Experience Does Matter:
Managerial Capital and the Dynamics of Entrepreneurship
Russell Toth,y
October, 2014
Abstract
A key question for entrepreneurship policy is how entrepreneurship-speci…c human capital
is acquired and its e¤ect on performance. This paper sheds light on this issue by providing
evidence on the causal returns to experience in small business entrepreneurship. We exploit
quasi-experimental variation in formal sector shocks in Indonesia, which induces exogenous
variation in the years of experience accumulated in entrepreneurship for individuals on the margin between wage work and informal-sector enterprise activity. Since the country’s formal sector
is relatively small the local average treatment e¤ect is policy relevant since a policymaker might
want to target entrepreneurial support to individuals on this margin. We provide a placebo test
to verify that labor-market shocks provide a valid instrument for occupational selection, excludable from the returns equation. We …nd that an extra year of overall experience running one’s
current business leads to an average increase in returns of about 2.7 percent per year, larger and
more likely to be signi…cant than for one’s total experience, consistent with …rm-speci…c human
capital. The e¤ect is also signi…cantly larger amongst younger individuals. The paper provides
some of the …rst evidence of a causal return to experience in entrepreneurship, complementing
the (largely experimental) literature on the returns to human capital in entrepreneurship in
developing countries.
I thank seminar audiences at the University of Southern California, Monash University, the University of Pittsburgh/Carnegie Mellon University and the Otago Development and Growth Workshop for useful feedback. For feedback on earlier iterations of this project and inspiration for the current work I thank Thomas Åstebro, Chris Barrett,
Jim Berry, AV Chari, David Easley, Corey Lang, Josh Lerner, Maximilian Mihm, Ted O’Donoghue, Albert Park,
Matthew Rhodes-Kropf, Viktor Tsyrennikov, discussants Jenny Aker, Richard Holden, Claus Schnabel, and L.G.
Thomas, and seminar audiences at Australian National University, Cornell University, Harvard Business School, the
Graduate Institute-Geneva, the University of Sydney, the University of Waterloo, NEUDC (MIT-Sloan), Harvard
Business School International Research Conference, the Third IZA Entrepreneurship Workshop, and the Workshop
on Emerging Economies (University of New South Wales). I thank Andre Syafroni, Maria Wihardja and KPPOD for
assistance with …eldwork. I acknowledge support from the Cornell University Institute for the Social Sciences, the
Mario Einaudi Center for International Studies, the Cornell University Graduate School, and NSF Expeditions grant
number 0832782. All errors are my own.
y
School of Economics, Room 363 Merewether Building H04, The University of Sydney, NSW 2006 Australia. Web:
https://sites.google.com/a/cornell.edu/russelltoth/. Email address: russell.toth@sydney.edu.au
1
Introduction
Does managerial capital matter? Bruhn et al. (2010) label the set of skills speci…c to managerial
performance "managerial capital" and discuss how it has received relatively little attention in the economics literature until quite recently. Bloom and van Reenen (2007, 2010) report on a cross-country
project that allows for the measurement of managerial practices. While they …nd that the adoption
of standardized management practices is correlated with better …rm performance, it is still di¢ cult
to know if this is a causal relationship, due to various potential sources of unobserved heterogeneity
in the relationship between management practices and …rm performance. Some recent studies have
attempted to overcome this selection problem through conducting randomized experiments providing management consulting and technical support services to …rms of various sizes, particularly in
the developing world (e.g., Bruhn et al. (2012), Bloom et al. (2013)). However, it is still di¢ cult
to identify whether the reported positive treatment e¤ects represent a long-lived increase in managerial capital per se, or merely a short-lived information or technical assistance shock. We take a
somewhat di¤erent approach, taking the accumulation of direct experience in running a small business as a proxy for managerial capital accumulation. We estimate the return to managerial capital
through exploiting a unique natural experiment that provides plausibly exogenous variation in years
of experience running a small business.
While a substantial literature in labor economics that similarly proxies human capital accumulation
through years of formal education and year of wage work experience generally points to a positive
e¤ect of human capital on earnings,3 rigorous empirical evidence focused on human capital speci…c to
entrepreneurship is relatively scarce. Key to identifying the stock of managerial capital is the ability
to disentangle it from other (generally unobservable) factors generating selection into entrepreneurial
occupations, such as a market opportunity, a "good idea" shock, unobserved ability or information,
and the composition of a …rm’s management team. To attempt to overcome this identi…cation
challenge,4 we exploit a uniquely-suitable natural experiment: shocks to the private labor sector that
vary the incentive to move from wage employment to single-proprietor enterprise activity yet do not
have a …rst-order e¤ect on the returns to running such an enterprise. This allows us to make use of
these shocks as instruments for occupational selection and hence the accumulation of experience, to
estimate the causal e¤ect of experience on returns.
We focus on Indonesia, where a uniquely rich panel dataset on entrepreneurial dynamics is avail3
See Card (1999) for a review of studies taking years of education as a proxy for generalized human capital
accumulation, and Angrist (1990) and Behrman & Rosenzweig (1999) for studies of human capital accumulation
through work experience.
4
Throughout the paper I use the terms "entrepreneurship", "self-employment", "running a business" and the like,
largely interchangeably. I acknowledge that this is not fully satisfying, while arguing that these distinctions are not
so important for the current context.
1
able spanning about 20 years (1988-2008). Beyond data availability, a large developing economy is
attractive for a few reasons. First, we can argue institutionally and historically, and corroborate
statistically, that the private labor sector is largely independent of the informal sector where much
of the small business activity occurs. This is critical for identi…cation: to satisfy our exclusion restriction we need (lagged) formal sector shocks have …rst-order e¤ects on selection into enterprise
activity, but not (subsequent) returns once there. Second, a developing country tends to have less
social protections that might distort selection processes after negative labor market shocks. Finally,
since the private labor sector is relatively small, there are individuals reasonably high in the talent
distribution who participate in informal-sector enterprise activity, and we estimate a local treatment
e¤ect on a relatively high-ability population, which adds policy relevance to the results.
After developing a stylized model that captures the mechanisms we have in mind, we carry out
the empirical exercise to estimate the causal e¤ect of experience on earnings. We use years of
experience in running a business as a proxy for managerial capital acquisition through learning-bydoing, parsing experience by "total experience" (all years of experience in running any business) and
"current experience" (years of experience in running one’s present business). By parsing these two
measures we can also shed light on the role of …rm-speci…c managerial capital. The key identi…cation
challenge is that years of experience is clearly endogenous in a regression of earnings on experience:
entrepreneurs with higher earnings for any number of reasons will be more likely to accumulate more
years of experience. In order to overcome this identi…cation challenge and provide causal evidence
on this issue, we estimate returns to experience through an instrumental variables strategy, using
labor market churning measures that are analogous to a Bartik instrument (Bartik (1991)). The
instrument set is a vector of prior shocks (the contemporaneous shock is never included), speci…ed
in two di¤erent ways based on order of lag or order of year as will be explained in more detail later.
We additionally control for province-year …xed e¤ects in all speci…cations.
Using either instrument set, we estimate an average return to current experience of 2.7% per year.
When we additionally restrict the sample to individuals who were born after 1968 and hence only
have their experience measured using the detailed, annual data in the survey, we obtain an estimate of
up to 8.4% per year in one instrument set. Hence the e¤ect of experience seems to be diminishing in
age. Perhaps the most obvious explanation for this is that individuals might acquire complementary
forms of human capital through other activities even if they aren’t running a business, diminishing
the marginal return to enterprise-speci…c experience. While returns to experience in the current
enterprise are intuitively reasonable and economically meaningful, returns to total experience are
not signi…cant in most speci…cations. In addition, when we instrument both measures in the same
regression as a kind of horse race, we again …nd that current experience carries a much larger e¤ect.
