abstract - druid · work experience an adverse treatment in the labour market; in their rst quarter...
TRANSCRIPT
-
Paper to be presented at the DRUID Academy Conference 2018at University of Southern Denmark, Odense, Denmark
January 17-19, 2018
Scarred by Failure?
Jeroen MahieuKU Leuven
Managerial Economics, Strategy and [email protected]
Francesca MelilloKU Leuven
Managerial Economics, Strategy and [email protected]
Toke ReichsteinCopenhagen Business School
Strategic Management and [email protected]
AbstractAre failed entrepreneurs penalized in the labourmarket? The answer is yes under certainconditions. Using a novel dataset of matchedentrepreneurs and employees in Belgium, weshow that: (i) on average, entrepreneurs returningto the labour market after a business failure arepenalized, but this effect is most pronounced forthose coming from the higher echelons of thewage distribution, and absent for those whoearned relatively little before enteringself-employment. (ii) the penalty is only presentfor entrepreneurs who fail quickly: entrepreneursreturning to the labour market after 5 or moreyears do not get penalized. (iii) entrepreneurs whofail fast and move to a new employer limit thewage penalty by changing industry compared tothe industry they were venturing in. These resultsare consistent with theories of adverse selection inthe labour market.
-
Scarred by Failure?
Jeroen Mahieu∗
KU Leuven
Francesca Melillo
KU Leuven
Toke Reichstein
CBS
January 3, 2018
Abstract
Are failed entrepreneurs penalized in the labour market? The answer is yes under
certain conditions. Using a novel dataset of matched entrepreneurs and employees in
Belgium, we show that: (i) on average, entrepreneurs returning to the labour market
after a business failure are penalized, but this effect is most pronounced for those com-
ing from the higher echelons of the wage distribution, and absent for those who earned
relatively little before entering self-employment. (ii) the penalty is only present for
entrepreneurs who fail quickly: entrepreneurs returning to the labour market after 5
or more years do not get penalized. (iii) entrepreneurs who fail fast and move to a new
employer limit the wage penalty by changing industry compared to the industry they
were venturing in. These results are consistent with theories of adverse selection in the
labour market.
Keywords: entrepreneurship, returns, matching
JEL Codes: J23, J24, L26
∗Department of Managerial Economics, Strategy, Entrepreneurship and Innovation, KU Leuven, Naam-sestraat 69, 3000 Leuven. E-mail: [email protected]
mailto:[email protected]
-
1 Introduction
A large body of literature has studied the pecuniary and non-pecuniary returns from en-
trepreneurship (e.g., Evans and Leighton, 1989; Hamilton, 2000; Georgellis et al., 2007;
Levine and Rubinstein, 2017). However, this literature provides an incomplete view of the
full set of returns since a substantial share of entrepreneurial spells is only of transitory
nature: half of the self-employment spells end within the first 5 years after their inception
(Daly, 2015; Dillon and Stanton, 2017; Kaiser and Malchow-Møller, 2011; Manso, 2016).
Surprisingly, until recently, little attention has been paid to examine the returns from
self-employment experience outside the entrepreneurial context (Burton et al., 2016). De-
spite this growing interest, current studies have provided mixed evidence on the question how
past entrepreneurial failure affects a person’s outcomes in the labour market upon return to
wage work: a number of studies find that former entrepreneurs tend to incur a wage cost
upon return to wage work (Bruce and Schuetze, 2004; Kaiser and Malchow-Møller, 2011;
Baptista et al., 2012; Failla et al., 2017), and receive fewer responses on job applications
(Koellinger et al., 2015). Others find no significant wage penalty (Hyytinen and Rouvinen,
2008; Manso, 2016), and suggest that former entrepreneurs move to a position higher in the
firm hierarchy when they are hired (Baptista et al., 2012).
In this paper, we confront these contrasting results by applying a contingency approach:
we document a number of conditions that moderate the relationship between past en-
trepreneurial failure and wages of entrepreneurs moving back to paid employment. Using
longitudinal linked employer-employee data from the Belgian Datawarehouse Labour Market
& Social Protection covering all quarters between 2000 and 2015, we construct a matched
sample of employees that transition into self-employment and workers who never leave paid
employment during the sample period. The matched group of workers who remain in wage
employment function as a counterfactual for what would have happened to the entrepreneurs
1
-
if they did not choose to become self-employed. From this matched sample, we select the
matched pairs of which the entrepreneur moves from a full-time job to self-employment and
back, and we compare the wages between the two groups right before and after the self-
employment spell.
We report 3 key findings: (1) On average, entrepreneurs who fail and move back to wage
work experience an adverse treatment in the labour market; in their first quarter in paid
employment after leaving self-employment, former entrepreneurs earn on average 6.6% less
than their counterparts without self-employment experience. This difference is non-existent
in the periods before entry into self-employment. However, the wage penalty is more se-
vere for entrepreneurs who were located in the upper end of the wage distribution before
entering self-employment, while there is no penalty for those coming from the lower tail. (2)
Only entrepreneurs who fail relatively quickly are penalized: entrepreneurs returning to the
labour market after 5 or more years in self-employment do not earn significantly less than
their matched counterparts. (3) Entrepreneurs who were located in the upper end of the
wage distribution before entering self-employment and who change industry compared to the
industry they were venturing in, receive a lower penalty than those who stay in the same
industry during and after self-employment.
These results are inconsistent with human capital based explanations of an adverse treat-
ment of former entrepreneurs in the labour market (cf. Williams, 2000), but are consistent
with the predictions of models of asymmetric information between employers and job seekers
(Greenwald, 1986; Gibbons and Katz, 1991) where entrepreneurial failure holds signalling
value: failing quickly as an entrepreneur is a signal of low (entrepreneurial) ability. Since
entrepreneurial ability is positively related to certain skills like managing a business and
workers, this explains the higher penalty for entrepreneurs coming from the right tail of the
wage distribution, because they are more likely to hold managerial positions within a firm.
2
-
Furthermore, part of the signal appears to be industry-related: failing as an entrepreneur in
one industry does not predict low ability in another industry.
This study contributes to the current debate around the returns from entrepreneurship by
showing that the labour market consequences from attempting self-employment are hetero-
geneous, and depend on a person’s pre-entrepreneurship position in the labour market and
how well she performs as an entrepreneur. Therefore, our findings caution against a simple
estimation of the average returns of entrepreneurial experience in the labour market, since
testing the entrepreneurial grounds appears to hold different implications for high-ability
individuals compared to less able ones. While high-ability workers are strongly punished
if they experience an entrepreneurial failure, low-ability individuals are not penalized for
testing the entrepreneurial waters. For them, entrepreneurship might hold the option value
of experimenting with new ideas (cf. Manso, 2016).
The remainder of this paper proceeds as follows: In section 2, we present an overview
of the empirical context of the Belgian labour market. We describe labour market mobility
and entrepreneurial activity in Belgium. Section 3 contains an overview of the data from the
Datawarehouse Labour Market & Social Protection and a discussion of the sampling and
matching procedures. In section 4, we describe the analytical framework and present the
estimates of the impact of entrepreneurial failure on wages upon return to paid employment.
In addition, section 5 offers supplementary analyses and robustness checks. Section 6 pro-
vides a discussion of the alternate mechanisms that might explain these results. Concluding
remarks are in section 7.
