Incentives of Private Equity General Partners from Future Fundraising * Ji-Woong Chung Ohio State University Berk A. Sensoy Ohio State University Léa H. Stern Ohio State University Michael S. Weisbach Ohio State University and NBER February 11, 2010 Abstract Incentives from the explicit fee structure (“two and twenty”) of private equity funds understate the actual incentives facing private equity general partners because they ignore the rewards stemming from the effect of current performance on the ability to raise larger funds in the future. We evaluate the importance of these implicit incentives in the context of a learning model in which investors use current performance to update their assessments of a general partner’s ability, and, in turn, decide how much capital to allocate to the partners’ next fund. Our estimates suggest that implicit incentives from expected future fundraising are about as large as explicit incentives from carried interest in the current fund. This implies that the performance-sensitive component of revenue is about twice as large as suggested by previous estimates based only on explicit fees. Consistent with the model, we find that these implicit incentives are stronger when abilities are more scalable and weaker when current performance is less informative about ability. Overall, the results suggest that implicit incentives from future fundraising have a substantial impact on general partners’ welfare and are likely to be an important factor in the success of private equity firms. * Contact information: chung_303@fisher.osu.edu; sensoy_4@fisher.osu.edu; stern_122@fisher.osu.edu; weisbach_2@fisher.osu.edu.
44
Embed
Incentives of Private Equity General Partners from Future Fundraising
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Incentives of Private Equity General Partners from
Future Fundraising∗
Ji-Woong ChungOhio State University
Berk A. SensoyOhio State University
Léa H. SternOhio State University
Michael S. WeisbachOhio State University and NBER
February 11, 2010
Abstract
Incentives from the explicit fee structure (“two and twenty”) of private equity fundsunderstate the actual incentives facing private equity general partners because theyignore the rewards stemming from the effect of current performance on the ability toraise larger funds in the future. We evaluate the importance of these implicit incentivesin the context of a learning model in which investors use current performance to updatetheir assessments of a general partner’s ability, and, in turn, decide how much capitalto allocate to the partners’ next fund. Our estimates suggest that implicit incentivesfrom expected future fundraising are about as large as explicit incentives from carriedinterest in the current fund. This implies that the performance-sensitive componentof revenue is about twice as large as suggested by previous estimates based only onexplicit fees. Consistent with the model, we find that these implicit incentives arestronger when abilities are more scalable and weaker when current performance is lessinformative about ability. Overall, the results suggest that implicit incentives fromfuture fundraising have a substantial impact on general partners’ welfare and are likelyto be an important factor in the success of private equity firms.
Compensation agreements in private equity (PE) partnerships typically give General Part-
ners (GPs) a management fee that is a percentage (usually 1 to 2%) of the amount of capital
committed to the fund, as well as “carried interest” equal to a percentage of the profits
(usually 20%). The carried interest, together with the GP’s own equity contribution to the
fund, aligns the incentives of the GPs with those of the investors in a private equity fund
to a much greater extent than is typical in public corporations. These explicit incentives to
make value-maximizing decisions are commonly thought to be an important driver of the
success of private equity firms (see, for example, Jensen (1989), Kaplan (1989), Kaplan and
Stromberg (2009)).
At the same time, Metrick and Yasuda (2010) report that approximately two-thirds of
expected revenue to PE partnerships comes from fixed-revenue components that are not sen-
sitive to performance. This has led some to suggest that perhaps PE partnerships’ incentives
to deliver high performance are not as strong as previously thought, or at least not as strong
as they should be.1
Missing from these arguments is the fact that application of the explicit terms of the part-
nership agreement omits an important source of implicit incentives facing general partners
in private equity funds: the effect of current performance on the ability of partnerships to
raise larger funds in the future, and consequently to earn higher fees on those funds. Young
partners often give up promising careers in other fields such as investment banking to manage
a relatively small fund, their hope being that high returns from such a first fund will enable
them to raise larger, more lucrative funds in the future. The quantitative importance of this
effect, however, is not known. In other words, for every extra percentage point of returns
(or every extra dollar) earned for the current fund’s investors, how much, in expectation,
does the lifetime or future income of the fund’s general partner change? How large are these
implicit incentives relative to the much-discussed explicit ones? Theoretically, what factors1“It’s the Fees, not the Profits”, The Wall Street Journal, Sept. 13, 2007.
1
ought to affect the size of change in partners’ lifetime incomes as a function of fund returns?
Do these predicted patterns appear true in the data? More generally, how do today’s returns
affect the ability of partnerships to raise capital subsequently?
This paper evaluates the importance of future fund-raising to the incentives of private
equity general partners. To do so, we formalize the logic by which good performance today
could lead to higher future incomes for GPs. We present a model in which a private equity
partnership potentially has an ability to earn abnormal returns for their investors, but this
ability is unknown. Given an observation of returns, investors update their assessment of
the GP’s ability, and, in turn, decide how much capital to allocate to the partners’ next
fund. We derive predictions about the relation between performance of a particular fund
and the fund’s partners’ abilities to raise capital in the future. Intuitively, the model implies
that the more informative the fund’s performance is about GPs’ abilities, the more sensitive
future fundraising should be to today’s performance. In addition, the way in which abilities
can be “scaled” will affect investors’ willingness to commit higher quantities of capital for
a given level of managerial ability. These larger funds will lead, in expectation, to higher
compensation for the partners, since compensation agreements almost always change linearly
with fund size. Given this setup, we derive an explicit formula calculating the effect of fund
performance today on expected future GP compensation.
We test these predictions using a sample of 838 partnerships who manage 1,726 buyout,
venture capital, and real estate funds. In particular, we evaluate the informativeness cri-
terion, which suggests that performance of later funds (for example, a partnership’s third
or fourth fund) should be less informative about ability and hence be less strongly related
to future inflows of capital than would similar performance in a partnership’s first fund.
In addition, the ability of managers to translate their skills to larger funds depends on the
nature of the production process. Given Metrick and Yasuda’s (2010) finding that buyout
funds are more scalable than venture funds, the model predicts that the future fundraising
of buyout funds should be more sensitive to performance than that of venture funds.
2
Our empirical results are consistent with these predictions. For buyout, venture capital,
and real estate funds, the estimated relation between the IRR of a partnership’s current fund
and the expected size of future funds is positive, consistent with Kaplan and Schoar (2005).
The magnitude of this relation varies with the scalability of the investments. Buyout funds,
which are the most scalable, have the strongest relation between IRR and future fund size,
while venture capital funds, which are the least scalable, have the weakest relation. Further
consistent with the model, younger partnerships have a stronger relation between future fund
sizes and current fund returns than older partnerships.
Given these estimates of the sensitivity of future fundraising to current performance, we
next turn to calculating the magnitude of general partners’ incentives. Our model provides an
explicit formula for the change in general partners’ lifetime incomes as a function of the return
of the curent fund. To perform the calculations, we use this formula, plausible parameters
chosen to reflect the characteristics of our sample of private equity funds, and estimates
of expected carried interest and management fees taken from Metrick and Yasuda’s (2010)
simulations. We calculate the expected incremental revenue to the general partners from both
an additional percentage point of returns (IRR) to limited partners (LPs) in the current fund
and an incremental dollar of profits returned to LPs. We break the expected incremental
revenue into two parts: the incremental direct compensation (from carried interest on the
current fund) and the incremental expected indirect ecompensation (from carried interest
and management fees from future funds, whose size is a function of the performance of the
current fund).
For an average sized buyout fund in our sample ($853 million), we estimate that for
an extra percentage point of return (IRR) to limited partners in the current fund, general
partners receive on average an extra $8.7 million in direct fees in the current fund. Estimates
of incremental fees on future funds for each additional percentage point of IRR in the current
fund vary from $6.6 million to $26.3 million depending on the estimation approach used, with
about half of our estimates being approximately twice the incremental direct fees. The level
3
of both direct and indirect fees varies with fund size but their ratio is independent of size.