One drawback of these estimates is that they do not control for the type of business that an
2
individual runs. Our dataset does not only record whether or not an individual was running the
business, but also the classi…cation of whether it was a business with no employees, only family
members or other unpaid employees, or paid wage workers. A number of authors (e.g., Liedholm
& Mead (1999); Schoar (2010)) have drawn important distinctions between these di¤erent kinds of
enterprises. If we don’t correct for selection between these di¤erent enterprise types, it is possible
to introduce bias in the population estimates in the return to experience, perhaps in particular from
individuals who "get stuck" making relatively low returns in the enterprise sector. To correct for
this we adopt the approach of Thomas & Strauss (1997), …rst instrumenting for selection between
enterprise types, then projecting the hazard rate in enterprise selection into the second stage equation,
and again instrumenting for years of experience. After making this correction we …nd larger e¤ect
sizes on current experience of 4.4-12.1% that are still statistically signi…cant, and furthermore we …nd
that in some speci…cations there is also a statistically signi…cant return to total experience (around
4.6%).
Finally, in order to corroborate the key identifying assumption, that lagged labor market shocks
do not have …rst-order e¤ects on current enterprise returns, we explore the hypothesis that in the
Indonesian context informal-sector enterprise returns are largely independent from labor market
shocks. We are not able to test this hypothesis directly by running enterprise returns on lagged
shocks, as this is the reduced form of our IV. As noted earlier, however, the contemporaneous value
of the shock is excluded from the instrument vector, so as a partial test we run contemporaneous
enterprise returns on contemporaneous values of the instrument. We otherwise apply the same
regression speci…cation. This test assumes (1) that the contemporaneous relationship provides the
strongest representation of potential shock transmission, and (2) that a (lack of) e¤ect of shocks on
the larger population of informal-sector enterprises is reasonably representative of the same e¤ect
for the kinds of individuals who are responsive to our instrument. We that that the statistical
relationship between contemporaneous shocks and returns is very weak (i.e., p-values are on the
order of 0.9-0.97), corroborating the identi…cation strategy.
The unique empirical setting that we study allows us to contribute to the entrepreneurship literature
on the returns to experience. One of the main puzzles in this literature is that observed returns to
entrepreneurship seem to be ‡at over the life-cycle of the enterprise (i.e., entrepreneurs do not see
a signi…cant increase in earnings over the life cycle of their business, especially relative to wage
workers). See in particular Astebro (2012); Tergiman (2013) is among the papers that document this
in the US context. This seems to go against a theory that involves accumulation of entrepreneurial
skill or human capital. However, there are a couple of notable caveats. First, this stylized fact is
applicable to studies on entrepreneurship in the broader population, which are dominated by micro
and small …rms, and often lack information on experience operating prior enterprises. If we focus on
3
the set of larger and/or venture capital-funded enterprises in the presence of data on prior founding
experiences we …nd more signi…cant returns to experience, whether in the lead entrepreneur (Stuart
& Abetti (1990); Gompers et al. (2010); Eesley & Roberts (2012)), or the founding team (Delmar
& Shane (2006)). Gompers et al. (2010) argue that experience might particularly matter for timing
the market, while Eesley & Roberts (2012) argue that experience matters most when facing more
uncertainty. While based on data from quite di¤erent contexts, these ideas about the kinds of
judgment and performance skills that can be acquired by experience are relevant for interpreting the
results herein.
In addition, the model of Kawaguchi (2003) suggests a possible theoretical rationale for why we
would observe little gradient between experience and returns in the broader population of entrepreneurs. In his model, in order to survive, new entrants need to meet a reasonable standard of business
performance, so they generally only startup when they have su¢ cient human capital, while they face
higher opportunity costs to human capital investment on the job. Hence when they startup by free
choice, they immediately achieve a relatively high income level, fairly close to their longer-run level,
or exit. Kawaguchi (2003)’s theory implies that in order to detect learning-by-doing or experience
e¤ects, we would need entrepreneurs to be "forced" to enter this occupation before they would freely
choose to. This is exactly the kind of natural experiment that is pursued herein, by estimating returns
to experience among entrepreneurs whose earnings aren’t truncated by the endogenous entry decision. To our knowledge this is the …rst paper to provide such evidence for returns to entrepreneurial
experience.5
The question of what drives enterprise dynamics has important implications for enterprise policy
and our understanding of entrepreneurship. This issue has taken on new urgency in the developing
world due to the pressures and opportunities of globalization and the stresses of economic crises, with
an increased hope that vibrant private-sector enterprise activity can be a source of economic growth
and poverty reduction. While a number of explanations have been suggested to explain enterprise
dynamics on the intensive and extensive margin, including credit constraints, government regulations,
market frictions, and the (…xed) distribution of entrepreneurial ability, we are only beginning to gain
a clearer sense of what factors might be relatively important.
The paper proceeds as follows. We then outline a simple, dynamic model of entrepreneurial selection
to motivate and present our empirical strategy in Section 2. We introduce and describe the data in
Section 3. We estimate the main causal e¤ect of experience on earnings and present the key placebo
5
Analogously Shu (2012) exploits long-term US business cycles as a source of exogenous variation to argue that
economic booms lead elite US university graduates from MIT into occupations where they acquire signi…cantly less
inventive human capital and produce less patents.
4
test in Section 4. Section 5 expands on these results with a number of additional speci…cations and
robustness checks. Section 6 concludes with a discussion of the broader relevance of the results.
Tables and …gures appear in the Appendix.
2
A Simple Model of Managerial Capital Accumulation and
Empirical Strategy
In this section we outline a simple model of occupation-speci…c human capital accumulation in two
sectors. The model provides a useful framework to illustrate the central mechanism and empirical
strategy we have in mind, and especially to aid in the interpretation of parameter estimates. It also
provides a couple of additional theoretical implications that we explore in the empirical analysis.
2.1
Model of Occupational Choice and Occupation-Speci…c Skill
Consider a model in which an agent makes a binary occupational choice and accumulates earnings
in each of two periods. The binary occupational choice is between working for a wage and obtaining
the return w, or running a enterprise and obtaining the return R. One can think of this as a dualeconomy setup: the wage earning activity is in the formal sector, while the enterprise activity is in
the informal sector, so occupation and sector are one and the same for our purposes. We suppose
that there is some degree of independence between the two sectors. Namely, the formal sector is
primarily made up of …rms with paid workers, often with access to credit from the larger …nancial
institutions, trading in a fairly broad, established markets, whereas informal sector enterprises are
generally small, often have no paid employees, have little access to credit from the larger …nancial
institutions, and primarily serve very localized markets. Hence at least to a …rst-order approximation
the two sectors face largely independent distributions of shocks, due to being integrated with quite
di¤erent markets. We parameterize the state of the formal sector shock in the most recent prior
period by , which has mean value .
Suppose that the agent is endowed with a stock of skill speci…c to each occupation: W represents
the stock of skill speci…c to wage employment, while M C denotes the managerial capital stock of
the agent. Denote the starting values of each stock by 0W and 0M C , respectively. These occupationspeci…c skill stocks can increase through direct experience in each respective occupation: experience
in the wage sector increases W and M C increases through experience in the enterprise sector.6 This
6
The skills could also be enhanced by occupation-speci…c training, but training interventions are outside the scope
of this paper.