3
-
2 Empirical Context
Unlike the US, the Belgian labor market is perceived to be highly rigid. According to the
OECD, Belgian ranks 3rd in terms of protection of permanent workers against individual
and collective dismissals, just behind Venezuela and China. Zimmer (2012) observes that
Belgium has a high mismatch between labour supply and demand, driven by a shortfall in
the relative share of highly-skilled job-seekers, and, conversely, a relatively high share of
low-skilled labour supply for which demand is rather weak.
Furthermore, the Belgian business landscape is characterized by low rates of entrepreneurial
entry and exit. In 2015, new businesses accounted for only 6.40% of the total share of all
businesses, the lowest percentage in Europe. Belgium does fairly well in terms of regulatory
framework, market conditions, access to finance, and entrepreneurial capabilities (De Mulder
and Godefroid, 2016), but has poor contract enforcement(Calvino et al., 2016), high paid-in
minimum capital requirements (Dreher et al., 2013), and minimal entrepreneurial culture
(De Mulder and Godefroid, 2016). One important factor that may hamper entrepreneurship
in Belgium is the relatively high administrative burdens it puts on starting businesses: ac-
cording to the OECD’s PMR indicators Database, Belgium ranks among the 10 countries
with the highest administrative burdens on startups.
At the same time, 71% of the Belgian businesses survive their first 3 years, and 63% sur-
vive their first 5 years1. While these numbers might suggest an above average performance
of Belgian entrepreneurs over time compared to other countries, it can also indicate low
thresholds of performance (Gimeno et al., 1997), and low levels of experimentation (Landier,
2005; Manso, 2016), especially if it is combined with little growth. Findings of (Geurts and
Van Biesebroeck, 2016) are suggestive of the latter explanation. Using Belgian data, they
show that de novo entrepreneurs contribute little to overall job creation, much less than
1source: Eurostat
4
-
commonly believed based on official statistics.
The characteristics described above make the Belgian situation comparable to various
(Western) European countries, such as France, Germany, Finland, and Sweden. Previous
studies examining the returns from entrepreneurial experience in paid employment have
usually relied on data from relatively flexible labour markets, and dynamic business environ-
ments, such as the US and Denmark (Kaiser and Malchow-Møller, 2011; Daly, 2015; Manso,
2016; Bruce and Schuetze, 2004; Williams, 2000). These contexts are generally characterized
by high tolerance for entrepreneurial experimentation and failure (Landier, 2005; Manso,
2016).
3 Data and Methodology
3.1 Data source and construction of matched sample
We analysed data on the Belgian labor market from the Data Warehouse Labour Market and
Social Protection (DWH LM&SP), maintained by the Crossroads Bank for Social Security
(CBSS). The DWH LM&SP is a linked, administrative dataset that combines quarterly data
from nearly 20 Belgian social security institutions, and covers the full population of legal
residents in Belgium. The data spans all quarters between the 1st quarter of 2000 and the
4th quarter of 2015, and contains detailed information about individuals’ demographics, as
well as employment status, and income.
The initial sampling frame consists of the population of full-time employees working in
one job in wage employment in the 1st quarter of 2004, who either transitioned into self-
employment at some point between 2004 and 2015 or who remained in wage employment
for that period. A person is assigned to the entrepreneur group if she is classified as self-
employed for at least 4 consecutive quarters between the 2nd quarter of 2004 and the 4th
5
-
quarter of 2015. The CBSS classifies a person as self-employed if at the last day of a given
quarter her main profession is self-employment and she is not registered as an employee.
Relying on this classification ensures that individuals who were affiliated with more than
one firm in the form of an entrepreneurial venture and wage work were excluded, since these
so-called hybrid transitions involve different dynamics (Folta et al., 2010).
To reduce the likelihood that the results might be attributable to confounding factors,
I impose several additional restrictions. First, we restrict the sample to individuals aged
22-49 in the 1st quarter of 2004 to eliminate biases due to censoring. Second, we exclude
those with self-employment experience between 2000 and 2003 because the dynamics of serial
entrepreneurship are likely to be distinct from first-time entrepreneurship. Third, we drop
individuals working in agriculture, fishing, mining, and quarrying in the 1st quarter of 2004.
Fourth, to avoid missing observations for individuals moving out of the country at a certain
point in time, we restrict the sample to inhabitants of Belgium between 2004 and 2015.
Identifying the potential wage cost of a spell of entrepreneurship poses an important in-
ferential challenge, since the transition into self-employment is endogenous. Various studies
have shown that individuals self-select into self-employment based on certain characteristics
and incentives, such as a preference for non-pecuniary benefits (Hamilton, 2000), a varied
skill set (Lazear, 2005), or a taste for variety (Åstebro and Thompson, 2011). Therefore,
naive estimations of the outcomes would likely result in high model dependence and heavy
reliance on extrapolation due to insufficient overlap in the covariate distribution (Ho et al.,
2007; Stuart, 2010).
To address potential biases due to self-selection on observables, we construct a matched
sample of employees from the control group who are similar to the entrepreneurs on a range
of observable characteristics. The fundamental counterfactual is that the matched employees
6
-
represent the outcomes for the entrepreneurs had they not chosen to become self-employed.
We apply a combination of exact matching and propensity score matching (Rosenbaum
and Rubin, 1983), in particular 1:1 nearest neighbour matching without replacement2. This
methodology has been used to address issues of self-selection in previous studies measur-
ing entrepreneurial outcomes (Campbell, 2013; Daly, 2015; Failla et al., 2017; Kaiser and
Malchow-Møller, 2011; Manso, 2016) Variables included in the estimation of the propensity
score were measured in the 1st quarter of 2004 before the treatment of interest, i.e. the
transition into self-employment. In the matching procedure a wide range of variables known
to be related to treatment and outcome were included. An individual is considered a good
match to an individual who chooses to transition into self-employment if in the 1st quarter
of 2004:
1. They are in the same age category, have the same gender, live in the same region, work
in the same industry, and have the same daily wage.
2. Their estimated propensity scores lie within a caliper of 0.002 from each other3.
3.1.1 Matching Variables
In the matching procedure, we include a battery of variables previous studies have shown to
be related to the transition into entrepreneurship and wages. One advantage of including
many different variables rather than a small set of ‘predictors of convenience’ is that it min-
imizes potential bias due to the omission of an important confounder (Stuart, 2010).
As a starting point, we include several demographic variables: most people who start up
a business do so when they are well into their thirties or older, after they have acquired some
experience in paid employment (Evans and Leighton, 1989). Second, entrepreneurs are more
2Other matching estimators, such as Coarsened Exact Matching produced similar results.3The use of the specified caliper distance also implies common support, i.e. there is complete overlap in
the distributions of the propensity scores between the treatment and control groups in the matched sample.
7
-
likely to be male (Manso, 2016). Therefore, we consider individuals to be a good match if
they are in the same age category, and of the same gender. Folta et al. (2010) suggests that
civil status affects the entrepreneurial entry choice. We include a civil status variable which
distinguishes between individuals with or without partner, and for how many children they
are responsible. Variables are included capturing whether a person’s partner has a job, and
if so, his average daily wage. Furthermore, to control for potential structural differences in
labour market dynamics across regions, we match individuals on the region they were living
in the 1st quarter of 2004.
