Professional Documents
Culture Documents
Entrepreneurial Risk and Market Entry
Entrepreneurial Risk and Market Entry
Management Science
Publication details, including instructions for authors and subscription information:
http://pubsonline.informs.org
This article may be used only for the purposes of research, teaching, and/or private study. Commercial use
or systematic downloading (by robots or other automatic processes) is prohibited without explicit Publisher
approval, unless otherwise noted. For more information, contact permissions@informs.org.
The Publisher does not warrant or guarantee the article’s accuracy, completeness, merchantability, fitness
for a particular purpose, or non-infringement. Descriptions of, or references to, products or publications, or
inclusion of an advertisement in this article, neither constitutes nor implies a guarantee, endorsement, or
support of claims made of that product, publication, or service.
INFORMS is the largest professional society in the world for professionals in the fields of operations research, management
science, and analytics.
For more information on INFORMS, its publications, membership, or meetings visit http://www.informs.org
MANAGEMENT SCIENCE informs ®
Brian Wu
The Wharton School, University of Pennsylvania, 2000 SH-DH, 3620 Locust Walk,
Philadelphia, Pennsylvania 19104, wux@wharton.upenn.edu
T his paper attempts to reconcile the risk-bearing characterization of entrepreneurs with the stylized fact
that entrepreneurs exhibit conventional risk-aversion profiles. We propose that the disparity arises from
confounding two distinct dimensions of uncertainty: demand uncertainty and ability uncertainty. We further
propose that entrepreneurs will be risk averse with respect to demand uncertainty, yet “apparent risk seeking”
(or overconfident) with respect to ability uncertainty. To examine this view, we construct a reduced-form model
of the entrepreneur’s entry decision, which we aggregate to the market level, then test empirically. We find
that entrepreneurs in aggregate behave as we predict. Accordingly, risk-averse entrepreneurs are willing to bear
market risk when the degree of ability uncertainty is comparable to the degree of demand uncertainty. Potential
market failures exist in instances where there is a high demand uncertainty but low performance dispersion
(insufficient entry), or low demand uncertainty but high performance dispersion (excess entry).
Key words: entry; risk; entrepreneur; uncertainty; overconfidence
History: Accepted by Pankaj Ghemawat, business strategy; received September 2, 2004. This paper was with
the authors 3 months for 2 revisions.
biases. We do not, however, assess whether the biases (Headd 2003) indicate that 51% of firms exit within
(if any) of entrepreneurs differ from those of wage their first four years. Thus, failure risk looms large
earners. for entrants and is relatively nonexistent for estab-
lished firms. An additional but related risk pertains
to invested capital. While shareholders can minimize
2. Entrepreneurship and Demand risk by diversifying their holdings precisely as Knight
Uncertainty (1921) suggests, entrepreneurs typically must invest
Downloaded from informs.org by [141.225.1.24] on 19 November 2014, at 10:57 . For personal use only, all rights reserved.
A fairly well-established theme in the entrepreneurial the bulk of their wealth in a single asset, the venture.
literature is that a key economic role of entrepreneurs Given that entrepreneurs perform a risk-bearing
is risk bearing. This view dates back to Cantillon role, most theoretical literature has assumed that
(1755) who characterized the economy as consisting entrepreneurs have greater risk tolerance than wage
of two classes of inhabitants (aside from the prince earners (McClelland 1961, Lucas 1978, Kanbur 1979,
and landowners): “hired people” on fixed wages and Kihlstrom and Laffont 1979). Empirical literature
“undertakers” who purchase inputs (including labor) has emerged to test this inference. The surprising
at fixed prices without assurance of profits. The key result has been that entrepreneurs do not appear
distinguishing feature of the second class is that it to differ from wage earners on this trait. In fact,
undertakes the risk of demand and price uncertainty
where there are differences they tend to indicate that
(which at the time of his writing must have been
entrepreneurs exhibit greater risk aversion than wage
quite high because one of the factors Cantillon consid-
earners (Brockhaus 1980, Masters and Meier 1988,
ered was the number of deaths of local inhabitants).
Miner and Raju 2004).
Included in the undertaker (entrepreneur) class were
One of the reasons studies may fail to find risk
farmers, merchants, shopkeepers, and master crafts-
tolerance is that the instruments used to test risk
men (even robbers).
vary across studies and each instrument operational-
This view was expounded by Knight (1921) and,
izes risk quite differently. Brockhaus (1980) and
in fact, the view of entrepreneur as risk taker is
Masters and Meier (1988) use a Kogan-Wallach Choice
probably most associated with Knight. Knight’s con-
Dilemma Questionnaire (CDQ) which asks respon-
tribution was to draw a distinction between risk,
dents what success threshold a given action would
which involves recurring events whose relative fre-
quency can be known from past experience, and require before they would recommend the action to
uncertainty, arising from unique events which can someone else. Both studies found no significant dif-
only be subjectively estimated. Risk is considered to ference between entrepreneurs and managers on the
be a relatively insignificant problem in that it can recommended thresholds. Of course, one possible rea-
be accommodated through pooling and insurance. son for this result is that people may be willing to
In contrast, uncertainty requires an economic func- undertake risk themselves while not recommending
tionary, the entrepreneur, whose job it is to decide it for others. However, results from 12 studies using
what to do and how to do it in the face of uncertain- a different instrument, the Miner Sentence Comple-
ties. Knight proposed that there is diversity among tion Scale (MSCS), all found entrepreneurs to be risk
individuals with regard to confidence in one’s judg- avoiding relative to managers (Miner and Raju 2004).
ment and the disposition to act on those judgments. In another study, Sarasvathy et al. (1998) found that
Those who are “confident and venturesome ‘assume entrepreneurs are risk averse relative to bankers in
the risk’ or ‘insure’ the doubtful and timid by guar- that they trade higher expected value projects for ones
anteeing to the latter a specified income in return for with narrower variance (particularly avoiding nega-
an assignment of the actual results” (p. 269). tive outcomes).
Anticipating later work by Camerer and Lovallo Results from economic tests are similarly equivocal.