5
will be a simple model of occupation-speci…c skill accumulation, ignoring cross-occupational skill
accumulation, depreciation of skills, and other mechanisms, but it is su¢ cient for our purposes. In
addition, the agent accumulates generalized human capital, a = (a), just through the course of
aging, where a represents one’s age. a is complementary to both W and M C in the respective occupational returns functions. In addition, suppose that there is an additional parameter representing
unobserved ability that is not perfectly controlled for or correlated with W , a or M C .7
Hence in each of the two periods the agent will select an occupation by solving the following simple
comparison:
max fw ( W ; a ; ; ) ; R ( M C ; a ; )g ;
(1)
where each return function is increasing in each of its arguments, and we can think of R ( ) as a
simpli…ed pro…t function (i.e., the value function of the …rm’s pro…t maximization problem) only
taking M C and a as main arguments. We assume that w and R are twice-di¤erentiable, with the
partial and cross-partial derivatives each taking positive sign.8
To simplify the exposition of the model we will abstract away from explicitly modeling the dynamics. One could imagine functions W , M C , mapping current W and M C , respectively, into
the next-period value of each. One could build on this structure to provide a fully forward-looking,
dynamic optimization model, which would deliver additional predictions on issues such as human
capital lock-in and human capital investment (e.g., one might be willing to sacri…ce earnings today
in return for accumulation human capital that will be more valuable tomorrow). However, our empirical framework is not well-suited to address these predictions, and there are reasons to question
the extent to which individuals have well-formed beliefs about human capital accumulation dynamics
such that they could act on them.
This simple framework captures the core intuition behind models of occupational choice, including
that the highest-ability individuals in a particular occupation might not select into that occupation,
an insight going back to at least Roy (1951). I.e., an individual could have a high value of W relative
to the population, but even abstracting from , that individual might select into the enterprise sector
if M C is su¢ ciently high. This is one potential rationale for the signi…cant cohort of low-skill, selfemployed individuals in developing countries, most of whose enterprises have low returns and grow
little.9
7
Although we have only presented the static version of the model so far, we can assume that could be time-varying
in a dynamic version of the model, so that …xed e¤ects are not su¢ cient.
8
I.e., w W ; w a ; w ; R M C ; R a > 0, w W a ; w W ; w a ; R M C a > 0.
9
This charactization is consistent with recent empirical evidence (e.g., Carter & Olinto (2003); de Mel et al.
(2008); Banerjee et al. (2009); Karlan & Zinman (2010)). Demand for capital ends up being relatively stronger
amongst wealthier or higher-ability individuals and hence individuals end up more responsive to positive …nancial
shocks.
6
2.1.1
Using the Model to Illustrate the Empirical Strategy
We can illustrate the core endogeneity problem and how we attempt to overcome it with the static
version of the model alone. In the context of the model, our core objective in the empirical analysis is
to identify the average value of @R=@ M C within subsets of the population, by making comparisons
of enterprise returns between individuals who may have di¤erent numbers of years in experience in
running an enterprise, but are otherwise identical. Even if we can control for W , a and M C ,
could still drive selection. I.e., it could be that for two individuals i and j that W , a and M C are
identical, but i has a higher value of , leading that individual to select into enterprise more often
than j. Since is driving returns, we would observe i selecting into enterprise more often, hence
accumulating greater experience in that occupation, and in addition enjoying higher returns when
running an enterprise. This would lead us to overestimate the true, causal returns to M C .
One of the key restrictions in the equation (1), the exclusion of , highlights how we propose to
overcome this endogeneity concern. For expositional purposes we use the sub-script t to index discrete
time periods. We use a proxy for as an instrumental variable that in‡uences occupational selection,
but not (future) returns. In fact, we use a vector of proxies of over multiple annual lags, i.e.,
t T ; : : : ; t 1 , excluding t for the mechanical reason that since the data is recorded on an annual
basis the shock that could lead to a switch in occupations in period t would have had to occur in period
t 1 or earlier. To see the role of the shock suppose, for example, that w ( W;t ; a;t ; ) = R ( M C;t ; a;t )
so the agent is indi¤erent between the two sectors when t 1 = , and then in period t dips to
value t < . In period t + 1 we would have w ( W;t+1 ; a;t+1 ; t ) < R ( M C;t+1 ; a;t+1 ), and hence
the individual chooses to accumulate experience in the informal enterprise sector. The source of
exogenous variation in M C that we exploit is due to the average variation in response to these
ongoing shocks. This exclusion of from R ( ) formalizes the key identi…cation assumption in the
empirical strategy: at least up to a …rst-order approximation, the lag of shocks, t 1 , that a¤ect the
formal sector only in‡uence selection between occupations and do not directly a¤ect returns in the
informal sector. Hence our thought exercise is that we are comparing two individuals with identical
characteristics but for the fact of living in regions with di¤erent shock histories that can potentially
drive them to have di¤erent values of M C .
This discussion is also suggestive of our main placebo test to attempt to corroborate the validity of
the instrumental variables strategy. As noted, we exclude t from the instrument set t T ; : : : ; t 1
(and, of course, from R ( )). This allows us to test whether t should appear in the returns function
R, for individuals who are currently running an enterprise, by testing whether t is statistically
signi…cant in a regression with R as the dependent variable. While this is not a perfect test of the
exclusion restriction, partly because the propogation of shocks from the formal sector to the informal
enterprise sector could take more than one year, and partly because shocks may a¤ect the general
7
population of informal-sector enterprises di¤erently than they a¤ect the population susceptible to our
instruments, it at least provides some con…rmation. Indeed, we will show the relationship between
t and R is far from statistically signi…cant.
Futhermore, this speci…cation of the shock also highlights a key caveat on the interpretation of
the empirical results: the treatment e¤ect we can estimate will necessarily be local to to individuals
whose occupational choices can be in‡uenced by . I.e., for some individuals with very weak prospects
in the wage sector, it could be that w ( W ; a ; )
R ( M C ; a ). Regardless of the value of ,
such individuals will never be induced to switch into the wage sector. Conversely, there could
also be individuals with such high returns to wage employment that they would never consider the
enterprise sector, again regardless of . We argue that while we cannot estimate a population-average
treatment e¤ect, the treatment e¤ect parameter that we can identify is interesting as individuals on
the margin between the two sectors are obvious targets for policy interventions meant to increase
entrepreneurship. We will revisit these issues around empirical identi…cation and interpretation when
we present the full empirical strategy.
Also note that W and M C each only appear in one returns function. It is reasonable to posit
that skills accumulated in one occupation might be useful in the other. For example, managerial
capital might be useful in supervisory roles in a wage setting, while various skills that are valuable in
running an enterprise might be obtained in a wage job. We can think of the model as either ignoring
such possibilities, or interpret M C and W as pure "position-speci…c" skills, with a capturing the
accumulation of generalized skills.
2.2
Further Implications
The speci…cation of the returns function for the enterprise occupation, R ( M C ; a ; ), suggests a
couple of additional implications. First, recall that we assume that the parameters of R are complements and a represents generalized human capital accumulated as a function of age, a. Hence for
two individuals, i and j, who are identical but for having been born in di¤erent years, ai > aj if
ai > aj . Hence it follows that, conditional on ai and aj ,
@R (
M Ci ; a i ; i )
@
<
M Ci
@R
M Cj ; a j ; j
@
;
M Cj
so long as i = j . I.e., the marginal return to managerial capital will be higher for individual j, who
is younger. We can test this by estimating the returns to managerial capital on di¤erent sub-samples
of the data based on age cuto¤s.