Entrepreneurs are typically depicted as being jacks-of-all-trades (Lazear, 2005), or as
having a preference for (job) variety (Åstebro and Thompson, 2011). We include a variable
measuring the number of jobs a person held between 2000 and 2004 as a proxy for these
traits. We also include a measure of how many quarters a person was unemployed between
2000 and 2004 to capture differences in capabilities.
Smaller firms spawn entrepreneurs at a higher rate, and entrepreneurs coming from small
firms perform differently than those coming from large firms (Elfenbein et al., 2010; Sørensen,
2007). Additionally, wages and wage growth are on average higher in larger firms (Oi and
Idson, 1999). Therefore, we include an employer size variable, measured as the number of
employees at the firm the individual was affiliated with in the 1st quarter of 2004. An indica-
tor for whether a person is working in the public or private sector is included, since the type
of sector a worker is employed in affects the likelihood of becoming an entrepreneur (Özcan
and Reichstein, 2009). To control for potential differences in switching costs emerging from
differences in firm- or industry-specific human capital (Becker, 1964; Neal, 1995), an indi-
vidual’s tenure at her current employer and tenure in her current industry were included. In
order to minimize concerns about cross-industry differences in wage and entrepreneurship
dynamics, we perform an exact match on the employer’s industry.
8
-
Åstebro et al. (2011) and Elfenbein et al. (2010) show that a person’s position in the wage
distribution is an important predictor of entrepreneurial entry. Therefore, two individuals
were matched if they earned the same average daily wage in the 1st quarter of 2004. The
CBSS provides information about an individual’s average daily wage per quarter, which is
defined as:
(Quarterly normal remuneration+forfatary remunaration)Nr. full−time remunerated days
All reported wages are gross wages, and are provided by the CBSS in classes of 10 Euros.
We also include a measure of an individual’s annual wage growth to take into account po-
tential negative wage shocks related to transitions into entrepreneurship (Folta et al., 2010).
Despite the wide range of covariates included in the matching procedure, there is no
data available on a person’s education. Although this is definitely a limitation in the data,
we tried to minimize potential concerns about confounding effects of unobserved ability by
performing close matches on variables that are significantly correlated with ability, such as
wage, wage growth, and time spent in unemployment.
3.1.2 Quality of the Matched Sample
Table 1 provides summary statistics for the variables used in the matching process. Columns
1 to 3 show summary statistics for the full sample prior to matching. The first column dis-
plays the average values for the entrepreneurs (treatment group). The second column gives
the corresponding value for the employees in the control group. The third column displays
the standardized percentage bias4. One advantage of this balance diagnostic is that it is sim-
ilar to an effect size (Rosenbaum and Rubin, 1985), and is preferred over the use of t-tests
4The standardised % bias is the % difference of the sample means in the treated and non-treated (fulland matched) sub-samples as a percentage of the square root of the average of the sample variances in thetreated and non-treated groups (Rosenbaum and Rubin, 1985).
9
-
or p-values to assess balance (Stuart, 2010).
Entrepreneurs are more likely to be male, between 25 and 34 years old in 2004, and to
have a partner with self-employment experience. They have a higher likelihood of working
in the private sector, and in construction, wholesale and retail trade, or real estate and
professional services. They are less likely to be employed in the public administration, de-
fence, or education sector. There are also noticeable differences in terms of employer size.
Entrepreneurs are much more likely to be employed in smaller firms, and are more than 50%
less likely to come from the largest firms (>= 1000 employees). Furthermore, entrepreneurs
have had a more varied employment history, lower tenure at their current employer, and have
spent more quarters in unemployment between 2000 and 2004. They have a slightly higher
daily wage, but are not different from the control group in regard to annual wage growth.
Columns 4 to 6 present summary statistics for the treatment and control group after the
matching procedure. Apart from perfect balance on the variables on which exact matching
was performed (age, region, industry, and daily wage), the matching process significantly
improves balance on the remaining covariates: for none of the variables is there a standardized
% bias greater than 2.3% in the matched sample, well below the standard 10% ‘rule of thumb’
(Rosenbaum and Rubin, 1985). The severe pre-matching imbalances which could influence
the reliability of the estimates have been removed. In total, we retain a matched sample of
64,946 individuals.
3.2 Additional Sample Restrictions
One shortcoming of relying on occupational data is that it does not provide information
about the nature of the self-employment spell. In order to mitigate concerns about necessity
entrepreneurship, we restrict the analysis to matched pairs of whom the entrepreneur transi-
tions into self-employment no later than 4 quarters after leaving paid employment, and who
10
-
re-enters the labor market no later than 4 quarters after leaving self-employment. Therefore,
individuals who enter (exit) self-employment after a full year of unemployment or inactivity
are excluded from the analysis, because those might become self-employed due to limited
opportunities on the labour market.
Furthermore, to avoid issues of different pay schemes between full- and part-time or
temporary jobs, we restrict the sample to individuals who transition into entrepreneurship
after leaving a full-time job, and re-enter the labor market via a full-time job. While this
restriction leads to a significant drop in the amount of observations, it allows me to make es-
timations which are less likely to be biased due to confounding effects of the different nature
of remuneration between these types of jobs. However, this also urges caution in generaliz-
ing the results to workers who enter and/or exit self-employment via part-time or temporary
jobs. After imposing these additional restrictions, I retain a sample of 5,472 entrepreneurs
and employees.
Table 2 displays summary statistics for the entrepreneurs and their matched counterparts
in the quarter of pre-entry, i.e. the last quarter in wage employment before transitioning
into self-employment. It is important to verify whether the two groups are still balanced on
the observed covariates, since the matching was performed in the 1st quarter of 2004 and not
in the pre-entry quarter. The two groups are still very equal in terms of wages: on average,
entrepreneurs and non-entrepreneurs earn almost 128 Euros per day, and the difference is
insignificant. All of the demographic variables are also well balanced, and not significantly
different between the two groups. However, entrepreneurs have a significantly lower tenure:
on average, they were employed in the same firm for just over 4 years, versus an average
tenure of about 5 years in the control group. Similarly, entrepreneurs also have held more
jobs in the quarter of pre-entry. This indicates that between the 1st quarter of 2004 and
the quarter of pre-entry, entrepreneurs experienced a higher rate of job mobility compared
11
-
to employees in the control group.
4 Results
4.1 Descriptive Analysis
The aim of this study is to examine the relationship between entrepreneurial experience and
wage (growth) at the moment of re-entry in the labour market. To measure the wage growth
between the moment right before entering self-employment and the moment of re-entering
wage employment, we calculate the percentage wage growth between the two periods and
divide it by the duration of the in-between self-employment spell, using following formula:
∆wi =ln(wagei1) − ln(wagei0)quarteri1 − quarteri0
where wagei0 is the wage of person i in the last quarter in wage employment before entering
self-employment, and wagei1 is the wage in the first quarter in wage employment after exit-
ing self-employment. Using this formula provides me with a predicted quarterly percentage
growth rate of the average daily wage. For the employees in the control group, we take the
quarters of pre-entry and re-entry of their matched counterparts in the entrepreneurs group
to obtain an estimate of the wage growth. So by construction, the spell lengths between
both groups are equal.
