inherited agglomeration effects 16feb11
TRANSCRIPT
1
Inherited Agglomeration Effects in Hedge Funds*
Rui J. P. de Figueiredo Jr. Philipp Meyer Evan Rawley University of California Berkeley The Wharton School The Wharton School Haas School of Business University of Pennsylvania University of Pennsylvania [email protected] [email protected] [email protected]
February 16, 2011
Abstract This paper studies inherited agglomeration effects, the economic benefits that accrue to managers while working at a parent firm in an industry hub that can be subsequently transferred to a spinoff, regardless of where the new venture is located. We test the evidence for inherited agglomeration effects in the context of the hedge fund industry and find that hedge fund managers who previously worked in New York and London outperform their peers who worked elsewhere previously by about 1.5% per year. The results are driven primarily by managers who worked in investment management positions previously, and are at least as large as traditional agglomeration effects that arise from being located in an industry hub contemporaneously. The evidence suggests that managers can transfer the benefits of agglomeration from parent firms to their spinoffs.
1. Introduction
New ventures benefit from the knowledge and connections managers transmit to the new
firm from their previous work experiences. Thus, it is not surprising that scholars have found
that when managers join new firms their previous employers’ (“parent firm”) characteristics
systematically influence the new venture’s performance (Agarwal, Echambadi, Franco and
Sarkar, 2004). This paper builds on the literature on human and social capital transmission
between parent firms and new ventures, by examining how a parent firm’s exposure to
agglomeration effects, or local economies of scale (Marshall, 1920/1890; Krugman, 1991),
benefits new ventures. By focusing on the transmission of agglomeration effects between parent
* We thank Raffi Amit, Iwan Barankay, Charles Baden-Fuller, Olivier Chatain, Gary Dushnitsky, Simone Ferriani, David Hsu, Sascha Klamp, Joris Knoben, Ethan Mollick, Chris Rider, Lori Rosenkopf, Harbir Singh, MB Sarkar, and participants at the Atlanta Competitive Advantage Conference, the EGOS conference in Lisbon, the DRUID conference, the Strategic Management Society conference in Rome and the Academy of Management conference in Montreal for helpful comments and suggestions. Elaine Ho, Bryan Hsu, and Austin Winger provided outstanding research assistance. We are grateful to the Wharton Entrepreneurship and Family Business Research Centre at CERT, the Kauffman Foundation, and the Rodney L. White Center for Financial Research at The Wharton School for their generous financial support.
2
firms and their spinoffs, the paper extends the entrepreneurship literature on agglomeration to
develop a set of simple testable hypotheses that link parent firm location to new venture
performance. The basic idea is that managers develop more human and social capital when they
work in an industry hub (Glaeser, Kallal, Scheinkman and Shleifer, 1992) and, therefore, they
have more to offer a new venture that performs a closely related activity. Thus, we expect that
when a new venture’s senior managers previously worked in an industry hub, the venture will
experience a positive performance effect even if the new venture itself is not located in an
industry hub.
While scholars have made progress toward understanding how agglomeration effects at
the parent firm-level influence entrepreneurial spawning rates, or the rate at which the parent
firm creates new ventures (Gompers, Lerner and Scharfstein, 2005; Klepper and Sleeper, 2005),
less is known about how parent firm-level agglomeration effects influence the performance of
their spawns.1 We call such parent firm-level agglomeration effects “inherited” agglomeration
effects to emphasize that human and social capital is transmitted from parent firms in industry
hubs to their entrepreneurial spawns via managers who leave parent firms to manage spawns,2
and to distinguish this effect from traditional concurrent agglomeration effects.
We test the implications of inherited agglomeration effects in the context of the global
hedge fund industry. The hedge fund industry is a good setting for a study of inherited
agglomeration effects for four main reasons. First, the industry is characterized by high rates of
new venture formation over the last three decades, which provides us with a wealth of new
ventures to study. Second, most hedge funds are founded and managed by individuals who
worked in closely related financial firms previously, which provides prima facie evidence that
1 We use the terms spawn and spinoff interchangeably throughout. 2 The use of the biological metaphor of heredity in the context of new firm formation is due to Nelson (1995) and Klepper and Sleeper (2005).
3
“inherited” knowledge and connections are important in this industry. Third, and related, as with
many professional service firms, hedge funds are knowledge capital intensive businesses; yet,
there is very limited formal intellectual property protection in the industry. Thus, it seems
straightforward to hypothesize that hedge fund managers’ human capital is transferable and
could meaningfully impact the performance of their firms. Finally, hedge fund performance can
be measured with a relatively high degree of accuracy, even for very young firms, which
provides us with a more precise measure of spawn performance than firm survival (Agarwal,
Echambadi, Franco and Sarkar, 2004) and pre-money valuation (Chatterji, 2009).
We find that hedge funds whose principal managers were previously employed by parent
firms located in financial services industry hubs—New York or London—outperform their peers
by about 1.5% per year, net of fees. The results are robust to the inclusion of fixed effects for the
major (“bulge bracket”) investment banks, and to controls for manager selection into New York
and London employment based on all observables. Most of the inherited agglomeration effect
stems from managers whose pre-founding employment was in investment management, the job
type most closely related to hedge fund management. We also find evidence in favor of
traditional agglomeration effects: hedge funds located in industry hubs achieve better
performance than hedge funds located outside of New York and London, though inherited
agglomeration effects appear to be at least as large as traditional agglomeration effects.
By developing and testing the concept of inherited agglomeration effects, the paper
contributes to the literature on agglomeration effects in entrepreneurial ventures. Although we
cannot control for all sources of unobservable heterogeneity in our empirical tests, we use micro-
data, along with a wide range of controls to deliver stylized facts about the relationship between
parent firm location and new venture performance, which suggest that inherited agglomeration
4
effects can be an important source of competitive advantage for new ventures.
2. Conceptual development
Helfat and Lieberman (2002) suggest that the performance of new firms is shaped by
entrepreneurs’ prior affiliations and knowledge gained through previous employment. A number
of papers have found evidence consistent with this idea. For example, Burton, Sorenson and
Beckman (2000) show that the entrepreneurial prominence of parent firms influences spawning
rates of Silicon Valley start-ups as well as the strategy and innovation output of new firms.
Phillips (2002) finds evidence that social capital is transferred between parent and spawns among
Silicon Valley law firms. Agarwal, Echambadi, Franco and Sarkar (2004) demonstrate that
technical knowledge influences spawning rates and spawn survival rates in the disk drive
industry. Building on the idea that managers transfer knowledge between parent firms and
spawns, Chatterji (2009) shows how non-technical knowledge influences spawn performance in
the medical device industry. This paper builds on the idea that parent firm characteristics are
transferable to new ventures and extends this literature by examining how agglomeration effects
at the parent firm level influence the performance of spawns.
Of course, the idea that new ventures can benefit from agglomeration effects is well
established. Braun and MacDonald’s (1982) pioneering work on the rise of the semiconductor
industry in Silicon Valley (many arising from Fairchild Semiconductor) pointed to an important
correlation between new firm formation and agglomeration effects due to knowledge spillovers
and social networks (for example, see p.129). Jaffe, Trajtenberg, and Henderson (1993)
provided large sample empirical support for the idea that agglomeration effects influence
knowledge spillovers, finding that patent citations tended to cluster locally. In an in depth
comparative study of Silicon Valley firms and firms along Route 128 in Massachusetts, Saxenian
5
(1994a) suggested that start-ups flourish in agglomerated environments because they are more
centrally embedded in the industry’s customers, supplier and support networks. In related work,
Saxenian (1994b) emphasized the role that close connections between spawns play in facilitating
the spillover of innovation producing knowledge. Sorenson and Audia’s (2000) study of the
footwear industry and Stuart and Sorenson’s (2003) study of biotechnology start-ups find large
sample quantitative support for Saxenian’s (1994a, 1994b) hypothesis that social networks
underlie agglomeration effects. This paper builds on the prior work on agglomeration effects in
entrepreneurial firms by considering how agglomeration influences the development and flow of
human and social capital in new firms. We extend the stream of research on agglomeration
effects in start-ups by focusing on the intergenerational transmission of agglomeration effects
between parent firms and spawns.