An alternative approach is to calculate the expected future general partner fees per extra
dollar retured to limited partners in the current fund. For every extra dollar returned to LPs
in the current fund, the GP earns $.25 in carry (assuming that the carry is “in the money”)2,
while indirect profits are between $.19 and $.76 per dollar of profits in today’s fund, for an
average sized buyout fund. Of course, expected income from future funds is riskier than
direct fees and occurs in the future, so while it is not obvious how to discount these fees
appropriately, their true value is substantially less than it their undiscounted expectation.
On the other hand, carries are not always in the money, and our econometric methodology
is likely to understate the human-capital benefits going to general partners for a number of
reasons. Given these estimates of the magnitude of additional expected income through the
fundraising channel, it appears that does the indirect benefits to partners of buyout funds
are of similar magnitude as the direct fees.
We also perform the same calculations for venture capital and real estate funds. Future
fundraising is less sensitive to current performance for these types of funds than for buyout
funds, with venture capital funds displaying the least sensitivity. For an average-sized venture
capital fund, estimates of future profits per dollar of returns in today’s fund range from $.05
to $.26, and from $.15 to $.77 for an average-sized real estate fund. This pattern is consistent
with the scalability arguments, and also suggests that no matter what type of fund one is
considering, an important component of general partner incentives is the effect of today’s
performance on a partner’s ability to raise future funds.
Finally, we consider the cross-sectional pattern of these incentives. Our model suggests
that the sensitivity of future fundraising to current performance should depend on the extent
to which current performance adds incremental information to the market’s assessment of
the general partners’ abilities. We expect that this sensitivity should be greatest for younger
partnerships. Empirically, we find that funds of younger vintages have greater sensitivities of2Using a typical carry of 20%, for LPs to receive an extra dollar, the fund must earn an extra $1.25 in
profits, with $.25 going to the GPs.
4
fund growth to today’s performance, and that this greater sensitivity leads to larger expected
future profits per dollar of today’s returns for these types of funds.
This paper is related to a number of different literatures. It is most directly related to
work on the reasons for value improvements in private equity transactions. Kaplan (1989)
and Smith (1990) document that operating profitability increases following buyouts, although
this pattern appears weaker for more recent buyouts (Guo, Hotchkiss and Song, 2010).
Jensen (1989) and Kaplan and Stromberg (2009) attribute these value increases in large
part to the incentives facing general partners, although both focus on direct rather than
indirect incentives. Kaplan and Schoar (2005), in perhaps the most related analysis to that
done here, emphasize the talent of particular partnerships and estimate regressions showing
that fundraising is related to historical performance. Still, none of this work attempts to
estimate the magnitude of the incentives facing private equity general partners implicit in
the effect of today’s performance on their future income.
Closely related to this work is a large literature on mutual fund inflows and their relation
to historical performance. Ippolito (1992), Chevalier and Ellison (1997), Sirri and Tufano
(1998), Barclay, Pearson and Weisbach (1998), and Sensoy (2009) all estimate regressions
predicting the inflows to mutual funds as a function of a fund’s historical performance, and
find a strongly positive (nonlinear) relation. Berk and Green (2004) explain these patterns
in the context of a learning model similar to the one presented below in which mutual
fund investors draw inferences about managers’ abilities from current performance. A key
difference between Berk and Green’s (2004) analysis and ours is that they focus on the way in
which fund inflows dissipate the ability of fund managers to generate abnormal returns, while
we focus on how the relation between fundraising and performance affects the incentives of
fund managers.
More generally, our work adds new empirical evidence to the idea pioneered by Fama
(1980) that career concerns can be an important source of incentives inside firms. Chevalier
and Ellison (1999) explore the risk-taking of young mutual fund managers compared to
5
old in light of career concerns. Hong, Kubik, and Solomon (2000) and Hong and Kubik
(2003) consider how security analysts’ forecasts may be influenced by career concerns. In
contrast to these papers, our focus is not on agency problems, but rather on quantifying the
importance of career concerns in providing private equity general partners with incentives
to create value for their limited partners. In doing so, our work is in the spirit of Gibbons
and Murphy (1992), who emphasize the importance of understanding total, rather than
only explicit, incentives. The explicit (and observable) compensation formulas in private
equity partnerships together with the empirical relation between fund performance and future
fundraising allows for quantification of these incentives, which in other contexts clearly exist
but are hard to measure.
The remainder of this paper proceeds as follows: Section 2 lays out the model described
above. Section 3 describes the database of private equity funds used in the analysis. Section
4 presents regressions estimating the effect of today’s fund returns on future fundraising.
Section 5 performs calculations that transform the coefficients from these regressions into
general partner incentives, using the analysis from the model in Section 2 as a basis for the
calculations. Section 6 discusses the implications of this work and concludes.
II. Model
In this section we construct a learning model in which investors assign cash flows to private
equity partnerships based on their perceptions of GPs’ abilities to earn abnormal profits.
Investors observe returns earned by partnerships and allocate their capital to partnerships’
subsequent funds based on their posterior estimate of their ability. Given that the compen-
sation system in private equity partnerships is almost always a linear function of fund size
(Gompers and Lerner (1999)), this capital allocation process leads to a strong relation be-
tween performance in a current fund and that fund’s general partners’ future compensation.
6
A. Setup
To formalize this idea, we assume that a particular GP manages a series of funds over time.
This GP has ability equal to α, which is a measure of his ability to earn abnormal returns.
We assume that α is unobservable and that there is symmetric information, so all agents,
including the GPs themselves, have the same estimate of its value. We also assume that
α is constant over time for a particular partnership, which abstracts away from issues of
changing partnership composition, investment environments, or changing ability over time
due to health or other considerations.3
Let i denote the sequence of funds managed by a given GP, ri be the return to LPs for
fund i, I i be the size (committed capital) of fund i, and Ii ∗k (ri) be the total revenue earned
by the GP, where k (r) is an increasing and differentiable function, representing the fraction
of the initial size of the fund that is earned by the GP if performance is r. The function k (r)
should be thought of as the total profits from running a firm that has a return equal to r,
including management fees, carried interest, and other income earned by the fund, such as
additional fees earned by funds for managing portfolio companies. We characterize the GP
compensation in this manner following Metrick and Yasuda (2010), who provide estimates
for k (.) using a simulation approach.
We assume that the fund returns are increasing (in expectation) with the GP’s ability,
α, and that ri ∼ N(α, 1
s
)for all i. [s is the precision of the distribution.] Before any returns
are observed, the commonly held prior assessment of α is α0 ∼ N(α, 1
τ
)4. After observing
3These assumptions greatly simplify the formal analysis but do neglect a number of interesting andpotentially important factors. The assumption that there is symmetric information about managers’ abilitiesdates to Holmstrom (1982), and has been used in similar learning models by Gibbons and Murphy (1992),Hermalin and Weisbach (1998, 2009), and others. Implicitly, the idea is that anyone who can become aGP is smart, hard-working, well-educated, etc., but the key factor that determining who can earn abnormalreturns is an unobservable match between the individual and the tasks associated with earning profits as ageneral partner.
4This should be thought of the expected skill of a particular GP conditional on all observable character-istics prior to any returns being observed. Different GPs will therefore have different values of α0 from oneanother and consequently can raise funds of different sizes.
7
the returns on i funds, the market’s updated assessment of α, αi, is given by:
αi =τα0 + s
∑i ri
τ + is(1)
for all i (DeGroot, 1970 provides a derivation of this Bayesian updating formula).
We assume that investors allocate more funds to managers they believe more able and
have higher values of α, so the size of fund i, Ii = c (αi−1), where c (.) is an increasing,
differentiable function. We assume the GP runs a total of N funds over his lifetime. N is
exogenously determined (e.g. a function of the GP’s initial age). To capture the idea that
at some point performance may be so bad, and the updated assessment of α so low that
the GP is not able to raise additional funds (and so does not actually run N funds), we can
think of the c (.) function as producing follow-on funds of size zero for those cases.
B. Cross-sectional implications
This simple learning model characterizes the way that fund returns affect future fundraising
and, consequently, the future expected compensation for the funds’ partners. Future fund
size is given by Ii+1 = c (αi)= c(τα0+s
∑iri
τ+is
).