Table 3 shows summary statistics of entrepreneurs and their matched counterparts in the
quarter of re-entry in paid employment. On average, former entrepreneurs earn about 10.5
Euros less per day, and their predicted quarterly wage growth between the periods before and
after self-employment is 0.07% lower. These findings indicate a significant wage penalty for
failed entrepreneurs. Entrepreneurs are also much more likely to change employer: almost
12
-
82% moves to a new employer after self-employment, whereas only 23% of the employees
change employer during the period of the entrepreneurial spell of their matched entrepreneur.
Similarly, about 44% of entrepreneurs change industry compared to the industry they were
employed in before self-employment, versus only 9.5% of the employees group. We divide
the entrepreneurs in three groups: Industry Stayers are those who move to an employer
in the same industry as they were self-employed in, regardless of which industry they were
employed in before entering self-employment. Industry Movers are those who move to an
employer in a different industry after failing in entrepreneurship. From the Industry Movers
group, we extract a third group called Hobos5; these entrepreneurs have started a business
in a different industry than they were employed in before, and move again to another in-
dustry after self-employment. The size of these groups is roughly the same, which indicates
significant cross-industry mobility by the entrepreneurs.
Figure 1 displays the distribution of the durations of the entrepreneurial spells. On
average, an entrepreneur remains around 14 quarters or 3.5 years in self-employment (cf.
table 3), but the distribution of spell lengths is very right-skewed: almost 50% of the spells
last between 1 and 2 years, and almost 30 % last between 3 and 4 years. These findings
confirm previous observations that entrepreneurial spells tend to be short (cf. Manso, 2016;
Dillon and Stanton, 2017). However, in this study we only take into account entrepreneurial
spells which effectively end, and not the full range of entrepreneurial spells. Therefore, the
numbers are not directly comparable.
4.2 Regression Analysis
To investigate the relationship between self-employment experience and wages more com-
pletely, we estimate the following 3 models:
5I borrow the terminology from Åstebro and Thompson (2011), who refer to entrepreneurs as ‘hobos’due to their taste for variety and job change
13
-
wi0 = αi + αSSEi + αXXi0 + αIIi0 + �i0 (1)
wi1 = βi + βSSEi + βWWi0 + βSW (SEi ∗Wi0) + βXXi1 + βIIi1 + υi1 (2)
∆wi = δi + δSSEi + δWWi0 + δSW (SEi ∗Wi0) + δXXi1 + δIIi1 + ξi1 (3)
where wi0 and wi1 are respectively the natural logarithm of the average daily wage in the
quarter of pre-entry, and in the quarter of re-entry. ∆wi is the wage growth using the formula
explained above. αi, βi, δi are individual fixed effects capturing observed and unobserved
characteristics (such as age, gender, or risk preferences). Xi0 and Xi1 are vectors containing
observed time-varying characteristics of individuals and their employers at the quarter of
pre-entry, and re-entry. Specifically, in each model we account for a person’s gender, age,
position in the household, region, occupation, employer sector, industry, size, and the length
of her self-employment spell. SEi is a dummy variable which takes on the value of one if
the individual belongs to the group of entrepreneurs. Wi0 represents the avg. daily wage
at the quarter of pre-entry, in deciles. To allow for a differential effect of self-employment
experience at the various levels of the pre-entry wage, we also include the interaction effect
SEi ∗Wi0.
Equation (1) counts as a robustness check that conditional on a range of observables,
there are no significant differences in terms of wages between the entrepreneur and non-
entrepreneur groups. If there are, issues of positive or negative selection into self-employment
could influence the results of the estimations of the wage and wage growth at the quarter
of re-entry in the labour market. Potential significant differences at the pre-entry quarter
should therefore be taken into account when interpreting the coefficients of wi1 and ∆wi. We
estimate equation (2) and (3) first without, and then with the inclusion of the interaction
effect.
14
-
Table 4 displays the results of OLS regressions of equations 1 - 3. Column 1 reports the
results of the relationship between entering self-employment and a person’s wage in the last
quarter in paid employment before transitioning to entrepreneurship. The results from the
conditional analysis confirm those from the unconditional comparison of means (cf. table
2): conditional on a variety of observed covariates, entrepreneurs do not earn significantly
less or more before transitioning to self-employment. These results indicate that there are
no observable differences in ability between entrepreneurs and their matched counterparts
ex ante self-employment.
Columns 2 and 3 of table 4 show the results for the wage in the quarter of re-entry in
paid employment. We find that conditional on the wage before entering self-employment, on
average, entrepreneurs earn 6.6% less than their counterparts in the control group (column
2). Furthermore, when we include an interaction effect between the entrepreneur dummy
and the pre-entry wage (column 3), the results indicate that the penalty is larger the more
an entrepreneur earned before entering self-employment. Figure 2 displays the predictive
margins of the wage at re-entry for entrepreneurs and (SE) and the matched employees (not
SE). The red line indicates the estimated margins for the entrepreneurs, the blue line for
the matched employees. Entrepreneurs coming from the lower tail of the pre-entry wage
distribution (decile 1 and 2) do not earn a lower wage than similar employees, but the wage
gap appears at the 5th, and is the widest at the 9th and 10th decile.
Columns 4 and 5 of table 4 display the results for the predicted quarterly wage growth
between the quarter pre-entry into self-employment and re-entry into paid employment. The
results confirm the findings that failed entrepreneurs incur a wage cost when returning to
the labour market. On average, their quarterly wage growth is around 0.006% lower per
quarter (column 4). However, it are the entrepreneurs who come from the upper tails of
15
-
the pre-entry wage distribution are punished most. Figure 3 displays the predictive margins
for the entrepreneurs and employees of the estimated regression in column 5 of table 4.
Entrepreneurs coming from upper tails of the pre-entry wage distribution not only have a
significantly lower wage growth than their matched counterparts, their wage growth is also
negative (!). Entrepreneurs in the top pre-entry wage decile have a predicted quarterly wage
loss of 2%.
4.2.1 Spell length
It is possible that the wage penalty of entrepreneurial failure differs for entrepreneurs who
fail quickly compared to those who manage to survive for a number of years. To investigate
this possibility, we estimate the following interacted model of wage growth:
∆wi = θi+θSSEi+θWWi0+θDDi+θSW (SEi∗Wi0)+θSWD(SEi∗Wi0∗Di)+θXXi1+θIIi1+ψi1
(4)
The parameters included in equation (4) are those of equation (3), but we add a three-way
interaction effect between the entrepreneur dummmy SEi, a person’s position in the pre-
entry wage distribution Wi0, and the duration of her entrepreneurial spell Di(or her matched
counterpart’s spell for the individuals in the employees group).
Table 5 displays the marginal effects obtained from the OLS regression of equation
(4). Each cell represents the difference of wage growth between entrepreneurs and non-
entrepreneurs for a certain spell duration (all spells, 1-2, 3-4, 5-6, and 7+ years) at a certain
level of the pre-entry wage (1st, 2nd, 5th, 9th, and 10th decile). The results show that for
entrepreneurs coming from the lowest pre-entry wage deciles (1st and 2nd) they never incur
a penalty from attempting self-employment. However, for those who earned a relatively high
wage before entering self-employment (5th, 9th, and 10th decile), the wage penalty is most
pronounced when they fail in the 1st two years after entry, and gradually declines over time.