While much of the literature on agglomeration effects in entrepreneurial firms has
focused on concurrent agglomeration effects, or benefits that accrue to start-ups when they locate
in an industry hub, two papers have explored inherited agglomeration effects. Klepper and
Sleeper (2005) show that agglomeration effects moderate knowledge transference between
parent firms and their spawns in the laser industry, which suggests that agglomeration effects can
be transmitted between firms. In related work, Gompers, Lerner, and Scharfstein (2005) study
founders of venture capital-backed firms, showing that individuals who work for entrepreneurial
parent firms located in industry hubs are exposed to suppliers, customers, and venture capitalists,
gaining entrepreneurial skills during their employment, thereby equipping them for founding of
their own businesses. While the extant work suggests that agglomeration effects are inheritable,
neither paper examines how inherited agglomeration effects influence spawn performance. Yet,
if initial founding conditions imprint organizations (Stinchcombe, 1965; Geroski, Mata, and
6
Portugal, 2010; Johnson, 2007), inherited agglomeration effects should have a lasting impact on
new firms and their performance. Thus, it seems natural to extend the literature one step further
by asking how inherited agglomeration effects influence new firm performance.
The extant literature makes two well supported predictions that form the building blocks
for our first hypothesis: agglomeration effects increase the value of an individual’s human and
social capital when they work in an industry hub, and new ventures inherit human and social
capital from their managers. Therefore, we expect that ceteris paribus new ventures led by
managers from an industry hub will outperform ventures led by managers from outside the
industry hubs. Thus, our inherited agglomeration effects hypothesis can be summarized as:
Hypothesis 1: New ventures will perform better when their principal managers
were previously employed by parent firms located in the relevant industry’s
geographical hub.
Hypothesis 1 builds on the central ideas in the agglomeration and entrepreneurial
spawning literatures. When managers work in industry hubs, they are exposed to valuable
knowledge, information, and ideas as well as people that managers on the periphery of the
industry are less likely to encounter. The unique access to such critical drivers of human and
social capital that geographic proximity provides not only makes a manager more likely to form
or join a new venture, it also gives them a competitive edge over their peers who work outside of
industry hubs when they decide to found and lead new ventures.
Inherited agglomeration effects differ from traditional agglomeration effects in the sense
that inherited effects are those received by new ventures from managers who join new firms
from existing firms, regardless of the new ventures location, while traditional agglomeration
effects are obtained by the new venture concurrently. The novelty of this paper rests on
7
developing and testing the idea of inherited agglomeration effects on new firm performance, but
we also test for traditional agglomeration effects, which predict that new venture performance
will be positively influenced by the new venture’s location in an industry hub. Therefore, we
also hypothesize:
Hypothesis 2: New ventures will perform better if they are located in the
industry’s geographical hub.
Inherited agglomeration effects and traditional agglomeration effects have much in
common intellectually, but they do make different predictions. In particular, inherited
agglomeration effects predict that new ventures will flourish when their senior managers worked
in an industry hub previously, regardless of the new ventures geographic location, while
traditional agglomeration effects predict that new ventures will flourish when the new venture is
located in the industry’s geographic hub, regardless of where their managers worked previously.
Figure 1 summarizes the points of comparison and contrast between Hypothesis 1 and 2.
-------------------------------------- Insert Figure 1 about here
--------------------------------------
While in principle inherited agglomeration effects could obtain via a manager who
worked in any job in an industry’s geographical hub prior to joining a new venture, the
mechanisms underlying the theory of inherited agglomeration effects suggest that some forms of
prior employment will be far more valuable than others. In particular, one should expect that a
new venture will benefit more when a manager’s previous job was more closely related to the
new venture’s operations because manager’s working in closely related functions will bring more
relevant knowledge and social capital to the new venture. Therefore, our third hypothesis is:
8
Hypothesis 3: Inherited agglomeration effects will be greater when a manager’s
pre-founding experience is more closely related to a new venture’s operations.
3. Empirical context: the hedge fund industry
Hedge funds are private investment vehicles that raise capital from high net worth
individuals and institutional investors to exploit investment opportunities. As private investment
vehicles, hedge funds are not subject to the same rules and regulations that govern mutual funds,
which gives them more investment flexibility. However, private investment vehicles are not
allowed to market themselves to the general public. We exploit the fact that hedge funds report a
substantial amount of information about their managers and performance as an indirect
marketing tool. We use this information to analyze how the location of hedge fund managers’
previous parent firm incumbent employer influences the post-founding performance of hedge
fund spawns.
The hedge fund industry emerged as an important sub-sector of the financial services
industry in the 1980s, though the first hedge fund was founded in 1949 by Alfred W. Jones. The
industry has subsequently undergone rapid expansion with compound annual growth in assets
under management above 15% to approximately $1.7 trillion in 2010 (Hedge Fund Research,
2010). It is particularly noteworthy that the hedge fund sector has witnessed significant
entrepreneurial activity over the past three decades: our estimates, based on industry data and
discussions with hedge fund managers, suggest that 10,000-12,000 hedge fund firms have been
founded since 1978.
Part of the reason for the remarkable number of new ventures formed in the hedge fund
industry is undoubtedly the lack of intellectual property (IP) protections over investment
strategies, which allows individuals to appropriate the value of the knowledge gained while
9
working at their parent firm by spawning a new venture. For example, traders at Goldman’s risk
arbitrage desk, famously led by Robert Rubin, spawned Farallon, TPG-Axon, Eton Park,
Taconic, Och Ziff, and Perry amongst others.3 Aside from factors such as more entrepreneurial
autonomy and flexibility, the key driver of individuals leaving their current jobs to found and
manage hedge funds is the attractiveness of the external environment; hedge fund managers are
amongst the most highly remunerated professionals in the world. The average annual
compensation for a partner of a hedge fund in 2009 was estimated to be $10 million. The
conditions in the hedge fund industry fit closely with Hellmann’s (2007) “entrepreneurial
equilibrium”—employees can appropriate the firm’s intellectual property and the external
environment is very attractive—where individuals explore their entrepreneurial ideas through
external ventures. Thus, our emphasis on manager’s opportunities to learn and network while
employees of parent firms in industry hubs as key drivers of new venture performance in the
hedge fund industry seems warranted. In this regard, as a knowledge intensive industry with
limited intellectual property protection, the hedge fund sector provides a particularly good setting
to study inherited agglomeration effects. Moreover, given the knowledge intensity and lack of
intellectual property protection, the hedge fund sector appears representative of many other
service sectors in the economy.
Broadly, hedge funds are classified into five broad investment styles: “macro funds”
invest in financial securities based on global macro-economic trends; “equity long/short funds”
invest in equity of value generating while short selling equity of value reducing companies;
“event-driven funds” invest in financial securities based on corporate events; “relative value
funds” exploit securities mispricing; “fund-of-Funds” invest in other hedge funds. However,
3 Other examples include Julian Robertson’s Tiger Management Corporation, which has spawned a large number of “Tiger Cubs”, including Maverick, Lone Pine, Touradji, Sumway, Lone Pine, and Millgate, amongst others
10
within each investment style a wide number of (often overlapping) trading strategies exist, which
dampens the importance of differences between firm’s stated trading strategies.