B. 1. Informativeness of Returns over a Sequence of funds for a given Partnership
The sensitivity of future fundraising to current performance is given by the derivative of I2
with respect to r1, which is equal to:
∂I2∂r1
=∂c (α1)
∂r1= c′ (α1)
s
τ + s. (2)
The sensitivity of the size of the GP’s third fund to the performance of his first fund, which
equals the sensitivity of the third fund with respect to the second fund is:
∂I3∂r1
=∂c (α2)
∂r1= c′ (α2)
s
τ + 2s. (3)
8
Each of these sensitivities given by equations (2) and (3) is the product of c′ (), multiplied
by a weighting factor that reflects the relative infomativeness of the return to the market’s
perception of the GP’s ability. Intuitively, the function c () represents the quantity of funds
the market chooses to invest in the fund as a function of their best estimate of α. A more
steeply sloped c () function means that for a small increase in managerial talent, the fund
can profitably invest relatively large increases in funds, i.e., the fund is more “scalable”. For
example, if a manager is shown to be talented at buying out companies and increasing value,
he or she can likely buy out larger companies and increase value similarly to what she has
done with smaller companies if the market is willing to fund these investments. In contrast,
if a manager has demonstrated that he is talented at investing in startup companies, he or
she is unlikely to be able to increase fund size much because the size of startup investments
is not scalable (and because it is not feasible to simply increase the number of investments
given that increasing value is a time-consuming process).5
The second term of equations (2) and (3) represents the weight given to returns in
forming the market’s posterior estimate of α. The larger the weighting term, the more
informative the signal and the higher the derivative of future fund size to today’s returns.
As partnerships progress through time, the partnership’s α becomes known more precisely,
so that the optimal updating rule means that subsequent αs do not change as much as earlier
αs for a given return. The overall effect measures the impact on the change in the market’s
best assessment of α for a given return times the amount of capital the market will choose
to invest in a particular fund, given this change in their assessment of α.
Comparing expressions (2) and (3), it seems likely that the sensitivity of future fund
sizes to returns is decreasing in fund sequence. The weighting term strictly decreases as the
numerator is s in each one, while the denominator increases with the sequence number. If
c () is linear or close to being linear, the prediction is unambiguous: the sensitivity of future5Consistent with this logic is the fact that the most successful buyout funds such as KKR and Blackstone
have steadily increased the size of their funds to the point where the largest funds are between $15 and$20 billion in committed capital, while the most successful Silicon Valley venture capitalists such as KleinerPerkins and Sequoia have remained at or under $1 billion in committed capital.
9
fund size to current performance is decreasing in the sequence of funds. However, if c () is
convex and α2 > α1, or if c () is concave and α2 < α1, the pattern may go the other way.
That said, even if c () is highly nonlinear, on average we would not expect α2 to differ much
from α1, so it seems likely that the weighting term effect will dominate. Consequently, in the
data we expect to observe a decreasing sensitivity of future fund size to current performance
as a given partnership manages subsequent funds.
B. 2. Informativeness of Returns across Partnerships
The model suggests that the effect of one fund’s return on the capital inflows to a partnership
depends on the extent to which the fund’s return leads the market to update its prior of
the partnership’s ability. This informativeness of this return depends on how precisely the
market knows the partnership’s α prior to the observation of the return. This idea can be
formalized by examining how the sensitivity of fund size with respect to initial returns varies
with s, the precision of α:
∂2I2∂r1∂s
=∂2c (α1)
∂r1∂s=
∂
∂s
[c′ (α1)
s
τ + s
]=
τ
(τ + s)2
[c′ (α1) +
s
τ + s(r1 − α0) c
′′ (α1)]. (4)
When does this match our intuition that it should be positive? If c () is linear, c′′ () is
equal to zero so the expression is unambiguously positive. If c () is nonlinear, the expression
is still likely to be positive because on average r1 − α0 should be close to zero.
The intuition for why the expression could be negative is the following: Suppose c () is
convex and performance is much worse than expected, so r1 < α0. Then the assessment
of ability adjusts downward, to a point where the slope of c () is smaller (because of the
convexity). The more informative the return r1, the greater the adjustment. Following
similar logic, the expression to be negative if c () is concave.
10
B. 3. Equal Informativeness of Returns within a Sequence
One element of the learning model is that, in each assessment of ability, all past returns are
equally weighted. This means that the marginal impact of any one historical return is the
same as that of any other. In particular, the model implies that:
∂I3∂r1
=∂I3∂r2
= c′ (α2)s
τ + 2s. (5)
This implication is potentially testable, though we do not test it here. The extent to which
it is true for a particular partnership is likely to depend on the amount that a partnership
raises and on how its investment strategy changes over time.
C. Lifetime compensation of GPs
The total revenue earned by the GP over his lifetime is given by:
TR = k (r1) c (α0) + k (r2) c (α1) + k (r3) c (α2) + . . .+ k (rN) c (αN−1) . (6)
This formulation assumes that, following practice, GPs are compensated with a combi-
nation of management fees, which are a function of committed capital, and carried interest,
which is a function of returns times the amount of capital in the fund. We think of the
k (.) function as incorporating these two elements, plus other fee income that is likely to be
proportional to fund size. The expected total revenue at any point in time would be equal to
the realization of k times the current fund size plus the expectation of k times the expected
fund sizes for the remainder of the funds (some of which are size zero with some probabil-
ity). In our empirical implementation, we base our estimates of expected revenue to the
GPs from the current fund on the standard 2% management fee plus 20% carried interest fee
structure. For expected revenue from future funds, we rely on Metrick and Yasuda (2010),
who calculate via simulartion the expected fraction of a fund’s capital that becomes GPs’
11
compensation. 6
We are interested in calculating the incentives of the general partners and decomposing
them into the direct and indirect components. In other words, how much do the partnerships
expect to keep from incremental revenue, and how much of this additional revenue comes
in the form of direct vs. indirect compensation. To perform, this calculation, there are two
potential ways to measure incremental revenue: one can calculate the incremental revenue
from an additional percentage point in returns or one can calculate the incremental revenue
for each extra dollar returned to the fund’s limited partners.
C. 1. Sensitivity of expected Lifetime GP Compensation to Percentage Return
of the GP’s first Fund
To calculate the additional revenue for each percentage point return, we differentiate the
expression for total return (equation (6)) by r1. This calculation leads to:
∂TR
∂r1= k′ (r1) c (α0)+k (r2) c
′ (α1)s
τ + s+k (r3) c
′ (α2)s
τ + 2s+. . .+k (rN) c′ (αN−1)
s
τ + (N − 1) s.
(7)
Our goal is to provide empirical estimates of this derivative. We obtain estimates of the
c′ (αi−1)s
τ+isterms using regression analysis.
1) One approach is to regress I2I1
on r1. The β from that regression is an estimate of∂(I2/I1)∂r1
:
β =∂ (I2/I1)
∂r1=∂ (c (α1) /c (α0))
∂r1=c′ (α1)
c (α0)
s
τ + s. (8)
6We will refer to revenue and compensation synonymously throughout the paper. In fact, private equitypartnerships do have some (but not many) costs that create a wedge between revenue and partner compen-sation. However, many of these costs, such as the costs of renting an office and hiring support staff, aremore or less fixed and do not affect marginal compensation. In addition, our focus is on the indirect aspectsof compensation and its size relative to direct compensation and it seems unlikely that this ratio would besubstantially affected by ignoring direct costs in our calculations.
12
So:
c′ (α1)s
τ + s= βc (α0) = βI1. (9)
2) Because fund growth rates tend to be skewed due to a few exceptionally popular
funds, it is possible that these equations could fit the data better if we use a logarithmic
transformation of fund growth rates. To use this method, we regress ln(I2I1
+ 1)on r1 (the
+1 is to avoid taking the log of zero). We have:
β =∂ln (c (α1) /I1 + 1)
∂r1=
c′ (α1) /I1c (α1) /I1 + 1
s
τ + s. (10)
So:
c′ (α1)s
τ + s= βI1 (1 + c (α1) /I1) = β (I1 + I2) . (11)
Regressions substituting later fund sizes for I2, and later returns for r1, provide estimates
of the other terms in the same way as 1) and 2) above.