16
-
After 5 years of survival in self-employment, there is no wage growth difference between
entrepreneurs and their matched counterparts, at any level of the pre-entry wage.
4.2.2 Industry Switching
Kaiser and Malchow-Møller (2011) suggest that a wage penalty for former entrepreneurs
occurs only when they also switch industries after self-employment, which could indicate a
loss of industry-specific human capital (Neal, 1995). To investigate the possibility that the
wage cost of quick entrepreneurial failure is moderated by a change in industry, we divide
the entrepreneurs in three groups: (1) Industry Stayers who move to an employer in the
same industry as the industry they were self-employed in. (2) Industry Movers who move
to an employer in a different industry. From the Industry Movers, we extract a third group
(3) Hobos who also change industry when they enter self-employment. We then estimate a
model similar to equation (4), but with 3 groups of entrepreneurs (Industry Stayer, Movers,
and Hobos) instead of 1.
Table 6 presents marginal effects of the wage growth regression of the difference between
the control group, and respectively the Industry Stayers, Industry Movers, and Hobos. We
only consider the short spell lenghts (< 5 years) since the analysis in the previous section
revealed that after 5 years in self-employment, entrepreneurs receive no penalty. The results
indicate that for the three groups, the wage growth difference is almost equal across different
levels of the pre-entry wage, although Stayers and Movers do not get penalized at the median
pre-entry wage, while Hobos do receive a significant penalty at this wage level. Furthermore,
Movers and Hobos receive a significantly lower penalty in the upper (10th) wage decile. While
it is not clear from this picture whether one group receives a lower penalty, it contradicts
the findings of Kaiser and Malchow-Møller (2011) that Industry Movers receive a stronger
punishment.
17
-
5 Robustness Checks
A central assumption in this study is that entrepreneurs who move back to wage work have
failed. While the data does not provide exact information about the specific underlying
reason of the exit out of self-employment, the pre-exit earnings profiles of the entrepreneurs
indicate this assumption is realistic. Figure 4 displays the average yearly income of the en-
trepreneurs in the 3 years before re-entry into the labour market, split up by different deciles
of the entrepreneurial earnings distribution. The slope of the earnings curve is negative for
all deciles. For example, entrepreneurs in the top 10% of the earnings distribution earn
almost 40,000 Euros per year 3 years before re-entry, but only slightly more than 20,000
Euros in their last quarter in self-employment. These findings suggest that, in general, the
entrepreneurs in the sample only leave self-employment when they perform badly. This is in
line with Landier (2005) model of entrepreneurial failure, where only the worst performing
entrepreneurs exit self-employment in a context of high costs of exit. As discussed in section
2, this is indeed an important characteristic of the Belgian business landscape.
6 Alternative Theoretical Interpretations
This section discusses possible theoretical explanations for our findings. Williams (2000) ar-
gues that a wage penalty for failed entrepreneurs is consistent with models of human capital
depreciation where entrepreneurs’ job-, firm-, or sector-specific skills atrophy while they are
self-employed. One prediction of these models is that the penalty should increase over time
in entrepreneurship, as long as the depreciation rate is positive. However, this prediction is
inconsistent with our findings that a penalty is only present if the entrepreneur fails within
the first 4 years after founding the business. Another prediction of these models is that
entrepreneurs who move to a different industry after failure would be punished more, since
they sacrifice industry-specific human capital. The findings presented in table 6 reject this
hypothesis; entrepreneurs who move to an employer in a different industry than they were
18
-
self-employed in, do not receive a stronger penalty. On the contrary; the findings suggest
that the penalty is actually lower for those who switch industries.
We argue that our findings are consistent with a broad class of models of adverse selec-
tion in the labour market (cf. Greenwald, 1986; Gibbons and Katz, 1991). Essentially, these
models predict that when employers have discretion about whom to lay off, the market will
infer that laid off workers are of low ability. We make the same argument for entrepreneurs:
since entrepreneurial performance depends on a person’s entrepreneurial capabilities, failure
is a signal of low ability to potential future employers.
Successfully operating a firm requires business skill. A worker acquires such skill through
on-the-job learning, which depends on the jobs to which she is assigned to (Lazear, 2004;
Liang et al., 2014). Workers placed in high-level positions that give them decision making
authority and management experience are more likely to acquire such business skill. Usually,
these positions are also highly paid. Therefore, individuals previously working in a high-pay,
high-level job should in theory have a high likelihood of becoming successful entrepreneurs.
However, if these entrepreneurs fail - and especially in the initial years of the business - the
market will interpret this as a signal of poor business acumen, which they were supposed to
have acquired in previous jobs. They can only avoid a penalty if they manage to keep their
venture alive for a long enough amount of time (ca. 5 years or more), which signals that
they actually do have the skills to run a business. On the contrary, entrepreneurs coming
from low-level, low-paid jobs are not expected to already possess the necessary human cap-
ital to successfully operate a firm. Hence, entrepreneurial failure will not have an impact
on their status in the labour market, also because when they return to paid employment
these individuals will probably not move to management positions. For them, testing the
entrepreneurial waters offers an opportunity to discover their business and management ca-
pabilities (cf. Manso, 2016).
19
-
Another possibility could be that entrepreneurs coming from a high-paid job have difficul-
ties in finding a similar position when they return to the labour market, because management
positions are more scarce than jobs in the lower levels of the firm hierarchy. However, this
is unlikely to be the case. First, it is a well established fact that the unemployment rate
is highest among the less able workers, which implies that higher ability workers can more
easily find jobs. Second, in a recent paper, Lazear et al. (2016) show that vacancy rates are
highest for high-paying, high-quality jobs, since high ability individuals are more flexible and
can do a wide variety of jobs, while low ability workers are not able to fill in high-quality
positions. These findings suggest that entrepreneurs coming from low-paid positions would
have more difficulties in finding a job when returning to paid employment. Given that we
find no penalty for entrepreneurs coming from low-paid positions, but a strong punishment
for those coming from high-paid jobs, our results are likely to be conservative.
7 Conclusion
Public policy often explicitly focuses on encouraging individuals to become self-employed,
highlighting the many benefits of entrepreneurship. For example, in its Entrepreneurship
2020 Action Plan, the European Commission explicitly states that: “To bring Europe back
to growth and higher levels of employment, Europe needs more entrepreneurs” (European
Commision, 2013). When evaluating the net benefits of such entrepreneurial support pro-
grammes, it is paramount to take into account how past entrepreneurial experience affects
an individual’s labour market outcomes in case she returns to paid employment after an en-
trepreneurial failure. Unfortunately, previous studies that have estimated the effect of former
entrepreneurial experience on subsequent earnings in wage work have provided mixed results.
In this paper, we contribute to this literature by documenting different conditions un-
20
-
der which failed entrepreneurs receive a wage penalty when returning to wage work. First,
we show that on average former entrepreneurs earn lower wages and experience lower wage
growth in the quarter of re-entry in paid employment compared to similar employees without
entrepreneurial experience. However, the penalty is stronger for entrepreneurs who earned
a relatively high wage before transitioning to self-employment, while there is no punishment
for entrepreneurs who earned relatively little pre-entry into self-employment. Second, we
find that the penalty resolves over time; after 5 years of successfully operating a business,
entrepreneurs who move back to wage work do not experience a wage penalty. Third, en-
trepreneurs who move to an employer in a different industry after failure do not receive a
higher penalty. On the contrary, the results indicate that industry movers are penalized
less. These findings highlight the potential negative spillovers from entrepreneurial failure in
the labour market. The results are consistent with predictions of a broad class of screening
models where employers use the event of an entrepreneurial failure as a signal of unobserved
capabilities and recognize that this signal is less informative when individuals have operated
a business successfully for a number of years. The results are not consistent with the pre-
dictions of models of human capital depreciation where the depreciation rate is steady over
time and equal for individuals with different abilities.