(1999), Knight asserts that “if men are poor judges Cramer et al. (2002) compare individuals’ valuations
of their own powers as well as ignorant of those for a lottery ticket and find that subjects who had ever
of others (entrepreneurs), the size of the profit share been self-employed exhibit greater risk tolerance than
depends on whether they tend on the whole to over- wage earners even after controlling for wealth effects
estimate or underestimate the prospects of business (the self-employed tend to have greater wealth and
operations” (p. 285). therefore bear less relative risk than wage earners). In
Indeed, entrepreneurs do bear greater risk than contrast, O’Brien et al. (2003) examine the impact of
wage earners. First, they bear income risk in that the demand uncertainty on entrepreneurial entry across
stream of income from new ventures is uncertain. In 57 industries and find that demand uncertainty has
the worst case, the firm fails and the income stream a significant negative impact on entry. Mazzeo (2004)
ceases. This risk of failure is considerable. Data from compares an entrepreneur’s choice between operating
the U.S. Census Business Information Tracking Series an independent establishment (sole ownership) and
Wu and Knott: Entrepreneurial Risk and Market Entry
Management Science 52(9), pp. 1315–1330, © 2006 INFORMS 1317
becoming a franchisee (sharing risk with the fran- entrepreneurs have greater risk tolerance; it is that
chisor) in the same market. He finds that franchising they do not view the business situations as being
increases with the local demand uncertainty, thus sug- risky (March and Shapira 1987, Palich and Bagby
gesting risk aversion. 1995, Sarasvathy et al. 1998). Instead, they believe that
In summary, while there is some evidence to the through skill and information they can condition the
contrary, the weight of the empirical evidence tends to odds they face.
indicate that entrepreneurs have comparable risk pro- This bias is related to the “Lake Wobegon” or “bet-
Downloaded from informs.org by [141.225.1.24] on 19 November 2014, at 10:57 . For personal use only, all rights reserved.
files to those of wage earners. This leaves the question ter-than-average” effect (Alicke et al. 1995), where
of what accounts for their willingness to bear the high Lake Wobegon refers to Garrison Keillor’s fictional
risk of failure. We believe the answer to that question hometown in Minnesota where “all the children are
lies in a second dimension of uncertainty. above average.” A frequently cited example of the
effect is that 60% of high-school seniors believe they
are in the top 10% in ability to get along with oth-
3. Entrepreneurship and Performance ers, while 25% believe they are in the top 1% (Myers
Dispersion 1982). If we extrapolate these numbers, it appears that
While the most prevalent view of entrepreneurial risk the entire population of seniors believes it is in the
pertains to classic risk aversion in that an individ- top 20%.
ual is more interested in a certain payoff than an A more pertinent example is engineers’ assessment
equivalent expected value from an uncertain pay- of their own performance relative to peers (Zenger
off, a second view emerges from the work of Knight 1994). That study found that in established compa-
(1921), Kanbur (1979), Jovanovic (1982), Lippman and nies, 32.4% of engineers believed they were in the top
Rumelt (1982), March and Shapira (1987), Camerer 5% of peer performance; while 61.8% believed they
and Lovallo (1999), and Norton and Moore (2002). were in the top 10% and fully 89.7% believed they
In this view, a critical source of economic uncer- were in the top 25%. These biases were even more
tainty pertains to entrepreneurial ability. For exam- pronounced for engineers in entrepreneurial firms
ple, entrepreneurs know the performance distribution where self-assessments within the top 5%, 10%, and
within a market but are uncertain about where they 25% of peers were, respectively, 42.0%, 73.3%, and
lie within that distribution. 92.5%. Thus, while all engineers are prone to over-
Performance dispersion has two potential effects confidence, those drawn to start-ups are particularly
on entry: an options effect and an overconfidence overconfident.
effect. We discuss each of these in turn. High degrees Although the better-than-average effect has been
of performance dispersion will lead to apparent demonstrated for both traits and behaviors, the bias
risk seeking even if entrepreneurs are risk neutral is more likely for traits that are perceived to be con-
(Lippman and Rumelt 1982). This occurs because trollable, e.g., fairness rather than intelligence (Alicke
entrepreneurial ventures have an options structure. 1985, Allison et al. 1989); or in cases of greater
The upside returns are uncertain, but the downside ambiguity. For example, the bias is more likely when
risk is limited by the entry fee. Accordingly, greater individuals compare themselves to an anonymous
profit dispersion leads to higher expected values net distribution rather than to a known set of individuals
of the entry fee. If entrepreneurs base decisions on net (Alicke 1985, Alicke et al. 1995, Allison et al. 1989).
present values, risk-neutral entrepreneurs will appear This qualitative characterization suggests a bias tied
to be risk seeking in that they will rationally prefer to dispersion of the trait in the referent population.
more disperse performance distributions. Thus, if and capture, respectively, the mean and
A more provocative view finds that under cer- standard deviation of ability, unbiased entrepreneurs
tain circumstances, entrepreneurs exhibit “apparent without private information should expect an ability
risk seeking” with respect to ability uncertainty. This draw equal to . If, however, entrepreneurs exhibit
arises from overestimating their capability. In this overconfidence (the better-than-average effect), their
view, entrepreneurs know the performance distribu- expected ability will have the form + A, where
tion and expected value of entry within a given mar- the bias A will depend on perceived controllability
ket but they believe their own ability is drawn from a captured by A, and the degree of ambiguity captured
narrower and biased distribution (March and Shapira by .