Second, the role of age adds important subtlety to the interpretation of results, as di¤erent-aged
8
individuals face a di¤erent pro…le of shocks, t 1 ; : : : ; t T and shocks have di¤erent e¤ects on individuals at di¤erent ages. As outlined in the previous paragraph in the context of the marginal
e¤ect of M C on R, in payo¤ functions in which arguments are complements, the accumulation of
generalized human capital will dampen the e¤ect of changes in the other argument of the function.
Hence this is also true in the context of the function w ( W ; a ; ; ): all things equal the e¤ect of
on w will be reduced for older individuals with a larger value of a . This means that there will
be a heterogenous response to on the age dimension, or in other words the variation in M C will
be correlated with age. While we initially present empirical results on the e¤ects of M C without
discriminating by age, e¤ectively an estimate of the treatment e¤ect average across the distribution
of heterogeneous responses to by age, we also present results in which we use the interaction
a
to construct the instrument set.
Finally, it is notable that the speci…cation of a stock M C is a relatively crude way to capture skill
in operating a business. In particular, it may be that entrepreneurial skill is at least partly speci…c
to the enterprise in question, due to speci…c know-how relevant to that particular enterprise, or the
industry it is in. Hence we might wish to parse out M C into curr
M C , managerial capital speci…c to
the current enterprise, lifetime managerial capital speci…c to the industry the current enterprise is
in, and total managerial capital across all enterprises. Our data will allow us to approximate these
distinctions by recording years of experience across di¤erent enterprises that an individual gains
experice with. This leads to the natural question of whether these di¤erent measures of managerial
capital have di¤erential e¤ects on returns. If enterprise-speci…c or industry-speci…c know-how matters
then we would expect to see higher estimated returns to these stock of managerial capital than total
enterprise experience.
2.3
Empirical Strategy
The stylized representation of the economy in the above model feeds into our empirical strategy.
While it is tempting to measure the returns to experience by simply correlating returns with years
of experience in the presence of some control variables, this approach su¤ers from the classic endogeneity problem in the human capital literature. Namely, it could be that individuals have certain
unobservable characteristics (e.g., ability) that make them more likely both to accumulate more human capital and experience higher returns, independent of the direct impact of human capital on
returns. The classic example is the measurement of the returns to education: high-ability individuals
may be both more likely to accumulate more years of schooling and earn higher subsequent wages,
with the latter potentially only a partial causal e¤ect of the former. The typical empirical strategy
employed to overcome this issue is to search for sources of exogenous variation in years of schooling,
from cost shifters such as distance to school to policy restrictions such as student quotas.
9
The version of this endogeneity issue that we consider herein is a twist on the classic problem.
In the case of schooling, individuals usually …rst accumulate human capital, and then subsequently
redeem its value in the labor market. Incentives to continue to accumulate years of schooling may
be linked in sensible ways to expectations about later earnings, but these connections are somewhat
loose because the earnings only occur in the future. By contrast, entrepreneurs get immediate
feedback on their performance along the way, through earnings, which we would expect to be at
least partially determined by unobserved ability. We naturally expect that this will lead to direct
behavioral response, with the particular concern being on the extensive margin – individuals may
completely exit running a business in response to returns. Hence we expect the observed sample of
entrepreneurs at any given time to be a direct product of such selection forces in the market.
In order to empirically identify MC, the ideal experiment would randomly assign MC to individuals,
orthogonal to all other characteristics, and then observe the resulting enterprise performance trajectories. Clearly such an experiment would be infeasible for a number of reasons, including endogenous
enterprise survival, and di¢ culties in "assigning" MC. Since the ideal experiment is not feasible in
practice, we …rst exploit a source of exogenous and unanticipated assignment into self-employment
(experience).10 In particular, the interest is in individuals who were ’pushed’into self-employment,
who would not have otherwise entered, which provides a source of a counterfactual to consider the
e¤ects of the quasi-random assignment of MC.
In order to have exogenous variation in years of experience we exploit cross-regional variation
in entry incentives in running an enterprise, using measures of formal-sector market churning as
instruments. These are the measures summarized in Table 3. Formally, the IV strategy is as follows.
In the …rst stage we instrument for total years of experience in running a business for person i in
community j in time period t, as follows,
expijt =
0
+ Instrjt + Xijt +
X
year province I
(year) I (province) + "ijt ;
where Instrjt is the set of instruments generated at the community level, Xijt are other controls, and
are dummy coe¢ cients on year*province …xed e¤ects. The second stage is then,
yijt =
10
0
0
+
0
e\
xpijt + Xijt
0
+
X
0
year province I
(year) I (province) +
ijt ;
It is true that a number of recent studies have used randomized control trials to estimate the causal e¤ect of
enterprise training or technical assistance on enterprise outcomes, and hence potentially uncover the value of enterprisespeci…c human capital. However, these studies su¤er from a couple of key identi…cation problems: …rst, they entangle
the quality of the intervention with the fundamental value of enterprise-speci…c capital; i.e., if there is a zero e¤ect
we do not know if it is a failure of the theory or a failure of the intervention. Second, it is not clear whether such
problems fundamentally in‡uence human capital, or merely provide a shot of information that may have temporary
value.
10
where yijt is the outcome of interest (reported net pro…ts), e\
xpijt is the projected experience value
coming from the …rst-stage, and the parameters have primes (0 ) to signify that they are di¤erent
estimands.
We consider speci…cations in which we normalize the value of returns relative to an index expressed
in dollar values, so we cancel out the role of currency ‡uctuations such as in‡ation. Also, by normalizing against market returns, the results are e¤ectively a form of "di¤erences in di¤erences," indexing
to the dynamic path of wage earnings in the formal sector, which is arguably a more appealing
reference for comparison.
The instrument set in the …rst stage is constructed as a set of lags of measures of market shocks.
We construct the lagged values in two ways. First, we vary the lag structure, so that the lags are
relative to the year in question. This speci…cation assumes that the e¤ects of shocks are primarily
determined by the number of years of lag from a particular year. Second, we …x the lag structure,
so that it is the particular shock year that matters, regardless of the year of the observation. We
remain agnostic on which is the preferred speci…cation, reporting results for both.
In addition, we consider the e¤ect of truncating the dataset on age. This is interesting for two
reasons. The …rst is simply technical: the most reliable data we have on occupational choices spans
1988-2008, so we are eliminating individuals who would have entered the workforce prior to that
date. Hence the results should be more reliable as they are based on full year-on-year elicitation of
occupational choices. Second, it is interesting to hone in on a sample of younger individuals. If we
think that there is a concave learning function, then we might expect marginal returns to be higher
for younger individuals.
3
Data and Preliminary Evidence
In this section we discuss the dataset that will be used throughout the paper, and present some
initial descriptive evidence from the dataset.
3.1
Data
Our primary dataset is the Indonesia Family Life Survey (IFLS).11 The data were collected as a
household panel survey in Indonesia, with data collection rounds in 1993, 1997-98, 2000-01 and 200708. The 1997-98 round directly preceded the Asian …nancial crisis. For the intervening years when
11
Various organizations and researchers have been involved in designing, collecting and funding the IFLS. For more
details, see Strauss et al. (2009), Strauss et al. (2004), Frankenberg & Thomas (2000), and Frankenberg & Karoly
(1995).
