Small Bus Econ (2017) 48:393–412

DOI 10.1007/s11187-016-9778-x

Do micro start-ups fuel job creation? Cross-country

evidence from the DynEmp Express database
Chiara Criscuolo . Peter N. Gal . Carlo Menon

Accepted: 7 June 2016 / Published online: 21 July 2016

Ó Springer Science+Business Media New York 2016

Abstract Exploiting a novel database recently built
from national business registers by the OECD with the
support of an international network of experts, this
paper investigates the growth dynamics of micro-firms
(employing less than ten workers) across 16 countries.
Results show that only a small proportion of micro-
firms manage to grow beyond ten employees, but those
contribute disproportionately to overall job creation.
Econometric analysis focusing in particular on the role
of age confirms that young micro-firms—especially
those below 3 years of age—are much more likely to
grow above ten employees than older firms. These
findings are remarkably stable over the three time
periods considered (2001–2004, 2004–2007, and
2007–2010), i.e., also during the Great Recession.
Disclaimer The views expressed in these papers are those of
the authors and do not necessarily reflect those of the OECD or
of the governments of its member countries.
C. Criscuolo  C. Menon (&)
Science, Technology and Innovation Directorate, OECD,
2, rue André Pascal, 75775 Paris Cedex 16, France
e-mail: chiara.criscuolo@oecd.org
C. Menon
e-mail: carlo.menon@oecd.org
P. N. Gal
Economics Department, OECD, 2, rue André Pascal,
75775 Paris Cedex 16, France
e-mail: peter.gal@oecd.org
Keywords Micro-firms  Start-ups  Employment
dynamics
JEL Classifications D22  L25  L26
1 Introduction: up-or-out dynamics
and the importance of start-ups for job creation
Start-ups are considered to be an important driver of
job creation in all countries, as a significant share of
jobs is created by new entrants; furthermore, incum-
bent young firms are also generally net job creators
(Criscuolo 2014a, b; Lawless 2014). However, it is
also known that a significant share of them die within
the very first years of life (Anyadike-Danes et al.
2013). Therefore, the net contribution of start-ups to
job creation needs to take into account both of these
phenomena that characterize the ‘‘up-or-out’’ dynam-
ics of young firms’ growth (Haltiwanger et al. 2013).
Cross-country evidence on the growth dynamics of
start-ups is extremely limited (with Anyadike-Danes
et al. 2013 being a noteworthy exception). This paper
investigates the growth dynamics of micro-firms, i.e.,
firms employing less than ten workers, over the time
periods 2001–2004, 2004–2007, and 2007–2010
across 16 countries.1 We look in particular at the
1
Included countries are Austria, Belgium, Brazil, Canada,
Finland, France, Hungary, Italy, Japan, Luxembourg, the