1987, Zenger 1994, Busenitz 1999, Camerer and Considering the two types of uncertainty (demand
Lovallo 1999, Norton and Moore 2002). Indeed, this uncertainty and ability uncertainty) jointly suggests
tendency was anticipated by Knight’s (1921) insight entrepreneurs could appear to be risk seeking when
that the distinguishing feature of entrepreneurs is in fact they are not. In particular, they may be risk
their level of confidence in being able to han- averse with respect to demand uncertainty but over-
dle unforeseeable events. Accordingly, it is not that confident with respect to ability uncertainty. If so,
Wu and Knott: Entrepreneurial Risk and Market Entry
1318 Management Science 52(9), pp. 1315–1330, © 2006 INFORMS
there may be settings where they will bear economic utility comparison is meaningless, however, without
risk associated with uncertain demand so long as the context of an actual market to enter. In most
there is sufficient ability uncertainty. cases, this is precisely the market (industry plus geo-
It is worth noting that we do not postulate that graphic location) of the individual’s prior employer.
entrepreneurs differ from wage earners on either This makes sense. Individuals accrue experience, mar-
dimension. Indeed, the empirical record discussed ket knowledge, and affiliations during the course of
previously indicates that entrepreneurs are indistin- their employment. The returns of this human and
Downloaded from informs.org by [141.225.1.24] on 19 November 2014, at 10:57 . For personal use only, all rights reserved.
guishable from wage earners with regard to risk aver- social capital are probably highest in the market
sion. In addition, we have just shown that high-school in which they were accrued. Over time individuals
seniors (the pool from which most entrepreneurs and observe market trends (the performance of all firms
wage earners are drawn) are overconfident. All we in the market) as well as their own performance
require to reconcile entrepreneurs’ risk aversion with relative to peers (both within the firm and within
their willingness to bear risk is that these dimensions the industry). These observations offer noisy signals
affect entry behavior and that there is heterogeneity about underlying demand and entrepreneurial abil-
along some dimension such that not all individuals ity. At any given time, an individual remains in wage
jump from wage employment to entrepreneurship at employment if his or her wage is higher than the
the same time. expected entrepreneurial profits in that market, given
the current estimates about the market and firms’
abilities.
4. A Model of Entrepreneurial Entry
4.1. Qualitative Description of the Entrepreneurial 4.2. Reduced-Form Model of the Entry Decision
Entry Decision We can formalize the decision process just described
There is substantial literature within management using the basic entry decision model from the
and economics regarding the decision to become an self-employment literature. The model examines an
entrepreneur. Taken together, the literature suggests individual’s choice between the expected value of
that individuals can be characterized as having under- entrepreneurial profits , and that of wage income W ,
lying propensities to become entrepreneurs. These given uncertainty about own ability (captured as
propensities are driven by human capital such as marginal cost).
wealth and education (Dunn and Holtz-Eakin 2000, 4.2.1. Risk Neutrality. We begin with a model of
Hamilton 2000), demographic characteristics such as risk-neutral and unbiased behavior. The functional
age, marital status, and whether your parents were form for profits follows Lippman and Rumelt
themselves entrepreneurs (Dunn and Holtz-Eakin (1982):
2000), and the quality of alternative options (Lofstrom
1 c0
2002). In addition to individual propensities, there are E = −K + qp − c dF c where (1)
exogenous shocks that often trigger entrepreneurship r 0
such as divorce, graduation, termination from a prior q = firm output,
job (Wooten et al. 1999), or working for a firm that p = market price,
becomes acquired (Stuart and Sorenson 2003). Finally, K = investment required to produce quantity q,
economic factors affect the level of entrepreneurship, r = discount rate,
e.g., tax rate, the unemployment rate, and the level of c = firm’s marginal cost, drawn from distribution F ,
technology (Blau 1987). and
The net effect of these factors is that at any given c0 = largest solution to p − c = 0.
time typically 12%–13% of the nonagricultural work Prospective entrepreneurs compare the expected
force is self-employed. However, there is considerable profit from their proposed venture to the alterna-
movement in and out of self-employment. Typically tive income stream from wage employment. Entry
2% to 3% of the workforce becomes entrepreneur each occurs when the expected profit stream exceeds the
year and, as a result, over their entire career approxi- entrepreneur’s opportunity cost (which comprises the
mately 30% of the workforce tests the entrepreneurial sunk cost of entry K, plus the foregone wage stream
waters at some point. These transitions occur both from the best employment alternative W ), as follows:
because individuals’ human capital and demographic c0
status are changing over time, and because their alter- qp − c dF c > rW + K (2)
0
natives are changing over time.
Accordingly, a reasonable way to envision the self- We assume that F c is normally distributed and,
employment decision is that individuals continually thus, can be fully characterized by its mean and dis-
assess whether they have higher utility in conven- persion . Accordingly, we can write a simple expres-
tional wage employment or self-employment. This sion for the expected value of annual profits Et ,
Wu and Knott: Entrepreneurial Risk and Market Entry
Management Science 52(9), pp. 1315–1330, © 2006 INFORMS 1319
given the mean cost s of surviving firms (those for do individual biases interact with market conditions
whom c ≤ c0 ) and the failure rate , representing the to affect the level of entry? Fortunately, our market-
entrepreneur’s ex ante probability of drawing a value level test of aggregate entry allows us to jointly test
for c > c0 . This expression characterizes the truncated both individuals’ biases and their economic impact.
distribution of long-run profits, where firms are To examine entry in market j, we aggregate indi-
forced to exit (c > c0 ) and obtain zero long-run profits, vidual propensity in Equation (4) over the pool of
and 1 − firms survive c ≤ c0 and obtain qp − c prospective entrants, nj . The pool includes those with
Downloaded from informs.org by [141.225.1.24] on 19 November 2014, at 10:57 . For personal use only, all rights reserved.
The second dimension, overconfidence with respect Figure 1 Trends in New Bank Charters by Bank Type
to ability uncertainty, requires us to define a func- Bank entry by type
tional form for overconfidence. Following the discus- 300
sion in §3, we assume that overconfidence takes the De novo
form A, where is ability (cost) dispersion, and A 250 Add to hold co
84
85
86
87
88
89
90
91
92
93
94
95
96
97
ability uncertainty), we replace Propositions 5A and
19
19
19
19
19
19
19
19
19
19
19
19
19
19
6A as follows:
over the past 10 years. Accordingly, a study of entry
Proposition 5B. Entry is increasing in cost dispersion in banking is potentially important in its own right.
(overconfidence).
Proposition 6B. Entry is decreasing in demand 5.2. De Novo Entry in Banking
uncertainty rm (risk aversion). Banking does not typically come to mind when nam-
ing entrepreneurial industries. However, there is sub-
Remember that we already capture the contribu- stantial entry in the banking industry over the period
tions of cost dispersion associated with the options we examine (1984–1997), as shown in Figure 1. The
effect through the mean cost of surviving firms s and figure also demonstrates that the dominant mode of
the failure rate . entry is de novo start-ups by entrepreneurs, rather
than expansion of existing bank holding companies.
