues by the next value counting inward from the extremes.

Panel B: Abnormal performance, IRR vs. PME

Performance measure na mean (median) std. dev. min max

(1) Abnormal performance 19.8 (15.37) 22.88 − 22.87 145 .63(2) IRR gross 56.1 (43.18) 46.62 0 179 .05(3) PME Sector 395 1.88 (1.37) 1.87 − 0.16 7.17(4) Cash in/cash out multiple 0.85 (0.70) 0.67 0.10 4.69(5) Cash in/cash out multiple sector 0.32 (0.30) 0.14 0.03 1.11(6) Abnormal cash in/cash out multiple 0.53 (0.37) 0.62 − 0.20 3.99

The table reports for all deals in the McKinsey and LP data set (our sample) summary statistics on (1) abnormalperformance, (2) IRR, and (3) the public market equivalent (PME) for each deal in the spirit of Kaplan andSchoar (2005). In the PME calculation, we discount all cash ows with the total sector return and then calculate

the ratio of discounted cash ows to the largest cash inow for the deal.a Due to a few but large outliers, we winsorize the IRR, PME, and cash-in/cash-out multiple distribution byreplacing the 5% highest and 5% lowest values by the next value counting inward from the extremes.

( ti d )


17/36


Table 3Continued

Panel C: Correlation matrix: Abnormal performance, IRR, PME, and sector returns

Abnormalperformance

IRR gross PMESector

Leveredsectorreturn

Cash-in/cash-out

multiple

Cash-in/cash-out

multiplesector

Abnormalcash-

in/cash-outmultiple

Abnormalperformance

1

IRR gross 0.73∗∗∗ 1PME sector 0.53∗∗∗ 0.53∗∗∗ 1Levered sector

return− 0.28∗∗∗ 0.17∗∗∗ − 0.32∗∗∗ 1

Cash-in/cash-outmultiple

0.55∗∗∗ 0.84∗∗∗ 0.70∗∗∗ 0.10∗∗∗ 1

Cash-in/cash-outmultiple sector

0.08 0.51∗∗∗ − 0.15∗∗∗ 0.76∗∗∗ 0.44∗∗∗ 1

Abnormal

cash-in/cash-outmultiple

0.57∗∗∗ 0.78∗∗∗ 0.79∗∗∗ − 0.07 0.97∗∗∗ − 0.18∗∗ 1

The table shows for all deals in the McKinsey and LP sample ( n = 395)a the correlation between abnormalperformance (scenario 1), IRR, the public market equivalent (PME), levered median sector returns, and differentmeasures of a cash-in/cash-out multiple.Note: signicance level (of the pairwise correlation coefcient) ∗p < 0.1, ∗∗p < 0.05, ∗∗∗p < 0.01.a Due to a few but large outliers, we winsorize the IRR, PME, and cash-in/cash-out multiple distribution byreplacing the 5% highest and 5% lowest values by the next value counting inward from the extremes.

deals in the McKinsey sample for which we were able to nd the exact cost of

deal debt in Dealogic. The key ndings are as follows:Out of the average IRR of 56.1% for all 395 deals, sector risk and leverage

amplication accounts for 8.5%. In other words, less than one-fth of the totalreturn is attributable either to the sector-picking ability of PE houses or simplyto pure luck. The incremental leverage effect of 27.9% is due to high dealleverage relative to a comparatively low sector leverage. The average abnormalperformance of 19.8% is statistically signicant (at the 1% level), conrmingthat large, maturePE houses do in fact generate higher (enterprise-level) returnscompared with benchmarks, and that not all of these are attributable to sector

exposure and nancial leverage. The medians tell a similar story.Abnormal performance stays statistically signicant (at the 1% level) when

we cluster the deals by entry date. Even in years with very low sector returns,as in row (4), PE was able to generate abnormal performance. The high returnfrom incremental leverage in row (6) might correspond to the availability of cheap debt nancing, a phenomenon believed to be at work especially for PEdeals struck between 2003 and mid-2007, and that is likely responsible for thesomewhat high valuation multiples paid by PE houses in that period (Acharya,Franks, and Servaes 2007; Kaplan and Stromberg 2008).