11
the survey is not …elded, signi…cant retrospective data are collected in the subsequent round. The
dataset was designed to be representative of 83% of the Indonesian population in 1993, covering 13
of the higher-population provinces generally in the western parts of the country, with over-sampling
of urban locations and locations outside Java island, the main economic hub. Data were collected at
the individual, household, and community level, and these three sources can be matched together.
More details on relevant parts of the dataset, including for enterprise activity, will be discussed in
more detail below.
The original 1993 round of the survey (IFLS1) surveyed 7224 households. Subsequent rounds
have involved re-sampling the original households, and then sampling all split-o¤s from the original
households. Attrition has been relatively minor, at less than 10% between rounds, and overall 87.6%
of the original households appear in all four rounds. Table 1 presents the number of individuals,12
households, household enterprises and communities appearing in each round of the survey. We see
that the sample expands in each subsequent round, as splits from the original households are tracked
and surveyed. In addition, the proportion of household members directly interviewed also increases
across rounds.
There is signi…cant geographic and size variation amongst the enterprises.13 Though the largest
…rm representations are from Java, the economic and population center of the country, the bias is
not overwhelming and a signi…cant proportion of …rms are observed from all of the main survey
provinces. This is true even if we focus on …rms with a relatively larger capital stock, above $1000
US (converted from Indonesian rupiah at the going exchange rate in a given survey year). It is
notable that the slightly larger proportion of …rms seems to be in rural areas. This …ts with evidence
in Liedholm & Mead (1999) and may be due to the fact that smaller …rms are more likely to service
demand in more remote areas. Also, we see that the sample contains a signi…cant number of …rms
exceeding the sizes observed in the vast majority of studies on micro and small enterprises from
developing countries, while …rm-level surveys looking at such …rms generally have little information
on the primary entrepreneur. Given that conversion to US purchasing power parity implies a multiple
of about 12, there are hundreds of enterprises with more than $25,000 US PPP equivalent in capital,
and dozens with 10, 15 or more workers.
Table 3 presents a summary of a number of community-level measures of market churning. In this
paper we focus on the variable ’Growth rate of employment’, which is most analagous to a Bartik
instrument.
12
Both adults and children (de…ned as those under age 15 at the time of the survey) are surveyed, though the
childrens’module is less extensive.
13
The distribution of enterprises is less even if we stratify by industry–the largest proportions of enterprises by far
are in the sectors of restaurant/food, and sales:non-food, at around 30% each. The next two largest sectors are food
processing, and services:transport.
12
3.1.1
Preliminary Evidence
In Table 2 we present summary statistics on the smaller population of individuals who enter selfemployment during the …nancial crisis. While it is a selection from the full population that enters due
to shocks, it exempli…es the type of individuals who tend to switch. There are 684 such individuals
who are eligible for the study due to entry during 1998, and 1355 eligible due to entry in 1999. We
put greater stock in the 1999 cohort as this is the cohort that was most clearly a¤ected by the crisis,
since the brunt of the crisis hit Indonesia during 1998, so resulting occupational changes would be
seen as of 1999. We see that both cohorts are highly likely to be married, and more likely to be
male. The main cohort of interest, the 1999 entrants, are relatively young,14 and signi…cantly more
educated, which is consistent with a relatively high-ability cohort being involuntarily pushed into
self-employment due to the crisis, the major shock in our observed time period.
4
The Return to Experience
The main results appear in Tables 4-7. The OLS results appear in Table 4. As noted above this
speci…cation su¤ers from potential endogeneity due to various sources of unobserved heterogeneity
and selection bias, and this might particularly lead to underestimates of the return to experience since
able individuals might endogenously start up a business only when ready to succeed, while less able
individuals might get stuck in low-skill enterprise activity. For current experience, the estimates vary
from a return to experience of 1.1-2% per year, while for total experience the e¤ects are sometimes
insigni…cant; signi…cant estimates range from 0.5% to slightly negative.
In Tables 5-7 we move to the instrumental variables estimation. In the …rst-stage estimation (not
reported here) we see that the instrument set generally strongly explains variation in the endogenous
variable, years of experience. This is con…rmed by a number of tests of weak identi…cation and
the …rst-stage F-test. All of the demographic controls are highly signi…cant. Age has the expected
concave functional form in virtually all speci…cations, and the other demographic controls generally
…t with expected patterns.
In Table 5 we present results from experience measured in terms of total experience (i.e., the total
number of years the individual has run a business, which could include multiple businesses). We see
that e¤ects are generally insigi…cant. Where signi…cant we see an average return to experience of
2.9% per year. In Table 6 we present results from experience measured in terms of experience in
one’s current business. We might expect these estimates to be larger if we think that …rm-speci…c
managerial capital matter, and indeed that is what the results suggest. We see that the IV estimates
14
This could be a re‡ection of the role of seniority, rather than ability, in worker separations during the crisis.
13
are up to twice as large as the OLS estimates, with an estimated return to experience of 2.7% per year
in the full sample. While the e¤ect is not signi…cant using instruments constructed by year (arguably
the less appealing instrument set), we see a larger return to experience for younger individuals, of
8.4% per year using instruments constructed by lag. This larger estimate is also consistent with the
idea that younger individuals would have larger marginal returns to managerial capital.
The importance of current experience relative to total lifetime experience is further emphasized in
Table 7, where we run the same speci…cations as in Table 5 and 6, but with the experience measures
together (jointly instrumented). There we see that current experience is signi…cant even in the
presence of a variable for one’s total experience.
4.1
Placebo Test
While the exogeneity tests provide supportive evidence, we might be concerned about weakness in
those tests. Another check on the identi…cation strategy is a placebo test for the critical identi…cation
assumption that the formal sector shocks only a¤ect selection between the formal and informal
sectors, and not returns in the informal sector. In Table 8 we present the test described earlier,
running contemporaneous returns to enterprise activity on the contemporaneous value of the shock
variable. Recall that in the main IV speci…cation we only use lags of shocks as instruments, so
this exercise is unique. Con…rmation for the identi…cation strategy would come in …nding a highly
insigni…cant estimate here, and indeed that is what we see: the statistical signi…cance on the main
relationship has p-values of 0.9 and above.
5
The Return to Experience: Additional Speci…cations and
Robustness
In Figure 1 we non-parametrically plot experience-earnings (net pro…t) pro…les across these three
qualitative categories, using a Lowess tri-cube smoother. There we see that while all three groups
enjoy an increase in earnings on average, the rate of increase is substantially higher for those running
the enterprise we would expect to be most complex: …rms with hired, wage workers. This bifurcation
in returns is suggestive of the select group of individuals running more complex enterprises "pulling
away" from the much larger group of individuals running enterprises in the other two categories. We
would expect that signi…cantly greater returns would enable signi…cantly greater capital accumulation. This motivates the use of a richer estimation strategy, in which we follow Thomas & Strauss
(1997) and attempt to correct for endogenous selection between running the more subsistence-type
14
enterprise (no employees, or only family/unpaid employees) and those which have employees.
To implement the method, we …rst run an instrumental-variables multinomial logistic model for
selection between these three enterprise types, and then include the …tted probability of occupational
selection for each individual ("hazard rate") as a control in a second-stage instrumental variables
strategy that parallels the models we have used earlier. Because of problems with project regressors and other issues with standard errors we bootstrap standard errors in both stages with 400
replications each time.