C. 2. Sensitivity of expected Lifetime GP Compensation to Dollar Return of the
GP’s first Fund
An alternative way of measuring incentives is to compute the change in general partners’
compensation for an extra dollar returned to the limited partners, which is the metric empha-
sized by Jensen and Murphy (1990). It is fairly straightforward to adjust the estimates we
obtain following the previous subsection to be in terms of dollar return rather than percent
return.
If the return measure used in the previous section is the total return for the fund, then we
just multiply by initial fund size. However, the usual return measure (and the one typically
quoted in the private equity databases, including the Preqin one used below), is the internal
rate of return (IRR) of the fund. The IRR is an annualized return measure which is subject
to well-known problems such as the implicit assumption that intermediate distributions are
reinvested at the IRR. To make things as simple as possible, and because the data used
13
below are not sufficient to make more accurate calculations, we can assume that all capital
is called at once and all distributions are made at once. That is, we can assume that each
fund has a single capital call and a single distribution, and the time (denoted T) between
the two matches the average length of time in the data between the call of a dollar and the
return of the profits associated with investing that dollar. For a private fund, this length is
typically between 3 and 6 years. Metrick and Yasuda (2010) use an expected time to exit
of 5 years for all buyout and venture investment, following Metrick’s (2007) evidence that 5
years is the median holding period for a first-round VC investment.
Under these assumptions, the total dollar return to limited partners in the first fund, D,
is given by:
D =[(1 + r1)
T − 1]I1, (12)
where r1 is the IRR of the first fund. Note that because IRR is a net-of-fee measure, D
represents the total dollars to limited partners, not the total dollars earned by the fund
(some of which go to the GP in the form of management fees and carried interest).
To obtain the sensitivity of the lifetime GP compensation to dollar return of the GP’s
first fund, we can use the estimate of ∂TR∂r1
from the previous subsection together with the
identity:
∂TR
∂D=∂TR
∂r1
∂r1∂D
. (13)
Which holds given that D is invertible:
r1 =[D
I1+ 1
]1/T− 1. (14)
So:
∂r1∂D
=1
TI1
[D
I1+ 1
](1−T )/T
=1
TI1(1 + r1)
1−T . (15)
14
Therefore:
∂TR
∂D=[
1
TI1(1 + r1)
1−T]∂TR
∂r1. (16)
Note that returning an extra dollar implies different marginal percentage returns depend-
ing on the baseline return. To get expectations, we will need to make assumptions about
the baseline return.
III. Data
To estimate the relation between fund performance and capital raising, we rely on fund-level
data provided by Preqin. We consider the largest three types of funds: buyout, real estate,
and venture capital. The total number of buyout, real estate, and venture capital funds is
9,523 in Preqin as of June 2009. Preqin claims to cover about 70% of all capital ever raised
in the private equity industry. In addition, in private communication Preqin informs us that
about 85% of their data is collected via Freedom of Information Act requests and thereby is
not subject to self-reporting biases. While we cannot directly verify these claims, our data
appear similar on key dimensions (notably performance) to that used in prior work.
In all of our analysis, we exclude funds without vintage year data (64) or without fund
size data (1,137). We also exclude 78 funds which are still being raised. To construct our
sample of preceding (i.e., “current”) funds, we require performance (IRR) data. Because
very small funds can grow at extremely high rates, we follow Kaplan and Schoar (2005) and
drop funds with less than $5m (in 1990 dollars) in committed capital. In addition, in order
to allow for sufficient time to elapse between a fund and its follow-on (should the latter
ultimately be raised), we drop funds raised after 2005.
Finally, when a private equity firm raises multiple funds in a given year, we aggregate
funds in that year and compute the fund size weighted IRR. The exception to this is a few
cases in which the same partnership manages, say, both buyout and real estate funds. In
15
those cases, we treat the partnership as two separate partnerships, one each for buyout and
real estate funds. We do so to avoid allowing, for example, a real estate fund to be a follow-on
fund to a buyout fund.
These sample screens leave us with a final sample of 1,726 preceding funds. The sample
consists of 848 (49%) venture capital funds, 640 (37%) buyout funds, and 238 (14%) real
estate funds. For each preceding fund, we ask whether we observe a follow-on fund in the
database. We define a follow-on fund as the next fund raised by the same partnership for
which we have information on fund size. Thus each preceding fund is allowed to have at
most one follow-on fund. If we observe a follow-on fund, we record the size of the follow-on
fund and compute the growth rate in fund size from the preceding fund to the follow-on
fund. If we do not observe a follow-on fund in the data, or if the data indicate follow-on
funds but do not provide size information, we treat this as if the partnership did not raise
a follow-on fund. The working assumption we use throughout the paper is that the absence
of a follow-on fund with size information in the data means the partnership was unable to
raise one, and thus we assume that the partnership raised a fund of size zero in these cases.7
Tables 1 through 4 report the descriptive statistics for the variables which enter the
multivariate regressions, i.e. preceding fund size, preceding fund performance, follow-on
fund size, and fund growth. Since the focus is on growth rates between funds of a given
partnerships, we present statistics by fund sequence, for both “preceding” and “follow-on”
funds, as well as with and without zero-sized funds.
Table 1 presents basic statistics about the funds in our sample. Since the database is
organized by fund pairs to focus on the effect of one fund’s performance on the subsequent
fund’s size, we present statistics for “preceding” funds and “follow-on” funds. Panel A shows
the number of partnerships, broken down by whether each was buyout, venture capital, or
real estate. It also presents statistics the number of “preceding” funds partnership for each7This assumption has the effect of downward-biasing our estimates. Undoubtedly some partnerships
do raise follow-on funds that are missing from the data because the data are incomplete. Additionally,in practice partnerships may dissolve for some reason even though the market would have been willing toprovide capital for a follow-on fund had the partnership desired one.
16
type. Note that these figures are particular to our study, as to be considered a “preceding”
fund, it is necessary to be founded by 2005 and have available data on fund IRR, so the
numbers in Panel A of Table 1 understate the true number of funds per partnership. The
distribution is clearly skewed, with many firms raising just one or two funds and a few
substantially more (the maximum in the sample is 12 funds). Panel B contains information
about fund size and contains statistics for “preceding” funds and “follow-on” funds, with the
latter presented both for the whole sample, which includes zeros for the cases where the
partnership did not raise a subsequent fund, and for the subsample in which they did raise a
follow-on fund. Even if we average in zeros for the cases where there was no follow-on fund,
funds grew substantially, and of course the rates are even higher if we restrict the sample to
those cases where there was a follow-on fund.
Table 2 breaks down the sizes by type of fund by sequence number. Panel A contains
information for “preceding” funds, Panel B for “follow-on” funds including zero-sized funds
for the cases when partnerships did not raise a subsequent fund, and Panel C for follow-on
funds not including these cases. There are substantial differences in size across types of
funds, with buyout funds being the largest, venture capital the smallest, and real estate
in between. In addition, higher sequence number funds are substantially larger than lower
sequence funds, both because they represent successful partnerships with a substantial alpha,
and also because they tend to be located later in time when funds were larger. Finally, the
numbers in Panel B tend to be larger than those in Panel A, indicating that funds grow
over time even when one averages in zeros for the partnerships who do not raise subsequent
funds. The numbers in Panel C are of course even larger because they do not average in the
“zero” funds.
Table 3 reports statistics on fund growth rates (the proportional difference between the
“preceding” fund and the “follow-on” fund). Once again we break down the growth rates by
type of fund and present the data with follow-on funds of size zero (with growth set equal
to “-1”) included in Panel A, and without those funds in Panel B. Similar to the numbers on
17
fund size, buyout funds have grown faster than venture capital funds, and the growth rates
are (by construction) much higher when the growth rates of -1 are not included in Panel B.