Abnormal performance is also positive and statistically signicant for thesixty-seven deals for which we have data on the cost of debt (via Dealogic),


18/36


data; that is, the cost of debt is typically available only for later deals, and, onthe whole, abnormal performance declines over our sample period.

Overall, the evidence points to outperformance of PE deals in our samplein a manner that is robust to alternative measures. In Table 3, panel B, we

provide IRR, PME, cash-in/cash-out multiple, and abnormal cash-in/cash-outmultiple as alternative measures of deal performance. Consistent with ourearlier results—which showed that PE rms in our sample generate positiveabnormal performance—our deals also display high IRRs, PME, and cash-in/cash-out multiples. For example, the average PME (relative to the sector)is 188%, the median being 137%, and abnormal cash-in/cash-out multiple(also relative to the sector) is 0.53, the median being 0.37. In Table 3,panel C, we see that the abnormal performance is positively correlated withIRR, PME, cash-in/cash-out multiple, and abnormal cash-in/cash-out multiple,

albeit imperfectly, with correlation coefcients of 0.73, 0.53, 0.55, and 0.57,respectively. Interestingly though, abnormal performance measures are weaklyor even negatively correlated with sector returns.

5. Operating Performance

5.1 Operating measuresThe next step in our analysis is determining if, at the enterprise level, abnormalnancial performance is related to abnormal operating performance. Abnormaloperating performance can be displayed in two ways: as a larger EBITDAgrowth rate of the portfolio company during PE ownership compared withpre-acquisition, or as a larger EBITDA growth rate after PE ownershipcompared with the sector. To disentangle the PE impact on EBITDA during PEownership, we focus on the effects on (i) sales and (ii) protability (margin =EBITDA/Sales). We capture the impact on the company after the PE ownershipperiod by analyzing (iii) the EBITDA multiple (Enterprise Value/EBITDA) attime of exit from the deal. Here, we have to rely on the assumption that marketexpectations are rational at exit, since we do not have operational gures afterthe PE phase for many of the deals (trade sales, for example).

The three measures we analyze in detail are: 14

(i) Sales , equal to operating revenues earned in the course of ordinaryoperating activities.

(ii) Margin (EBITDA/Sales). EBITDA (Earnings before Interest, Taxes,Depreciation, and Amortization) is equal to Operating revenues – COGS(cost of goods sold) – SG&A (selling, general, and administrativeexpenses) – Other (e.g., R&D) = Operating income. Note that EBITDAis commonly used as it shows a company’s fundamental operational


19/36


earnings potential. However, EBITDA is not a dened measureaccording to Generally Accepted Accounting Principles (GAAP) orIFRS/IAS. In the present article, we dene EBITDA as excluding “Non-operating income.” 15 This measure is often more precisely referred

to as EBITDA(E): Earnings Before Interest, Taxes, Depreciation,Amortization (and Exceptionals).(iii) EBITDA multiple (Enterprise Value/EBITDA). In our data, Enterprise

Values (EV) are available only at acquisition and at exit; PE rms alsogave values for equity (E) and net debt (D) at entry and exit, whereE + D = EV. For the ve bankrupt deals, equity values are assumed to bezero at the time of bankruptcy (exit).

Note that to identify the impact of PE on operating performance between

pre-acquisition and during PE ownership, it is crucial to have access to aconsistent data set for both periods. Probably the only data without a structuralinconsistency are those collected by the PE rms themselves, during the courseof due diligence prior to their ownership and through monitoring their efforts.These are the data we collected from PE houses and which we use in this article.

5.2 PE impact on operating performance

Table 4, panel A, provides a snapshot of the operating performance change forthe deals in our sample prior to acquisition and hence, for the three operatingmeasures ( x ), reports: the difference ( x i,pre = xit − x it − 1) from two yearsbefore PE ownership ( t − 1) to the last pre-acquisition year ( t ). We also showthese gures for the corresponding sector companies ( x s,pre = xst − xst − 1) anduse median sector changes, given that there are generally fewer than 100companies in each three-digit ICB sector. As described in Kaplan (1989),and also according to the deals in our sample with sufcient operational dataavailable ( n = 69), PE does not seem to pick companies that were exposed to anegative idiosyncratic shock, which in better times would revert to the mean,potentially allowing the target to be sold with an upside. Target companiesshow a robust pre-acquisition increase in nominal sales, together with constantprotability. Importantly, in terms of performance trends, PE-owned companiesdo not differ in a statistically signicant way from their sector peers during thepre-acquisition phase.