The results are presented in Table 9. There we see that while the correction is not statistically
signi…cant in any speci…cation, this correction does lead to intuitive results, with somewhat higher
returns to experience measured in each speci…cation, including a higher likelihood of measurable
e¤ects even from total experience.
We run a number of other robustness checks (not reported), including di¤erent constructions of the
dependent variable, and di¤erent subsets of years. We …nd that the results are generally consistent.
5.1
What is Being Accumulated?
While much of the preceding analysis seems to paint a convincing picture that learning-by-doing
through direct experience matters for entrepreneurs, it is still not quite clear what this human
capital might be, or if it is entangled with other valuable forms of capital that one might accumulate
over time, such as reputation and business network connections.
As an initial, simple way to try to get at this issue, we study the industry distribution of individuals
who enter during the crisis and start their …rst business during that period. We would expect that
if they are primarily high-skill individuals pursuing relatively human-capital intensive enterprise
activities, then they would sort into industries with this kind of character. The IFLS provides a fairly
crude, 10-industry classi…cation, including Agriculture, forestry, …shing and hunting, Manufacturing,
Construction, Wholesale, retail, restaurants and hotels, Transportation, storage and communications,
and Social services. We propose that Wholesale, retail, restaurants and hotels, and Transportation
(presuming it is basic transport like a bicycle carrier) are relatively low-skill occupations that can be
quickly learned with relatively less need for specialized human capital. In looking at the evolution of
industry shares over time, comparing the new entrants to the larger population of entrepreneurs, we
see that the new entrants are less likely to enter the posited low-skill segments like Wholesale, retail,
restaurants and hotels, and Transportation, and more likely to enter apparent higher-skill sectors like
Social services. While there is still potentially signi…cant variation within these industry categories,
this at least provides some support for the idea that the entrants are truly of relatively higher-ability,
and that they accumulate signi…cant human capital.
15
As an additional test, we look at the propensity of individuals to run what is plausibly the most
skill-intensive enterprise category: those which hire outside workers. These enterprise are generally
more developed on a range of measures –from formality to number of employees to capital stock. If
the story that relatively high-ability individuals who would not otherwise go into enterprise activity
do so during the crisis period holds, then we would expect the crisis-period entrants to be more likely
to be running relatively more complex enterprises a few years later. And indeed, this is what we see:
the crisis-period entrants are roughly twice as likely to be running enterprises with outside workers
9 years after the crisis, compared to those from other entry cohorts.
6
Concluding Discussion and Implications
In this paper we develop and test a microeconomic theory of managerial capital (MC) accumulation.
The key channel for acquiring MC is posited to be through learning-by-doing. The theory is tested
through exploiting a natural experiment that induces exogenous variation in the years of experience
accumulated by small business owners in Indonesia. This provides a source of exogenous selection
into enterprise activity, focusing on e¤ects for relatively high-ability individuals, as the shock is
focused on the formal sector. We estimate an average return to current enterprise-speci…c experience
of 2.7% per year, which can be more than doubled for younger individuals. These results suggest the
importance of modeling entrepreneurial dynamics in a way that incorporates the role of endogenous
human capital accumulation.
Throughout the paper the results on the returns to experience have been interpreted as a kind
of "managerial capital", which is an interpretation that …ts with other literature. Yet at the same
time, it is not clear exactly what kind of human capital this is, or even if it is human capital
at all. For example, the main return to experience running a business could be in building up a
reputation with customers and suppliers, or in operational improvements that come about from a
more experienced workforce. While we argue that the apparent learning dynamics are relatively less
consistent with latter this explanation, it cannot be ruled out entirely. It quickly becomes apparent
that disentangling the entrepreneurial and managerial skill of the …rm’s management team from
other dynamic, intangible, assets of the …rm is not trivial. However, to be fair, this is a problem
that plagues parallel literatures on the returns to experience, such as the literatures on the returns
to education and wage work experience.
More broadly, this work raises a number of questions about what kind of human capital is accumulated and how. This is a potentially exciting topic for future research, looking in more detail
at what and how entrepreneurs learn. Learning more about this would require even richer data on
entrepreneurial behavior, which isolates in more detail the kinds of interactions and experiences the
16
entrepreneur has, and the ways in which ownership and managerial behaviors are altered in response
to these experiences. Further questions might include: What are the most important high-level entrepreneurial skills? Are they complementary to each other, or are certain skills critical at certain
stages? How can such skills be e¤ectively transmitted?
Given the caution that must be attached to the interpretation of the results, it is very di¢ cult
to make concrete policy prescriptions. Hence we will brie‡y discuss some tentative ideas for further
investigation. First, the existing approaches to the identi…cation of high-potential entrepreneurs in
the literature are largely static in nature, based on the analysis of a …xed set of characteristics.
The implicit assumption from this approach is that the optimal …nancing of the entrepreneur is
…xed from the time they enter the market. Of course asymmetric information about underlying
characteristics will prevent this determination from occurring, yet even if this screening problem
could be overcome a static approach would still be sub-optimal. Instead of trying to customize
services to an entrepreneurial "type" that is assumed to be unchanging, policymakers, banks, etc.,
should consider the process of entrepreneurial development of the individual entrepreneur, and how
…nancing and other services should evolve over time as the entrepreneur’s own abilities evolve.
Secondly, if further research building on this work were able to show more robustly that learningby-doing is indeed a crucial channel in entrepreneurial development, then it might raise the need for
more high-ability individuals to get the opportunity to gain experience running an enterprise. In most
countries the primary institution for the formation of skills for the wage sector is formal education,
which can last twelve or more years. While some writers, notably Schultz, have suggested that
education might be an important venue for the formation of entrepreneurial skill, such a hypothesis
is not well supported by the evidence in this paper. Instead, the results suggest that entrepreneurial
skills are more speci…c and require more focused and sustained exposure to enterprise activity itself.15
Hence this suggests the potential for specialized institutions for the transfer of MC. In most developing
countries, the existing institution seems to be the family unit, at least those households in which
the parents have a signi…cant stock of MC that can be transferred to their children. More formal
arrangements, such as internships and …xed-term job placements in entrepreneurial …rms, might be
a way to give potential entrepreneurs much more exposure to enterprise activity.
There have attempts at various forms of entrepreneurial training, including recent tests in the
economics literature based on RCT designs, but based on the results in this paper it is not so
surprising that the results from short-term training have been mixed at best. While many of the
existing programs are focused on transferring low-level entrepreneurial skills (keeping records, basics
of managing …nances, etc.), it seems that high-level entrepreneurial skills (sales, marketing, risk
15
This is not to suggest that education is not useful in general, particularly for pushing up the overall level of
human capital in the population. However, the evidence herein, based on within-population variation in education
and entrepreneurial human capital, suggests that MC is a more important relative factor in enterprise outcomes.
17
judgment, product development, etc.) may be signi…cantly more important, particularly for growthoriented …rms. It may be that a more intensive, sustained mix of direct experience and perhaps
mentorship from more experienced and successful entrepreneurs is needed to enable the emergence
of higher-potential entrepreneurs and the transfer of high-level entrepreneurial skills.
18
References
Angrist, Joshua. 1990. Lifetime earnings and the vietnam era draft lottery: Evidence from social
security administrative records. The american economic review, 80(3), 313–336.
Astebro, Thomas. 2012. The returns to entrepreneurship. Oxford University Press. Pages 45–108.
Banerjee, Abhijit, Duflo, Esther, Glennerster, Rachel, & Kinnan, Cynthia. 2009.