123
394
differential growth patterns of micro-firms according
to their age profile. This is possible thanks to a novel
database called DynEmp Express, recently constructed
using national business registers by the OECD with the
support of an international network of experts. It
contains detailed information on transition dynamics
of firms according to their age and size profiles.
The results show that only a tiny proportion—on
average around 5 %—of micro-firms grow above ten
employees over a 3-year period, and that this share is
almost twice as much for younger firms (\3 years old)
than for firms aged 11 years or more. The substantially
higher dynamism of start-ups is remarkably constant
over time. In particular, it is not significantly affected
by the crisis, during which we show that firms across
all age groups adjusted primarily through becoming
inactive in larger numbers, rather than through lower
growth. In addition, we also shed light on the sectoral
dimension of this mechanism and we find that
manufacturing young firms tend to grow faster than
their counterparts among services.
The reason for looking at micro-firms is manifold.
First, micro-firms are an important component of
industrialized economies. They represent 70–90 % of
the firm population across 18 countries in the sample,
and they account for 10–40 % of employment
(Criscuolo 2014a, b). Second, existing evidence also
suggests that among micro-firms, those that are
young—and start-ups in particular—contribute sig-
nificantly to net job creation. Third, the nature of the
data for micro-firms in the DynEmp Express database
is particularly well suited for such a targeted analysis,
since it relates to a narrowly defined and at the same
time densely populated group of firms. Potential
measurement issues related to firm age are also
somewhat mitigated in the case of this group, as new
firm creation by mergers and acquisitions is more
likely to occur at larger sizes.
There are important cross-country differences in
the degree to which young firms contribute to
Footnote 1 continued
Netherlands, Norway, New Zealand, Portugal, Sweden and the
United Kingdom. As data suppression due to confidentiality
restrictions is more binding for the United States, results
including at will appear only as a robustness check and for those
descriptive charts where the issue does not pose a problem.
123
C. Criscuolo et al.
employment growth. While entrants do not show
remarkable differences in average size across coun-
tries, enabling the most productive start-ups to scale
up is key to aggregate employment and productivity
growth (Criscuolo 2014a, b; Hsieh and Klenow 2012).
Also, start-up rates present some interesting differ-
ences across countries and a decreasing trend over
time in most countries (Haltiwanger et al. 2013 for the
USA; Criscuolo 2014a, b for evidence across 18
countries).
By inspecting the growth pattern of micro-firms,
and especially micro start-ups, our paper provides
useful evidence for economic policies aimed at
increasing business dynamism. In a dynamic economy
with low entry barriers, small start-ups should be
characterized by an ‘‘up-or-out’’ pattern. This means,
on the one hand, that many ventures enter the market,
experimenting with new business ideas. On the other
hand, most entrants exit few years after birth, while
only a handful of very successful businesses grow very
rapidly (Haltiwanger et al. 2013). Conversely, in more
stagnant economies with high entry barriers the
number of entrants is smaller; and the few new
businesses who do manage to enter tend to be bigger in
order to be able to cover the larger entry costs.
Moreover, stronger economic frictions and obstacles
to resource reallocations may limit their ability to
attract employment and capital and to grow if
successful (Andrews et al. 2014). This results in lower
exit rates but also lower growth rates, both at the firm
and at the aggregate level.
Our work is also closely related to the debate on the
‘‘cleansing’’ vs. ‘‘scarring’’ effects of recessions.
Recessions may spur a cleansing process in the
economy, as unproductive firms exit the market and
free up resources which could be employed by more
productive firms, consistent with a more intense
process of ‘‘creative destruction’’ (Schumpeter
1942). There is some evidence for the USA suggesting
that this process is actually in place. For instance,
Davis and Haltiwanger (1992, 1999) and Davis et al.
(2006, 2012) find evidence of increased reallocation in
the manufacturing and service sector during recessions
which occurred in the second half of the twentieth
century.
On the other hand, recessions may also have a
‘‘scarring’’ effect on the economy, impeding the
developments of potentially successful new firms
and hampering the process of reallocation, e.g.,
Do micro start-ups fuel job creation?
because of frictions in the financial market. The
disproportionate death rate among infant businesses
which is often found during recessions may be highly
detrimental to growth also in the medium to long run.
The reason is that some of these young firms are the
ones who have the highest potential to achieve high
productivity growth over time, even if appearing less
productive in the short run (Ouyang 2009). Evidence
from the USA suggests that the last recession, contrary
to what happened during the previous ones, the
intensity of reallocation fell rather than rose; further-
more, the reallocation that did occur was less produc-
tivity enhancing than in prior recessions (Foster et al.
2014). Our paper tests for differential effects of the
crisis along the age dimension and does not find clear
patterns in differences in business dynamics related to
the crisis for young firms that might indicate system-
atic effects of the crisis on young businesses across
countries.
Section 2 below describes the data, followed by the
descriptive and econometric analysis in Sects. 3 and 4,
respectively. The final section concludes.
2 Data
2.1 DynEmp: a distributed micro-data analysis
approach
The OECD has embarked on a project called the
Dynamics of Employment (DynEmp) which utilizes
confidential firm level data from national business
registers. The project aims at providing new empirical
evidence on firm employment growth dynamics; the
contribution to job creation and destruction by differ-
ent groups of firms; and the upscaling of young
businesses, with a particular emphasis on the cross-
country dimension of both the data collection exer-
cise—and therefore on harmonization—and of the
analysis. It is achieved by distributing a centrally
written software code among national representatives
having access to business registers or comparable data
sources. The code implements calculations using a
common methodology and also relies on previously
collected detailed metadata information on each
country’s datasets. This methodology is often referred
to as distributed microdata analysis (Bartelsman et al.
2004). A detailed description of the project is available
in Criscuolo (2014a, b).
395
2.2 The DynEmp Express database and the sample
used in the analysis
The dataset used for this paper has been produced
during the first phase of the DynEmp project and is
based on the ‘‘DynEmp Express’’ Stata routine that
was run by national representatives during the first
quarter of 2013.2 The ‘‘DynEmp Express’’ database
includes data for the following economies: Austria,
Belgium, Brazil, Canada, Finland, France, Hungary,
Italy, Japan, Luxembourg, the Netherlands, New
Zealand, Norway, Portugal, Spain, Sweden, the
United Kingdom, and the USA. The default period is
2001–2011; however, for some countries the late or
early years are not available. These and other quali-
fications regarding the coverage are reported in
Table 8 in Appendix.
In particular, it should be noted that Spain could not
be included in the analysis underlying this paper as the
part of the database used for this paper (the transition
matrices) has been produced in a different format and
thus was not exactly comparable with those of other
countries. For the United Kingdom, the period
2004–2007 is not available in the transition matrices
database due to confidentiality restrictions. The USA
is also excluded from the main sample because the
confidentiality blanking is much more prevalent than
in all other countries.3 To the extent that coverage for
the USA is selected by the blanking in a non-random
manner, this could lead to a bias in the results. We
were therefore forced to exclude USA from the main
database to ensure comparability, but we report the
results when including it in the robustness sec-
tion. Therefore, the main sample for the regression
analysis covers 16 countries.
DynEmp Express collects statistics aggregated to
the level of the three broad macroeconomic sectors
(manufacturing, construction, and market services
excluding financial services), by year, size class or/
and age class and distinguishing between entry,
incumbent, and exiting firm status. The employment
variable is based on headcounts, while full time
2
See the documentation of the DynEmp v2 routine in Criscuolo
et al. (2014b), which also includes as an option the DynEmp
Express run.
3
Confidentiality blanking for US follows more complex
criteria that are themselves confidential, and that do not only
depend on the number of units contained in the cell.
123
396
equivalents or hours worked are generally not avail-
able. This data limitation does not allow capturing the
part of employment variation that affect the number of
hours worked for employee (e.g., part-time contracts),
rather than the number of employees.
2.2.1 The transition matrices
This paper focuses on one specific output of the
DynEmp Express routine: a set of transition matrices
across firm size classes for three time periods
(2001–2004; 2004–2007, and 2007–2010). For each
age class, these matrices report the number of units
completing a transition from employment size class
j to size class k, their total employment at the
beginning and at the end of the period, and their
median size. The ‘‘entering’’ category includes those
firms that did not exist at the beginning of the period;
symmetrically, the ‘‘exiting’’ category at the end of the
period includes those firms that do not exist or are
inactive at the end of the 3-year period under
consideration.
2.3 Caveats
Despite our best efforts to homogenize the data
collection by use of a single program routine, the
dataset comes with some caveats stemming from the
inherent difficulties in harmonizing certain definitions
across countries. One of this involves the issue of
confidentiality blanking. For the great majority of
countries in the database cells that are scarcely
populated—i.e., which contain less than a specified
number of units (e.g., 5 or 10)—are blanked. In
addition, the data released also have to conform to
secondary disclosure and dominance rules, i.e., taking
into account residual confidentiality and the presence
of high concentration among very few firms. When
looking at micro-firms the issue is generally negligi-
ble, due to the sheer size of the reference population.
As mentioned above, the USA is a significant excep-
tion, which led us to exclude the country from the main
sample.
A second caveat relates to the age variable and to
the definition of entry and exit. Firms that ‘‘enter’’,
i.e., that appear for the first time in the business
registers, may or may not indicate genuine eco-
nomic births of new businesses, in the traditional
sense of being created de novo by entrepreneurial
123
C. Criscuolo et al.
managers (Geurts and Van Biesebroeck 2014). They
may reflect spurious births due to problems of
coding errors in the source data, or breaks in the
register coding. They may also reflect the birth of
new legal entities such as spin-offs from a larger
company; a creation of a new firm as part of an
enterprise group (e.g., a greenfield investment); the
merger of more companies; the restructuring of an
existing firm or the renaming of a company under a
different name. Similar issues apply for the defini-
tion of exit. To the extent that in some countries,
sectors, or time periods, spurious entry and/or exit
events are more frequent, the data may show more
churning activity and a younger business population
than what would actually be recorded with better
quality data. However, the work done so far on the
DynEmp Express database (Criscuolo 2014a, b) has
shown that most of the general findings are robust to
limiting the sample to those countries for which the
entry and exit information is considered to be more
precise. Moreover, the group of micro-firms below
ten employees—the main focus of the current
analysis—is less prone to some of these issues. In
particular, they are less likely to have the financial
resources to acquire other firms, and therefore to
grow by acquisition rather than by organic growth.
They are more likely, however, be the target of
acquisitions, which would be reflected in the data in
some spurious exits.
Finally, a third caveat relates specifically to the
transition matrices. The category of firms that are
classified as ‘‘inactive’’ at the end of the period
includes firms that are not in the business register both
because they have ended business, and because their
employment record is missing in that particular year.
Those firms could be only temporarily inactive. As a
consequence, the fraction of firms appearing as
inactive may overstate true exit rates. This is the
reason we prefer to use the term inactive rather than
exiting for this group of firms. However, the focus of
our analysis is on within country patterns by firm age
and over time and not on the comparison of inactivity
rates per se across countries.
3 Descriptive analysis
The analysis of the transition matrices across employ-
ment in different size classes allows for a detailed
Do micro start-ups fuel job creation?
description of the growth dynamics of firms of
different age, initial size, sector, and country.
3.1 The dynamics of micro start-ups
This section focuses on a specific subset of the
transition matrix dataset, i.e., the transition dynamics
of start-ups (firms aged 0–2 at the beginning of the
period) with less than ten employees.
In a dynamic economy, a substantial share of start-
ups exit within the first few years of activity, but those
who survive grow very rapidly. In turn, this group of
fast-growing firms has the potential to more than
compensate for the jobs destroyed by those that fail.
These ‘‘up-or-out’’ dynamics have been recently
documented for young firms in the USA (Haltiwanger
et al. 2013), but Fig. 1 shows the importance of this
phenomenon in all 16 countries considered as well as
differences across them both in terms of number of
firms and jobs involved.
Figures 2 and 3 further disentangle the analysis for
micro start-ups for the three cohorts over time. Panel
(a) of Figs. 1 and 2 reports the share of firms in each
transition group (moving above ten employees, stay-
ing below ten employees, becoming inactive) in the
total number of firms, while panel (b) of Figs. 1 and 3
reports the relative contribution to net job creation
(CNJC) for each transition group i, calculated as
follows:
NetJobVari
CNJCi ¼ P ð1Þ
i jNetJobVari j
where NetJobVari is the net job variation (equal to the
difference between gross job creation and gross job
destruction) in the transition group i, and the denom-
inator is the sum of the absolute values of the net job
variations across the three groups.
Three main features are noteworthy: firstly, very
few micro start-ups—between 2 and 9 %—grow
above ten employees, but their contribution to
employment change ranges from 21 to 47 %; sec-
ondly, stable firms still create a reasonable amount of
jobs, but their contribution is less than proportional to
their weight in terms of number of firms, and in
exceptional cases, they can actually contribute nega-
tively to net job creation; thirdly, the extent to which
micro start-ups survive is very different across coun-
tries (barely above 60 % for the UK and the USA,
397
while around 80 % for Belgium, Finland, Italy, and
Sweden).
3.1.1 Changes over time
The probability of becoming inactive has increased in
the period 2007–2010, affected by the outburst of the
international financial crisis in 2008 and the subse-
quent enduring recession or slow recovery in many
countries—as shown in panel (a) of Fig. 4, and even
more remarkably in panel (b) illustrating the share of
jobs involved. On the contrary, the share of growing
jobs (and to some extent, the share of growing firms)
appears to be more stable over the three periods. This
pattern, therefore, suggests that the crisis had a
stronger impact on the exit margin in the economy,
with a harsher selection process leading to higher exit
rates and only minor effects on upscaling start-ups.
This seems in line with the presence of a cleansing
process whereby low productivity firms are forced out
of the market. The subsequent econometric analysis
takes a closer look at these patterns.4
In aggregate terms across countries (excluding
those for which data are not available in all three
periods: France, Japan, Portugal, and the United
Kingdom; and the USA) in the regression sample,
the youngest group of micro-firms created around
339,000 jobs in the 2001–2004 period, 394,000 in the
2004–2007 period, and 211,000 in the 2007–20010
period; therefore, this group of firms kept creating jobs
even during the Great Recession. Conversely, micro-
firms aged three or more are always net job destroyers
in aggregate terms, with those aged 11 or more
destroying more than one million jobs in the period
2007–20010 (Fig. 5, panel a). To put these numbers
into perspective, we note that micro-firms of all age
classes account for around 24 million jobs in 2007 in
our sample, out of the total number of jobs at 90
million (in the last period). Given their limited weight
in the economy, their relative contribution to net job
creation in is also quite limited, never going above the
4
To answer the question of whether the increased exit rates
were productivity enhancing one should look at whether the
firms that exited were on average the least productive, and
whether average productivity increased after the recession.
Recent evidence from the US suggest that the crisis was
accompanied by productivity enhancing reallocation; for
Europe recent evidence (Andrews et al. 2015) suggests that
for Europe this was also the case.
123
398
Fig. 1 Three-year survival and growth performance of micro
start-ups. Notes Average over the three 3-year periods
(2001–2004; 2004–2007; 2007–2010) for all countries except
France, Portugal, and the United Kingdom (see Table 8).
Figures refer to the group of firms which have between 0 and
9 employees in the beginning of the period and which are
\3 years old. Inactive firms do not report information on
employment at the end of the 3-year period, either because they
are temporarily inactive or because they have permanently
Fig. 2 Three-year survival and growth performance of micro
start-ups, 2001, 2004, 2007. Notes stable firms belong to the
same size category (0–9) at the end of the 3-year period;
growing firms belong to a higher size category at the end of the
3-year period; inactive firms do not report information on
employment at the end of the 3-year period, either because they
are temporarily inactive or because they have permanently
exited. Sectors covered are: manufacturing, construction, and
123
C. Criscuolo et al.
exited. Sectors covered are: manufacturing, construction, and
non-financial business services. Owing to methodological
differences, figures may deviate from officially published
national statistics. For Japan data are at the establishment level,
for other countries at the firm level. The reported figures for the
USA are prone to measurement problems due to confidentiality
blanking. Source OECD DynEmp database. a Share of firms,
b share of jobs affected
non-financial business services. Owing to methodological
differences, figures may deviate from officially published
national statistics. For Japan data are at the establishment level,
for other countries at the firm level. The reported figures for the
USA are prone to measurement problems due to strict rules for
blanking cells due to confidentiality reasons. Source OECD
DynEmp database
Do micro start-ups fuel job creation?
Fig. 3 Three-year survival and growth performance of micro
start-ups, 2001, 2004, 2007. Notes stable firms belong to the
same size category (0–9) at the end of the 3-year period;
growing firms belong to a higher size category at the end of the
3-year period; inactive firms do not report information on
employment at the end of the 3-year period, either because they
are temporarily inactive or because they have permanently
exited. Sectors covered are: manufacturing, construction, and
Fig. 4 Three-year survival and growth performance of micro
start-ups over time. Notes Average across countries, excluding
those for which data are not available in all three periods
(France, Japan, Portugal, and the United Kingdom) and the
USA. Figures refer to the group of firms which have between 0
and 9 employees in the beginning of the period and which are
\3 years old. Inactive firms do not report information on
employment at the end of the 3-year period, either because they
399
non-financial business services. Owing to methodological
differences, figures may deviate from officially published
national statistics. For Japan data are at the establishment level,
for other countries at the firm level. Data for Canada refer only to
organic employment changes and abstract from merger and
acquisition activity. The reported figures for the USA are prone
to measurement problems due to strict rules for blanking cells
due to confidentiality reasons. Source OECD DynEmp database
are temporarily inactive or because they have permanently
exited. Sectors covered are: manufacturing, construction, and
non-financial business services. Owing to methodological
differences, figures may deviate from officially published
national statistics. For Japan data are at the establishment level,
for other countries at the firm level. Source OECD DynEmp
database. a Share of firms, b share of jobs affected
123
400 C. Criscuolo et al.