In summary, our results suggest that attempting entrepreneurship is not per se penalized
in case of failure, but can lead to severe punishment in the labour market for individuals who
are supposed to possess the capabilities to successfully operate a business. Previous studies
have shown that there is excessive entry into entrepreneurship from both tails of the ability
and wage distribution (e.g. Åstebro et al., 2011; Elfenbein et al., 2010), where entrepreneurs
coming from the upper end of the distribution are considered as the stars, whereas the least
able are misfits who are unable to work productively with other workers (cf. Åstebro et al.,
2011). However, the findings presented here indicate in a sense that stars might turn into
misfits if they do not live up to the expectations of the market in terms of entrepreneurial
21
-
survival.
There are several limitations that affect interpretation and generalizability of the findings
and suggest future research. First, the matching algorithm will produce biased outcomes if
the transition into entrepreneurship is related to unobservable factors not captured in the
model to estimate the propensity score. Although we match individuals on an extensive
range of variables, there is no information in the data about a person’s educational back-
ground which previous studies have shown is an important predictor of entrepreneurial entry
(Åstebro et al., 2011). To minimize potential confounding effects of unobserved ability dif-
ferences between the group of entrepreneurs and the counterfactual employees we perform
close matches on wage and wage growth, but the possibility remains that we are not able
to fully control for unobserved heterogeneity. Moreover, the Belgian business landscape is
characterized by low entrepreneurial dynamism, i.e. relatively low entry and exit. Therefore,
a potential avenue for future research is to verify whether the results can be replicated in
more dynamic contexts, such as the US and Denmark.
22
-
References
Åstebro, T., Chen, J., and Thompson, P. (2011). Stars and Misfits: Self-Employment and
Labor Market Frictions. Management Science, 57(11):1999–2017.
Åstebro, T. and Thompson, P. (2011). Entrepreneurs, Jacks of all trades or Hobos? Research
Policy, 40(5):637–649.
Baptista, R., Lima, F., and Preto, M. T. (2012). How former business owners fare in the
labor market? Job assignment and earnings. European Economic Review, 56(2):263–276.
Becker, G. S. (1964). Human Capital: A Theoretical and Empirical Analysis, with Special
Reference to Education. The University of Chicago Press, 2nd edition.
Bruce, D. and Schuetze, H. J. (2004). The labor market consequences of experience in
self-employment. Labour Economics, 11(5):575–598.
Burton, M. D., Sørensen, J. B., and Dobrev, S. D. (2016). A Careers Perspective on En-
trepreneurship. Entrepreneurship Theory and Practice, 40(2):237–247.
Calvino, F., Criscuolo, C., and Menon, C. (2016). No Country for Young Firms? Technical
report, OECD Science, Technology and Industry Policy Papers.
Campbell, B. A. (2013). Earnings Effects of Entrepreneurial Experience: Evidence from the
Semiconductor Industry. Management Science, 59(2):286–304.
Daly, M. (2015). The long term returns of attempting self-employment with regular employ-
ment as a fall back option. Labour Economics, 35:26–52.
De Mulder, J. and Godefroid, H. (2016). How to stimulate entrepreneurship in Belgium ?
Economic Review, (ii):63–80.
Dillon, E. and Stanton, C. (2017). Self-Employment Dynamics and the Returns to En-
trepreneurship. National Bureau of Economic Research Working Paper Series, No. 23168.
23
-
Dreher, A., Gassebner, M., Dreher, A., Gassebner, ·. M., and Gassebner, M. (2013). Greasing
the wheels? The impact of regulations and corruption on firm entry. Public Choice,
155(155).
Elfenbein, D. W., Hamilton, B. H., and Zenger, T. R. (2010). The Small Firm Effect and the
Entrepreneurial Spawning of Scientists and Engineers. Management Science, 56(4):659–
681.
European Commision (2013). Entrepreneurship 2020 action plan: reigniting the en-
trepreneurial spirit in Europe.
Evans, D. S. and Leighton, L. S. (1989). Some Empirical Aspects of Entrepreneurship.
Failla, V., Melillo, F., and Reichstein, T. (2017). Entrepreneurship and employment stabil-
ity Job matching, labour market value, and personal commitment. Journal of Business
Venturing, 32(2):162–177.
Folta, T. B., Delmar, F., and Wennberg, K. (2010). Hybrid Entrepreneurship. Management
Science, 56(2):253–269.
Georgellis, Y., Sessions, J., and Tsitsianis, N. (2007). Pecuniary and non-pecuniary aspects of
self-employment survival. The Quarterly Review of Economics and Finance, 47(1):94–112.
Geurts, K. and Van Biesebroeck, J. (2016). Firm creation and post-entry dynamics of de
novo entrants. International Journal of Industrial Organization, 49:59–104.
Gibbons, R. and Katz, L. F. (1991). Layoffs and Lemons. Journal of Labor Economics,
9(4):351–380.
Gimeno, J., Folta, T. B., Cooper, A. C., and Woo, C. Y. (1997). Survival of the Fittest?
Entrepreneurial Human Capital and the Persistence of Underperforming Firms. Adminis-
trative Science Quarterly, 42(4):750–783.
24
-
Greenwald, B. C. (1986). Adverse Selection in the Labour Market. The Review of Economic
Studies, 53(3):325.
Hamilton, B. H. (2000). Does Entrepreneurship Pay? An Empirical Analysis of the Returns
of Self-Employment. Journal of Political Economy, 108(3):604.
Ho, D., Imai, K., King, G., and Stuart, E. (2007). Matching as Nonparametric Preprocessing
for Reducing Model Dependence in Parametric Causal Inference. Political Analysis, 15.
Hyytinen, A. and Rouvinen, P. (2008). The labour market consequences of self-employment
spells: European evidence. Labour Economics, 15(2):246–271.
Kaiser, U. and Malchow-Møller, N. (2011). Is self-employment really a bad experience?
Journal of Business Venturing, 26(5):572–588.
Koellinger, P. D., Mell, J. N., Pohl, I., Roessler, C., and Treffers, T. (2015). Self-employed
But Looking: A Labour Market Experiment. Economica, 82(325):137–161.
Landier, A. (2005). Entrepreneurship and the Stigma of Failure. SSRN Electronic Journal.
Lazear, E., Shaw, K., and Stanton, C. (2016). Who Gets Hired? The Importance of Finding
an Open Slot. Technical report, National Bureau of Economic Research, Cambridge, MA.
Lazear, E. P. (2004). Balanced Skills and Entrepreneurship. American Economic Review,
94(2):208–211.
Lazear, E. P. (2005). Entrepreneurship. Journal of Labor Economics, 23(4):649–680.