Hedge funds firms derive their revenue from management and incentive fees.
Management fees are annual asset management fees based on the net asset value of the assets
under management (AUM). The usual management fee in our sample period was around 2% of
AUM. Incentive fees entitle the hedge fund firm to a percentage of the achieved investment
return, usually subject to high water marks (no incentive fees are paid until an investor’s net
asset value exceeds their initial investment). Typically, incentive fees are 15-20% of gross
returns. In our analysis, we focus on hedge fund performance as reported to their investors. In
other words, we analyze returns net of management and incentive fees, rather than attempting to
calculate returns to the hedge fund managers themselves. While estimating returns to managers
would also be interesting it would require additional assumptions about investment timing, high
water marks and cost structure. Because we do not observe these factors, we confine ourselves
to analyzing a firm’s performance to its investors.
4. Data and empirical design
4.1. Sample construction
Our unit of analysis is the hedge fund. Data on hedge fund performance, location, size
and inception date were obtained by combining the two most extensive and widely used hedge
fund databases: Lipper-TASS (“TASS”) and Hedge Fund Research (HFR). The data sets include
data on over 12,000 individual funds from 3,113 hedge fund firms during the period 1978 to
2006. Though the datasets are self-reported, they are widely believed to be broadly
11
representative of the global hedge fund industry.4
We hand-collected detailed biographical information about the top two hedge funds
managers in 684 hedge fund firms from Barclays’s 2004 (second quarter) MARhedge database,
including manager name, educational history and previous two employers. While all the
managers in our dataset are not founders, all are senior executives who can be reasonably
assumed to have a significant influence on the performance of the hedge fund.5 MARhedge only
sporadically reports graduation and employment dates so, unfortunately, we cannot exploit time
variation in the biographical data.
MARhedge does not track defunct funds, but we verified that the MARhedge data was
drawn from a statistically equivalent pool of hedge funds as the TASS and HFR datasets by
comparing the means of the common variables. There were no meaningful statistical differences
between the data sets with respect to firm size, location, and fees charged to investors. We then
merged the MARhedge data with the TASS and HFR datasets, which resulted in 1,585 unique
hedge fund manager-previous employer pairs of which 1,058 pairs had full hedge fund
performance and previous employer location information.
In order to conduct the empirical tests with meaningful controls on previous employer
characteristics, we restricted the test sample to include only job spells with previous employers
4 While the TASS dataset is free of survivorship bias, we run our main tests on a pooled sample of TASS and HFR. We confirmed that our results are robust to dropping the firms that are listed in HFR only. 5 Some hedge fund firms in MARhedge list only one senior manager, typically the founder and CEO, though many list two co-founders. Other firms list multiple top executives, like the CEO and CFO or the CEO and chief portfolio manager without indicating which managers were founders. Where more than two managers are listed in MARhedge, we used the associated biographical reports to identify the two most senior executives. The results are consistent if we limit our test sample to the firm’s top manager instead of using the firm’s top two managers. As the hedge fund managers’ biographies were only available in an on-screen format, we had two individuals enter the data independently of each other into a database program, and afterwards checked for consistency between the two entries. If there was a discrepancy, we had our research assistants double-check the entry with the original data. For job spells where a previous employer was taken-over by another firm (e.g. Salomon Brothers/Citi) and where we could not ascertain whether the manager worked for the previous employer when it was independent or when it was already affiliated with the new parent company, we coded the job spell as being part of the new (acquiring) organization. However, our results are robust to coding the previous employer as independent (pre-take-over), or to dropping these observations.
12
that were listed on public stock exchanges in the United States and United Kingdom between
1978-2007 including NYSE, NASDAQ, AMEX and the LSE. The resulting data set consists of
658 hedge jobs spells, our unit of observation, from 548 hedge fund managers at 414 hedge
funds that were spawned from 95 unique previous employers. Table 1 shows the characteristics
of the 25 most prolific parent firms in our sample terms of the numbers of job spells associated
with managers who left the parent firm to become a senior manger of a hedge fund.
-------------------------------------- Insert Table 1 about here
--------------------------------------
4.2. Dependent variables
We use two different dependent variables to measure hedge fund performance: 8-factor
average abnormal returns (“ALPHA”) and the 8-factor Information Ratio (“IR”). Abnormal
returns are estimated, using the standard approach as the difference between fund i’s actual
return fund at time t and the fund’s expected return, using equation (1):
(1) Rit = ai + Rft +XtBi + eit,
where Ri is a fund’s monthly raw return, net of fees charged to investors, and the vector X
contains factors that form the fund’s expected return. Equation (1) captures the abnormal return
of hedge funds by taking into account eight hedge fund specific factors in the vector X, seven
factors from Fung and Hsieh (2004) and a liquidity risk exposure factor from Pástor and
Stambaugh (2002).6 In equation (1), the term ai is the time invariant component of a fund’s
6 The seven factors are Rm-Rft (from the CAPM), SMB (from Fama and French 1996), the excess returns on portfolios of straddle options on currencies, on commodities, and on bonds, the yield spread of the duration adjusted U.S. 10-year Treasury bond over the 3-month T-bill, the change in the credit spread of the duration adjusted Moody’s BAA bond over the 10-year Treasury bond (BAA); The Fung and Hsieh factors are available at http://faculty.fuqua.duke.edu/~dah7/HFData.htm. The Pastor-Stambaugh liquidity factors are available at http://finance.wharton.upenn.edu/~stambaug/ liq_data_1962_2008.txt. The results are robust to using raw (unadjusted) returns and to estimating alphas and information ratios based on the standard model used for capturing
13
performance (fund alpha) and e is the residual. We estimate fund alphas and the coefficients on X
by running fund-level longitudinal regressions. After estimating the fund-level alphas, we
compute the fund’s information ratio at the by dividing the fund alpha by the standard deviation
of the fund’s excess returns (fund alpha plus the residual). Since the information ratio adjusts a
fund’s performance measured relative to systematic risk exposure by its idiosyncratic risk
exposure, the information ratio offers some conceptual advantages over fund alpha in our
context. To control for outliers, we windsorized 8-factor fund alphas and information ratios at
the 1st and 99th percentile. We then compute a firm-level alpha and information ratio by
averaging the measures across all the funds in the same firm on an AUM weighted basis.
4.3. Explanatory variables
Our key independent variable to test Hypothesis 1 measures the location of the previous
employer of the hedge fund firm’s principal managers. Specifically, we capture whether the
headquarters location of the previous employer (parent firm) of a hedge fund’s principal manager
was in one of the geographical hubs of the financial service industry—New York or London.7
We measure parent firm location as a binary variable, PARENT_CENTER, which equals 1 if the
location of the previous employer is New York City or London and 0 otherwise. In the absence
of more detailed information on the exact location of the manager’s previous employment, the
previous employer’s headquarters location is a frequently used proxy in the entrepreneurial
spawning literature for the location of the entrepreneur’s previous employment (Klepper and
abnormal returns in the mutual fund industry: the 4-factor model (i.e. the three Fama and French (1996) factors and Carhart’s (1997) momentum factor. 7 New York and London account for the world’s largest equity, debt and derivatives markets and have been ranked as the top two global financial centers by the Global Financial Centers Index and the Worldwide Centers of Commerce Index in every year these indices were compiled. The results are robust to allowing New York and London location dummies to enter separately.