Fig. 5 Aggregate net job

variation of micro-firms by
period and age class. Note
the graph reports the total
net job variation (equal to
gross job creation minus
gross job destruction) of
micro-firms (below ten
employees) by period and
age class. It includes the
contribution from entrants
and exiting firms as well as
continuing firms. Aggregate
values across countries,
excluding those for which
data are not available in all
three periods (France, Japan,
Portugal, and the United
Kingdom) and the USA.
Source OECD DynEmp
database. a Number of
employees, b percentage of
total employment at the
beginning of each period

0.5 % of the total employment in aggregate (Fig. 5,
panel b). It is also noteworthy that the group of micro-
firms as a whole (i.e., including old firms) is always a
net job destroyer, even during the expansionary period
(2004–2007).
4 Econometric analysis
This section explores the descriptive findings illus-
trated so far in more detail, by a simple set of
regression models. The unit of observation is each
age class 9 macro-sector 9 country 9 time period
combination (or cell) available in the dataset for
the size class of 1–9 employees. The dependent
variables are the shares of the different transition
groups (moving above 10; staying below 10;
becoming inactive) and their contribution to net
job variation. The main independent variables are a
set of dummies for age categories and possibly
their interactions with dummies for time periods or
sectors.
More formally, these are the estimated equations:
yikct ¼ a þ b  logðavsizeÞikct þ ci þ dk þ uc þ st
þ eikct ð2Þ
where i indexes age classes, k macro-sectors, c coun-
tries, and t time periods, c is a set of age class
dummies, with age classes equal to 0–2; 3–5; 6–10;
and 11 or more. d is a set of macro-sector dummies