Levine, R. and Rubinstein, Y. (2017). Smart and Illicit: Who Becomes an Entrepreneur and
Do They Earn More? The Quarterly Journal of Economics, 132(2):963–1018.
Liang, J., Wang, H., and Lazear, E. (2014). Demographics and Entrepreneurship. Technical
report, National Bureau of Economic Research, Cambridge, MA.
25
-
Manso, G. (2016). Experimentation and the Returns to Entrepreneurship. Review of Finan-
cial Studies, 29(9):2319–2340.
Neal, D. (1995). Industry-Specific Human Capital: Evidence from Displaced Workers. Jour-
nal of Labor Economics, 13(4):653–677.
Oi, W. Y. and Idson, T. (1999). Firm size and wages. In Handbook of Labor Economics,
volume 3, Part B, pages 2165–2214. Elsevier.
Özcan, S. and Reichstein, T. (2009). Transition to Entrepreneurship from the Public Sector:
Predispositional and Contextual Effects. Management Science, 55(4):604–618.
Rosenbaum, P. R. and Rubin, D. B. (1983). The central role of the propensity score in
observational studies for causal effects. Biometrika, 70(1):41–55.
Rosenbaum, P. R. and Rubin, D. B. (1985). Constructing a Control Group Using Multivari-
ate Matched Sampling Methods That Incorporate the Propensity Score. The American
Statistician, 39(1):33.
Sørensen, J. B. (2007). Bureaucracy and Entrepreneurship: Workplace Effects on En-
trepreneurial Entry. Administrative Science Quarterly, 52(3):387–412.
Stuart, E. A. (2010). Matching methods for causal inference: A review and a look forward.
Statistical science : a review journal of the Institute of Mathematical Statistics, 25(1):1–21.
Williams, D. R. (2000). Consequences of self-employment for women and men in the United
States. Labour Economics, 7(5):665–687.
Zimmer, H. (2012). Labour market mismatches. National Bank of Belgium Economic Review,
ii:55–68.
26
-
Tables
Table 1: Balance of Full and Matched Samples
Full Sample Matched Sample
Treatment Control Std. % bias Treatment Control Std. % bias
Age
22 - 24 0.0445 0.1030 -22.5 0.0863 0.0863 0.0
25 - 29 0.2931 0.1525 34.3 0.3208 0.3208 0.0
30 - 34 0.2567 0.1857 17.2 0.2668 0.2668 0.0
35 - 39 0.1759 0.2125 -9.2 0.1734 0.1734 0.0
40 - 44 0.1115 0.2319 -32.4 0.1034 0.1034 0.0
45 - 49 0.0597 0.1729 -35.9 0.0492 0.0492 0.0
Female 0.2501 0.3329 -18.3 0.2072 0.2072 0.0
Region
Flanders 0.6654 0.6516 2.9 0.7861 0.7861 0.0
Wallonia 0.2416 0.2850 -9.9 0.1943 0.1943 0.0
Brussels 0.0929 0.0633 11.0 0.0195 0.0195 0.0
Household position
Living with parents 0.1614 0.1231 11.0 0.1911 0.1928 0.4
Single 0.1445 0.1184 7.7 0.1216 0.1198 0.6
Cohabiting - 0 children 0.1991 0.1647 8.9 0.2014 0.2062 -1.2
Cohabiting - 1 child 0.1715 .1923 -5.4 0.1798 0.1792 0.2
Cohabiting - 2 children 0.1933 0.2480 -13.2 0.1991 0.1983 0.2
Cohabiting - 3> children 0.0756 0.0929 -6.2 0.0635 0.06402 -0.2
Head 1 parent family - 1 child 0.0134 0.0201 -5.2 0.0085 0.0087 -0.2
Head 1 parent family - 2> children 0.0107 0.0174 -5.7 0.0067 0.0068 -0.1
Other 0.0301 0.0227 4.6 0.0263 0.0257 0.4
Occupation
Blue-collar 0.3492 0.3399 2.0 0.4714 0.4752 0.8
White-collar/Govt. official 0.6508 0.6600 -2.0 0.5248 0.5285 -0.8
Employer Sector
Public 0.0800 0.2605 -49.5 0.0878 0.0885 -0.3
Employer Industry
Manufacturing 0.1812 0.2372 -13.8 0.2529 0.2529 0.0
Electricity, gas, water 0.0023 0.0088 -8.6 0.0004 0.0004 0.0
Construction 0.1445 0.0643 26.5 0.1906 0.1906 0.0
Wholesale and retail trade 0.2065 0.1313 20.2 0.2068 0.2068 0.0
Hotels and restaurants 0.0338 0.0105 12.9 0.0099 0.0099 0.0
Transport, storage, communication 0.0699 0.0989 -10.4 0.0638 0.0638 0.0
Financial institutions 0.0460 0.0541 -3.7 0.0329 0.0329 0.0
Real estate and professional services 0.1735 0.0892 22.3 0.1312 0.1312 0.0
Continued on next page
27
-
Full Sample Matched Sample
Treatment Control Std. % bias Treatment Control Std. % bias
Public administration, defence 0.0229 0.0998 -32.5 0.0285 0.0285 0.0
Education 0.0237 0.0939 -30.2 0.0257 0.0257 0.0
Healthcare and support services 0.0543 0.0849 -12.1 0.0455 0.0455 0.0
Social and cultural services 0.0392 0.0254 7.8 0.0112 0.0112 0.0
Households as employers 0.0003 0.0005 -0.9 0.00006 0.00006 0.0
Employer size
< 5 0.1625 0.0449 39.3 0.1014 0.0981 1.1
5 - 9 0.1154 0.0461 25.6 0.1009 0.0977 1.1
10 - 19 0.1196 0.0628 19.8 0.1136 0.1118 0.6
20 - 49 0.1551 0.1122 12.6 0.1653 0.1652 0.0
50 - 99 0.0821 0.0753 2.5 0.0845 0.0866 -0.8
100 - 199 0.0700 0.0774 -2.8 0.0749 0.0787 -1.5
200 - 499 0.0819 0.1053 -8.0 0.0934 0.0983 -1.7
500 - 999 0.0518 0.0743 -9.3 0.0617 0.0611 0.2
>= 1000 0.1612 0.4012 -55.4 0.2041 0.2020 0.5
Nr. of jobs 2.0411 1.6561 35.2 1.7019 1.7146 -1.3
Nr. of quarters unemployed 2.3338 0.9669 40.5 1.2817 1.3079 -0.9
Employer tenure 10.25 11.964 -29.3 12.263 12.149 2.0
Industry tenure 12.224 14.544 -46.4 14.032 13.938 2.0
Daily wage 124.88 118.51 7.3 109.19 109.19 0.0
Wage growth 0.0699 0.0686 0.2 0.0632 0.0489 2.3
Partner works 0.2973 0.2371 13.6 0.2835 0.2860 -0.6
Partner works#partner daily wage 31.572 24.17 9.5 27.355 26.727 1.3
Partner self-employment experience 0.0526 0.0306 11.0 0.0296 0.0314 -1.0
Observations 87614 835969 32473 32473
Comparison of means between treatment and control group in the full and matched sample. The standardised
% bias is the % difference of the sample means in the treated and non-treated (full and matched) sub-samples
as a percentage of the square root of the average of the sample variances in the treated and non-treated groups.