14
Simons, 2000; Gompers, Lerner and Scharfstein, 2005). In our sample, this proxy seems
reasonable as the vast majority of individuals worked for wholesale, corporate and investment
banking divisions of financial institutions and these divisions are frequently co-located with the
institutions’ headquarters. Although there is undoubtedly some measurement error in our
explanatory variable PARENT_CENTER because some individuals in our sample did not work at
their firm’s headquarters, the effect of misclassification errors in PARENT_CENTER is to bias
our results toward zero (Aigner, 1973).8
The key independent variable to test Hypothesis 2 captures the location of the hedge
fund; HF_CENTER is a binary variable set equal to one if the hedge fund is based in New York
or London, and zero otherwise. In order to test Hypotheses 3, which predicts that inherited
agglomeration effects will increase with closely related work experience, we characterize hedge
fund managers’ previous as being very closely related to hedge fund operations or otherwise.
While almost all of the managers in our sample worked in finance jobs in some capacity before
starting or joining a hedge fund, interviews with industry experts suggest that investment
management jobs, investing in financial securities or active management of investments, are the
type of jobs most closely related to hedge fund management. To create an investment manager
dummy variable, we coded as unity any parent firm job spell in financial security trading (third
party or proprietary traders), mutual fund management, asset management, trust management,
wealth management, private banking, pension fund management or other institutional investor
8 Aigner (1973) shows that misclassification of a binary explanatory variable always leads to a coefficient that is biased toward zero as long as the classification error is uncorrelated with the regression disturbance term. In our case, the most plausible way in which this assumption would be violated would be if firms headquartered outside of New York and London systematically assigned workers with the highest unobservable (to the econometrician) abilities to their New York or London offices. However, even if misclassification is correlated with the disturbance term in this way, the bias on the coefficient on PARENT_CENTER will be toward zero because the effect of this type of misclassified high ability workers will be to increase the mean performance of worker’s job spells in the control (non-PARENT_CENTER) population. Of course, heterogeneous unobservable ability might bias our results in other more pernicious ways as we discuss in more detail below, especially in section 5.2.
15
fund management and hedge fund and alternative investment management. Non-investment
management job spells were coded as zero, including any other finance or non-finance related
job, most notably corporate finance professionals, brokers, research analysts, commercial
bankers, consultants, lawyers, accountants and academics.9 We then interacted the investment
manager dummy variable with PARENT_CENTER in order to create a variable that measures the
marginal effect of closely related experience, conditional on inherited agglomeration effects, on
hedge fund performance.
Table 2 shows the summary statistics for the dependent, independent and control
variables at the job spell level. The average information ratio is 0.27, or 27 basis points per
month (3.24%/year) with a standard deviation of 28 basis points per month. 58% of job spells
were in New York or London (PARENT_CENTER), 46% of (parent firm) job spells led to future
employment at a hedge fund located in New York or London (HF_CENTER) and 42% of job
spells were in investment management.
-------------------------------------- Insert Table 2 about here
--------------------------------------
4.4 Control variables
As indicated by the three sections in Table 2, we include three types of control variables
in our analyses: manager controls, job spell controls, and hedge fund controls. We use
educational background measures to control for managers’ ability before their employment spells
with parent firms. Chevalier and Ellison’s (1997) find that mutual fund manager performance is
correlated with the median SAT score of the undergraduate educational institution they attended.
We, therefore, include in all of our specifications the median SAT score of each manager’s
9 Two research assistants coded job types independently of each other. We checked for consistency of the two coding entries, and had the individuals re-code any discrepancies.
16
undergraduate educational institution as well as a set of dummy variables for their highest
educational level achieved.10 36% of the job spells in our sample were associated with managers
who held an MBA degree, 4% were associated with managers with Ph.Ds, 2% were associated
with managers with JD, and 11% held other postgraduate degrees.
We control for parent firm quality using two direct measures of parent firm performance,
whether the parent firm was ranked as a top 25 securities trading firm in any year 2000-2007 by
Institutional Investor Magazine (RANKED), and the parent firm’s long-run average Tobin’s Q.
The Institutional Investor ranking is considered a good benchmark by industry experts for
identifying high-quality trading firms. Since hedge funds are typically heavily engaged in
securities trading, we include this measure to control for the quality of a manger’s parent firm’s
trading capabilities. 74% of the job spells in our sample were associated with firms on the
Institutional Investor list (Table 2). Tobin’s Q provides a measure of how much market value a
firm has created historically, which offers another measure of parent firm quality. The average
Tobin’s Q in our sample was 1.21, which means the average job spell was associated with a firm
whose market value was 21% higher than the book value of the firm’s assets. We also include a
control for the effects of industry relatedness between a parent firm and their spawns’
performance, using a dummy variable based on whether the manager’s job spell was in the
financial industry, measured by the first digit of the parent firm’s primary SIC code (SIC6).11
Our quality and relatedness measures are good, but incomplete, proxies for the direct
effect of a parent firm’s impact on their employees’ ability to manage a hedge fund. Ideally, we
10 The mean SAT score by job spell is 1335 for the 47% of job spells we have data on (Table 2). 16% of the job spells are associated with individuals who matriculated from an undergraduate institution outside of the U.S., and we do not have educational records for 37% job spells in our sample. We, therefore, use a spline of the quintiles of institutions’ SAT scores along with a missing SAT score dummy variable in our empirical tests. 11 SIC codes beginning with 6 are in the Finance, Insurance and Real Estate division. 94% of the job spells were associated with firms whose primary SIC code began with a 6. Unfortunately, multi-digit SIC codes appear to be rather imprecise in the financial services industry so we did not include more refined SIC code dummies in our main analysis. However, similar results were obtained using two-digit SIC code dummies.
17
would like to run our specifications with a full set of parent firm fixed effects to control for all
the observable and unobservable characteristics of a manager’s previous employers that might
influence a hedge fund spawn’s performance. We are particularly concerned that firms located
in industry hubs might provide additional human and social capital in the form of higher quality
training, better connections, and by providing stronger signals about employees’ unobservable
quality to potential future employers. If the provision of human and social capital is correlated
with the parent firm’s location, and are not well controlled for with other proxy variables, they
could induce a spurious correlation between our hypothesized inherited agglomeration effects
and hedge fund spawn performance due to omitted variable bias. However, since our key
explanatory variable PARENT_CENTER does not vary over time, we cannot specify a model
with a complete set of parent firm fixed effects. Instead, we include a set of dummy variables for
firms that are most likely to create value for their employees that may not be correlated with
public measures of firm quality. Specifically, we include a firm fixed effect for each of the
“bulge bracket” investment banks: Citi, Credit Suisse, Goldman Sachs, JP Morgan, Lehman
Brothers, Merrill Lynch, and Morgan Stanley (Fang, 2005). Bulge bracket banks offer a widely
recognizable implicit endorsement that is potentially valuable to employees, and are the most
likely providers of training programs and providers of broad access to key players in the
financial industry that might be valuable to nascent hedge fund managers.12
We also include hedge fund controls in our specifications, including controls for firm
scope, size, age and location as of 2007. Scope is measured as the log average number of funds
in the hedge fund firm (equal weighted by month), size is measured as the monthly average firm-
level aggregate assets under management (AUM) and age is measured as years since the firm
12 These firms account for just under 50% of the observations in our sample. Our results are also robust to including
dummy variables for each of the top five or ten spawning parent firms (see Table 1 for a list of top spawning firms).