123
Do micro start-ups fuel job creation?
(manufacturing, non-financial business services, con-
struction). s is a set of dummies for each of the three
time period considered: 2001–2004, 2004–2007,
2007–2010. Finally, the variable size measures the
average size of the firms in the specific country-year-
size-age cell and is meant to control for the higher
probability of bigger firms to move up.
The dependent variable is either (1) the share of
firms moving up; (2) the share of firms staying in the
same size class; or (3) the share of firms becoming
inactive:
j contributions to net job creation, firms moving up
j Firmsikct
Shareikct ¼ ð3Þ
Firmsikct
j contribution of around 2 %, while firms turning
where Firmsikct is the number of firms moving up
(j ¼ upÞ, staying below ten employees (j ¼ stayÞ, or
becoming inactive (j ¼ outÞ at the end of the 3-year
period, and Firmsikct is the total number of firms in the
cell at the beginning of the period. The analogs of
these measures for job variation are also potential
dependent variables: (1) the net job contribution of
firms moving up; (2) of those staying in the same size
class; or (3) of those becoming inactive. Net job
contributions are defined as the net job variation of the
specific transition group over the total employment
(average of initial and final value) of all transition
groups in the same country, sector, period, age, and
size class:
j concerns the role of age: Younger firms are system-
j NJVikct
NJCikct ¼1
2 ðEmploymentikct þ Employmentikct1 Þ
ð4Þ
j
where NJVikct;tþ3 is net job variation coming over the
3-year period t from firms moving up (j ¼ upÞ, staying
below ten employees (j ¼ stayÞ, or becoming inactive
(j ¼ outÞ: Note that taking the average of the initial
and final value of employment as the normaliza-
tion term in Eq. (4) has the advantage of avoiding
the ‘‘regression to the mean’’ phenomenon, which
would automatically bias upwards the growth rates of
small firms (see more on this in Haltiwanger et al.
2013).
4.1 Main results for micro-firms (0–9)
The equation is estimated by a two-limit Tobit model
when the dependent variables are the shares of firms,
401
as those are bounded between zero and one. When the
dependent variable is instead the job contribution, a
traditional ordinary least squares (OLS) model is used.
Table 1 reports the summary statistics of the
regression sample relative to micro-firms (less than
ten employees at the beginning of the period). The
table shows that only around 5 % of firms grow above
the ten employees threshold over the 3-year period.
The bulk of firms within the cell—around 74 % on
average—are stable, while around 21 % of them
become inactive at the end of the period. In terms of
contribute positively for 14 % of the total job varia-
tion, on average; stable firms have also a positive
inactive negatively contribute for around 20 % of the
total job variation.
The baseline results from the estimation of Eq. (2)
are reported in Table 2. Note that the sum of the
estimated coefficients of all variables in columns 1, 3,
and 5, i.e., of the regressions with share of firms as
dependent variable, always sum up to zero by
construction, as the sum of the three shares is
obviously always equal to 100. This instead is not
true when summing the estimates in columns 2, 4, and
6, for the regressions with job contributions as
dependent variable, as in this case the total is not
equal to a constant value.
The first clear message from the regressions
atically more likely to move up. Firms in the age class
0–2 are on average four percentage points (p.p.) more
likely to reach or overcome the ten employees
threshold than firms aged 11 or more (the excluded
category in the regression). Firms in the age class 3–5
and 6–10 are two and one p.p. more likely to move up,
respectively, than firms in the 11? class.
Younger firms are also less likely to stay in the
same size class, and more likely to become inactive at
the end of the period. Namely, firms aged two or less
are 13 p.p. less likely to be stable than firms aged 11 or
more, while they are 9 p.p. more likely to be inactive at
the end of the period.
The higher probability of younger firms to move up
is also reflected in the analysis of the job variation
contribution. Among the group of firms that move up,
those in the age class 0–2 contribute to job variation by
15 p.p. more than firms in the age class 11?. The same
is true for firms aged 3–5 and 6–10, although the
123
402 C. Criscuolo et al.

Table 1 Descriptive statistics, size class 1–9 employees

Sector Period Job Job contr. Job contr. Share Share Share Average
contr. up stable inactive up stable inactive size

Manufacturing 2001–2004 No. obs. 55 55 55 55 55 55 55

Mean 0.16 0.02 -0.18 0.06 0.76 0.18 3.77
SD 0.13 0.06 0.09 0.03 0.10 0.08 0.79
p25 0.07 -0.03 -0.20 0.04 0.74 0.14 3.20
p50 0.13 0.01 -0.15 0.05 0.78 0.15 3.71
p75 0.21 0.06 -0.13 0.07 0.81 0.20 4.21
2004–2007 No. obs. 51 51 51 51 51 51 51
Mean 0.16 0.03 -0.19 0.06 0.74 0.20 3.47
SD 0.10 0.07 0.08 0.03 0.07 0.07 0.53
p25 0.09 -0.02 -0.22 0.04 0.70 0.16 3.03
p50 0.14 0.01 -0.18 0.05 0.75 0.20 3.41
p75 0.22 0.10 -0.13 0.07 0.78 0.23 3.88
2007–2010 No. obs. 51 51 51 51 51 51 51
Mean 0.12 -0.01 -0.23 0.05 0.72 0.23 3.59
SD 0.10 0.07 0.09 0.03 0.09 0.08 0.81
p25 0.05 -0.06 -0.27 0.03 0.67 0.17 3.09
p50 0.09 -0.02 -0.21 0.04 0.73 0.22 3.36
p75 0.17 0.03 -0.16 0.05 0.79 0.29 3.90
Services 2001–2004 No. obs. 51 51 51 51 51 51 51
Mean 0.13 0.03 -0.19 0.04 0.78 0.19 3.11
SD 0.07 0.05 0.08 0.01 0.08 0.07 0.47
p25 0.07 0.00 -0.23 0.03 0.74 0.13 2.76
p50 0.12 0.03 -0.18 0.04 0.78 0.18 3.10
p75 0.16 0.06 -0.14 0.05 0.83 0.22 3.59
2004–2007 No. obs. 52 52 52 52 52 52 52
Mean 0.14 0.04 -0.21 0.04 0.75 0.21 3.01
SD 0.07 0.06 0.08 0.01 0.07 0.07 0.46
p25 0.08 0.00 -0.25 0.03 0.70 0.16 2.64
p50 0.12 0.03 -0.20 0.04 0.76 0.20 2.93
p75 0.18 0.09 -0.14 0.05 0.80 0.26 3.34
2007–2010 No. obs. 48 48 48 48 48 48 48
Mean 0.11 0.01 -0.22 0.03 0.74 0.23 2.96
SD 0.06 0.07 0.08 0.01 0.08 0.07 0.38
p25 0.06 -0.04 -0.26 0.03 0.68 0.18 2.70
p50 0.10 0.00 -0.21 0.03 0.74 0.22 2.88
p75 0.15 0.05 -0.16 0.04 0.79 0.27 3.13
Constructions 2001–2004 No. obs. 51 51 51 51 51 51 51
Mean 0.14 0.04 -0.18 0.05 0.77 0.18 3.34
SD 0.11 0.07 0.14 0.04 0.13 0.12 0.70
p25 0.07 -0.01 -0.18 0.03 0.76 0.12 2.74
p50 0.10 0.03 -0.15 0.04 0.80 0.15 3.30
p75 0.16 0.08 -0.11 0.06 0.83 0.19 3.85

123
Do micro start-ups fuel job creation? 403

Table 1 continued
Sector Period Job Job contr. Job contr. Share Share Share Average
contr. up stable inactive up stable inactive size

2004–2007 No. obs. 52 52 52 52 52 52 52

Mean 0.18 0.06 -0.19 0.06 0.73 0.21 3.20
SD 0.17 0.07 0.08 0.04 0.11 0.09 0.65
p25 0.08 0.00 -0.23 0.03 0.70 0.16 2.74
p50 0.12 0.04 -0.18 0.05 0.75 0.19 3.10
p75 0.19 0.10 -0.14 0.08 0.79 0.23 3.71
2007–2010 No. obs. 47 47 47 47 47 47 47
Mean 0.14 0.00 -0.23 0.05 0.72 0.24 3.16
SD 0.18 0.09 0.08 0.04 0.11 0.09 0.57
p25 0.05 -0.07 -0.30 0.02 0.67 0.17 2.78
p50 0.08 -0.02 -0.02 0.03 0.73 0.22 3.04
p75 0.11 0.05 -0.16 0.05 0.79 0.30 3.43
Total No. obs. 458 458 458 458 458 458 458
Mean 0.14 0.02 -0.20 0.05 0.74 0.21 3.30
SD 0.12 0.07 0.09 0.03 0.10 0.09 0.66
p25 0.07 -0.02 -0.24 0.03 0.70 0.15 2.80
p50 0.11 0.01 -0.18 0.04 0.76 0.19 3.18
p75 0.17 0.07 -0.14 0.06 0.81 0.25 3.69

