Abstract

We present a newly-developed database that measures the distribution of firm growth across twelve countries, building on confidential microdata from business registers in the US and Europe that cover the universe of firms. We document not just average growth but also the growth rate for all the percentiles of the growth distribution, breaking it down by size, sector and age. We identify a number of stylized facts that hold across countries, but also show that there are significant differences in their growth distributions. Firms in the US grow and shrink more rapidly than in Europe, which has a much larger share of static firms, for which employment does not vary much (up or down). Finally, we find that a more dynamic growth distribution is associated with faster productivity growth, and this relationship becomes stronger as countries converge to the global technology frontier (specifically, a 5pp higher share of static firms is associated with 1pp lower productivity growth).

JEL Classification: D22, L11, O49 Keywords: growth, firm dynamics, productivity

The data in this paper was collected as a part of a joint Nesta-FORA project in collaboration with Henrik Lynge Hansen, Glenda Napier and Ditte Petersen, and with support from the International Consortium for Entrepreneurship (ICE). I am grateful to Alex Coad, Chiara Criscuolo and Mariarosa Lunati, as well as conference participants at the AEA Annual Meetings 2010, Kauffman Foundation, Bruegel, EU Concord and the OECD for very valuable comments. This initiative would not have been possible without the generous collaboration of many researchers and statistical agencies in the participating countries that provided the data underlying this database. For this I would like to thank Werner Hölzl (Austria), Sonja Djukic, Chris Johnston and Chris Parsley (Canada), Henrik Lynge Hansen (Denmark), Henri Kahonen, Petri Rouvinen and Mika Pajarinen (Finland), Stavroula Maroulaki and Theano Tyfoxylou (Greece), Patrizia Cella and Caterina Viviano (Italy), Rico Konen (Netherlands), Geoff Mead (New Zealand), Svein Myro and Christian L. Wold Eide (Norway), Valentín Llorente Garcia (Spain), David Brown, Ronald Davis and Javier Miranda (US) and, last but not least Michael Anyadike-Danes and Mark Hart (UK). Joseph Alberti and Rodrigo Lluberas provided excellent research assistance. The views expressed herein are those of the author and do not necessarily reflect the views of Nesta. Correspondence address: Nesta, 1 Plough Place, EC4A 1DE, London, UK. Email: [email protected].

The Nesta Working Paper Series is intended to make available early results of research undertaken or supported by Nesta and its partners in order to elicit comments and suggestions for revisions and to encourage discussion and further debate prior to publication (ISSN 2050-9820). Year 2016 by Nesta. Short sections of text, tables and figures may be reproduced without explicit permission provided that full credit is given to the source. The views expressed in this working paper are those of the author(s) and do not necessarily represent those of Nesta.

1 Introduction

Creative destruction is one of the driving forces of economic growth. ¹ But despite the importance of firm growth dynamics in this process, there is limited cross-country comparable data on them. This paper aims to fill this gap. We collect harmonized micro aggregated data from official business registers, which provide quasi-universal coverage of business activity in all sectors of the economy, and develop a new database of firm growth distributions for 12 countries, including the US and several small and large European economies. ² With it, we present a series of stylized facts that hold across the sample, identify differences in firm growth dynamics across countries (particularly between Europe and the US), and examine its effect on productivity growth.

The interest in firm growth dynamics is not new. Gibrat's law, which asserts that firm growth is independent of firm size (Gibrat, 1931), has been intensively scrutinized. ³ Resource-based and evolutionary theories (Penrose, 1959; Nelson & Winter, 1978), learning and selection models (Jovanovic, 1982; Hopenhayn, 1992) and, more recently, financial constraints models (Cooley & Quadrini, 2001; Cabral & Mata, 2003; Arellano, Bai, & Zhang, 2009), customer acquisition models (Arkolakis, 2011; Dinlersoz & Yorukoglu, 2012) and ex-ante heterogeneity theories (Holmes & Stevens, 2010; Hurst & Pugsley, 2011) ⁴ have been used to explain several stylized facts on firm growth. ⁵ Recent work has also explored the shape of the firm growth distribution (Bottazzi & Secchi, 2006). And a large literature has examined jobs flows and the contribution of different groups of firms to job creation (see Haltiwanger 2012 for a review). Much of this research has used single country datasets, often based on small sample surveys or incomplete commercial databases, and frequently only capturing manufacturing sectors. ⁶ Thus, we exploit this more comprehensive database to discuss several stylized facts on firm dynamics and their robustness across countries and sectors.