18
was founded until exiting the sample or until the end of the sample period (2005), whichever was
later. The mean scope, size and age, by job spell, in our sample are 4.1 funds, $66M of AUM,
and 7.35 years, respectively (Table 2). We also include a binary variable, “hedge fund near
center,” set equal to one if the hedge fund is within 100 miles of New York or London, and zero
otherwise, as many hedge funds are located close to New York City, or to London. 14% of the
job spells in our sample were associated with hedge funds that located near (but not in) New
York City or London (Table 2).13
4.5 Baseline empirical specification
Although we are concerned with endogeneity, the baseline tests of our first hypothesis are
simple OLS regressions of hedge fund performance measures (ALPHA or IR) on managers’ job
spell location measures PARENT_CENTER and HF_CENTER for hedge fund firm i and job spell
(or parent firm) j as in:
(2) PERFORMANCEi = a + β1 PARENT_CENTERj + β2HF_CENTERj + Xcβc + ei,
where X is a vector of controls that might plausibly influence hedge fund performance, described
above in section 4.2, including parent-firm fixed effects for bulge bracket investment banks, and
c indexes individual, job spell and hedge fund controls. Standard errors are clustered at the
parent-firm (j) level. In our test of Hypothesis 3, we also include in equation (2) two interaction
terms: PARENT_CENTER x previous job in investment management and HF_CENTER x
previous job in investment management.
While including parent firm fixed effects for bulge bracket investment banks reduces the
risk of omitted variable bias from parent firm effects, specification (2) is vulnerable to sorting
13 Because the common risk factors included in (1) absorb the effects of standard trading strategies, and self-described investment style is often imprecise, we exclude controls for investment style, though we verify that including them has no meaningful effect on our results.
19
effects stemming from employment choices made by managers and parent firms. For example,
the best future hedge fund managers may self-select into companies located in New York or
London, precisely because they want to start their own hedge fund in the future. Thus, a positive
coefficient on PARENT_CENTER may confound selection and treatment effects.
4.6 Matched sample
To deal with selection effects that are correlated with all the observable information
about firms and workers, we use a propensity score matching approach. In the absence of
unobserved heterogeneity, propensity score matching controls for selection bias by creating a
matched sample of treatment and control observations that are similar with respect to all
observable characteristics (Rosenbaum and Rubin, 1983). While propensity score matching
cannot control for unobservable differences between workers and firms, it can alleviate selection
effects by reducing the observable differences between treatment (job spells in New York or
London) and control groups (job spells outside of New York or London). Even if propensity
score matching does not completely eliminate the effect of unobservable heterogeneity, for
example the impact of unobservable ex ante worker quality on hedge fund spawn performance,
the matched sample results eliminate job spells that are non-comparable between treatment and
control groups, helpding focus the set of plausible interpretations of any correlation between
PARENT_CENTER and hedge fund performance. Moreover, compared to the standard approach
of adding controls to a linear regression, the propensity score methodology is superior in that it
makes fewer functional form assumptions and eliminates the influence of non-comparable
control and treatment group observations.
To implement propensity score matching, we estimate a probit of the individual and
20
employers’ joint decision to enter into an employment relationship in New York or London (i.e.
in the industry hub) and use fitted values from that model as estimates of the propensity score:
Pr(PARENT_CENTERi = 1|Xij), where Xij includes all observable characteristics of individuals
and their employer firms that might plausibly have an effect on either party’s decision to enter
into the employment relationship. The vector Xij includes all the individual and parent firm
covariates from the OLS specification (2), except the bulge bracket firm fixed effects, the size of
parent firm (measured by the logarithm of the number of employees), the rank of the manager’s
undergraduate institution (by quintile), as well as whether a manager’s previous job spell was in
the industry hub (pre-founding). After trimming off extreme values and observations off the
common support of the propensity score distribution, we match each treatment to a single control
group observation without replacement to obtain our matched sample.
-------------------------------------- Insert Figure 2 about here
--------------------------------------
Figure 2 reveals why matching matters. Figure 2A shows that the distributions of the
propensity scores before matching are quite different between the treatment and control groups.
Figure 2B shows the distributions of the propensity scores after matching. Visually there is a
much tighter fit between the two groups after matching. Table 3 corroborates this result
statistically. Column 1 shows the coefficients (marginal effects) from the probit model
estimating the likelihood a job spell would occur in the industry hub conditional on a manager
subsequently becoming a senior member of a hedge fund, while columns 2 and 3 show the
differences in the means of the covariates for the control and treatment groups before and after
matching. Conditional on an individual becoming a senior manager of a hedge fund, achieving a
post-graduate degree other than an MBA, PhD or JD is positively correlated with job spells
21
occurring in the industry hub, and the incidence of other post-graduate degrees is significantly
higher in the treatment (PARENT_CENTER=1) population than in the control
(PARENT_CENTER=0) population. Also, job spells at employers that are ranked by
Institutional Investor as leading trading houses (RANKED) are far more likely to be in the
industry hub and are much more prevalent in the treatment group than in the control group.
While no other variables are both statistically significant in predicting a job spell being in the
industry hub and have statistically different means in the treatment and control groups before
matching, SIC6, Tobin’s Q, and number of employees are statistically different between the two
populations, and the F-test for the joint difference in means between the two groups before
matching is statistically significant at the 1% level.
-------------------------------------- Insert Table 3 about here
--------------------------------------
After matching the differences between control and treatment group decrease
substantially. Only the difference in the means on RANKED remains significant at the 5% level,
and the F-test for the joint significance of the differences in means between the treatment and
control groups is not significant at the 10% level (p-value of 0.39). In other words, the matching
approach appears creates treatment and control group job spells that are similar along all
observable dimensions ex ante.
5. Results and discussion
5.1 Results
Table 4 shows the results from the baseline OLS regressions on hedge fund spawn
performance. In column 1, using the 8-factor alpha as the dependent variable, the coefficient
estimate of PARENT_CENTER is 15 basis points per month (0.15%/month), and is significant at
22
the 5% level, while the coefficient on HF_CENTER is negative and not statistically significant.
While alpha is a standard measure of hedge fund performance, the information ratio offers a
better sense of a fund’s true economic performance as it controls for idiosyncratic risk exposure.
In the regression with the 8-factor information ratio as the dependent variable, the coefficient on
PARENT_CENTER is 10 basis points per month and significant at the 1% level, while the
coefficient on HF_CENTER is 4 basis points per month and significant at the 5% level (column
2). The correlation between parent firm location and hedge fund performance suggests that
inherited agglomeration effects are positive, economically and statistically significant, and even
larger than conventional agglomeration effects (though the difference between the coefficients is
only on the margin of statistical significance).14
-------------------------------------- Insert Table 4 about here
--------------------------------------
In column 3, we test whether agglomeration effects are larger when managers have
related experience in investment management by interacting PARENT_CENTER with the
investment management dummy. The marginal effect of investment management experience on
performance, conditional on inherited agglomeration effects is 8 basis points, but is only
significant at the 10% level.
Table 5 shows the matched sample performance results. The PARENT_CENTER
coefficient in the 8-factor alpha regression is 26 basis points per month (column 1), and 14 basis
points per month in the information ratio regression (column 2). Both estimates are significant at
the 1% level. The coefficient on HF_CENTER is negative, but not significant in the 8-factor
alpha regression, and 6 basis points per month in the information ratio regression. The results
14 Because the factor models, shown in equation (1) above, pick up most of the variation in performance, the R2s in
our estimating regressions are relatively low.
23
lend further support to our key contention that inherited agglomeration effects are important, at
least in the hedge fund context. Indeed, the results suggest that inherited effects are at least of
the same magnitude as traditional agglomeration effects, even after controlling for systematic
and idiosyncratic risk exposure and parent-firm fixed effects for bulge bracket firms.