15 The reason for the exclusion of “Non-operating income” is that this measure contains income derived from a

source other than a company’s regular activities and is by denition nonrecurring. For example, a company mayrecord as non-operating income the prot gained from the sale of an asset other than inventory (which can belarge in relation to the operating income). From a practitioner’s perspective, an EBITDA multiple including“N ti i ” ld t b h l f l t d t d th i id i l ti t th t


20/36


21/36


Table 4, panel B, captures the change during PE ownership and shows thedifference ( x i =xiT − xit ) from the last pre-acquisition year ( t ) to the last PE-ownership year ( T ). We analyze the annual change by dividing the difference,

x i , by the number of years of PE ownership ( T − t ).

In the rst set of columns, we report the changes for all deals with sufcientoperational data available ( n =255). In the second and third columns, weseparate deals with organic strategies ( n = 151) from those with inorganicstrategies ( n =104)—the latter of which include major, follow-on M&A eventsduring the private phase—so that we can analyze x i by strategy. This alsoallowsus to control for the effects ofM&Aevents, which cause sales to increase,and EBITDA margins and multiples to either increase or decrease, dependingon whether the ratios of the acquired entities are higher or lower than those of the acquirer. It is important to note that the rst M&A event tends to happen

after oneyear, on average, for deals in the McKinsey sample (for which detailedinformation on M&A events is available). This tends to suggest that, given theplanning required to execute an acquisition,M&Aeventsareplanned rather thana reaction to performance that is better or worse than anticipated. We also test if thechanges aredifferent from zero and, in thespiritof a difference-in-difference(DiD), test for differences between deal and median sector changes.

Overall, PE ownership tends to have a positive impact on the operatingperformance in our sample.Asshown inTable4, panel B, column(i) on average,all deals show a margin improvement of 1% and a multiple improvement of 1.1

during PE ownership, relative to their sector peers. Interestingly, the marginimprovements of the deals in our sample are slightly lower than the 1.4% to3.8% reported in Kaplan (1989). In contrast, deals do not outperform the sectorin terms of annual sales growth, although portfolio company sales do grow bya statistically signicant 7.9% during PE ownership.

Another important result shown in Table 4, panel B, is that large M&Atransactions do not cause our ndings on underlying operating improvementsto change: separately, both organic and inorganic deals, columns (ii) and (iii)respectively, display improvements in both EBITDA margin and multiple,

relative to their sectorpeers.16

Only inorganic deals seem to increasesalesabovesector, but this isn’t surprising due to the large, post-deal M&A transactionsinvolved, which naturally boost revenue.

6. Abnormal Performance and Operating Performance

Having separately identied abnormal nancial and operating performance of PE deals relative to their quoted peers, we can now investigate the relationship

16 In a robustness check in the McKinsey data set, we also include deals with M&A events after two years of PEownership in the organic deal set and nd our results qualitatively unchanged. We include those deals, since lateM&A t i ht b d l d t i d b th b d f f th d l M&A t i th


22/36


Table 5Financial and operating performance

Panel A: Abnormal performance and operating performance changes

Independent variables Dependent variable: Abnormal performance in %

(i) organic and inorganic deals (ii) organic deals (iii) inorganic deals

(1)a (2)a,b (3)a (4)a,b (5)a (6)a,b

xi − xs log sales 0.44∗∗ 0.37∗∗ 0.84∗∗∗ 0.70∗∗∗ 0.02 0.13(2.40) (2.12) (3.61) (3.16) (0 .04) (0.41)

xi − xs margin 1.94∗∗∗ 1.49∗∗∗ 2.09∗∗∗ 1.78∗∗∗ 1.83∗∗∗ 0.79(5.82) (4.34) (4.00) (3.30) (2 .96) (1.49)

xi − xs mult 2.23∗∗∗ 2.02∗∗∗ 2.47∗∗∗ 2.28∗∗∗ 1.40 0.50(4.55) (4.10) (4.05) (3.82) (1 .17) (0.37)

duration − 3.64∗∗∗ − 4.00∗∗∗ − 4.69∗∗∗(− 4.66) (− 3.88) (− 3.07)

log value − 0.14 − 0.34 0.63 0.21 − 1.33 − 1.27(− 0.20) (− 0.50) (0.68) (0.27) (− 1.01) (− 1.04)