5. Empirical Approach We are interested exclusively in the de novo start-ups.
5.1. Industry One of the primary drivers behind de novo entry
To test the propositions, we need a setting that (1) is the Community Reinvestment Act (U.S. Congress
allows us to characterize the cost distribution over 1977), designed to encourage banks to meet the credit
the full set of firms, (2) has substantial entry and needs of local communities. The Act has spawned
provides reliable counts of that entry for firms of all a growing industry of consultants, service providers,
sizes, and (3) comprises a large number of compara- and websites dedicated to small and start-up bank-
ble markets that share common technology and com- ing (e.g., http:// www.denovobanks.com). This “de
mon demand functions. We could find only one such novo bank support industry” facilitates new banks
industry—commercial banking post deregulation. The both with the start-up process and with ultimate
banking industry is ideal because it is fragmented operations. The consultants allow banks to compile
with localized competition and is marked by signif- necessary documentation (business plans and bank
icant de novo entry; also, the Federal Deposit Insur- charter applications) and resources (human capital,
ance Corporation (FDIC) collects complete cost data physical capital, and financial capital) for start-up
on all firms. Fragmentation allows us to compare dis- more quickly. The service providers allow community
crete markets with a common demand function and banks to compete more effectively with large banks.
common technology. Thus, we can compare variance These providers offer online banking, check clearing,
in market conditions (demand volatility and cost dis- account maintenance, and statement processing on a
persion) while controlling for other factors that differ fee basis. Accordingly, community banks can emulate
across industries. We can also control for differences large banks without costly investments in technology.
in the level of market demand through time-varying This allows them to reach profitability more quickly,
differences in market conditions. both through higher demand (from improved service
In addition to the quasi-experimental advantages of offerings) and lower fixed cost (outsourced technol-
banking for our purposes, banking is one of the most ogy).
important industries in the economy. Financial ser- Perhaps one reason entrepreneurship seems incon-
vices’ output is roughly $2.1 trillion (7% of U.S. total). gruous with banking is because of industry regu-
Furthermore, even though banking is one of the old- lation. One important aspect of that regulation is
est industries, it has been growing at 6.5% annual rate approval of new charters by either the respective state
banking department or its federal counterpart, the
2
In the empirical test, A will be captured by the coefficient on . Office of the Comptroller of Currency. In either case,
Wu and Knott: Entrepreneurial Risk and Market Entry
Management Science 52(9), pp. 1315–1330, © 2006 INFORMS 1321
banks also require approval from the FDIC for deposit for the overconfidence effect. An ideal measure of
insurance. The two major components of the appli- overconfidence would compare an entrepreneur’s
cation to the chartering agencies as well as the FDIC expected cost to the firm’s realized cost. We do not,
are the business plan and the summary of the man- however, know the entrepreneur’s expected cost. All
agement team. The business plan provides details for we can infer is that all entrepreneurs anticipate hav-
the bank’s first three years of operations and must ing cost below the survivor threshold—they expect
demonstrate that operations are consistent with mar- the new firm to survive. Thus, if we see failure, we
Downloaded from informs.org by [141.225.1.24] on 19 November 2014, at 10:57 . For personal use only, all rights reserved.
ket demand, customer base, competition, and general can assume some form of error in the entry decision.
economic conditions, and that the bank will achieve Figure 2a demonstrates that there is substantial failure
acceptable leverage ratios by the end of three years. in banking.
The management team summary must demonstrate To further demonstrate that the error is overcon-
that the team has substantial expertise in bank admin-
fidence, we need to show that failure is associated
istration, commercial and consumer lending, bank
with high cost (inefficiency). While this has been doc-
operations, and investment/funds management. In
umented in the literature on banking (e.g., Berger and
particular, at least two members should have previ-
Humphrey 1992), we can also demonstrate it for the
ous financial institution experience and the majority
of directors must have close ties to the community set of firms examined here. To do so, we take the cost
(Federal Reserve Bank of Atlanta 2005). efficiencies we derive in Stage 1 analysis (described
An interesting question is why there is any de novo in §§5.4 and 6). We then create distributed lag mod-
entry in an industry that is otherwise experienc- els for exiting firms to track their efficiency preced-
ing dramatic consolidation. Interviews with industry ing failure. Figure 2b presents graphical results of
experts reveal that consolidation actually provides the
impetus for entry. Merger activity has three effects.
Figure 2a Substantial Failure in Banking
First, it creates the “market void” that is required for
approval of a new charter, which is typically small 1,200
Bank exits: 1985 – 1992
100%
business lending (Goldberg and White 1998). Sec-
Frequency
ond, mergers create “liquidity events” (Stuart and Cumulative %
90%
1,000
Sorenson 2003), wherein displaced bank executives 80%
are both in search of new employment and flush 70%
800
with proceeds from the merger.3 This displacement
Cumulative
60%
Frequency
this analysis.4 The figure indicates that exits come We use the translog cost function to accommodate
from high-cost firms. Failing firms not only have the complex array of bank inputs and outputs. In
higher cost than the industry overall (26% higher addition, the translog form accommodates trade-offs
than an average bank of comparable scale),5 but their in both market strategies (product mixes and prices)
cost deteriorates as they approach their exit year. and operational strategies (input mixes). The basic
While these effects hold for both failures and merg- translog cost function mimics the linear programming
ers, they are more pronounced for failures. Taken problem for firm i in year t, that is minimizing total
Downloaded from informs.org by [141.225.1.24] on 19 November 2014, at 10:57 . For personal use only, all rights reserved.
together, the failure rates and the cost data suggest cost cit by choice of output levels yit , taking input
that entrepreneurs are overconfident about their likely prices wit as given:
performance.