difference is quantitatively much smaller (equal to five
and three p.p., respectively). This implies that, among
younger firms, the contribution of firms moving up to
net job growth is relatively more important than
among older firms. A back-of-the-envelope calcula-
tion aimed at quantifying the net job creation of the
youngest micro-firms suggests that in the regression
sample they contributed on average by about 12
million jobs more than firms aged 11 or above. To put
this number into perspective, note that altogether over
the entire period, micro-firms destroyed about 3.3
million and all firms about 17 million jobs.
The results on the relative job contribution also show
that, among stable firms, i.e., firms that do not overcome
the threshold of ten employees over the 3-year period,
younger firms are growing much more than older firms.
This can be deducted from the positive coefficients on
net job contribution on age class dummies. Put differ-
ently, although there are fewer stable firms in younger
age classes, they create relatively more employment (or
destroy less) than firms aged 11 or more.
Table 2 also shows interesting differences across
time periods, enriching the patterns presented in Fig. 3.
In the expansion period (2004–2007), there was a tiny—
but statistically significant—increase in the number of
firms moving up, and a larger increase—of around three
p.p.—in the number of inactive firms, counterbalanced
by a contraction of around four p.p. of the share of
stable firms. In the period mostly dominated by the
Great Recession (2007–2010), there was a marginal
contraction of the share of firms moving up, and a more
significant decrease in the share of stable firms, mirrored
by a substantial increase (four p.p.) in the share of firms
becoming inactive. The decrease in the share of firms
moving up and of stable firms is also reflected by a
correspondent decline in their respective job creation;
symmetrically, the increase in the share of firms
becoming inactive is reflected by a negative coefficient
on the job contribution of inactive firms during the crisis
(the net job contribution of firms going inactive is
always non-positive by definition, thus the negative
coefficient implies that they destroy more jobs).
Last but not least, the sum of the three coefficients on
the different age class dummies across the three
transition groups gives the total contribution to net job
creation with respect to the excluded category, i.e., the
11? age class. The value is equal to around 15 p.p. for
the age class 0–2, and it is very close to zero for the 3–5
and 6–10 age classes. This therefore confirms the
finding of Haltiwanger et al. (2013) and of Criscuolo
(2014a, b), among others, that young firms are respon-
sible for most—if not all—net job creation. Table 9 in

123
404 C. Criscuolo et al.

Table 2 Regression results, baseline estimates: initial size class: 0–9 employees
1 2 3 4 5 6
Transition group Share of moving up Share of stable Share of inactive
Measure Firms Job contr. Firms Job contr. Firms Job contr.

Log average size 0.117*** 0.053 -0.192*** -0.173*** 0.075** -0.055

(0.016) (0.048) (0.029) (0.026) (0.030) (0.038)
Services -0.002 -0.023* -0.021*** -0.012** 0.023*** -0.026***
(0.003) (0.012) (0.006) (0.005) (0.006) (0.008)
Construction 0.009*** -0.001 -0.027*** -0.000 0.018*** -0.017**
(0.003) (0.011) (0.006) (0.005) (0.006) (0.008)
Period 2004–2007 0.008*** 0.013 -0.041*** 0.009** 0.033*** -0.017***
(0.002) (0.009) (0.005) (0.004) (0.005) (0.006)
Period 2007–2010 -0.004** -0.025*** -0.048*** -0.037*** 0.053*** -0.042***
(0.002) (0.009) (0.005) (0.005) (0.005) (0.007)
Age 0–2 0.046*** 0.154*** -0.136*** 0.098*** 0.091*** -0.104***
(0.005) (0.016) (0.010) (0.009) (0.010) (0.013)
Age 3–5 0.021*** 0.053*** -0.096*** 0.039*** 0.075*** -0.086***
(0.003) (0.011) (0.007) (0.006) (0.006) (0.008)
Age 6–10 0.014*** 0.029*** -0.047*** 0.018*** 0.033*** -0.037***
(0.003) (0.010) (0.006) (0.005) (0.005) (0.006)
Constant -0.119*** 0.020 1.083*** 0.208*** 0.035 -0.040
(0.024) (0.073) (0.042) (0.038) (0.044) (0.056)
Country dummies Yes Yes Yes Yes Yes Yes
Number of obs. 458 458 458 458 458 458
Robust standard error in parenthesis. The excluded categories are macro-sector manufacturing; age class 11?; period 2001-2004.
The regressions results reported in column 1, 3, and 5 are estimated with a Tobit model; the regressions results reported in column 2,
4, and 6 are estimated with OLS
***p \ 0.01; **p \ 0.05; *p \ 0.1

Appendix shows that differences across time periods do
not tend to have a marked sectoral component, as sector-
period interaction dummies turn out to be not significant
in most cases.
4.1.1 Period–age interactions
Table 3 further explores the inter-temporal differ-
ences, assessing whether the different age groups are
showing distinctive patters over the three time period
by means of interacted period and age class dummies.
The excluded (baseline) category in this case is the
11? age group in the first (2001–2004) period. As the
manufacturing dummy and the first-period dummy are
also still excluded, the reference group for the
interacted dummies is the 11? age group in the
manufacturing sector in the first period.
The results show that the overall pattern that younger
firms are more likely to move above ten employees and
create more jobs holds robustly in each of the three
periods, even including the crisis. In fact, there are only a
few significant coefficients among the interacted age-
time period variables which would indicate a change in
this phenomenon. The most noteworthy among them is
that the youngest (0–2) group of stable firms creates even
more jobs—as compared to older firms in the same
period—both during the expansion period (2004–2007)
and the crisis period (2007–2010) compared to the initial
period (2001–2004). Further, the oldest age class still
below 10 years old (6–10) shows a relative increase in
the firms becoming inactive during the crisis, at the cost
of the stable firms. All in all, however, the most important
finding from Table 3 is that, across all age groups, the
share of firms moving up and their contribution to net job
creation is remarkably stable across time.

123
Do micro start-ups fuel job creation? 405

Table 3 Regression results, period–age interactions: initial size class: 0–9 employees
1 2 3 4 5 6
Transition group Share up Share stable Share inactive
Measure Firms Job contr. Firms Job contr. Firms Job contr.

Log average size 0.117*** 0.050 -0.206*** -0.192*** 0.090*** -0.065

(0.017) (0.050) (0.030) (0.026) (0.031) (0.041)
Services -0.002 -0.023** -0.023*** -0.015*** 0.025*** -0.027***
(0.003) (0.012) (0.006) (0.005) (0.006) (0.008)
Construction 0.009*** -0.001 -0.028*** -0.002 0.019*** -0.018**
(0.003) (0.011) (0.006) (0.005) (0.006) (0.008)
Period 2004–2007 0.004 -0.000 -0.032*** -0.003 0.028*** -0.015
(0.003) (0.013) (0.009) (0.006) (0.008) (0.010)
Period 2007–2010 -0.005 -0.027** -0.041*** -0.047*** 0.046*** -0.032***
(0.004) (0.014) (0.009) (0.009) (0.008) (0.010)
Age 0–2 0.043*** 0.140*** -0.148*** 0.071*** 0.106*** -0.116***
(0.008) (0.022) (0.016) (0.011) (0.016) (0.024)
Age 3–5 0.021*** 0.051*** -0.084*** 0.034*** 0.064*** -0.075***
(0.005) (0.015) (0.011) (0.008) (0.011) (0.013)
Age 6–10 0.011** 0.022 -0.031*** 0.015** 0.020** -0.024**
(0.004) (0.014) (0.011) (0.007) (0.010) (0.011)
Period 2004–2007 9 age 0–2 0.007 0.025 0.009 0.030*** -0.016 0.019
(0.006) (0.024) (0.015) (0.011) (0.015) (0.020)
Period 2004–2007 9 age 3–5 0.002 0.011 -0.024* 0.003 0.022* -0.014
(0.005) (0.023) (0.013) (0.009) (0.013) (0.015)
Period 2004–2007 9 age 6–10 0.008 0.017 -0.025* 0.009 0.018 -0.015
(0.005) (0.020) (0.013) (0.008) (0.011) (0.012)
Period 2007–2010 9 age 0–2 0.002 0.013 0.017 0.038*** -0.019 0.011
(0.007) (0.025) (0.015) (0.014) (0.016) (0.022)
Period 2007–2010 9 age 3–5 -0.003 -0.006 -0.018 0.003 0.020* -0.025
(0.005) (0.022) (0.013) (0.012) (0.012) (0.015)
Period 2007–2010 9 age 6–10 0.002 0.003 -0.028** -0.005 0.026** -0.027**
(0.005) (0.021) (0.013) (0.012) (0.011) (0.012)
Constant -0.125*** 0.008 1.089*** 0.199*** 0.036 -0.042
(0.024) (0.073) (0.042) (0.038) (0.043) (0.055)
Country dummies Yes Yes Yes Yes Yes Yes
Number of obs. 458 458 458 458 458 458
Robust standard error in parenthesis. The excluded categories are macro-sector manufacturing; age class 11?; period 2001–2004.
The regressions which results are reported in column 1, 3, and 5 are estimated with a Tobit model; the regressions which results are
reported in column 2, 4, and 6 are estimated with OLS
***p \ 0.01; **p \ 0.05; *p \ 0.1

4.1.2 Sector–age interactions
Table 4 assesses whether young firms show different
transition patterns across sectors. This is done by adding
interactions of the sector dummies with the age category
dummies to the right-hand side of the regression
equation. Again, as the manufacturing dummy and the
first-period dummy are also still excluded, the reference
group for the interacted dummies is the 11? age group
in the manufacturing sector in the first period.
The results show that younger firms (age 0–2) in the
construction sector are less dynamic than in

123
406 C. Criscuolo et al.