PE house Yes Yes Yes Yes Yes Yesentry period Yes Yes Yes Yes Yes Yesintercept Yes Yes Yes Yes Yes Yes

no. of deals 234 234 140 140 94 94R 2 adjusted 0.25 0.34 0.34 0.43 0.13 0.29

The table relates cross-sectional nancial abnormal performance to changes in operating measures with OLSregressions forall exited deals in theMcKinseyand LPdata set. Forthe operational changes of a deal we calculatefor EBITDA margin, log sales, and log EBITDA multiple the average difference ( xi =xiT − xit ) from the lastpre-PE-ownership year ( t =0) to last PE-ownershipyear ( T ). We divide thedifference for logsales by the numberof PE ownership years ( T − t ) to get annual nominal sales growth. In the same way we calculate median changesin the deal corresponding sector companies ( xs = xsT − xst ). We add to the regressions the difference betweendeal changes and sector changes xi − xs .First, in regressions (1)–(2) we use organic and inorganic deals. Second, in regressions (3)–(4) we showregressions for only organic (without major M&A events) deals, and in regressions (5)–(6) for only inorganic(with major M&A events) deals.In the lower part of the table we control for deal duration and size. We also add entry time period (1991–93,1994–97, 1998–2000, 2001–02, 2003–07) and PE house dummies for time period and fund xed effects.Note: OLS regressions, t -stats in parentheses with robust standarderrors; signicance level ∗p < 0.1, ∗∗p < 0.05,∗∗∗p < 0.01.a Due to a few but large outliers, we winsorize the IRR distribution (before deriving abnormal performance) byreplacing the 5% highest and 5% lowest values by the next value counting inward from the extremes (and forthe operational measures by replacing the 10% highest and 10% lowest values).b Including deals with entry and exit EBITDA multiple available only.

(continued )

between the twomeasures in Table 5, panelA. Specically, we regress abnormalperformance on the increase in EBITDAmargin, growth in sales, and change inEBITDAmultiple. Columns (1) and (2) present results for the whole samplebut,once more, we distinguish between organic (columns (3) and (4) and inorganicdeals (columns (5) and (6)). We also control for duration and deal size at entrydate, and include dummies for different entry time periods and PE houses, inorder to control for time and the xed effects of PE rms on their acquisitions. 17

Another potential driver of abnormal performance is that PE houses may have

17 However, in unreported results, we nd that the signicance and size of theestimates on operating improvementsi i i ll ff t d b itti ti d i f t W d t i l d d i f th i f th


23/36


Table 5Continued

Panel B: IRR and operating performance changes

Independent variables Dependent variable: IRR in %


(1)a (2)a,b (3)a (4)a,b (5)a (6)a,b

xi − xs log sales 0.71∗ 0.51 1.78∗∗∗ 1.40∗∗∗ − 0.62 − 0.35(1.85) (1.54) (3.47) (3.22) (− 0.85) (− 0.70)

xi − xs margin 3.24∗∗∗ 1.92∗∗∗ 3.27∗∗∗ 2.40∗∗ 3.30∗∗∗ 0.81

(5.25) (3.12) (3.33) (2.51) (2 .88) (0.75)xi − xs mult 2.35

∗∗ 1.75∗ 2.29∗ 1.78 3.62∗ 1.47(2.38) (1.97) (1.84) (1.51) (1 .78) (0.76)

duration − 10.51∗∗∗ − 10.95∗∗∗ − 11.25∗∗∗

(− 6.48) (− 4.91) (− 4.15)log value − 1.83 − 2.42∗ 0.68 − 0.48 − 5.09∗ − 4.94∗

(− 1.32) (− 1.73) (0.42) (− 0.33) (− 1.83) (− 1.90)