-------------------------------------- Insert Table 5 about here
--------------------------------------
In the matched sample, the marginal effect of investment management experience on
performance (the information ratio) is 11 basis points and significant at the 5% level, conditional
on inherited agglomeration effects (column 3). The point estimate on the main effect of inherited
agglomeration effects remains positive, but is statistically indistinguishable from zero.
Figure 3 illustrates the impact of relatedness on performance graphically by splitting
PARENT_CENTER, non-PARENT_CENTER and previous job in investment management and
PARENT_CENTER observations into deciles and comparing their means. In all 10 deciles the
PARENT_CENTER mean is bigger than the non-PARENT_CENTER mean, with the difference
being significant at the 5% level in 8 of the deciles. Moreover, in 9 deciles the investment
management x PARENT_CENTER mean is bigger than the PARENT_CENTER mean, and the
difference is significant at the 5% level in 8 deciles. The results suggest that, at least in the
context of hedge funds, inherited agglomeration effects operate primarily through managers with
closely related work experience, perhaps because these managers have developed the most
relevant human and social capital while working in the industry center.
-------------------------------------- Insert Figure 3 about here
--------------------------------------
Discussion 5.2
In the ideal experiment, we would randomly assign individuals to job spells at parent
24
firms and to senior management positions in hedge funds. In practice, we must base our
statistical tests on self-selected populations. Though our results are robust to a wide range of
controls, we cannot control for all sources of unobservable heterogeneity between the treatment
and control populations. In particular, we are unable to control directly for a manager’s
unobservable ex ante ability, which might be systematically correlated with having a job spell in
New York or London. Thus, our results might be confounded by managers with higher ability
selecting into job spells in an industry hub.15 Given the absence of a true experiment, we must
interpret the correlation between parent firm location and hedge fund performance cautiously. It
appears that inherited agglomeration effects positively influence hedge fund performance, but the
effect could be due, in part, to selection effects based on managers’ unobservable ability.
While acknowledging the limitations of our empirical design we note that it seems
somewhat unlikely that our results are being driven selection effects for two reasons. First, the
inclusion of firm fixed effects for each of the bulge bracket firms absorbs the influence of the
premier investment banks. If selection on unobservables were driving our results, one would
expect that the top investment banks, like Goldman Sachs and Morgan Stanley, would be the
firms driving our results, as these banks would presumably be the destination for the most
15 In fact, any effect of ex ante ability that is not correlated with the observable characteristics of firms and their managers may confound our results. While selection in to parent firm job spells in New York or London based on unobservable skill would seem to be the most pressing threat to identification in our empirical context, manager selection into hedge funds could also bias our results. For example, if managers sort from parent firms in to hedge funds at different rates for systematic reasons that are correlated with location, but uncorrelated with our controls we would also have an omitted variable bias problem. In particular, we might worry if relatively low (high) skill managers from parent firm job spells outside (inside) of New York and London systematically joined hedge funds that the coefficient estimate on PARENT_CENTER would be biased upward. While we cannot definitively rule out this form of omitted bias, as a practical matter it seems that in the context of hedge funds, we would have the opposite problem: ceteris paribus managers in New York and London would have better opportunities to sort into hedge funds, and would, therefore, have lower levels of unobservable ability on average. Thus, one should expect that it would take a higher compensating level of skill to sort into a hedge fund from outside of New York or London. And indeed, we find that New York and London based firms are far more likely to spawn new hedge funds than firms outside of New York and London (results available upon request). Based on the nature of the hedge fund industry and the spawning rate result, we conclude that any bias from selection into hedge funds based on unobservable ability is unlikely to pose a problem for our empirical strategy.
25
capable individuals. Yet, our results are consistent for managers who come from the less exalted
financial firms in New York and London, firms that would not appear to hold any obvious
advantages over top financial firms that are not headquartered in New York or London. Second,
the point estimates on PARENT_CENTER increase once we control for selection on observables.
Although it is certainly possible that the effect of selection on unobservables works in the
opposite direction as the effect of selection on observables, it is comforting to know that the
known bias in our OLS results works against us.
Taken together, the evidence is consistent with the idea that managers acquire greater
human and social capital in industry hubs, and then transmit these agglomeration-based
capabilities from parent firms to their spawns. The results suggest that inherited agglomeration
effects have a meaningful causal impact on new venture performance, and the effect is
particularly important when the new venture’s managers worked in a closely related line of work
previously.
6. Conclusion
This paper develops the logic for inherited agglomeration effects: economic benefits that
accrue to managers while working at a parent firm in an industry hub that can be subsequently
transferred to an entrepreneurial spawn when managers leave incumbent firms to manage a
spawn. We test the performance predictions of the theory in the context of the global hedge fund
industry. The results show that hedge funds whose principal managers were previously
employed by parent firms located in financial services industry hubs—New York or London—
outperform other hedge funds by about 1.5% per year, a result that is robust to controlling for a
hedge fund’s systematic and idiosyncratic risk exposure, whether the manager’s previous
26
experience was with a bulge bracket parent-firm (i.e., Goldman Sachs, Morgan Stanley, etc.),
and for selection on observables. Interestingly, the impact on performance from inherited
agglomeration effects is at least as large as the effect of traditional agglomeration effects. As
predicted by the theory, the core result is driven by hedge fund managers who previously worked
in closely related investment management positions in New York and London.
The results suggest that meaningful variation in the performance of a new venture can be
traced to agglomeration-related human and social capital senior managers’ transfer to a new firm
from their parent firms. By explicating one important link between parent firm characteristics
and new venture performance, this paper offers some insight into the mechanisms behind the
sources of competitive advantage in young firms.
27
References
Aigner, D.J. 1973. Regression with a Binary Independent Variable Subject to Errors of
Observation. Journal of Econometrics 1(1): 49-60.
Agarwal, R., Echambadi, R., Franco, A. M., and Sarkar, MB. 2004. Knowledge transfer through
inheritance: Spin-out generation, development, and survival. Academy of Management
Journal 47(4): 501-522.
Braun, E., and Macdonald, S. 1982. Revolution in miniature: The history and impact of
semiconductor electronics re-explored. Cambridge University Press: Cambridge, England
Burton, M.D., J.B. Sorensen, and Beckman, C.M. 2002. Coming from Good Stock: Career
Histories and New Venture Formation. In Research in the Sociology of Organization,
Lounsbury, M. and Ventresca, M. J. (eds.). Elsevier: Oxford, England; 229-262.
Carhart, M.M. 1997. On persistence in mutual fund performance. Journal of Finance 52(1): 57-
82.
Chevalier, J., and Ellison, G. 1999. Are some mutual fund managers better than others? Cross-
sectional patterns in behavior and performance. The Journal of Finance 54(3): 875–899.
Chatterji, A. K. 2009. Spawned with a silver spoon? Entrepreneurial performance and
innovation in the medical device industry. Strategic Management Journal 30(2): 185-206.
Fang, L.H. 2005. Investment bank reputation and the price and quality of underwriting services.
Journal of Finance 60(6): 2729-2761.
Fama, E.F. and French, K.R. 1993. Common risk factors in the returns on stocks and bonds.
Journal of Financial Economics 33(1): 3–56.
Fung, W., and Hsieh, D. A., 2004, Hedge fund benchmarks: A risk based approach. Financial
Analysts Journal 60(5): 65–80.
Geroski, P.A., Mata, J., and Portgula, P. 2010. Founding conditions and the survival of new
firms. Strategic Management Journal 31 (5): 510-529.