j 1 j j
5.4. Empirical Model cit = !0 + !1j yit + !2k witk + ! y y
j k
2 j j 3ij it it
Our empirical approach compares aggregate entry
across markets to characterize entrepreneurs’ behav- 1 j
+ ! wk wk + !5jk yit witk
ior along each dimension of uncertainty. This allows 2 k k 4kk it it j k
us to examine if “apparent risk seeking” (overconfi-
dence) with respect to ability uncertainty might com- + uit + eit where (6)
pensate for risk aversion with respect to demand
uncertainty. If so, risk-averse entrepreneurs may be cit = log observed firm cost,
j
willing to bear economic risk in certain markets. Note yit = vector of log output levels; j indexes output
that our approach examines aggregate behavior in elements,
each market; thus, it ignores characteristics of indi- witk = vector of log input prices; k indexes input
vidual entrepreneurs. Accordingly, we are unable to elements,
answer the question of whether entrepreneurs’ biases uit = cost inefficiency with truncated normal distri-
differ from wage earners. We can only answer the bution, and
question of whether entrepreneurs on average appear eit = error term with normal distribution.
to be risk averse and/or overconfident. We pool data for all firms over 14 years using the
Analysis proceeds in two stages. In the first stage, stochastic frontier model to capture firm-year mea-
we model an industry cost frontier to collect mea- sures of cost inefficiency relative to a fixed global cost
sures of cost efficiency for each firm in each year. This frontier. We collect the estimates of the expected value
allows us to characterize s and for each market in of firm-year cost inefficiency in Stage 1, Euit eit ,
each year. In the second stage, we model Equation (5) which for convenience we continue to label as uit ,
to test Propositions 5B and 6B that entrepreneurs and use the estimates to form the cost distribution for
are risk averse with respect to demand uncertainty each market in each year. The frontier analysis gen-
and overconfident with respect to ability (cost) uncer- erates values that have a truncated normal distribu-
tainty. tion, whereas the model in Equation (3) assumes that
5.4.1. Stage 1: Characterizing Firm Cost Effi- cost is normally distributed. Accordingly, we trans-
ciency. We follow convention in studies of bank effi- form values of uit by taking their natural log, lnuit .
ciency by modeling a stochastic cost frontier using a 5.4.2. Stage 2: Variables. To test the propositions
translog cost function (Cebenoyan et al. 1992, Herma- derived from Equation (5), we need to make assump-
lin and Wallace 1994, Berger et al. 1993, Mester 1993). tions about how entrepreneurs assess demand uncer-
Stochastic frontier analysis, developed by Aigner et al. tainty rjt , the cost distributions sjt and jt , and the
(1977), is based on the econometric specification of a failure rate jt , as well as opportunity costs for the
cost frontier. The stochastic frontier model assumes entry fee Kjt and foregone wage Wjt in their respective
that the log of firm i’s cost in year t, cit , differs from markets j.
the cost frontier c min by an amount that consists of two Demand uncertainty. Our measure of demand uncer-
distinct components: a standard normally distributed tainty follows the conventional approach (Stock and
error term eit , and a cost inefficiency term modeled as
Watson 1998, Morgan et al. 2004), where risk is approx-
a nonnegative random variable uit , with a truncated
imated by variability in demand. Demand uncer-
normal distribution.6
tainty is measured as the root mean squared error
4 (RMSE) from the regression lnmarket demandt /
Coefficient estimates are available from the authors.
5
market demandt−1 − 1 = $0 + $1 ∗(trend) over a mov-
The cost inefficiency (in ln form) increases from the mean of 0.18
to 0.35. This corresponds to a 26% cost increase. ing 10-year window. We measure demand in two
6
Other distributional assumptions are also possible, the most com-
alternative ways. Our primary measure, state per-
mon of which are the half-normal and exponential distributions. sonal income, uses economic conditions as a proxy
All results are robust to these alternative distributions. for demand. While gross state product GSP sounds
Wu and Knott: Entrepreneurial Risk and Market Entry
Management Science 52(9), pp. 1315–1330, © 2006 INFORMS 1323
like a better proxy for market conditions, the two are Figure 3 Bank Assets (Physical Plus Financial Less Deposits) vs. Age
defined to be equal. By familiar identity, output must 40
be equal to the sum of payments to inputs (wages,
35
rent, and profits). We use state income rather than
($million)
for the years both were available was 99.7%, indi-
Downloaded from informs.org by [141.225.1.24] on 19 November 2014, at 10:57 . For personal use only, all rights reserved.
20
cating that the identity holds for our data. In addi-
tion to the demand volatility, RMSE(state income) or 15
survivor pool sjt−1 , and opportunity cost Wjt−1 and banking efficiency in log thousands of constant 1996
Kjt−1 , while controlling for time-varying market fac- dollars. The dependent variable is total cost c—total
tors (growth and concentration) as well as market interest and noninterest expenses. The six indepen-
fixed effects %j and industrywide year effects &t : dent variables are divided between input prices and
output quantities. Input prices are (a) labor price w1
entryjt = !0 + !1 ∗ incumjt−1 + !2 ∗ jt−1 + !3 ∗ sjt−1 (salary divided by the number of full-time equivalent
+!4 ∗jt−1 +!5 ∗jt−1 ∗sjt−1 +!6 ∗growth employees), (b) physical capital price w2 (occupancy
Downloaded from informs.org by [141.225.1.24] on 19 November 2014, at 10:57 . For personal use only, all rights reserved.
Table 1 Stage 1 Data Summary The most important result of the Stage 1 frontier
estimation is the value of the inefficiency terms uit .
Variable Description Obs. Mean Std. dev. Min Max
The distribution of firm cost inefficiency over all mar-
c Cost 174,869 798 1.27 −082 1676 ket years is given in Figure 4a, and the mean value
w1 Price_labor 174,673 297 0.28 −485 939
over time is depicted in Figure 4b. The mean uit over
w2 Price_physical 173,999 −164 0.73 −993 563
capital the entire period is 0.171, which indicates that the
w3 Price_capital 173,789 −354 0.41 −1258 332 mean firm has cost 18.6% above that of a firm on
Downloaded from informs.org by [141.225.1.24] on 19 November 2014, at 10:57 . For personal use only, all rights reserved.
y1 Mortgage 172,503 947 1.58 −137 1786 the cost frontier. The data also indicate that while the
y2 Other loans 173,373 963 1.40 −070 1855 mean cost inefficiency changes over time in response
y3 Securities 173,828 958 1.47 −127 1751
to changing technologies and demand conditions, the
Variable c w1 w2 w3 y1 y2 y3 general trend (particularly over the mid-1990s) is
toward increasing efficiency (decreasing cost). This is
c 1
w1 01929∗ 1 consistent with prior studies of the industry.