Thetable relates IRR(in %) to operational changes ( xi and xs ) with OLSregressions for allexited deals in theMcKinseyand LPdataset. We calculate the PME in thespiritof Kaplanand Schoar(2005).Weuse as independentvariables thedifferencebetweendealchangesand sector changes xi − xs . Forthe operationalchanges ofa dealwe calculate for EBITDA margin, log sales, and log EBITDA multiple the average difference ( xi = xiT − xit )from the last pre-PE-ownership year ( t =0) to last PE-ownership year ( T ). We divide the difference for log salesby the number of PE ownership years ( T − t ) to get annual nominal sales growth. In the same way we calculatemedian changes in the deal corresponding sector companies ( xs = xsT − xst ).First, in regressions (1)–(2) we use all organic and inorganic deals. Second, in regressions (3)–(4) we showregressions for only organic (without major M&Aevents) deals, and in the regressions (5)–(6) for only inorganic(with major M&A events) deals.In the lower part of the table we control for deal duration and size. We also add entry time period (1991–93,1994–97, 1998–2000, 2001–02, 2003–07) and PE house dummies for time period and fund xed effects.Note: OLS regressions, t -stats in parentheses with robust standard errors; signicance level ∗p < 0.1, ∗∗p < 0.05,∗∗∗p < 0.01.a Due to a few but large outliers, we winsorize the distribution of the dependent variable by replacing the 5%highest and 5% lowest values by the next value counting inward from the extremes (and for the operationalmeasures by replacing the 10% highest and 10% lowest values).b Including deals with entry and exit EBITDA multiple available only.

(continued )

been lucky on some deals simply because they bought them at the right timewhen the margins or multiples in the sector were growing. We therefore includethe change above sector for all three operating measures.

Of the three measures of operating performance, the two that we identied asbeing improved during PE ownership—EBITDA margin and multiple—alsoappear to be signicant determinants of abnormal performance: a change ineithermeasure has a positive and economically meaningful impacton abnormalperformance (columns (1) and (2)). Sales growth also relates to abnormalperformance, even though it is not generally improved above sector by PE

(as shown previously in Table 4, panel B).For organic deals, as described in Table 5, panel A, regression (3), abnormal


24/36


Table 5Continued

Panel C: PME and operating performance changes

Independent variables Dependent variable: PME


(1)a (2)a,b (3)a (4)a,b (5)a (6)a,b

xi − xs log sales 0.05∗∗∗ 0.05∗∗∗ 0.08∗∗∗ 0.07∗∗∗ 0.06 0.06∗

(3.55) (3.40) (3.94) (3.83) (1 .64) (1.71)xi − xs margin 0.18

∗∗∗ 0.17∗∗∗ 0.18∗∗∗ 0.16∗∗∗ 0.25∗∗∗ 0.24∗∗∗

(5.65) (4.86) (3.70) (3.31) (3 .78) (3.27)xi − xs mult 0.22

∗∗∗ 0.21∗∗∗ 0.15∗∗ 0.14∗∗ 0.30∗∗∗ 0.30∗∗

(4.58) (4.39) (2.51) (2.38) (2 .71) (2.48)duration − 0.10 − 0.18∗ − 0.02

(− 1.19) (− 1.88) (− 0.10)log value − 0.13∗ − 0.14∗ − 0.08 − 0.09 − 0.20 − 0.20

(− 1.87) (− 1.85) (− 0.87) (− 1.00) (− 1.42) (− 1.41)



Thetable relates public market equivalent (PME) to operational changes ( xi and xs ) with OLSregressions forall exited deals in the McKinsey and LPdata set. We calculate the PME in the spirit of Kaplan and Schoar (2005).We use as independent variables the difference between deal changes and sector changes xi − xs . For theoperational changes of a deal we calculate for EBITDA margin, log sales, and log EBITDAmultiple the averagedifference ( xi = xiT − xit ) from the last pre-PE-ownership year ( t =0) to last PE-ownership year ( T ). We dividethe difference for log sales by the number of PE ownership years ( T − t ) to get annual nominal sales growth. Inthe same way we calculate median changes in the deal corresponding sector companies ( xs = xsT − xst ).First, in regression (1)–(2) we use all organic and inorganic deals. Second, in regression (3)–(4) we show

regressions for only organic (without major M&A events) deals, and in the regression (5)–(6) for only inorganic(with major M&A events) deals.In the lower part of the table we control for deal duration and size. We also add entry time period (1991–93,1994–97, 1998–2000, 2001–02, 2003–07) and PE house dummies for time period and fund xed effects.Note: OLS regressions, t -stats in parentheses with robust standarderrors; signicance level ∗p < 0.1, ∗∗p < 0.05,∗∗∗p < 0.01.a Due to a few but large outliers, we winsorize the distribution of the dependent variable by replacing the 5%highest and 5% lowest values by the next value counting inward from the extremes (and for the operationalmeasures by replacing the 10% highest and 10% lowest values).b Including deals with entry and exit EBITDA multiple available only.