Glaeser, E. L., Kallal, H. D., Scheinkman, J. A., and Shleifer, A. 1992. Growth in cities. The
Journal of Political Economy 100(6): 1126-1152.
28
Gompers, P., Lerner, J., and Scharfstein, D. 2005. Entrepreneurial spawning: Public
corporations and the genesis of new ventures, 1986 to 1999. The Journal of Finance 60(2):
577-614.
Hedge Fund Research, 2010. Global Hedge Fund Industry Report, First Quarter
Helfat, C. E., and Lieberman, M. B. 2002. The birth of capabilities: Market entry and the
importance of pre-history. Industrial and Corporate Change 11(4): 725–760.
Hellmann, T. 2007. When do employees become entrepreneurs? Management Science 53(6):
919-933.
Jaffe, A. B., Trajtenberg, M., and Henderson, R. 1993. Geographic localization of knowledge
spillovers as evidenced by patent citation. Quarterly Journal of Economics 108(3):577-598.
Johnson, V. 2007. What is organizational imprinting? Cultural entrepreneurship in the founding
of the Paris Opera. American Journal of Sociology 113(1): 97-127.
Klepper, S., and Simons, K.L. 2000. Dominance by birthright: Entry of prior radio producers
and competitive ramifications in the U.S. television receiver industry. Strategic Management
Journal 21(10-11): 997-1016.
Klepper, S., and Sleeper, S. 2005. Entry by spin-offs. Management Science 51 (8): 1291–1306.
Krugman, P. 1991. Increasing returns and economic geography. Journal of Political Economy
99(3): 483–499.
Marshall, A. 1920/1890. Principles of Economics. Macmillan Publishers: London, England
Nelson, R.R. 1995. Recent evolutionary theorizing about economic change. Journal of Economic
Literature 33(1): 48-90.
Pástor, L, and Stambaugh, R. F. 2002. Mutual fund performance and seemingly unrelated assets.
Journal of Financial Economics 63(3): 315-349.
Phillips, D. 2002. A Genealogical Approach to Organizational Life Chances: The Parent-
Progeny Transfer among Silicon Valley Law Firms, 1946-1996. Administrative Science
Quarterly 47(3): 474-506.
Rosenbaum, P. R., and Rubin, D. B. 1983, The Central Role of the Propensity Score in
Observational Studies for Causal Effects. Biometrika 70(1): 41–55.
29
Saxenian, A. 1994a. Regional Advantage: Culture and Competition in Silicon Valley and Route
128. Harvard University Press: Cambridge, MA
Saxenian, A. 1994b. Lessons from Silicon Valley. Technology Review, 97(5): 42–51.
Sorenson, O., and P.G. Audia. 2003. The social structure of entrepreneurial activity:
Geographic concentration of footwear production in the United States, 1940-1989. The
American Journal of Sociology 106(2): 424-461.
Stinchcombe, A.L. 1965. Social Structure and Organizations. In Handbook of Organizations,
March, J. (Ed.). Rand McNally & Co: Chicago, IL; pp. 142–193.
Stuart, T., & Sorenson, O. 2003. The geography of opportunity: Spatial heterogeneity in
founding rates and the performance of biotechnology firms. Research Policy 32(2): 229-253.
30
Figure 1 Comparison of predicted inherited agglomeration effects and traditional agglomeration effects
PARENT
INDUSTRY HUB PERIPHERY
INDUSTRY HUB
NEW VENTURE
PERIPHERY
Hypothesis 1 predicts inherited agglomeration effects.
Hypothesis 2 predicts traditional agglomeration effects.
Hypothesis 1: POSITIVE Hypothesis 2: POSITIVE
Hypothesis 2: POSITIVE
Hypothesis 1: POSITIVE No prediction
I II
III IV IV III
31
Figure 2 Distributions of the probabilities of job spells occurring in the center before and after matching
Figure 2A: Kernel density distributions of the probability of a job spell occurring in the center
before matching (n=658)
Figure 2B: Kernel density distributions of the probability of a job spell occurring in the center after
matching (n=462)
De
ns
ity
dis
trib
uti
on
0 .2 .4 .6 .8 1Propensity Score
non-CENTER CENTER
De
ns
ity
dis
trib
uti
on
.4 .5 .6 .7 .8Propensity score
non-CENTER CENTER
32
-.5
0.5
1
(mean) ten_meani (mean) ten_meani
(mean) ten_meani upper/lower
Figure 3
8-factor information ratio by decile
Matched sample results (n=466). Statistical difference in 8- factor IR means: * significant at the 10% level; ** significant at the 5% level; *** significant at the 1% level
CENTER
Standard Deviation Non-CENTER
8-factor Information Ratio Mean
*** *** *** *** *** *** *** ***
*** *** *** *** *** *** *** ***
Deciles 1st 2nd 3rd 4th 5th 6th 7th 8th 9th 10th
Inv. Mgmt. CENTER
Center minus non-Center Center minus non-Center
Inv. Mgmt. Center minus Center
33
Table 1
Top 25 parent firms by number of job spells spawned
Firm HQ Location Ranked SIC
code Employees
(‘000)
# of job
spells
spawned
Citigroup New York Y 61 387 68
JP Morgan New York Y 60 181 54
Merrill Lynch New York Y 62 64 50
Lehman Brothers New York Y 62 29 38
Deutsche Bank Frankfurt Y 60 78 33
Morgan Stanley New York Y 62 48 32
Goldman Sachs New York Y 62 31 31
UBS Zurich Y 62 84 29
Bear Stearns New York Y 62 14 25
Bankers Trust New York N 60 21 21
Bank of America Charlotte Y 60 210 21
Credit Suisse Zurich Y 62 48 18
RBS Edinburgh Y 60 200 17
Oppenheimer Toronto N 62 3 12
ING Amsterdam N 63 120 12
Wells Fargo San Francisco Y 60 160 10
Schroders London N 62 3 10
Barclays London Y 60 156 10
Allianz Munich N 63 181 9
RBC Montreal/Toronto Y 60 65 6
Alliance Bernstein New York Y 67 6 6
Prudential Financial Newark Y 63 41 6
CIB Toronto Y 60 41 6
HSBC London Y 60 313 5
Franklin Templeton San Mateo N 62 9 5
34
Table 2
Summary statistics
n=658 mean SD min max
Hedge fund firm level variables
ALPHA, 8-factor model (%/month) 0.55 0.63 -1.33 3.47
Information Ratio, 8-factor model 0.27 0.28 -0.58 1.18
HF_CENTER 0.46 n/a 0 1
Hedge fund location near fin. center 0.14 n/a 0 1
Hedge fund age (years) 7.35 4.24 2 24
Avg. number of hedge funds 4.14 5.19 1 53
Lifetime average AUM (in $m) 66.1 135.1 0.5 13,500 Parent firm level variables
PARENT_CENTER 0.58 n/a 0 1
SIC6 0.94 n/a 0 1
Tobin’s Q 1.21 0.59 0.88 8.11
RANKED in the top 25 by Institutional Investor ‘00-07
0.74 n/a 0 1
Individual manager level variables
Previous Job in Investment Management 0.42 n/a 0 1
MBA degree dummy 0.36 n/a 0 1
PhD degree dummy 0.04 n/a 0 1
JD degree dummy 0.02 n/a 0 1
Other postgraduate degree dummy 0.11 n/a 0 1
Median SAT score of highest education institution attended
1335.3 124 834 1495
Interaction variables
Previous Job in Investment Management x
PARENT_CENTER 0.24 n/a 0 1
Previous Job in Investment Management x HF_CENTER
0.19 n/a 0 1
Note: there are no significant correlations between the variables
35
Table 3
Propensity score matching
Dependent variable: Job spell in the industry hub (pre-founding)
(1) (2) (3)
Probit coefficients
∆ means pre-match
(t-stats.)