Table 4 Regression results, age–sector interactions: Initial SIZE class: 0–9 employees
1 2 3 4 5 6
Transition group Share up Share stable Share inactive
Measure Firms Job contr. Firms Job contr. Firms Job contr.

Log average size 0.119*** 0.054 -0.196*** -0.170*** 0.078*** -0.058

(0.016) (0.049) (0.029) (0.026) (0.030) (0.038)
Services 0.006* -0.008 -0.027*** 0.000 0.021** -0.021*
(0.004) (0.014) (0.009) (0.008) (0.009) (0.011)
Construction 0.016*** 0.017 -0.023** 0.001 0.007 0.000
(0.004) (0.016) (0.009) (0.008) (0.007) (0.008)
Period 2004–2007 0.008*** 0.014 -0.042*** 0.009** 0.033*** -0.017***
(0.002) (0.009) (0.005) (0.004) (0.005) (0.006)
Period 2007–2010 -0.004** -0.025*** -0.048*** -0.036*** 0.053*** -0.042***
(0.002) (0.009) (0.005) (0.005) (0.005) (0.007)
Age 0–2 0.060*** 0.194*** -0.137*** 0.106*** 0.077*** -0.081***
(0.006) (0.020) (0.011) (0.011) (0.011) (0.014)
Age 3–5 0.023*** 0.057*** -0.098*** 0.047*** 0.075*** -0.084***
(0.004) (0.014) (0.008) (0.008) (0.008) (0.010)
Age 6–10 0.017*** 0.028** -0.047*** 0.022*** 0.030*** -0.033***
(0.004) (0.013) (0.009) (0.007) (0.007) (0.008)
Construction 9 age 0-2 -0.023*** -0.056*** 0.013 -0.021* 0.009 -0.017
(0.005) (0.021) (0.013) (0.011) (0.012) (0.015)
Construction 9 age 3-5 -0.003 -0.005 0.006 -0.018** -0.003 0.001
(0.004) (0.020) (0.012) (0.009) (0.012) (0.014)
Construction 9 age 6–10 -0.005 -0.001 0.002 -0.010 0.003 -0.004
(0.004) (0.018) (0.012) (0.009) (0.010) (0.011)
Services 9 age 0–2 -0.020*** -0.066** -0.015 0.001 0.035** -0.057***
(0.007) (0.026) (0.015) (0.014) (0.015) (0.021)
Services 9 age 3–5 -0.003 -0.005 -0.000 -0.004 0.004 -0.007
(0.006) (0.025) (0.013) (0.012) (0.012) (0.014)
Services 9 age 6–10 -0.004 0.002 -0.002 0.000 0.006 -0.008
(0.006) (0.022) (0.013) (0.011) (0.011) (0.011)
Constant -0.125*** 0.008 1.089*** 0.199*** 0.036 -0.042
(0.024) (0.073) (0.042) (0.038) (0.043) (0.055)
Country dummies Yes Yes Yes Yes Yes Yes
Number of obs. 458 458 458 458 458 458
Robust standard error in parenthesis. The excluded categories are macro-sector manufacturing; age class 11?; period 2001–2004.
The regressions which results are reported in column 1, 3, and 5 are estimated with a Tobit model; the regressions which results are
reported in column 2, 4, and 6 are estimated with OLS
***p \ 0.01; **p \ 0.05; *p \ 0.1

manufacturing, as the share of moving-up firms is 2
p.p. lower. The same is true for the services sector, for
which in addition there is also a relatively higher share
of start-ups becoming inactive. Both dynamics are also
reflected in the job contributions of moving-up firms,
with micro start-ups in the construction and services
sector contributing around 5–6 p.p. less to net job
creation than in the manufacturing sector. All in all,
these results point to higher incidence of fast-growing
firms in manufacturing, and to a higher inactivity rate
at the end of the 3-year periods in the service sector.
This may be explained by the fact that there are more

123
Do micro start-ups fuel job creation? 407

Table 5 Descriptive Stats Share up Share stable Share inactive Share down Log av. size
statistics, initial size class
10–49 No. obs. 458 458 458 458 458
Mean 0.033 0.645 0.163 0.159 2.942
SD 0.022 0.110 0.086 0.049 0.085
p25 0.020 0.578 0.103 0.124 2.880
p50 0.029 0.658 0.145 0.160 2.942
p75 0.040 0.719 0.203 0.189 3.007
Stats Job contr. up Job contr. stable Job contr. inactive Job contr. down

No. obs. 458 458 458 458

Mean 0.074 0.046 -0.178 -0.077
SD 0.091 0.051 0.110 0.038
p25 0.028 0.012 -0.215 -0.098
p50 0.048 0.045 -0.154 -0.072
p75 0.088 0.073 -0.105 -0.048

upscaling possibilities in manufacturing due to easier
tradability of its output than in services. It also
suggests that the rate of failure among starting
businesses is higher in services.
4.2 Comparison with small firms (10–49)
In this section, we compare the results obtained for the
size class 1–9 with the results obtained by estimating
exactly the same specification on the size class 10–49.
Compared with the size class 1–9, the size class 10–49
is characterized by a lower share of firms moving up,
although this is to some extent probably an immediate
result from the latter size class being much wider (see
Table 5).
Another substantial difference with respect to the
size class 1–9 is that now firms can also ‘‘move down’’,
i.e., they can go below the ten employees threshold.
The results from the estimation of Eq. (2) on the size
10–49 sample for the baseline model are reported in
Table 6. The new set of results is remarkably similar;
in particular, younger firms are still significantly more
likely to move to a higher size class (in this case, above
the 49 employees threshold) at the end of the 3-year
period. Furthermore, younger firms are less likely to
be stable and to move down and are substantially more
likely to become inactive.
The results with contributions to net job variation
by the different transition groups as dependent vari-
ables basically mirror those with shares of firms and
are again aligned with the correspondent results from
the sample with the 1–9 size class. Younger firms
moving up—especially those less than three years
old—contribute significantly more to job creation than
older firms in the same transition group. Symmetri-
cally, younger firms becoming inactive contribute
much more to job destruction than older firms sharing
the same trajectory. Among stable firms, younger
firms show a higher net job contribution than older
firms, similarly to the 1–9 size class.
4.3 Other robustness tests
Although the results are remarkably stable and con-
sistent across specifications, we run a couple of
additional robustness test, which are related to differ-
ent temporal coverage across countries and to confi-
dentiality blanking for the USA, as mentioned above.
The first test consists in excluding France from the
sample, as data are unavailable for the last period
(2007–2010). The results—not reported here but
available from the authors upon request—are almost
identical to the baseline estimations.
The second test consists in adding to the sample
also the USA, excluded from the main regressions due
to the substantial and erratic blanking of around 30 %
of cells. Those results are reported in Table 7. The
main conclusions are confirmed, but generally coef-
ficients are lower in absolute values, consistently with
an attenuation bias originating from a standard mea-
surement error. This results in a nonsignificant differ-
ence between the share of jobs destroyed by young

123
408 C. Criscuolo et al.

Table 6 Regression results, baseline estimates, small firms: initial size class: 10–49 employees
1 2 3 4 5 6 7 8
Transition group Share up Share stable Share inactive Share down
Measure Firms Job contr. Firms Job contr. Firms Job contr. Firms Job contr.