w2 −00418∗ 02792∗ 1 While a discussion of the estimated coefficients of
w3 01464∗ 00921∗ 00452∗ 1 the frontier model is outside the scope of this paper,
y1 08640∗ 01151∗ −01322∗ −00463∗ 1 the coefficient estimates are consistent with expec-
y2 09146∗ 01493∗ −00594∗ 01201∗ 07711∗ 1
y3 07603∗ 00821∗ −00922∗ 00259∗ 06883∗ 06934∗ 1 tations as (a) total costs appear to rise with output
and increases in the price of capital, and (b) firms
Note. Units: ln(thousand—1996 dollars). *Significant at 5%.
substitute labor and physical capital in response to
changing prices for these inputs. These results rein-
such de novo entry comprises 75.8% of new charters force confidence in the frontier technique.
over the period we examine (see Figure 1). Table 1
provides variable descriptions and summary statistics 6.2. Test of Propositions
of the data used in Stage 1. The firm cost efficiencies from Stage 1 were trans-
formed (by taking natural logs) to convert their values
from the truncated normal distribution in frontier
6. Results analysis to the normal distribution assumed by Equa-
6.1. Characterizing Firm Cost Efficiency tion (3). The transformed cost efficiencies were then
We estimate the Stage 1 stochastic frontier model pooled by market year and used to create estimates
assuming a truncated normal distribution for the inef- of the cost distribution for each market j in each
ficiency term and a normally distributed error term. year t. We characterized each ex ante cost distribu-
Results from the Stage 1 analysis using Equation (6) tion by its dispersion jt . We then excluded firms less
are given in Table 2. than three years old to characterize the means of the
Notes. Absolute value of t statistics are in parentheses. + Significant at 10%. ∗ Significant at 5%. ∗∗ Significant at 1%.
Wu and Knott: Entrepreneurial Risk and Market Entry
1326 Management Science 52(9), pp. 1315–1330, © 2006 INFORMS
Figure 4a Histogram of Firm-Year Efficiency Metrics entry cost K, and foregone wage W . This could occur
Cost inefficiency density function for two reasons. First, both variables are correlated
8 with state income, which is the basis for our measures
for demand uncertainty and growth. Accordingly,
market growth likely captures some opportunity cost
6
and foregone wage effects. An alternative explana-
tion for foregone wage is that equilibrium mod-
Downloaded from informs.org by [141.225.1.24] on 19 November 2014, at 10:57 . For personal use only, all rights reserved.
Density
6. Entry costjt 8,265.6 65,533.9 −005 −022 009 008 0.19 100
7. Failure ratejt 0007 0.020 −008 −010 −029 010 0.07 −006 1.00
8. Merger ratejt 0043 0.041 007 −007 002 010 0.06 005 0.02 100
9. F cind jt ∗ 0180 0.073 003 −038 005 043 0.16 024 0.21 005 100
10. ∗ failure ratejt ∗ 0002 0.005 −007 −013 −027 015 0.08 −005 0.95 003 030 1.00
11. RMSE(state incomejt−1 to jt−11 ) 0023 0.012 −014 −008 −022 −007 0.01 −013 0.16 −010 −009 0.15 100
12. F cind jt 0552 0.162 009 −025 017 029 0.30 043 0.01 007 066 0.06 −020 1.00
13. F chold co jt 0551 0.222 000 −027 −008 031 0.35 040 0.09 005 038 0.13 −002 0.34 1.00
Models 2 and 3 reflects the cost distribution of other entrepreneurs compare themselves to other indepen-
independent banks; Model 4 adds the cost dispersion dent entrepreneurs rather than to the large bank hold-
for members of bank holding companies. The coef- ing companies.
ficient for holding company dispersion is near zero,
while that for independent banks remains positive 6.3. Robustness Checks
and highly significant. This suggests that potential We conducted a number of robustness checks of the
main results reflecting concerns expressed in §§5.4.2
and 5.4.3. These checks are presented in Table 5
Table 4 Test of Propositions alongside the main model (Model 3 from Table 4).
(1) (2) (3) (4) The checks include sensitivity to changes in each
∗∗ ∗∗
of the following: the measure used for the market
2 failurejt−1 −3717 −3717 −3518∗∗
277 282 269 demand variables, the functional form of the options
Mergersjt−1 090 087 108 effect, lags for the explanatory variables, definition
074 072 090 of the survivor pool, the distributional assumption in
3 (costs )jt−1 −506∗∗ −667∗∗ −672∗∗ Stage 1 frontier analysis, and the econometric speci-
323 425 425 fication of the main model. Results are robust to all
5 (costs )jt−1 ∗ failurejt−1 1503∗∗ 1539∗∗ 1509∗∗ these changes.
287 301 297
7 RMSE(state incomejt−1 to jt−11 ) −2580∗ −3320∗∗
234 294 7. Discussion
4 (costind )jt−1 195∗∗ 202∗∗ The goal of this paper is to reconcile the risk-
375 386
bearing role of entrepreneurs with the stylized fact
(costhold co )jt−1 010
031
that entrepreneurs exhibit conventional risk-aversion
1 ln(incumbents)jt−1 −094∗∗ −088∗∗ −116∗∗ −139∗∗
profiles. We proposed that the disparity arises from
346 305 313 362 confounding two distinct dimensions of uncertainty:
6 growth(state incomejt−1 to jt−11 ) 2564∗∗ 2545∗∗ 2043∗∗ 1943∗∗ demand uncertainty and ability uncertainty. We fur-
1073 963 708 676 ther proposed that entrepreneurs will be risk averse
CR4jt−1 −193∗ −183∗ −133∧ −189∗ with respect to demand uncertainty, while “apparent
254 236 168 226 risk seeking” (overconfident) with respect to ability
8 (mean wage)jt−1 001 001 001 001
uncertainty. To examine this view, we constructed a
076 082 047 051
reduced-form model of the entrepreneurial entry deci-
9 (entry cost)jt−1 000∧ 000 000 000
176 145 128 138 sion, which we then aggregated to the market level. In
the model, entrepreneurs compare the expected value
Year effects
Constant 670∗∗ 736∗∗ 939∗∗ 112∗∗
of an uncertain profit stream against the opportunity
357 372 358 404 cost of continuing in wage employment. The baseline
Log likelihood −77453 −76606 −75641 −74261 model anticipates that with rational options behavior,
Chi square 262.76 274.83 307.92 317.08 entry will be increasing in mean cost, and decreas-
Observations 637 637 637 629 ing in the failure rate as well as the entrepreneur’s
Number of markets 49 49 49 49 opportunity cost. The model with risk preferences
Notes. Dependent variable is entryjt . Absolute value of z statistics are in adds variables for demand uncertainty as well as abil-
parentheses. ∧ Significant at 10%. ∗ Significant at 5%. ∗∗ Significant at 1%. ity (cost) uncertainty. We tested the aggregate model
Wu and Knott: Entrepreneurial Risk and Market Entry
1328 Management Science 52(9), pp. 1315–1330, © 2006 INFORMS
2 failurejt−1 −3717∗∗ −4762∗∗ −4505∗∗ −2102 −3981∗∗ −4012∗∗ −1548 −3712∗∗ −1163 −1811
282 326 3192 152 315 314 121 282 050 077
Downloaded from informs.org by [141.225.1.24] on 19 November 2014, at 10:57 . For personal use only, all rights reserved.