by roughly2.1%; when themultiple fromentry toexitgrows by1, abnormalper-formance increases by roughly 2.5%; and a 1% sales growth above sector altersabnormal performance by 0.8%. Our results are qualitatively unchanged whenwe include duration in the regression in column (4). For inorganic deals, how-ever, there is little evidence to suggest that changes in operational performancedrive abnormal performance: only margin improvements seem to contributeto abnormal performance (although the size and signicance is lower than fororganic deals), but the result is inconclusive, as deals with signicant, follow-onM&A are subject to an error term due to distortion by the acquired entity.

The economic contribution of these operating improvements is substantialfor explaining abnormal performance. In the previous sections we identied a


25/36


and a multiple increase of approximately 1 above sector (Table 4, panel B(ii)).Based on the estimated coefcients in regression (3), operational performancechanges explain nearly one-third of the abnormal performance (4.6/15.4).

Our ndings arealso robust to alternative measures of nancial performance.

In Table 5, panel B, we simply replace the dependent variable abnormal performance with IRR and, in panel C, with PME based on sector. In panelC, regression (3), margin, multiple, and sales improvements above sector aresignicant explanatory factors of PME based on sector. All results that followare also qualitatively robust to employing a cash-in/cash-out multiple and anabnormal cash-in/cash-out multiple as performance measures (excluded herefor thesakeofbrevity but available fromtheauthors upon request). We concludethat operational improvements explain abnormal performance and distinguishgood deals from others in terms of nancial value creation. This is a potentially

important result: it provides insight that operating value creation strategiesmight be at play in different PE deals, as we explore below.

7. Human Capital Factors of Deal Partners

7.1 Financial performance and deal partner backgroundWe now show that the “t” between a deal partner’s background (professionalexperience in nance vs. operations) and deal strategy (organic vs. inorganic)correlates with deal performance. This could be considered evidence thatPE partners’ task-specic skills are relevant to deal performance. We followthe simple but plausible idea of Gibbons and Waldman (2004): that mainlytask-specic learning-by-doing leads to an accumulation of human capital.In particular, we hypothesize that deal partners with nancial backgroundshave learned skills via investment banking activities, or similar, and those withoperational backgrounds have learned revenue growth and cost-cutting skills.If this is true, more partners with nancial backgrounds should be matched todeals with M&A activity, and partners with operational backgrounds shouldbe matched to organic deals, assuming that PE rms match the skills of theirpartners to the needs of their deals.

For the LP sample, we identify the leading deal partner as the partner rstmentioned in the execution of the deal. For the McKinsey sample, we base ouranalysis on additional deal-level information collected during interviews withgeneral partners (GPs). We interviewed seventy-two deal partners involved inour deals; 18 for an additional thirty deals in which no interview was available,we identied the leading deal partner from a number of sources, includingCapital IQ, PE house websites, or press articles. We used the same three sourcesto identify the professional backgrounds (either operational, including industry

18 We veried if the intervieweewas indeed the single leading partner for thedeal with information from theCapital


26/36


or managementconsulting experience, or non-operational, typically investmentbanking andaccounting); their education (science background, MBA, charteredaccountant); and experience in the PE industry (number of years in the PEindustry at deal entry date). To verify the representativeness of our sample of

deal partners, we gathered information on the professional backgrounds foradditional deal partners from two hundred randomly drawn PE investmentsmade by funds with a size similar to the funds in our sample. We found that theshare of operating versus non-operating partners was not signicantly differentfrom our sample (nontabulated results).