Table 4 presents fund-level returns (Panel A) as well as statistics on the time between
fundraisings (Panel B). The mean annual return over the entire sample is 16% for buyout
funds, 14% for venture capital funds and 15% for real estate funds, with somewhat lower
median returns. These numbers are similar to those in Kaplan and Schoar (2005), who
report average returns of 19% for buyout funds and 17% for venture capital funds (p. 1798).
The similarity with Kaplan and Schoar (2005), who use a different data source (Venture
Economics) and a different time period (their sample ends in 2003), is reassurance that our
data do not suffer from important biases missing from data used in prior work.
The time between fundraising averages 3.3 years for the entire sample, with somewhat
longer periods for buyout funds (3.76 years) and shorter periods for real estate funds (2.36
years). As funds become more established, they tend to raise funds more quickly, with an
average time period of 3.95 years between the first and second fundraisings, and averages of
less than 3 years for all fundraisings after the fourth.
IV. The Empirical Relation between today’s Returns and
Future Fundraising
A. Calculating indirect incentives for different types of funds
We next estimate regressions that predict the growth in the capital committed to the subse-
quent fund of a particular partnership as a function of the return in the partnership’s prior
fund. In doing so, we have two goals: first, the model’s main empirical predictions concern
the relation between one fund’s performance and future fundraising, so these equations allow
for testing of these implications. Second, we can use these estimates to calculate the implicit
incentives for general partners arising from the possibility of future fund raising.
18
Table 5 contains the results of regressions predicting the growth rate between a partner-
ship’s current fund and its next one as a function of the IRR of the current fund. Panel
A uses the raw growth rate, while Panel B reports results using the natural logarithm of
the growth rate as the dependent variable, similar to the specification used in Table 10 of
Kaplan and Schoar (2005). We report results separately for buyout, venture, and real estate
funds, as well as results pooling all types (with type dummies included). Each cell reports
three specifications: the first uses IRR by itself as the independent variable, the second con-
tains vintage year fixed effects, while the third uses IRR adjusted by the benchmark return
provided by Preqin (the median IRR of all other funds of the same type with the same
vintage year and geographic focus). Because the dependent variable is bounded below at a
growth rate of -1 (reflecting partnerships that do not raise subsequent funds), we estimate all
regressions using Tobit.8 In all specifications, we cluster standard errors at the partnership
level.
The results in Table 5 suggest that regardless of the specification used, a higher IRR in
the current fund leads to a higher fund size in the future. The coefficients on IRR are all
positive and statistically significantly different from zero. This pattern is true for both the
specifications using raw growth and the log of growth as a dependent variable, although the
R2 values are substantially higher for the logarithmic specifications, indicating that these
specifications fit the data better.
However, there are substantial differences across types of funds in the estimated sensitiv-
ity of future fund size to current fund IRR. The coefficients on buyout funds are noticeably
larger than those on real estate funds, which are themselves larger than the coefficients for
venture capital funds. These differences suggest that buyout funds are the most aggressive
at increasing fund size in response to good performance while venture capital funds the least
aggressive. This pattern is very persistent and is consistent with the idea that buyout funds
are the most scalable; in the context of the model presented above, the results suggest that8The pattern is similar but the coefficients on IRR are much larger if we eliminate these observations and
estimate the equations using ordinary least squares.
19
the buyout funds have the highest c′ ().
Several measurement issues are of concern with these regressions and potentially cloud
their interpretation. First of all, the measure of performance used, the fund’s IRR, is calcu-
lated at the end of the fund’s life, and will not be known precisely by market participants
at the time the subsequent fund is raised, although they are likely to have a reasonably
good idea of the fund’s performance through the exits of its first few deals. This problem is
likely to be most severe when the follow-on fund is raised very quickly following the initial
one. In addition, some large fund companies operate multiple types of funds simultaneously,
such as one focusing on American buyouts and another focusing on European ones. The
performance of the American funds is likely to be reflective of the ability of the American
partners and relatively uninformative about their European counterparts. In this case, there
are likely to be multiple fundraisings shortly after one another but the reason for one being
able to raise a large fund is likely to have little relation to the performance of some of the
partnership’s other funds’ performance.
While a perfect solution to these issues is impossible because we do not have access to the
precise information available to market participants at the time of the fundraising and we do
not know the identity of the actual partners involved with particular deals, it seems likely
that both of these issues are particularly problematic when funds are raised shortly after one
another. Consequently, we repeat the analysis conducted in Panels A and B, eliminating all
preceding-follow-on fund pairs that are raised less than three years apart.9
We present the results including only funds with a three-year gap in fundraising in Panels
C and D of Table 5. This restriction causes us to lose about a third of our observations for
buyout and venture funds and well over half of our real estate funds since real estate funds9Suppose funds were raised in years 1994, 1996, 1997, and 1998 (situations like this are not common in
our data). For our main analysis, we code the 1996, 1997, and 1998 funds as follow-on funds to the 1994,1996, and 1997 funds, respectively. For this robustness analysis, we drop all three pairs because no pairinvolves at least a three-year gap. An alternative approach would be to treat the 1994 fund as the precedingfund for the other three funds, and use the performance of the 1994 fund to predict the fund size of thefollowing three funds. Another alternative would be to count the 1997 fund as the only follow-on fund to the1994 fund. Both of these alternative screening methods yield qualitatively similar results to those presentedin Panels C and D of Table 5.
20
typically raise new funds relatively quickly. Nonetheless, the coefficients on the IRR of the
initial fund are all still positive, mostly still statistically significant, and of approximately
the same size as those in Panels A and B. Consequently, it appears that the issues of not
knowing performance perfectly at the time of the fundraising and of unrelated funds being
grouped in the same partnership are not serious issues in interpreting the coefficients from
our analysis.
B. Indirect incentives of older and younger partnerships
One prediction of the model presented above is that the sensitivity of fundraising to perfor-
mance decreases over time, as a partnership’s ability becomes more well-known to investors.
To examine whether this prediction holds in the data, in Table 6 we add to the regressions in
Table 5 variables for the fund’s sequence number as well as the sequence number interacted
with IRR.
The results in Table 6 indicate that the effect of sequence is two-fold: First, higher
sequence funds tend to grow the fastest as the coefficient on Sequence number is positive
and statistically significantly different from zero. Second, the interaction of sequence number
with IRR is negative and significantly different from zero in most specifications. This pattern
of coefficients is consistent with the model. Higher sequence numbers are associated with
funds that have done well historically and hence have high αs, so they grow the fastest, but
α is estimated more precisely over time, so the marginal impact of current returns on future
fundraising grows smaller over time.
C. Indirect incentives and fund size
In Table 7 we present analogous regressions to those in Table 6 for fund size instead of
sequence, entering the natural logarithm of the preceding fund size and the interaction of this
variable with preceding fund IRR. If fund size is positively related to the precision with which
the market estimates α, we would expect that larger funds would have a lower sensitivity
21
of future fundraising to current IRR than smaller ones, because current performance would
not lead to as much updating of α.
The results in Table 7 are not particularly supportive of this idea. The coefficients on
size-interacted IRR tend to be negative, but not significantly different from zero. These
results suggest that fund size is more reflective of the market’s assessment of α but that the
precision of that assessment does not vary strongly with firm size.
V. General Partner Incentives Implied by the Regression
Estimates
A. Basic results
The positive relation between a fund’s performance and subsequent fundraising indicates
that fund managers do receive some indirect, market-based incentives from the possibility of
raising funds in the future. But how large are these incentives, both in absolute terms and
relative to the direct incentives offered by the carried interest in the current fund? What
factors determine the magnitude of these incentives across funds?
To answer these questions, one must convert the coefficients from Tables 5 and 6 to a
more useful metric for making comparisons. We consider two such metrics, the expected
change in dollar GP revenue for each additional percentage point of IRR in the initial fund,
and the expected change in dollar GP revenue for each dollar returned to the investors.
Equations (9) and (11) from the model presented above provide a method of converting the
coefficients in this manner. In this section we do this conversion and calculate the magnitude
of implied indirect incentives to general partners.