(Failurejt−1 )2 10200
1565
Mergersjt−1 087 193 096 216∧ 085 082 151 087 681∧ 396
072 148 0793 167 071 069 123 072 172 094
3 (costs )jt−1 −667∗∗ −788∗∗ −1132∗ −470∗∗ −672∗∗ −698∗∗ −665∗∗ −670∗∗ −807∧ −1990∗∗
425 483 2533 280 421 440 327 425 190 395
((costs )jt−1 )2 883
1121
5 (costs jt−1 ∗ failurejt−1 1539∗∗ 1595∗∗ 14670∗∗ 9999 1631∗∗ 1641∗∗ 6481 1544e∗∗ 4704 7650
301 287 3039 180 341 338 134 301 049 078
7 RMSE(state incomejt−1 to jt−11 ) −2580∗ −2408∗ −3467∗∗ −2560∗ −2617∗ −2770∗ −2581∗ −4324∗ −5052∧
234 2161 292 232 238 246 234 221 194
RMSE(agg bank outputjt−1 to jt−11 ) −125
095
4 (costind jt−1 195∗∗ 287∗∗ 187∗∗ 176∗∗ 189∗∗ 188∗∗ 108∗∗ 194∗∗ 459∗∗ 759∗∗
375 549 3578 301 368 369 239 374 282 403
1 ln(incumbents)jt−1 −116∗∗ −059∗ −108∗∗ −137∗∗ −113∗∗ −115∗∗ −095∗∗ −116∗∗ 104∗∗ −432∗∗
313 228 2775 344 312 315 283 313 293 276
6 growth(state incomejt−1 to jt−11 ) 2043∗∗ 1978∗∗ 1968∗∗ 2038∗∗ 2018∗∗ 2053∗∗ 2043∗∗ 3553∗∗ 2979∗∗
708 6725 651 707 701 699 708 475 382
Growth(agg bank outputjt−1 to jt−11 ) 234∗∗
394
CR4jt−1 −133∧ −155∗ −136∧ −141 −135∧ −137∧ −121 −133∧ 098 −773∗∗
168 198 1708 161 170 173 153 168 057 269
8 (mean wage)jt−1 001 −003 000 −001 001 001 001 001 003 001
047 156 0347 081 045 049 072 047 074 018
9 (entry cost)jt−1 000 000∧ 000 000 000 000 000∧ 000 000 000∗∗
128 182 1478 124 144 156 193 128 067 294
Year effects
Constant 939∗∗ 605∗∗ 952∗∗ 1108∗∗ 927∗∗ 950∗∗ 820∗∗ 938∗∗ −364 284∗∗
358 332 3414 382 360 365 354 358 133 332
Log likelihood −75641 −78777 −754691 −67152 −75602 −75527 −76197 −75644
Chi square 307.92 217.01 311.101 312.43 308.21 311.57 292.52 307.83 144.10
Observations 637 637 637 588 637 637 637 637 663 663
Number of markets 49 49 49 49 49 49 49 49 51
Notes. Dependent variable is entryjt . Absolute value of z statistics are in parentheses. ∧ Significant at 10%. ∗ Significant at 5%. ∗∗ Significant at 1%.
across markets and over time in the banking indus- fident. It is noteworthy that we obtain these results
try. We found that entrepreneurs behave rationally to from real entry behavior rather than with personality
the options structure of payoffs, but that in addition instruments. Thus, this test is similar to lab experi-
they appear to be risk averse with respect to demand ments in that we observe how people actually behave
uncertainty and “risk seeking” (overconfident) with rather than how they report they will behave. More-
respect to cost uncertainty. over, this behavior occurs in real settings with sub-
The test was constructed in a manner that allowed stantial investments (approximately $11 million per
us to draw inferences at both the individual and mar- entry), thereby offering greater external validity than
ket levels. For the characterization of entrepreneurial lab experiments.
personality, our results using large sample data It is worth repeating that our results about individ-
(1,635 entrants) confirm observations made previ- ual biases pertain exclusively to entrepreneurs who
ously for small samples—namely, that entrepreneurs actually enter a banking market. Thus, we are unable
as a group appear to be both risk averse and overcon- to say anything about how (if at all) entrepreneurs’
Wu and Knott: Entrepreneurial Risk and Market Entry
Management Science 52(9), pp. 1315–1330, © 2006 INFORMS 1329
one dimension. Accordingly, what may distinguish Vredenburg. 1995. Personal contact, individuation and the
entrepreneurs from wage earners in this setting are better-than-average effect. J. Personality Soc. Psych. 68 804–825.
simple differences in human capital and wealth. Allison, S. T., D. M. Messick, G. R. Goethals. 1989. On being better
but not smarter than others: The Muhammad Ali effect. Soc.