Table 6 gives an overviewof thepartners’backgrounds andtheir performanceby deal strategy. For each deal partner, we assign dummy variables forprofessional background, education, and experience (for a detailed descriptionof the variables, see Table A1 in the appendix). The operation dummy records

whether a leading deal partner has ever worked in consulting or industry—thecase for only one-third of the leading deal partners (95 out of 295); most leadingdeal partners are pure bankers or accountants. For organic deals, operatingpartners generate higher abnormal performance on average (22.4%) comparedwith non-operating (nance) partners (16.4%); the reverse is true for inorganicdeals, where operating partners generate lower abnormal performance (12.9%)compared with non-operating partners (20.9%). The same observations aretrue for IRR and PME. We also observe, from the ratio of partner backgroundsby deal strategy, that PE rms appear, on average, to do little, if any, skill

matching: in our sample, 34% (60/177) of partners assigned to organic dealshave operatingbackgrounds, but 30%(35/118)of partners assigned to inorganicdeals have operating backgrounds too. More generally, to examine partnerimpact by deal strategy, we estimate the following regression:

Y i =ϕ+ x ′i β + θ 1operation i + θ 2organic i + θ 3 (operation i ∗inorganic i )+ ε i , (3)

in which Y i is a measure of (out)performance (abnormal performance, IRR, orPME) of deal i . The vector x represents a set of control variables that includeobserved deal characteristics—for instance, holding period in years, deal size,entry period, or PE house dummies—because we want to control for time-based variations and PE house xed-effects in nancial returns. We capture theperformance differences of the deal partner, depending on the strategy, withthree dummy variables. First, operation i is a dummy equal to one for deals withan operating deal partner in the lead (and zero for a non-operating, or nance,deal partner in the lead). Second, organic i is a dummy that equals one for dealswithout major M&A events during PE ownership (and zero otherwise). Third,operation i *organic i is the interaction term of both. Hence, the base groupof deals includes those with nance deal partners and inorganic strategies,and all effects are measured relative to the performance of this group. Finally,φ , β , θ 1 , θ 2, and θ 3 are coefcients to be estimated and ε i is the regression error.The coefcients are estimated by cross-sectional variation because we observe


27/36


6 r m a n c e o v e r v i e w

b y

d e a

l p a r t n e r

b a c

k g r o u n

d a n

d P E s t r a t e g y

L e a d i n g p a r t n e r b a c k g r o u n d

( i ) o r g a n i c a n d i n o r g a n i c d e a l s

( i i ) o r g a n i c d e a l s

( i i i ) i n o r g a n i c d e a l s

n

( 1 )

a b n o r m .

p e r f o r m .

( 2 )

I R R

( 3 ) P M E

n

( 1 )

a b n o r m .

p e r f o r m .

( 2 )

I R R

( 3 )

P M E

n

( 1 )

a b n o r m .

p e r f o r m .

( 2 )

I R R

( 3 )