∆ means Matched (t-stats)
MBA dummy 0.02 0.66 -0.54
PhD dummy -0.03 -0.32 0.47
JD dummy -0.08 0.63 0.13
Other postgraduate degree dummy 0.20 *** -2.46 ** -0.54
Graduate degree information missing 0.15 * -0.86 -0.28
Avg. SAT undergrad.—quintile 5 (top) -0.08 0.13 0.31
Avg. SAT undergrad.—quintile 4 0.01 -1.34 -0.87
Avg. SAT undergrad.—quintile 2 -0.05 -0.16 -0.73
Avg. SAT undergrad.—quintile 1 -0.09 1.29 0.17
Avg. SAT score missing -0.07 -0.13 0.70
Rank undergrad. institution quint5 (top) -0.04 1.05 0.41
Rank undergrad.—quintile 4 -0.08 1.35 -0.18
Rank undergrad.—quintile 2 -0.08 -0.30 -0.91
Rank undergrad.—quintile 1 0.04 -0.97 -0.08
Rank missing -0.01 -0.41 0.73
SIC6 0.20 -5.01 *** -1.38
RANKED by Institutional Investor 0.35 ** -9.10 *** -2.69 ***
Tobin’s Q 0.01 3.30 *** -0.79
Tobin’s Q missing 0.20 -0.40 -0.29 Number of Employees (logged) -0.01 -3.08 *** -0.49
Number of Employees missing -0.29 4.10 *** 0.00
Previous job spell in the industry hub 0.76 *** -0.92 0.51
Previous job spell missing 0.85 *** -0.96 -0.52
Previous job in Inv. Mgmt. 0.04 0.56 0.50
Previous job type missing 0.08 1.01 0.87
N 658 658 466
Pseudo-R2 0.15
F-stat. for joint test of differences 5.42 *** 1.06
Marginal effects displayed for probit model in column (1) * significant at the 10% level; ** significant at the 5% level; *** significant at the 1% level ∆ means are calculated as non-PARENT_CENTER population means minus PARENT_CENTER population means.
36
Table 4
Abnormal returns to investors
(1) (2) (3)
DV 8-factor α 8-factor IR 8-factor IR
PARENT_CENTER 0.15
(0.07)
** 0.10
(0.03)
*** 0.06
(0.04)
HF_CENTER -0.04
(0.05)
0.04
(0.02)
** 0.03
(0.03)
PARENT_CENTER x Previous Job 0.08 *
in Investment Management (0.04) SIC6 -0.10
(0.15) -0.05
(0.05) -0.03
(0.05)
RANKED by Inst. Investor 0.04 (0.06)
0.03 (0.03)
0.02 (0.03)
Tobin’s Q 0.03 (0.04)
0.00 (0.02)
-0.01 (0.02)
Hedge fund near center 0.02 (0.11)
0.04 (0.02)
0.00 (0.04)
Hedge fund firm age -0.01 (0.00)
* -0.01 (0.00)
*** -0.01 (0.00)
***
Log avg. number of funds -0.09 (0.04)
** 0.00 (0.02)
-0.00 (0.02)
Log lifetime avg. AUM (size) 0.03 (0.02)
0.01 (0.01)
0.01 (0.01)
Previous Job in Investment Mgmt. -0.08 (0.06)
-0.01 (0.03)
-0.07 (0.05)
HF_CENTER x Previous Job 0.02 in Investment Management (0.04) 7 Bulge bracket banks dummies Included Included Included SAT score quintile dummies Included Included Included Graduate degree dummies Included Included Included Missing data dummies Included Included Included Constant 0.13
(0.39) 0.12
(0.21) 0.14
(0.21)
N 658 658 658 R2 0.08 0.11 0.12
* significant at the 10% level; ** significant at the 5% level; *** significant at the 1% level Standard errors (in parentheses) are robust and clustered at the parent-firm level. The monthly 8-factor abnormal return is measured net of the seven factors proposed by Fung and Hsieh (2004), including exposure to: the S&P 500 index; a firm size factor; returns to portfolios of straddle options on currencies, commodities, and bonds; the yield spread of the duration adjusted U.S. 10-year Treasury bond over the 3-month T-bill; the change in the credit spread of the duration adjusted Moody’s BAA bond over the 10-year Treasury bond), as well as Pástor and Stambaugh’s liquidity factor (2002). Abnormal returns are averaged over time to compute the standard measure of persistent excess returns (fund “alpha”) at the fund level. Alphas are then averaged on an AUM weighted basis by fund to compute the firm’s equal weighted average monthly abnormal return (e.g., firm “alpha”). The Information ratio divides the obtained monthly ALPHA by the standard deviation of ALPHA.
37
Table 5 Matched sample results
(1) (2) (3)
DV 8-factor α 8-factor IR 8-factor IR
PARENT_CENTER 0.26
(0.09)
*** 0.14
(0.05)
*** 0.09
(0.06)
HF_CENTER -0.05
(0.08)
0.06
(0.03)
** 0.06
(0.04)
*
PARENT_CENTER x Previous Job 0.11 **
in Investment Management (0.06)
SIC6 -0.02
(0.15) -0.13
(0.08) -0.09
(0.08)
RANKED by Inst. Investor 0.12 (0.12)
0.06 (0.06)
0.06 (0.06)
Tobin’s Q -0.03 (0.05)
0.02 (0.03)
0.01 (0.04)
Hedge fund near center 0.03 (0.13)
0.00 (0.05)
0.00 (0.05)
Hedge fund firm age -0.02 (0.01)
* -0.02 (0.00)
*** -0.02 (0.00)
***
Log avg. number of funds -0.05 (0.05)
0.02 (0.02)
0.02 (0.02)
Log lifetime avg. AUM (size) 0.00 (0.03)
0.00 (0.01)
0.00 (0.01)
Previous Job in Investment Mgmt. -0.04 (0.06)
0.03 (0.03)
-0.04 (0.06)
HF_CENTER x Previous Job -0.01
in Investment Management (0.05)
7 Bulge bracket bank dummies Included Included Included SAT score quintile dummies Included Included Included Graduate degree dummies Included Included Included Missing data dummies Included Included Included Constant 0.53
(0.41) 0.35
(0.24) 0.38
(0.24)
N 466 466 466 R2 0.09 0.14 0.15
* significant at the 10% level; ** significant at the 5% level; *** significant at the 1% level Standard errors (in parentheses) are robust and clustered at the parent-firm level. The monthly 8-factor abnormal return is measured net of the seven factors proposed by Fung and Hsieh (2004), including exposure to: the S&P 500 index; a firm size factor; returns to portfolios of straddle options on currencies, commodities, and bonds; the yield spread of the duration adjusted U.S. 10-year Treasury bond over the 3-month T-bill; the change in the credit spread of the duration adjusted Moody’s BAA bond over the 10-year Treasury bond), as well as Pástor and Stambaugh’s liquidity factor (2002). Abnormal returns are averaged over time to compute the standard measure of persistent excess returns (fund “alpha”) at the fund level. Alphas are then averaged on an AUM weighted basis by fund to compute the firm’s equal weighted average monthly abnormal return (e.g., firm “alpha”). The Information ratio divides the obtained monthly ALPHA by the standard deviation of ALPHA.