28
-
Table 2: Summary statistics of entrepreneurs and their
matched employees in the quarter pre-entry into self-
employment
Entrepreneurs Matched employees
mean sd mean sd
Average daily wage 127.654 52.35 127.767 46.76
Age ref. ref.
22-24 0.019 0.14 0.023 0.15
25-29 0.199 0.40 0.192 0.39
30-34 0.294 0.46 0.298 0.46
35-39 0.234 0.42 0.223 0.42
40-44 0.161 0.37 0.170 0.38
45-49 0.076 0.27 0.074 0.26
50-54 0.016 0.13 0.019 0.14
55-59 0.001 0.03 0.000 0.00
Male 0.894 0.31 0.894 0.31
Household position ref. ref.
Living with parents 0.109 0.31 0.123 0.33
Single 0.125 0.33 0.124 0.33
Cohabiting - 0 children 0.179 0.38 0.167 0.37
Cohabiting - 1 child 0.196 0.40 0.197 0.40
Cohabiting - 2 children 0.254 0.44 0.250 0.43
Cohabiting - 3 > children 0.092 0.29 0.088 0.28
Head 1 parent family - 1 child 0.008 0.09 0.009 0.10
Head 1 parent family - 2 >children 0.008 0.09 0.010 0.10
Household resident 0.011 0.10 0.013 0.11
Other 0.019 0.14 0.018 0.13
Region ref. ref.
Flanders 0.766 0.42 0.766 0.42
Wallonia 0.217 0.41 0.218 0.41
Brussels 0.017 0.13 0.016 0.13
Employer tenure 16.275 11.80 20.417*** 11.93
Nr. of jobs held 2.792 2.26 2.165*** 1.72
Occupation ref. ref.
Continued on next page
29
-
Entrepreneurs Matched employees
mean sd mean sd
Blue-collar worker 0.505 0.50 0.494 0.50
White-collar worker 0.455 0.50 0.450 0.50
Govt. official 0.040 0.20 0.056** 0.23
Employer size ref. ref.
< 5 0.136 0.34 0.094*** 0.29
5-9 0.117 0.32 0.093** 0.29
10-19 0.119 0.32 0.107 0.31
20-49 0.179 0.38 0.171 0.38
50-99 0.083 0.28 0.095 0.29
100-199 0.064 0.24 0.084** 0.28
200-499 0.084 0.28 0.114*** 0.32
500-999 0.050 0.22 0.057 0.23
>= 1000 0.167 0.37 0.186 0.39
Employer sector ref. ref.
Private 0.927 0.26 0.914 0.28
Employer Industry ref. ref.
Manufacturing 0.205 0.40 0.240** 0.43
Electricity, gas, water 0.001 0.03 0.003 0.06
Construction 0.241 0.43 0.224 0.42
Wholesale and retail trade 0.213 0.41 0.204 0.40
Hotels and restaurants 0.012 0.11 0.010 0.10
Transport, storage, communication 0.088 0.28 0.088 0.28
Financial institutions 0.032 0.18 0.033 0.18
Real estate and professional services 0.130 0.34 0.119 0.32
Public administration, defence 0.032 0.18 0.039 0.19
Education 0.019 0.14 0.018 0.13
Healthcare and support services 0.013 0.11 0.012 0.11
Social and cultural services 0.011 0.11 0.009 0.10
Observations 2736 2736
30
-
Table 3: Summary statistics re-entry quarter
Entrepreneurs Matched Employeesmean sd mean sd
Avg. daily wage 136.583 59.63 147.086*** 54.46∆ wage 0.003 0.03 0.010*** 0.02Spell length 14.325 8.72 14.325 8.72Employer change 0.817 0.39 0.230*** 0.42Industry change 0.439 0.50 0.094*** 0.29Industry Switching ref.Industry Stayers 0.316 0.47Industry Movers 0.335 0.47Hobos 0.349 0.48Observations 2736 2736
31
-
Table 4: OLS regressions on avg. daily wage at the quarter of pre-entry, re-entry, andquarterly wage growth.
(1) (2) (3) (4) (5)VARIABLES ln(wage0) ln(wage1) ln(wage1) ∆ln(wage) ∆ln(wage)
SE 0.003 -0.066*** 0.008 -0.006*** 0.002(0.006) (0.007) (0.014) (0.001) (0.002)
W0, decile = 2 0.065*** 0.089*** -0.008*** -0.005***(0.010) (0.011) (0.001) (0.001)
W0, decile = 5 0.185*** 0.232*** -0.015*** -0.010***(0.011) (0.012) (0.001) (0.001)
W0, decile = 9 0.535*** 0.627*** -0.023*** -0.013***(0.017) (0.016) (0.002) (0.002)
W0, decile = 10 0.790*** 0.890*** -0.031*** -0.018***(0.020) (0.020) (0.002) (0.003)
SE#W0 -0.039* -0.004*decile = 2 (0.019) (0.002)SE#W0 -0.082*** -0.009***decile = 5 (0.020) (0.002)SE#W0 -0.171*** -0.020***decile = 9 (0.029) (0.003)SE#W0 -0.181*** -0.022***decile = 10 (0.032) (0.004)Constant 4.316*** 4.445*** 4.385*** 0.007 0.001
(0.059) (0.114) (0.114) (0.017) (0.017)
Observations 5,471 5,471 5,471 5,471 5,471R-squared 0.474 0.640 0.646 0.152 0.171
Additional control variables (not displayed): age, gender, household position,region, occupation, employer sector, employer industry, employer size, spelllength, quarter fixed effects. Robust standard errors in parentheses. *p
-
Table 5: Quarterly wage growth over different spell durations: marginal effects of SE atdifferent deciles of avg. daily wage pre-entry.
Spell duration (years)Pre-entry wage (decile) All 1-2 3-4 5-6 7+
1st 0.003* 0.006 0.003 0.002 -0.001(0.002) (0.003) (0.002) (0.002) (0.002)
2nd -0.001 -0.004 0.001 0.002 -0.001(0.001) (0.003) (0.002) (0.002) (0.002)
5th -0.006*** -0.010*** -0.002 -0.003 -0.002(0.002) (0.003) (0.002) (0.002) (0.002)
9th -0.017*** -0.022*** -0.013** -0.005 0.001(0.003) (0.004) (0.004) (0.003) (0.005)
10th -0.018*** -0.027*** -0.010** -0.008 -0.004(0.004) (0.006) (0.004) (0.004) (0.003)
Observations 5,471 5,471 5,471 5,471 5,471
Standard errors in parentheses. *p
-
Figures
Figure 1: Duration of self-employment spells of returnees
34
-
Figure 2: Plot of predictive margins of ln(wage1) at different levels of the avg. daily wageat the pre-entry stage: Self-employed vs. non self-employed
Figure 3: Plot of predictive margins of ∆wage at different levels of the avg. daily wage atthe pre-entry stage: Self-employed vs. non self-employed
35
-
Figure 4: Avg. yearly income from self-employment by time to re-entry in wage employment
36
IntroductionEmpirical ContextData and MethodologyData source and construction of matched sampleMatching VariablesQuality of the Matched Sample
Additional Sample Restrictions
ResultsDescriptive AnalysisRegression AnalysisSpell lengthIndustry Switching
Robustness ChecksAlternative Theoretical InterpretationsConclusionTablesFigures