Log average size 0.206*** 0.410*** 0.109 -0.180*** -0.168*** 0.316*** -0.132*** 0.067**
(0.022) (0.091) (0.078) (0.036) (0.062) (0.088) (0.039) (0.031)
Services 0.017*** 0.053*** -0.035*** -0.001 -0.006 0.020* 0.027*** -0.018***
(0.002) (0.011) (0.009) (0.004) (0.008) (0.011) (0.005) (0.003)
Construction 0.012*** 0.040*** -0.040*** 0.004 0.005 0.000 0.024*** -0.017***
(0.003) (0.014) (0.011) (0.005) (0.009) (0.013) (0.006) (0.004)
Period 0.008*** 0.002 0.010 0.021*** -0.009* 0.017** -0.009*** 0.005**
2004–2007 (0.002) (0.010) (0.007) (0.004) (0.005) (0.007) (0.004) (0.003)
Period -0.002 -0.013 -0.033*** -0.023*** 0.015** -0.024*** 0.021*** -0.014***
2007–2010 (0.002) (0.010) (0.007) (0.004) (0.006) (0.009) (0.004) (0.003)
Age 0–2 0.032*** 0.106*** -0.161*** 0.054*** 0.121*** -0.129*** 0.008 -0.025***
(0.003) (0.017) (0.010) (0.006) (0.008) (0.011) (0.005) (0.004)
Age 3–5 0.021*** 0.052*** -0.097*** 0.036*** 0.064*** -0.065*** 0.011** -0.019***
(0.003) (0.015) (0.010) (0.005) (0.007) (0.010) (0.005) (0.004)
Age 6–10 0.013*** 0.029** -0.052*** 0.020*** 0.028*** -0.024*** 0.011** -0.012***
(0.002) (0.013) (0.008) (0.004) (0.006) (0.008) (0.005) (0.004)
Constant -0.609*** -1.243*** 0.475** 0.546*** 0.591*** -1.045*** 0.496*** -0.223**
(0.069) (0.288) (0.240) (0.112) (0.192) (0.270) (0.123) (0.095)
Country Yes Yes Yes Yes Yes Yes Yes Yes
dummies
Number of obs. 458 458 458 458 458 458 458 458
Robust standard error in parenthesis. The excluded categories are macro-sector manufacturing; age class 11?; period 2001–2004.
The regressions which results are reported in column 1, 3, and 5 are estimated with a Tobit model; the regressions which results are
reported in column 2, 4, and 6 are estimated with OLS
***p \ 0.01; **p \ 0.05; *p \ 0.1

firms becoming inactive and old firms becoming
inactive. The significantly higher upscaling and job
creation effects are well preserved, however.
5 Conclusions
In this paper, we present new evidence on the growth
dynamics of micro-firms. The analysis builds on the
DynEmp Express database, recently assembled by the
OECD with the support of a network of experts in 18
countries. In particular, this paper exploits a specific
section of the database reporting information on the
transition dynamics of firms of different size, age, and
sectors over the period 2001–2004, 2004–2007, and
2007–2010.
The focus of the empirical analysis is on micro-
firms, i.e., on firms below ten employees at the
beginning of each of the 3-year period. This choice is
justified both by the importance of this group of firms
in the economy, and by the higher data quality in the
DynEmp database due to the large size of the reference
population, which substantially reduces the incidence
of confidentiality blanking.
The main finding of the paper relates to the age
dimension: Young micro-firms—especially those
below 3 years of age—are much more likely to
overcome the ten employees threshold than older
firms. For instance, firms \3 years old are 13–14 p.p.
less likely to employ still less than ten workers at the
end of the 3-year period (i.e., to belong to the ‘‘stable’’
category) than firms aged 11 or more and on the other
hand, the former are 9 p.p. more likely to be inactive at
the end of the period than the latter. This pattern
confirms that the ‘‘up-or-out’’ dynamics identified for
the USA hold for a much wider set of economies.

123
Do micro start-ups fuel job creation? 409

Table 7 Regression results, baseline estimates including US: initial size class 0–9
1 2 3 4 5 6
Transition group Share up Share stable Share inactive
Measure Firms Job contr. Firms Job contr. Firms Job contr.

Log average size 0.133*** 0.074 0.001 -0.185*** -0.129 0.418**

(0.016) (0.046) (0.086) (0.025) (0.085) (0.186)
Services 0.001 -0.020* 0.034* -0.014*** -0.033* 0.089**
(0.003) (0.011) (0.019) (0.005) (0.019) (0.041)
Construction 0.008** -0.005 -0.004 -0.005 -0.003 0.022
(0.003) (0.011) (0.014) (0.005) (0.014) (0.030)
Period 2004–2007 0.008*** 0.012 -0.025** 0.006 0.017 0.018
(0.003) (0.009) (0.012) (0.004) (0.011) (0.024)
Period 2007–2010 -0.004* -0.026*** -0.036*** -0.039*** 0.040*** -0.019
(0.002) (0.009) (0.011) (0.004) (0.011) (0.024)
Age 0–2 0.046*** 0.148*** -0.085*** 0.090*** 0.040 0.005
(0.005) (0.015) (0.030) (0.008) (0.030) (0.066)
Age 3–5 0.022*** 0.054*** -0.060*** 0.035*** 0.039* -0.003
(0.004) (0.011) (0.022) (0.006) (0.021) (0.045)
Age 6–10 0.013*** 0.024** -0.017 0.016*** 0.004 0.021
(0.003) (0.010) (0.017) (0.005) (0.016) (0.033)
Constant -0.139*** -0.002 0.785*** 0.230*** 0.347*** -0.746***
(0.024) (0.069) (0.129) (0.037) (0.126) (0.278)
Country dummies Yes Yes Yes Yes Yes Yes
Number of obs. 492 492 492 492 492 492
Robust standard error in parenthesis. The excluded categories are macro-sector manufacturing; age class 11?; period 2001-4. The
regressions which results are reported in column 1, 3, and 5 are estimated with a Tobit model; the regressions which results are
reported in column 2, 4, and 6 are estimated with OLS
***p \ 0.01; **p \ 0.05; *p \ 0.1

This paper also contributes to the debate of the
effects of recessions. Our analysis shows that during
and in the aftermath of the Great Recession (i.e., over
the period 2007–2010) there was a marginal contrac-
tion of the share of firms moving up, and a more
significant contraction of the share of stable firms,
counterbalanced by a substantial increase in the share
of firms becoming inactive. Importantly, this pattern is
substantially undifferentiated across age groups.
This study opens a number of avenues for future
research. First, it would be interesting to explore these
dynamics at a more disaggregated sectoral level (e.g.,
two-digit industries) in order to assess whether indus-
try-specific characteristics—e.g., capital intensity,
natural barrier to entries, innovation activity—are
linked to higher and lower dynamism and to different
transition dynamics. Second, it would be extremely
informative to better characterize the scaling-up pro-
cess of cohorts of small start-ups over a longer time
horizon across countries, in order to assess whether the
‘‘up-or-out’’ process documented in great detail for the
USA also holds in other countries. Finally, since we
know that fast-growing start-ups contribute dispropor-
tionally to job creation, it would be of paramount
importance to better gauge how national polices and
framework conditions affect their birth and growth
pattern.
Acknowledgments The authors wish to thank the following
people for making the DynEmp project possible: Mariagrazia
Squicciarini, Javier Miranda, Werner Hölzl, Hilde Spinnewyn,
Chantal Kegels, Michel Dumont, Gabriel Lopes de Ulyssea,
Carlos Henrique Leite Corseuil, Fernanda de Negri, Pierre
Therrien, Jay Dixon, Anne-Marie Rollin, John Baldwin, Mika
Maliranta, Lionel Nesta, Flora Bellone, Adrienn Szep,

123
410 C. Criscuolo et al.