The miracle of micro…nance? evidence from a randomized evaluation. MIT working paper.
Bartik, Timothy. 1991. Who bene…ts from state and local economic development policies? W.E.
Upjohn Institute for Employment Research: Kalamazoo, MI.
Behrman, Jere, & Rosenzweig, Mark. 1999. ’ability’biases in schooling returns and twins: A
test and new estimates. Economics of education review, 18(2), 159–167.
Bloom, Nick, Eifert, Benn, McKenzie, David, Mahajan, Aprajit, & Roberts, John.
2013. Does management matter: Evidence from india. The quarterly journal of economics, 128(1).
Bruhn, Miriam, Karlan, Dean, & Schoar, Antoinette. 2010. What capital is missing in
developing countries? American economic review papers and proceedings, 2, 629–633.
Bruhn, Miriam, Karlan, Dean, & Schoar, Antoinette. 2012. The impact of consulting
services on small and medium enterprises: Evidence from a randomized trial in mexico. Working
paper.
Card, David. 1999. The causal e¤ect of education on earnings. Vol. 3A. Elsevier.
Carter, Michael, & Olinto, Pedro. 2003. Getting institutions right for whom? credit constraints and the impact of property rights on the quantity and composition of investment. American
journal of agricultural economics, 85(1), 173–186.
de Mel, Suresh, McKenzie, David, & Woodruff, Christopher. 2008. Returns to capital:
Results from a randomized experiment. Quarterly journal of economics, 123(4), 1329–1372.
Delmar, Frederic, & Shane, Scott. 2006. Does experience matter? the e¤ect of founding
team experience on the survival and sales of newly founded ventures. Strategic organization, 4(3),
¼
215âAŞ247.
Eesley, Charles, & Roberts, Edward. 2012. Are you experienced or are you talented? when
does innate talent versus experience explain entrepreneurial performance? Strategic entrepreneur¼
ship journal, 6, 207âAŞ219.
19
Frankenberg, Elizabeth, & Karoly, Lynn. 1995 (November). The 1993 indonesia family life
survey: Overview and …eld report. Tech. rept. RAND Corporation, Santa Monica, CA.
Frankenberg, Elizabeth, & Thomas, Duncan. 2000 (March). The indonesia family life survey
(IFLS): Study design and results from waves 1 and. Tech. rept. RAND Corporation, Santa Monica,
CA. DRU-2238/1-NIA/NICHD.
Gompers, Paul, Kovner, Anna, Lerner, Josh, & Scharfstein, David. 2010. Performance
persistence in entrepreneurship. Journal of …nancial economics, 96, 18–32.
Karlan, Dean, & Zinman, Jonathan. 2010 (January). Expanding microenterprise credit access:
Using randomized supply decisions to estimate impacts in manila. Working paper.
Kawaguchi, Daiji. 2003. Human capital accumulation of salaried and self-employed workers. Labour
economics, 10, 55–71.
Liedholm, Carl, & Mead, Donald. 1999. Small enterprises and economic development: The
dynamics of micro and small enterprises. New York, NY: Routledge.
Roy, A.D. 1951. Some thoughts on the distribution of earnings. Oxford economic papers (new
series), 3, 135–146.
Schoar, Antoinette. 2010. The divide between subsistence and transformational entrepreneurship.
Vol. 10. Chicago, IL: The University of Chicago Press.
Shu, Pian. 2012. The long-term impact of business cycles on innovation: Evidence from the massachusetts institute of technology. Working paper.
Strauss, John, Beegle, Kathleen, Sikoki, Bondan, Dwiyanto, Agus, Herawati, Yulia,
& Witoelar, Firman. 2004 (March). The third wave of the indonesia family life survey (IFLS).
Tech. rept. RAND Corporation, Santa Monica, CA. DRU-2238/1-NIA/NICHD.
Strauss, John, Witoelar, Firman, Sikoki, Bondan, & Wattie, Ana Marie. 2009 (April).
The fourth wave of the indonesian family life survey (IFLS4): Overview and …eld report. Tech.
rept. WR-675/1-NIA/NICHD.
Stuart, Robert, & Abetti, Pier. 1990. Impact of entrepreneurial and management experience
on early performance. Journal of business venturing, 5, 151–162.
Tergiman, Chloe. 2013. Entrepreneurship does pay. Working paper.
Thomas, Duncan, & Strauss, John. 1997. Health and wages: Evidence on men and women in
urban brazil. Journal of econometrics, 77, 159–185.
20
A
A.1
Appendix A: Figures and Tables
Figures
Note: The Figure records net pro…ts of enterprises that startup in the three employment categories ((i) no
employees, (ii) only family/unpaid workers, (iii) having waged employees) against years of experience of the
individual entrepreneur running the enterprise.
21
A.2
Tables
Table 1: Summary Statistics on IFLS Rounds
Survey Round
IFLS4
Year
2007-08
Individuals
44103 (50580)
Households
13536
Enterprises
6186
Communities
313
IFLS3
2000
38433 (43649)
10435
5452
311
IFLS2
1997
22019 (33081)
7619
2625*
313
IFLS1
1993-94
22019 (33081)
7224
2439*
312
Overall
66784
(unique)
*In IFLS1 and IFLS2 households only report on one, "primary" enterprise.
22
Table 2: Summary statistics on individual entrants
1998 Entrants
Age
Marriage (married=1)
Gender (male=1)
Education (years)
N
684
684
684
684
Mean
33.81
0.87
0.71
5.58
SD
12.78
0.34
0.46
6.04
P25
25
1
0
0
Median
30
1
1
0.5
P75
40
1
1
12
P95
60
1
1
15
P99
73
1
1
19
N
1355
1355
1355
1355
Mean
27.67
0.87
0.68
7.14
SD
10.72
0.33
0.46
6.15
P25
20
1
0
0
Median
25
1
1
9
P75
33
1
1
12
P95
49
1
1
19
P99
59
1
1
19
1999 Entrants
Age
Marriage (married=1)
Gender (male=1)
Education (years)
23
Table 3: Summary statistics on
Year 1998
Variable
Proportion workers switch to self-empl. (lag)
Proportion workers switch to self-empl.
Proportion switch any occupation (lag)
Proportion switch any occupation
Growth rate of employment (lag)
Growth rate of employment
Change in formal employment (lag)
Change in formal employment
Community uempl rate (lag)
Community uempl rate
Year 1999
Variable
Proportion workers switch to self-empl. (lag)
Proportion workers switch to self-empl.