As a benchmark, we first calculate the change in the direct income the general partners
receive from an additional percentage point of IRR. We do this calculation assuming that
the fund is of mean size for each type of fund, earns the mean IRR and General Partners
22
carried interest is “in the money”, so that the partners receive the standard 20% of the gross
profits, (which equals 25% of the amount of incremental income returned to the limited
partners, who receive 80% of the gross profits). Because it assumes the carried interest is “in
the money”, this calculation is therefore an overestimate of the expected change in general
partners’ compensation for an addition dollar returned to the limited partners; for a fund
whose capital is not yet returned to investors or who has not met the “hurdle” (typically
8%), the incremental effect of an percentage point return on general partners’ compensation
is zero.
These estimates of the direct effect of an incremental percentage point return are pre-
sented in Panel B of Table 8 for the base-case equation taken from Table 5. For an average
buyout fund with $853m in capital, an additional percentage point in IRR means an ad-
ditional $8.7m in carried interest for the general partners. For an average-sized ($215m)
venture capital firm the effect is smaller, equal to $2.1m for each percentage point return.10
The calculation of the indirect effect for our base case regressions is the product of four
terms: β·I·N·k. First, for the regressions with raw fund growth (log fund growth) as the
dependent variable, we use the coefficients associated with lag(IRR) from Table 5, Panel A
(Panel B) for each type of fund using specification (1). Second, we multiply this regression
coefficient by the mean size of preceding funds (when the dependent variable is log of fund
growth, we also add the mean size of follow-on funds, as a proxy for expected follow-on size).
The third term in this product is N, the number of future funds raised. Finally, we multiply
by k, the fraction earned by the GP.
There are a number of factors that affect the calculation. First, and perhaps most
important, is the type of fund. As previously discussed, more scalable funds are more able
to utilize increased capital commitments profitably, leading to a stronger empirical relation
between fund performance and subsequent growth in capital. We present calculations of
these indirect incentives for each type of fund, and also using both the raw growth rate and10The large difference between the two comes simply from differences in fund sizes and serves to emphasize
the importance of fund size in determining general partners’ compensation.
23
logarithmic specifications.
A key component of this calculation is the “k” parameter, defined as the fraction of
fund value that will be paid out to the general partners in expectation as a combination of
management fees and carried interest. The appropriate value of k is not obvious as it depends
on the fee structure as well as the entire distribution of returns, which matters because it
affects the likelihood of the carried interest being “in the money” and the amount that it will
be worth conditional on being in the money. Our analysis relies on the work of Metrick and
Yasuda (2010), who perform simulations using details of the compensation structure in the
partnership agreements of venture capital and buyout partnerships as well as data on the
distribution of fund returns. Metrick and Yasuda provide estimates of the distribution of k
for venture capital and buyout funds, and we use similar values for real estate funds (not
considered by Metrick and Yasuda) and the overall sample of funds.
In addition, the number of funds a partnership will manage over its lifetime clearly has
an important impact on the importance of future fundraising. This variable is unobservable;
most successful partnerships are still active albeit sometimes with different individuals as
partners. We calculate indirect incentives for varying values of “N”, the parameter that repre-
sents the number of future funds the partnership represents. Finally, the indirect incentives
depend on the size of the fund, as the dollar incentives for a given percentage point return
are magnified when a fund is larger.
Panel A of Table 8 calculates the magnitude of the indirect effect, in terms of dollars in
compensation to general partners per percentage point of return, and also per dollar returned
to the investors. A convenient way of characterizing this effect is to take the ratio of these
indirect incentives to the direct incentives from the carried interest in the current fund. We
present these ratios in Panel C of Table 8.
It is evident from Panel C that the ratios are large. Recall that the direct incentives are
substantial and are commonly credited as an important source of value in leveraged buyouts.
Yet, the ratios reported in Panel C of Table 8 are sizeable; most values are greater than 1
24
for buyout and real estate funds, with many greater than 2. Clearly, the incentives implicit
in the ability to raise capital in the future are large, and should be much more emphasized
in discussions of the drivers of private equity value creation.
The observed ratios are (by construction) increasing in both “N”, the number of future
funds to be raised, and “k”, the fraction of fund value that goes to the general partners. While
it is impossible to know which is the “correct” value of each, the findings do suggest, not
surprisingly, that the indirect incentives are most important for younger partnerships (who
have more years to reap the benefits of a reputation for α) and partnerships with higher fee
structures. In addition, the ratios are substantially higher for buyout partnerships than for
venture capital partnerships, with real estate partnerships in the middle. This effect comes
directly from the higher regression coefficients for buyout than for venture capital from Table
5, and is likely due to the higher scalability of buyout investments than venture capital ones.
B. Indirect incentives over the partnership’s life
The model presented above suggests that incentives from future fundraising are most im-
portant, not only for younger partnerships, who have the most time left in their careers,
but also for newer partnerships, who have not had time to establish a very precise α. To
consider the quantitative importance of this effect, we present calculations of indirect incen-
tives for partnerships by fund sequence, using the estimated equations presented in Table 6.
Table 9 contains these calculations for a fund with “Sequence” equal to 1, 3 and 5, using the
estimated equations for “All Funds” from Table 6.
The results in Table 9 suggest that the incentives from future fundraising are substantial
for new partnerships and decline rapidly. For funds with “Sequence = 1”, the incentives
from fundraising are substantial, with ratios of about 1 relative to direct incentives. The
estimates clearly indicate that the career-oriented incentives are substantial, expecially for
newer partnerships made up of younger individuals who have the potential to raise many
funds in the future.
25
The fundraising incentives decline dramatically over time. By the fifth fund in a partner-
ship, they are close to zero, as indicated by the low ratios in Panel C of Table 9. This pattern
is consistent with the model presented above, in which alpha is known fairly precisely after
several observations of returns, so each additional return has little incremental information.
This pattern is illustrated in Figure 1, which contains a graph of the estimated ratio of
the indirect to direct incentives as a function of sequence number for varying “N”. For the first
fund in the partnership (“Sequence” = 1), the ratio is highest for higher N’s. It declines with
the Sequence number for each value of N, becoming negative for each around fund 5. While
theoretically, this ratio should never be negative (as that would imply negative incentives
from future fundraising), this pattern is likely due to the imposition of a linear functional
form in sequence number. In future work we hope to explore this issue more carefully using
nonlinear specifications.
VI. Discussion and Conclusion
The analysis presented above suggested a mechanism by which the possibility of future
fundraising can provide incentives to general partners in private equity funds. Partnerships
establish a reputation for being able to generate returns early in their careers, which can
be lucrative later on as it allows partners to raise larger funds subsequently. We provide a
formal model of this process and empirical work consistent with the predictions of the model.
An important innovation is that the model provides explicit formulas for expected lifetime
compensation for private equity partners that allow calculation of the magnitudes of the
incentives implicit in the development of a partnership’s reputation. These incentives are of
the same order of magnitude as the direct incentives from carried interest that have been
commonly credited with much of the value creation in private equity. They are particularly
important for newer partnerships who have yet to establish a reputation and for younger
partnerships, who have the potential to reap the benefits of the reputation for a long period
26
of time.
Clearly there are a number of measurement issues that affect the interpretation of the
results. The numbers discussed throughout the paper are undiscounted, which overstates
their value since the payoffs to a better reputation are both risky and in the future. On
the other hand, it seems likely that aside from the discounting issue, the ratios presented
in Tables 8 and 9 understate the true career-oriented incentives in private equity. The
calculation of the direct effect is likely an overestimate, as it assumes that general partners
keep 20% of each incremental dollar while in fact they get less than that in expectation. In
contrast, the calculation of the indirect incentive effect ignores the possibility that individual
partners can use their personal reputations to raise new funds on their own, or join other
existing firms for lucrative salaries. These possibilities, which do occur fairly regularly, are
examples of valuable reputational capital acquired by general partners through earning high
returns that the formal analysis in this paper does not consider.
This paper contributes to the debate about the incentives of private equity managers
and their effect on value creation. Metrick and Yasuda (2010) find that roughly two-thirds
of the compensation in private equity partnerships comes from fixed rather than variable
components of compensation. Our results suggest that their calculations understate the total
incentive compensation that general partners have, and that incentive-based compensation
in private equity partnerships is larger than previously thought.