Szollosine, Erzsebet Eperjesi Lindnerne, Gabor Katay, Peter indebted to Nick Johnstone, Dirk Pilat, and Andy Wyckoff for
Harasztosi, Stefano Costa, Kyoji Fukao, Kenta Ikeuchi, Leila comments on an earlier draft, and to two anonymous referees for
Peltier—Ben Aoun, Anne Dubrocard, Michael Prombo, very useful suggestions.
Michael Polder, Lynda Sanderson, Richard Fabling, Arvid
Raknerud, Eivind Lorentzen, Jorge Portugal, Valentin Llorente
Garcia, Luis Esteban Barbado Miguel, Eva Hagsten, Chiara Appendix: Additional tables
Rosazza-Bondibene, Catherine Robinson, Albert Bravo-Biosca,
Steve Bond, and James Stainer. The financial contribution of See Tables 8 and 9.
NESTA (UK) is gratefully acknowledged. The authors are also

Table 8 Data coverage of the DynEmp Express database

Country Period covered Note

Austria 2001–2010 Companies have the choice of reporting either at the enterprise
or at the establishment level see Stiglbauer (2003)
Belgium 2001–2011
Brazil 2002–2010
Canada 2001–2011 Only a subset of statistics has been produced and therefore the
country could not be included in all graphs; see graph notes
for further detail
Finland 2001–2011
France 2002–2007
Hungary 2001–2011
Italy 2001–2010
Japan 2001–2009 Sectoral coverage is limited to manufacturing in the case of
yearly job flow data for most years in the sample. Data are at
establishment level
Luxembourg 2001–2010
Netherlands 2001–2011 Results are preliminary and data might be affected by breaks in
the longitudinal structure of the business register
Norway 2001–2010
New Zealand 2001–2009
Portugal 2006–2011
Spain 2003–2009 No transition matrices available
Sweden 2001–2010
United Kingdom 2001–2011 No transition matrices available for the 2004–2007 period
USA 2001–2011

Table 9 Regression results, period–sector interactions: initial size class: 0–9 employees
1 2 3 4 5 6
Transition group Share up Share stable Share inactive
Measure Firms Job contr. Firms Job contr. Firms Job contr.

Log average size 0.118*** 0.054 -0.192*** -0.173*** 0.075** -0.055

(0.016) (0.048) (0.029) (0.027) (0.030) (0.038)
Services -0.003 -0.035* -0.025*** -0.012* 0.028*** -0.031***
(0.004) (0.018) (0.009) (0.007) (0.008) (0.011)
Construction 0.007 -0.028* -0.023** 0.000 0.016 -0.022
(0.005) (0.017) (0.011) (0.007) (0.012) (0.016)

123
Do micro start-ups fuel job creation? 411

Table 9 continued
1 2 3 4 5 6
Transition group Share up Share stable Share inactive
Measure Firms Job contr. Firms Job contr. Firms Job contr.

Period 2004–2007 0.007** -0.006 -0.043*** 0.009 0.036*** -0.024***

(0.004) (0.016) (0.007) (0.006) (0.007) (0.009)
Period 2007–2010 -0.007* -0.045*** -0.046*** -0.036*** 0.053*** -0.046***
(0.004) (0.015) (0.007) (0.006) (0.007) (0.009)
Age 0–2 0.046*** 0.154*** -0.136*** 0.098*** 0.091*** -0.104***
(0.005) (0.016) (0.010) (0.009) (0.010) (0.014)
Age 3–5 0.021*** 0.053*** -0.096*** 0.039*** 0.075*** -0.086***
(0.003) (0.011) (0.007) (0.006) (0.006) (0.008)
Age 6–10 0.014*** 0.029*** -0.047*** 0.018*** 0.033*** -0.037***
(0.003) (0.010) (0.006) (0.005) (0.005) (0.006)
Construction 9 period 2004-2007 -0.002 0.014 0.008 -0.005 -0.006 0.006
(0.004) (0.020) (0.011) (0.008) (0.010) (0.012)
Construction 9 period 2007-2010 0.005 0.023 0.003 0.004 -0.009 0.011
(0.004) (0.018) (0.011) (0.009) (0.010) (0.012)
Services 9 period 2004-2007 0.005 0.044** -0.003 0.005 -0.002 0.013
(0.006) (0.022) (0.013) (0.010) (0.013) (0.017)
Services 9 period 2007-2010 0.001 0.038* -0.010 -0.007 0.009 0.001
(0.006) (0.023) (0.013) (0.012) (0.013) (0.018)
Constant -0.118*** 0.032 1.083*** 0.207*** 0.034 -0.037
(0.024) (0.074) (0.042) (0.038) (0.043) (0.056)
Country dummies Yes Yes Yes Yes Yes Yes
Number of obs. 458 458 458 458 458 458
Robust standard error in parenthesis. The excluded categories are macro-sector manufacturing; age class 11?; period 2001–2004.
The regressions which results are reported in column 1, 3, and 5 are estimated with a Tobit model; the regressions which results are
reported in column 2, 4, and 6 are estimated with OLS
***p \ 0.01; **p \ 0.05; *p \ 0.1

References
Andrews, D., Bartelsman, E., & Criscuolo C. (2015). Firm
dynamics and productivity growth in Europe. In OECD
working paper, Forthcoming.
Andrews, D., Criscuolo, C., & Menon, C. (2014). Do resources
flow to patenting firms? Cross-country evidence from
microdata. In OECD economics department working
paper, no. 1127.
Anyadike-Danes, M., Bjuggren, C. M., Gottschalk, S., Hölzl,
W., Johansson, D., Maliranta, M., & Grinde Myrann, A.
(2013). Accounting for job growth: disentangling size and
age effects in an international cohort comparison. In IFN
Working paper, no. 1019. Stockholm: Research Institute of
Industrial Economics (IFN).
Bartelsman, E., Haltiwanger, J., & Scarpetta S. (2004).
Microeconomic evidence of creative destruction in indus-
trial and developing Countries. In The World Bank, policy
research working paper series (p. 3464).
Criscuolo, C., Gal, P. N., & Menon, C. (2014a). The dynamics of
employment growth: new evidence from 18 Countries,
OECD science. In Technology and industry policy papers,
no. 14, OECD Publishing.
Criscuolo, C., Gal, P. N., & Menon, C. (2014b). DynEmp: a
StataÒ routine for distributed micro-data analysis of busi-
ness dynamics. In OECD science, technology and industry
working papers 2014/2, OECD Publishing.
Davis, S. J., Faberman, R. J., & Haltiwanger, J. (2006). The flow
approach to labor markets: new data sources and micro-
macro links. Journal of Economic Perspectives, 20(3), 3–26.
Davis, S. J., Faberman, R. J., & Haltiwanger, J. (2012). Labor
market flows in the cross section and over time. Journal of
Monetary Economics. Carnegie-NYU-Rochester confer-
ence series on public policy at New York University on
April 15–16, 2011, 59(1), 1–18.
Davis, S. J., & Haltiwanger, J. (1992). Gross job creation, gross
job destruction, and employment reallocation. The Quar-
terly Journal of Economics, 107(3), 819–863.

123
412
Davis, S. J., & Haltiwanger, J. (1999). Gross job flows. In O.
Ashenfelter & D. Card (Eds.), Handbook of labor eco-
nomics (1 ed., Vol. 3, pp. 2711–2805) (chapter 41).
Amsterdam, The Netherlands: Elsevier.
Foster, L., Grim, C., & Haltiwanger, J. (2014). Reallocation in
the great recession: Cleansing or not? In A. Mas & D. Card
(Eds.), The labor market in the aftermath of the great
recession. Washington: The National Press Club.
Geurts, K., & Van Biesebroeck, J. (2014). Job creation, firm
creation, and de novo entry. In CEPR discussion paper no.
DP10118.
Haltiwanger, J., Jarmin, R. S., & Miranda, J. (2013). Who cre-
ates jobs? Small versus large versus young. The Review of
Economics and Statistics, 95(2), 347–361.
123
C. Criscuolo et al.
Hsieh, C.-T., & Klenow, P. J. (2012). The life cycle of plants in
India and Mexico. National Bureau of Economic Research,
Working paper no. 18133.
Lawless, M. (2014). Age or size? Determinants of job creation.
Small Business Economics, 42(4), 815–830.
Ouyang, M. (2009). The scarring effect of recessions. Journal of
Monetary Economics, 56(2), 184–199.
Schumpeter, J. (1942). Capitalism, socialism, and democracy.
New York: Harper & Bros.
Stiglbauer, A., Stahl, F., Winter-Ebmer, R., & Zweimüller, J.
2003. Job creation and job destruction in a regulated labor
market: the case of Austria. Empirica, 30(2), 127–148.