Proportion switch any occupation (lag)
Proportion switch any occupation
Growth rate of employment (lag)
Growth rate of employment
Change in formal employment (lag)
Change in formal employment
Community uempl rate (lag)
Community uempl rate
community-level sources of variation
N Mean SD P25 Median P75 P95 P99
61796
0.01 0.02
0
0 0.02 0.06 0.09
52853
0.01 0.04
0
0 0.02 0.05 0.11
61796
0.02 0.14
0
0
0
0
1
52853
0.06 0.23
0
0
0
1
1
61797
0.01 0.14 -0.04
0 0.04 0.26 0.48
61796 -0.24 0.6 -0.92
-0.13 0.23 0.64 0.97
48712
0.12 0.8 -0.2
0
0 1.5
4
48886 -0.29 1.05
-1
-0.75
0
2
3
61797
0.07 0.06 0.03
0.06 0.1 0.21 0.3
52938
0.04 0.07
0
0.02 0.05 0.12 0.33
N Mean SD P25 Median P75 P95 P99
52853
0.01 0.04
0
0 0.02 0.05 0.11
60671
0.08 0.1 0.01
0.05 0.09 0.29 0.36
52853
0.06 0.23
0
0
0
1
1
60671
0.08 0.28
0
0
0
1
1
61796 -0.24 0.6 -0.92
-0.13 0.23 0.64 0.97
52853
0.1 1.17 0.02
0.07 0.12 0.29
1
48886 -0.29 1.05
-1
-0.75
0
2
3
31873
0.02 0.19
0
0
0 0.33
1
52938
0.04 0.07
0
0.02 0.05 0.12 0.33
60750
0.03 0.06
0
0.01 0.04 0.11 0.25
24
Table 4. OLS Estimates of the Effect of Experience on Returns
(1)
Full sample
Enterprise experience
Total
0.005***
(0.001)
(2)
Born >=
1968
0.002
(0.002)
Enterprise experience
Current
Demographics
Province*Year Effects
Observations
R2
F
YES
YES
26428
0.125
48.629
YES
YES
7211
0.124
14.430
(3)
Full sample
(4)
Born >=
1968
(5)
Full sample
0.001
(0.001)
(6)
Born >=
1968
-0.007***
(0.003)
0.012***
(0.001)
0.014***
(0.002)
0.011***
(0.001)
0.020***
(0.003)
YES
YES
26428
0.128
49.648
YES
YES
7211
0.128
14.809
YES
YES
26428
0.128
49.158
YES
YES
7211
0.129
14.722
Note: Standard errors in parentheses, * p<0.05, ** p<0.01, *** p<0.001
Note: p-values in square brackets
Note: Demographics include: age, age^2, gender, years of education, urban location
Table 5. IV Estimates of the Effect of Experience on Returns for Total Experience
(1)
Full sample
Enterprise experience
Total
0.029***
(0.010)
(2)
Born >=
1968
0.002
(0.032)
Constant
0.385***
(0.143)
YES
YES
26428
0.063
42.214
17.315
0.138
125.422
0.000
-0.092
(0.189)
YES
YES
7211
0.124
14.397
20.793
0.053
32.521
0.002
Demographics
Province*Year Effects
Observations
R2
F
Hansen's J
Hansen's J (p-value)
Overident stat
Overident stat (p-value)
(3)
Full sample
0.018
(0.012)
(4)
Born >=
1968
0.048
(0.045)
0.239
(0.172)
YES
YES
26428
0.107
45.731
14.756
0.255
69.866
0.000
-0.163
(0.200)
YES
YES
7211
0.060
13.346
8.858
0.715
20.379
0.086
Note: Standard errors in parentheses, * p<0.05, ** p<0.01, *** p<0.001
Note: Demographics include: age, age^2, gender, years of education, urban location
Note: columns (1)-(2) based on instruments-by-year, columns (3)-(4) on instruments-by-lag.
Table 6. IV Estimates of the Effect of Experience on Returns for Current Experience
(1)
Full sample
Enterprise experience
Current
0.027**
(0.011)
(2)
Born >=
1968
0.037
(0.028)
Constant
0.139*
(0.081)
YES
YES
26428
0.120
47.541
20.407
0.060
157.927
0.000
-0.245
(0.217)
YES
YES
7211
0.117
14.228
20.721
0.055
62.587
0.000
Demographics
Province*Year Effects
Observations
R2
F
Hansen's J
Hansen's J (p-value)
Overident stat
Overident stat (p-value)
(3)
Full sample
0.027**
(0.013)
(4)
Born >=
1968
0.084**
(0.038)
0.140
(0.088)
YES
YES
26428
0.120
47.532
12.623
0.397
114.932
0.000
-0.445*
(0.247)
YES
YES
7211
0.024
12.653
5.506
0.939
42.395
0.000
Note: Standard errors in parentheses, * p<0.05, ** p<0.01, *** p<0.001
Note: Demographics include: age, age^2, gender, years of education, urban location
Note: columns (1)-(2) based on instruments-by-year, columns (3)-(4) on instruments-by-lag.
Table 7. IV Estimates of the Effect of Experience on Returns for Both Measures
(1)
Full sample
Enterprise experience
Total
0.032**
(0.016)
(2)
Born >=
1968
-0.050
(0.041)
Enterprise experience
Current
-0.004
(0.019)
0.070*
(0.037)
0.079*
(0.043)
0.127**
(0.061)
Constant
0.399**
(0.159)
YES
YES
26428
0.053
41.062
17.017
0.107
44.471
0.000
-0.304
(0.222)
YES
YES
7211
0.085
13.532
19.002
0.061
16.886
0.154
-0.275
(0.343)
YES
YES
26428
0.081
35.267
8.537
0.665
10.327
0.587
-0.516*
(0.269)
YES
YES
7211
0.028
11.992
4.154
0.965
12.995
0.369
Demographics
Province*Year Effects
Observations
R2
F
Hansen's J
Hansen's J (p-value)
Overident stat
Overident stat (p-value)
(3)
Full sample
-0.051
(0.041)
(4)
Born >=
1968
-0.069
(0.074)
Note: Standard errors in parentheses, * p<0.05, ** p<0.01, *** p<0.001
Note: Demographics include: age, age^2, gender, years of education, urban location
Note: columns (1)-(2) based on ** p<0.05
*** p<0.01"
Table 8. Placebo Test: Effects of Current Shock of Current Returns
Growth employment instrument
Demographics
Province*Year Effects
Observations
R2
F
(1)
0.002
(0.016)
[0.901]
(2)
-0.000
(0.016)
[0.979]
YES
NO
30320
0.011
48.613
YES
YES
29421
0.029
9.233
Note: Standard errors in parentheses, * p<0.05, ** p<0.01, *** p<0.001
Note: p-values in square brackets
Note: Demographics include: age, age^2, gender, years of education, urban location
-0.000
(0.000)
-0.548**
(0.265)
YES
YES
6807
0.019
0.059*
(0.033)
(4)
Born >=
1968
0.000
(0.000)
0.483**
(0.233)
YES
YES
25289
-0.101
0.046**
(0.019)
(5)
Full sample
Note: Standard errors in parentheses, * p<0.05, ** p<0.01, *** p<0.001
Note: Estimates based on 400 bootstrap replications in each stage
Note: Demographics include: age, age^2, gender, years of education, urban location
Note: columns (1)-(4) based on instruments-by-year, columns (5)-(8) on instruments-by-lag.
Demographics
Year Effects
Observations
R2
Constant
-0.000
(0.000)
-0.248
(0.216)
YES
YES
6807
0.063
-0.000
(0.000)
0.086
(0.079)
YES
YES
25289
0.047
0.000
(0.000)
0.494***
(0.150)
YES
YES
25289
-0.109
Hazard rate
(3)
Full sample
0.044***
(0.015)
0.047***
(0.011)
(2)
Born >=
1968
0.017
(0.038)
Enterprise experience
Current
Enterprise experience
Total
(1)
Full sample
-0.000
(0.000)
-0.290
(0.231)
YES
YES
6807
0.033
(6)
Born >=
1968
0.035
(0.050)
-0.000
(0.000)
0.143
(0.094)
YES
YES
25289
0.006
0.057***
(0.020)
(7)
Full sample
Table 9. Occupational Selection Corrected IV Estimates of the Effect of Experience on Returns
-0.000
(0.000)
-0.904***
(0.323)
YES
YES
6807
-0.206
0.121***
(0.043)
(8)
Born >=
1968