The analysis in this paper could be easily applied to other forms of organization. Perhaps
the most straighforward would be to other parts of the money management industry, because
the explicit fee structures in this industry allow for straightforward calculation of the returns
to managing a larger quantity of funds. Hedge funds have a somewhat different institutional
structure than private equity funds with their infinite lives and their compensation system
based on the “high-water mark”. Nonetheless, it should be possible to do similar calculations
for hedge funds as was done in Section 5 of this paper for private equity funds. In addition,
mutual funds and private management of large institutional money, although with a different
27
fee structure, have the same implicit incentives operating through the channel of inflows
chasing returns. Calculating the incentives implicit in this inflow-return relation would be
an important addition to our understanding of these industries.
Most generally, the paper provides some empirical content to the idea started by Fama
(1980) and Holmstrom (1982, 1999) that career concerns can be an important source of
incentives inside firms. Holmstrom in particular argues that in an intertemporal setting,
agents will take actions to maximize people’s perception of their abilities, which can but do
not necessarily coincide with increasing a firm’s profitability. The advantage of focusing on a
private equity setting as we do here is that it is possible to quantify the long-term pecuniary
benefits to agents from these perceptions. Private equity is nonetheless an industry where
incentives, both direct and indirect, are particularly important. The extent to which indirect,
market-based incentives are important in other industries both in absolute terms and relative
to direct incentives, is likely to be an important topic of future research.
28
References
Barclay, M. J., N. D. Pearson, and M. S. Weisbach. 1998. Open-End Mutual Funds and
Capital Gains Taxes. Journal of Financial Economics 49:3-43.
Chevalier, J., Ellison, G., 1997. Risk-taking by mutual funds as a response to incentives.
Journal of Political Economy 105: 1167-1200.
Chevalier, J., Ellison, G., 1999. Career Concerns of Mutual Fund Managers. Quarterly
Journal of Economics 114: 389-432.
DeGroot, M. H. 1970. Optimal Statistical Decisions. New York, NY. MacMillan.
Fama, E. F. 1980. Agency Problems and the Theory of the Firm. The Journal of Political
Economy 88(2):288-307.
Gibbons, R., and Murphy K. J. 1992. Optimal Incentive Contracts in the Presence of Career
Concerns: Theory and Evidence. Journal of Political Economy 100: 468-505.
Gompers, P. and J. Lerner. 1999. An Analysis of Compensation in the U.S. Venture Capital
Partnership. Journal of Financial Economics 51:3-44.
Guo, S., E. S. Hotchkiss, and W. Song. 2010. Do Buyouts (Still) Create Value? Journal of
Finance, forthcoming.
Hermalin, B. E., and M. S. Weisbach. 1998. Endogenously Chosen Boards of Directors and
Their Monitoring of Management. American Economic Review 88:96-118.
Hermalin, B. E., and M. S. Weisbach. 2009. Information Disclosure and Corporate Gover-
nance. Working Paper.
Holmstrom, B. 1999. Managerial Incentive Problems: A Dynamic Perspective. Review of
Economic Studies 66(1):169-182.
29
Hong, H., Kubik, J. 2003. Analyzing the Analysts: Career Concerns and Biased Earnings
Forecasts. Journal of Finance 58:313-351.
Hong, H., Kubik, J., Solomon, A. 2000. Security Analysts’ Career Concerns and Herding of
Earnings Forecasts. RAND Journal of Economics 31:121-144.
Ippolito, R., 1992. Consumer reaction to measures of poor quality: Evidence from the
mutual fund industry. Journal of Law and Economics 35, 45-70.
Jensen, M. C. 1989. Eclipse of the Public Corporation. Harvard Business Review 67:61-74.
Jensen, M. C., and K. J. Murphy. 1990. Performance Pay and Top-Management Incentives.
Journal of Political Economy 98(2):225-64.
Kaplan, S. N. 1989. The effects of management buyouts on operations and value. Journal
of financial economics 24:217-54.
Kaplan, S. N., and A. Schoar. 2005. Private equity performance: Returns, persistence, and
capital flows. Journal of Finance 60:1791-823.
Kaplan S. N., and P. Strömberg. 2009. Leveraged Buyouts and Private Equity. Journal of
Economic Perspectives 23:121-46.
Metrick, A. 2007. Venture Capital and the Finance of Innovation. Wiley.
Metrick, A., and A. Yasuda. 2010. The Economics of Private Equity Funds. Review of
Financial Studies, forthcoming.
Sensoy, B. 2009. Performance Evaluation and Self-Designated Benchmark Indexes in the
Mutual Fund Industry. Journal of Financial Economics 92:25-39.
Sirri, E., Tufano, P., 1998. Costly search and mutual fund flows. Journal of Finance 53:1589-
1622.
30
Smith, A. J. 1990. Corporate ownership structure and performance: the case of management
buyouts. Journal of Financial Economics 27:143-64.
31
Table 1. Descriptive statistics
Panel A. Descriptive statistics for the number of preceding funds per partnership
The table reports the number of funds per partnership by fund type in Panel A. These statistics are computed for funds whose fund size information is available, preceding fund size is greater than or equal to $5 M in 1990 dollars, andpreceding fund performance (IRR) information is available. If a fund is raised in 2005 or before and does not have a follow-on fund, then we assume that the fund failed to raise a follow-on fund and, therefore, the follow-on fund size is setto zero. Panel B reports descriptive statistics for fund size (in $M), growth, performance (IRR), and the time between successive funds.
Fund type # of partnerships# of preceding funds per partnership
Fund size data is provided by Preqin. Preceding funds smaller than $5 M in 1990 dollars, or raised after 2005, are excluded from the analysis. If a fund is raised in 2005 or before and does not have a follow-on fund, then we assume that thefund failed to raise a follow-on fund and, therefore, the follow-on fund size is set to zero. Panel A reports the descriptive statistics for fund size of all preceding funds. Panels B reports descriptive statistics for fund size of all follow-on funds,and Panel C reports the same statistics conditional on raising a fund.
Fund size data is provided by Preqin. Preceding funds smaller than $5 M in 1990 dollars, or raised after 2005, are excluded from the analysis. If a fund is raised in 2005 or before and does not have a follow-on fund, then we assume that thefund failed to raise a follow-on fund and, therefore, the follow-on fund size is set to zero. Panel A reports descriptive statistics for fund growth, defined as (It/It-1)-1, where It-1 is preceding fund size and It is follow-on fund size. Panel Breports the same statistics conditional on raising a fund.
Buyout Venture capital Real Estate All funds
Table 4. Fund Performance and Time between Successive Funds
Panel A. Descriptive statistics for preceding fund performance
Fund performance data is provided by Preqin. Preceding funds smaller than $5 M in 1990 dollars, or raised after 2005, are excluded from the analysis. If a fund is raised in 2005 or before and does not have a follow-on fund, then we assumethat the fund failed to raise a follow-on fund and, therefore, the follow-on fund size is set to zero. Panel A reports the descriptive statistics for fund performance of all preceding funds, as measured by their IRR. Panel B reports descriptivestatistics, by sequence number, for the time elapsed before raising a follow-on fund.
Buyout Venture capital Real Estate All funds
Table 5. Future Fundraising and Current Performance
Panel D: Log fund growth with a three year gap restriction
Panel C: Raw fund growth with a three year gap restriction
Buyout Venture
Buyout Venture
The table presents Tobit regression estimates of the following specifications: (Fund Growth)t = αt + βt (IRR)t-1 + εt. For Panels A and C, the dependent variable is fund growth, defined as (It /I t-1 )-1,and ln(It/It-1 +1) for Panels B and D, where It-1 is preceding fund size and It is follow-on fund size. Model (1) uses raw fund IRR (Preceding fund IRR), (2) includes vintage year fixed effects, and(3) uses raw IRR minus the benchmark IRR. The benchmark IRR is the median IRR of other funds with same vintage year, fund type, and geographical focus. In “All Funds” regressions, fundtype fixed effects are included. Also, IRRs are demeaned by fund type. In all regressions, we exclude preceding funds smaller than $5 M in 1990 dollars, or raised after 2005. If a fund is raised in2005 or before and does not have a follow-on fund, then we assume that the fund failed to raise a follow-on fund and, therefore, the follow-on fund size is set to zero, i.e., fund growth is -1. InPanels C and D, we require there be at least a three year gap between two consecutive fund raisings. Standard errors are clustered at the PE firm level. *, **, and *** indicate statistical significanceat the 10, 5, and 1% level, respectively.