P M E

s i o n a l b a c k g r o u n d

o p e r a t i o n =

1

9 5

1 8 . 9

5 4 . 8

1 . 9

6 0

2 2 . 4

6 5 . 5

2

3 5

1 2 . 9

3 6 . 4

1 . 7

o p e r a t i o n =

0

2 0 0

1 8 . 3

5 0 . 4

1 . 7

1 1 7

1 6 . 4

4 9 . 3

1 . 4

8 3

2 0 . 9

5 2

2 . 2

n / a

3 4

2 0

5 6 . 3

1 . 8

2 1

1 9 . 1

6 1 . 7

2

1 3

2 1 . 6

4 7 . 7

1 . 4

ti o n

s c i e n c e =

1

7 0

1 6 . 7

5 7

1 . 5

4 8

1 8 . 9

6 1 . 1

1 . 4

2 2

1 1 . 7

4 7 . 9

1 . 7

s c i e n c e =

0

2 2 8

1 8 . 2

4 8 . 9

1 . 8

1 3 1

1 7 . 2

5 0 . 5

1 . 6

9 7

1 9 . 6

4 6 . 6

2 . 1

n / a

3 1

2 6 . 1

6 7 . 1

2 . 2

1 9

2 6 . 1

7 5 . 8

2 . 4

1 2

2 6 . 2

5 3 . 2

1 . 8

m b a =

1

1 0 3

1 7 . 7

5 0 . 9

1 . 7

6 3

1 9

5 6 . 7

1 . 5

4 0

1 5 . 6

4 1 . 7

2 . 1

m b a =

0

2 0 9

1 8 . 5

5 1 . 6

1 . 7

1 2 6

1 7 . 8

5 3 . 3

1 . 6

8 3

1 9 . 6

4 9 . 1

1 . 9

n / a

1 7

2 5 . 7

6 9 . 1

2 . 3

9

2 4

7 8 . 8

2 . 9

8

2 7 . 5

5 8 . 3

1 . 8

c a =

1

5 7

1 6 . 5

4 1 . 8

1 . 2

4 3

1 7 . 4

4 3 . 5

1 . 2

1 4

1 3 . 7

3 6 . 4

1 . 2

c a =

0

2 5 6

1 8 . 8

5 3 . 8

1 . 9

1 4 7

1 8 . 7

5 8 . 1

1 . 7

1 0 9

1 8 . 9

4 8

2 . 1

n / a

1 6

2 3 . 7

6 6

2 . 2

8

1 9 . 9

7 3 . 7

2 . 6

8

2 7 . 5

5 8 . 3

1 . 8

p e r i e n c e

> 5 y r s

1 7 9

1 7 . 1

5 3

1 . 7

1 2 1

1 6 . 8

5 6 . 2

1 . 6

5 8

1 7 . 9

4 6 . 2

2

u p t o 5 y r s

1 0 1

1 8 . 6

4 7 . 3

1 . 7

5 0

1 9 . 2

4 9 . 6

1 . 5

5 1

1 8

4 5

1 . 8

n / a

4 9

2 4 . 3

6 0 . 2

2 . 2

2 7

2 5

6 3 . 4

2 . 4

2 2

2 3 . 3

5 6 . 3

2 . 1

b l e g i v e s a n o v e r v i e w o f t h e b a c k g r o u n d o f t h e p a r t n e r s ( w h o l e d t h e d e a l s i n o u r s a m p l e ) . T

h e “ o p e r a t i o n ” d u m m y i n d i c a t e s i f a l e a d i n g d e a l p a r t n e r h a s a t l e a s t o n c e i n h i s o r h e r

s i o n a l c a r e e r w o r k e d i n a n c o n s u l t i n g o r i n d u s t r y - r e l a t e d j o b . T h e “ s c i e n c e ” d u m m y

i n d i c a t e s a u n i v e r s i t y d e g r e e i n s c i e n c e , a

n d t h e “ m b a ” d u m m y i n d i c a t e s a

m a s t e r o f b u s i n e s s

i s t r a t i o n . T

h e “ c a ” d u m m y s t a n d s f o r c h a r t e r e d a c c o u n t a n t s . I n t h e l a s t p a r t o f t h e t a b l e w e r e p o r t t h e t e n u r e o f t h e d e a l p a r t n e r i n t h e P E i n d u s t r y i n y e a r s a t e n t r y d a t e . C

o l u m n ( i )

ts t h e b a c k g r o u n d v a r i a b l e s f o r a l l o r g a n i c a n d i n o r g a n i c d e a l s i n t h e M c K i n s e y a n d L P

s a m p l e , a n d c o l u m n ( i i ) r e p o r t s f o r d e a l s w i t h o r g a n i c a n d ( i i i ) i n o r g a n i c s t r a t e g y s e p a r a t e l y

. F o r

a t e g o r y , w e r e p o r t t h e m e a n o f ( 1 ) a b n o r m a l p e r f o r m a n c e , (

2 ) I R R

, a n d ( 3 ) P M E .

S i m p l e a v e r a g e s , m e d i a n s i n p a r e n t h e s e s , a b n o r m a l p e r f o r m a n c e , a

n d I R R i n p e r c e n t .

t o a f e w b u t l a r g e o u t l i e r s , w e w i n s o r i z e t h e d i s t r i b u t i o n o f t h e v a r i a b l e b y r e p l a c i n g t h e 5 % h i g h e s t a n d 5 % l o w e s t v a l u e s b y t h e n e x t v a l u e c o u n t i n g i n w a r d f r o m t h e e x t r e m e s .


28/36


Table 7Abnormal performance by deal partner background and PE strategy

Independentvariables

(1) Dependent variable: Abnormal performance in % a

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

Documents

Acharya et al (2012).pdf