The test does, however, allow us to draw inferences Cognition 7 275–296.
about the market-level implications of entrepreneurs’ Berger, A. N., D. B. Humphrey. 1992. Measurement and efficiency
biases. In particular, our results offer a possi- issues in commercial banking. Z. Griliches, ed. Output Measure-
ble reconciliation of observed risk aversion with ment in the Service Sectors. Vol. 56. University of Chicago Press,
the entrepreneur’s role as economic risk bearer. Chicago, IL, 245–279.
Entrepreneurs are willing to bear economic risk when Berger, A. N., D. Hancock, D. Humphrey. 1993. Bank efficiency
derived from the profit function. J. Banking Finance 17(2, 3)
the degree of performance dispersion dominates the 317–347.
degree of demand uncertainty. In those instances, Blau, D. M. 1987. A time-series analysis of self-employment in the
overconfidence can compensate for risk aversion to United States. J. Political Econom. 95(3) 445–467.
achieve sufficient entry. This appears to be the case Brockhaus, R. 1980. Risk-taking propensity of entrepreneurs. Acad.
in banking. Potential market failures exist, however, Management J. 23(3) 509–520.
in instances where there is either a high degree of Busenitz, L. W. 1999. Entrepreneurial risk and strategic decision
making: It’s a matter of perspective. J. Appl. Behav. Sci. 35(3)
demand uncertainty but low performance dispersion 325–340.
(insufficient entry) or low demand uncertainty but Camerer, C., D. Lovallo. 1999. Overconfidence and excess entry:
high performance dispersion (excess entry). An experimental approach. Amer. Econom. Rev. 89(1) 306–318.
There are, of course, caveats to our results. The test Cameron, A. C., P. K. Trivedi. 1998. Regression Analysis of Count
was conducted in the banking industry because of Data. Cambridge University Press, Cambridge, UK, and New
its high rate of entry and nice experimental proper- York.
Cantillon, R. 1755. Essay on the Nature of Commerce. Higgs transla-
ties. One question, then, is how results for banking
tion, 1931. Macmillan, London, UK.
might generalize to other settings. The most obvi-
Cebenoyan, A., G. Papaioannou, N. Travlos. 1992. Foreign takeover
ous distinction between entry in banking versus other activity in the United States and wealth effects for target firm
industries is that it is regulated. A review of charter shareholders. Financial Management 21(3) 58–68.
requirements suggests that this has three effects. First, Cramer, J. S., J. Hartog, N. Jonker, C. M. van Praag. 2002. Low
bank founders are highly qualified. To gain char- risk aversion encourages the choice for entrepreneurship: An
empirical test of a truism. J. Econom. Behav. Organ. 48(1) 29–36.
ter approval, they must have both extensive banking
Dunn, T., D. Holtz-Eakin. 2000. Financial capital, human capital,
experience and strong ties to the local community. and the transition to self-employment: Evidence from intergen-
Second, banks require substantial up front investment erational links. J. Labor Econom. 18 282–305.
of approximately $11 million in physical and financial Evans, D., L. Leighton. 1989. Some empirical aspects of entrepre-
assets for our sample. Third, the chartering agencies neurship. Amer. Econom. Rev. 79(3) 519–535.
“match” the number of entrants to market demand Federal Reserve Bank of Atlanta. 2005. Banking Applications. Re-
through the requirement that the new bank satisfies trieved July 1, 2006, http://www.frbatlanta.org/ba.
Goldberg, L. G., L. J. White. 1998. De novo banks and lending to
an unmet need. The implication of all three effects
small businesses. J. Banking Finance 22 851–867.
is that there should be less entry in banking than in Greene, W. H. 1997. Econometrics Analysis, 3rd ed. Prentice-Hall,
other industries, both through the supply effect (fewer Upper Saddle River, NJ.
people will satisfy the high human capital and phys- Hamilton, Barton H. 2000. Does entrepreneurship pay? An empir-
ical or financial capital requirements or both required ical analysis of the returns to self-employment. J. Political
by charters) and through the demand effect (new Econom. 108(3) 604–631.
charters must satisfy unmet needs). Headd, B. 2003. Redefining business success: Distinguishing
between closure and failure. J. Small Bus. Econom. 21 51–61.
Hermalin, B., N. Wallace. 1994. The determinants of efficiency and
Acknowledgments solvency in savings and loans. RAND J. Econom. 25(3) 361–381.
The authors thank Tim Folta, Andrew King, Dan Lovallo, Jovanovic, B. 1982. Selection and the evolution of industry. Econo-
Barry Rubens, Mirjam Van Praag, Govert Vroom, the asso- metrica 50(3) 649–670.
ciate editor, and two anonymous referees, as well as sem- Kanbur, S. M. 1979. On risk taking and the personal distribution of
inar participants at Dartmouth College, Purdue University, income. J. Political Econom. 87(4) 760–797.
the University of Maryland, the United States Association Kihlstrom, R. E., J. J. Laffont. 1979. A general equilibrium entrepre-
for Small Business and Entrepreneurship, and The Wharton neurial theory of firm formation based on risk aversion. J. Polit-
School for helpful comments. ical Econom. 87(4) 719–748.
Wu and Knott: Entrepreneurial Risk and Market Entry
1330 Management Science 52(9), pp. 1315–1330, © 2006 INFORMS
Knight, F. H. 1921. Risk, Uncertainty and Profit. Houghton Mifflin, O’Brien, J. P., T. B. Folta, D. R. Johnson. 2003. A real options per-
New York. spective of entrepreneurial entry in the face of uncertainty.
Knott, A. M., H. Posen. 2005. Is failure good? Strategic Manage- Managerial Decision Econom. 24(8) 515–533.
ment J. 26(7) 617–641. Palich, L. E., D. R. Bagby. 1995. Using cognitive theory to
Lippman, S. A., R. P. Rumelt. 1982. Uncertain imitability: An analy- explain entrepreneurial risk-taking—challenging conventional
sis of interfirm differences in efficiency under competition. Bell wisdom. J. Bus. Venturing 10(6) 425–438.
J. Econom. 13(2) 418–438. Petersen, M., R. Rajan. 2002. Does distance still matter? The infor-
Lofstrom, M. 2002. Labor market assimilation and the self- mation revolution in small business lending. J. Finance 57(6)
Downloaded from informs.org by [141.225.1.24] on 19 November 2014, at 10:57 . For personal use only, all rights reserved.