Panel A: Raw fund growth
Panel B: Log fund growth
Real Estate All Funds
Real Estate All Funds
Table 6. Future Fundraising, Current Performance, and Fund Sequence
The table presents Tobit regression estimates of the following specifications: (Fund Growth)t = αt + βt (IRR)t-1 + γt (Fund Sequence)t + δt (Fund Sequence)t-1 (IRR)t-1 + εt. The dependent variable isfund growth, defined as (It/It-1)-1 for Panel A and ln(It/It-1 +1) for Panel B, where It-1 is preceding fund size and It is follow-on fund size. Model (1) uses raw fund IRR (Preceding fund IRR), (2)includes vintage year fixed effects, and (3) uses raw IRR minus the benchmark IRR. The benchmark IRR is the median IRR of other funds with same vintage year, fund type, and geographicalfocus. In “All Funds” regressions, fund type fixed effects are included. Also, IRRs are demeaned by fund type. In all regressions, we exclude preceding funds smaller than $5 M in 1990 dollars,or raised after 2005. If a fund is raised in 2005 or before and does not have a follow-on fund, then we assume that the fund failed to raise a follow-on fund and, therefore, the follow-on fund size isset to zero, i.e., fund growth is -1. Standard errors are clustered at the PE firm level. *, **, and *** indicate statistical significance at the 10, 5, and 1% level, respectively.
Panel A: Raw fund growthBuyout Venture Real Estate All Funds
Table 7. Future Fundraising, Current Performance, and Fund Size
The table presents Tobit regression estimates of the following specifications: (Fund Growth)t = αt + βt (IRR)t-1 + γt (Fund Size)t + δt (Fund Size)t-1 (IRR)t-1 + εt. The dependent variable is fund growth, definedas (It/It-1)-1 for Panel A and ln(It/It-1 +1) for Panel B, where It-1 is preceding fund size and It is follow-on fund size. Model (1) uses raw fund IRR (Preceding fund IRR), (2) includes vintage year fixed effects,and (3) uses raw IRR minus the benchmark IRR. The benchmark IRR is the median IRR of other funds with same vintage year, fund type, and geographical focus. In “All Funds” regressions, fund type fixedeffects are included. Also, IRRs and fund sizes are demeaned by fund type. In all regressions, we exclude preceding funds smaller than $5 M in 1990 dollars, or raised after 2005. If a fund is raised in 2005 orbefore and does not have a follow-on fund, then we assume that the fund failed to raise a follow-on fund and, therefore, the follow-on fund size is set to zero, i.e., fund growth is -1. Standard errors areclustered at the PE firm level. *, **, and *** indicate statistical significance at the 10, 5, and 1% level, respectively.
Table 8. Future Revenue and Current PerformanceThis table presents the estimates of the impact of the current fund's performance on the expected change in total revenue for the GP. Panel A reports the calculation, by fund type and for all funds, of the indirect effect (from subsequent funds)on total revenue of an extra 1% in IRR. Two model specifications are used. In specification (1), the dependent variable is fund growth, whereas in specification (2), we use the log of fund growth. We also report the sensitivity of GP totalrevenue to an extra dollar returned in the current fund (δTR/δD). We use the regression coefficients from (1) in Table 5. For specifications (1) and (2), the indirect δTR/δIRR are equal to β·N·k·Ii, and β·N·k·(Ii +Ii+1) respectively, where β is theregression coefficient with respect to (IRR)t-1, N is the number of subsequent funds raised by the GP, I i is the size of current fund i (in $M), and k is the revenue share from future funds. For buyout and venture funds, we used the distributionof k as estimated in Metrick and Yasuda (2010). Panel B reports the calculation, by fund type and for all funds, of the direct effect on total revenue (from the current fund) of an extra 1% in IRR. We assume in this calculation that the fund is "inthe money" and that the fund has a unique cashflow in, and a unique cashflow out. We use the mean IRR of our sample as the baseline IRR. The direct effect is therefore computed as follows: Ii·k·{[1+IRRi+1%]^(Ti)-1] - [(1+IRRi)^(Ti)-1]},where k is the revenue share from the current fund to the GP (in this case equal to 25%), I i is the mean size of the current fund (in $M), IRR i is the mean IRR, and T i is the expected lifetime of the current fund. Panel C reports the ratio of theindirect effect over the direct effect for the mean fund size: δTR/δIRR (Indirect) / δTR/δIRR (Direct).
BUYOUT FUNDS
Panel A - Indirect effect
N=2 N=4N=3 N=5
Table 8. Future Revenue and Current Performance (continued)
Panel B - Direct effect
Buyout Venture Real Estate All funds
Mean Ii (in $M) 852.82 214.89 476.87 487.56Mean IRRi 16.35% 14.18% 15.07% 15.11%Ti 3 3 3 3Revenue share (k) 25% 25% 25% 25%ΔTR 8.733 2.120 4.777 4.887
Panel C - Ratio of indirect effect over direct effect
Table 9. Future Revenue, Current Performance and Fund SequencePanel A reports the calculation, for all funds, of the indirect effect (from subsequent funds) on total revenue of an extra 1% in IRR, controlling for the sequence number of the fund. Two model specifications are used. In specification (1), the dependentvariable is fund growth, whereas in specification (2), we use the log of fund growth. We also report the sensitivity of GP total revenue to an extra dollar returned in the current fund (δTR/δD). We use the regression coefficients from (1) in Table 6. Forspecifications (1) and (2), the indirect δTR/δIRR are equal to k·{(β1 + (i+1)β3)Ii + (β1 + (i+2)β3)Ii+1 + (β1 + (i+3)β3)Ii+2} and k·{(β1 + (i+1)β3)·(Ii + Ii+1) + (β1 + (i+2)β3)·(Ii+1 + Ii+2) + (β1 + (i+3)β3)·(Ii+2 + Ii+3)} respectively, where β1 is the regressioncoefficient with respect to IRR and β3 the regression coefficient with respect to IRR·Sequence#. I i is the size of current fund i (in $M), and k is the revenue share from future funds. N is the number of subsequent funds raised by the GP. For buyout andventure funds, we used the distribution of k as estimated in Metrick and Yasuda (2010). Panel B reports the calculation, for all funds, of the direct effect on total revenue for the GP, of an extra 1% in the IRR of the current fund. To see how sensitive thisestimate is to the sequence number of the fund, we successively consider funds with sequence number 1, 3 and 5 to be the current fund. We assume that the fund is "in the money" and that it has a unique cashflow in, and a unique cashflow out. We use themean IRR of our sample as the baseline IRR. The direct effect is therefore computed as follows: Ii·k·{[1+IRRi+1%]^(Ti)-1] - [(1+IRRi)^(Ti)-1]}, where k is the revenue share from the current fund to the GP (in this case equal to 25%), I i is the mean sizeof the current fund (in $M), IRR i is the mean IRR of the current fund, and T i is the expected lifetime of the current fund. Panel C reports the ratio of the indirect effect over the direct effect for the mean fund size: δTR/δIRR (Indirect) / δTR/δIRR(Direct).
Figure 1. Importance of incentives from future fundraising over the partnership's lifeThe figure illustrates the future fundraising incentives dynamic for all funds. Values from specification (2) Panel C inTable 9 are used to plot the evolution of the ratio of indirect incentives from future fundraising over direct incentives asa function of the sequence number of the fund (i ). N is the parameter representing the number of future funds to beraised. The fraction of fund value going to the GP is assumed to be 18% (k =18%).