Differences in firm growth dynamics across countries have also attracted much interest, particularly among policy makers. Most cross-country research has focused however on firm entry and exit rather than growth due to data constraints, so the evidence is limited. Bartelsman, Haltiwanger and Scarpetta (2004) conclude that entry and exit rates across developed countries are fairly similar, and instead highlight the large US-Europe differences in the growth rate of surviving new entrants. ⁷ European countries have fewer “gazelles” and high-growth firms than the US (OECD, 2008). And, while the US and Europe have a similar number of companies in the ranking of the world's 500 largest companies by market capitalization, only 3 of the European companies in the list were founded after 1975, in sharp contrast with 25 in the US (Véron, 2008). ⁸ We expand upon this work by examining the full distribution of firm growth across countries. In other words, not just looking at the “average firm” or at a subset of firms, whether the young, the largest, or the fastest growing ones, but instead providing a complete picture of how firms expand and shrink in each economy, using comparable data extracted with the same methodology and definitions in partnership with national statistical offices or local researchers. We identify sizeable differences, with US firms growing and shrinking much faster across sectors and sizes, while European firms being much more likely to remain stable.

Firm growth dynamics can help explain differences in aggregate productivity growth across countries, such as the widening productivity gap between Europe and the US over the last two decades (Ark, O'Mahony, & Timmer, 2008). The reallocation of output and labour towards more productive plants accounts for about half of total factor productivity growth in US manufacturing (Baily, Hulten, Campbell, Bresnahan, & Caves, 1992; Haltiwanger, 1997). ⁹ A more dynamic growth distribution implies faster resource reallocation. It is also a signal of higher competitive pressures, which force firms to improve their performance and raise within-firm productivity growth (Bartelsman, Haltiwanger, & Scarpetta, 2004). We use a differences-in-differences approach to test whether country-industry pairs with fewer static firms are associated with faster productivity growth and find this to be case. Both a larger share of growing and shrinking firms are associated with faster labour and total factor productivity growth.

Schumpeterian growth models also predict that experimentation and selection become more important as industries converge to the global technology frontier (Acemoglu, Aghion, & Zilibotti, 2006). While far from the frontier firms can improve their productivity by imitating what others have already invented, at the frontier they need to innovate. But innovation is risky and the outcome uncertain, so only the successful few expand while the unsuccessful ones shrink. Our findings support this hypothesis. A very static business growth distribution has a particularly strong negative effect on productivity growth the closer industries are from the global technology frontier.

Finally, a growing literature examines the role of factors such as regulation, finance, institutions or culture in explaining differences in firm dynamics across countries. ¹⁰ But while entry, expansion, contraction and exit are all part of the resource reallocation process, much of this work has focused exclusively on entrepreneurial activity. This database thus creates a new resource to examine what the drivers and barriers to firm growth are, which Bravo-Biosca, Criscuolo & Menon (2016) exploit to address some of these questions.

Structure of the paper

This paper is organized as follows. Section 2 describes the database, the data collection process and the construction of the growth distributions. Section 3 presents some stylized facts on the distribution of firm growth and employment dynamics, while Section 4 discusses some of the cross-country differences observed. ¹¹ Section 5 examines the links between firm growth dynamics and productivity growth and Section 6 concludes.

A separate appendix available online includes additional information on the database and extensive supplementary tables and figures, which provide data for all the indicators discussed here (and others) at a more disaggregated level. A version of the database as well as “non-technical" summaries are also available for download. ¹²

2 The database

We measure the distribution of firm growth using confidential microdata extracted from official business registers in 12 developed countries: Austria ¹³, Canada ¹⁴, Denmark, Finland, Greece, Italy, the Netherlands, New Zealand, Norway, Spain, the UK ¹⁵ and the US.

Business registers are assembled with data collected from social security records, tax records, censuses and/or other administrative sources. ¹⁶ Thus, they provide the most comprehensive coverage of economic activity in any country, basically covering the universe of firms (in contrast to commercial providers, whose coverage of business activity is limited and differs across countries). However, access to the data is very restricted, so we follow the approach used by other researchers (Bartelsman, Haltiwanger, & Scarpetta, 2004; Brandt, 2004; OECD, 2008) and partner with each country's national statistical offices or, alternatively, with researchers based there with authorized access to the microdata. We provide them with a methodology manual and a code file to extract data, building whenever feasible on the Eurostat-OECD Business Demography Manual (2007), which most business registers are required to follow. The datasets submitted are then scrutinized to identify potential inconsistencies and, if necessary, subjected to a process of revisions with each country partner.

2.1 Data collected

We collect data on employment growth between 2002 and 2005. In addition, whenever feasible we also collect data for other 3-year periods (2004-07 or 2005-08), for longer time periods (5-year growth or longer) and for turnover growth. The population of firms consists of all active employer enterprises (with at least one employee) in the private sector (ISIC sector 10_74) that have survived during the measurement period. In addition to non-survivors, enterprises born in the initial year are also excluded from the analysis. ¹⁷ The overall number of firms in the participating countries that satisfy these criteria is six million, which employed over 120 million people in 2002 (Table 1 provides a breakdown of the sample by country).

Table 2 provides a breakdown of the sample by firm size. Firms with less than 50 employees account for 95% or more of all surviving firms in all countries, even if typically represent less than half of employment. There are however large differences in the size distribution across countries, also within sectors (see appendix). For instance, countries like Greece, Spain and Italy have the highest proportion of employment in micro firms. Despite being much larger countries, average Spanish or Italian firms are smaller than Danish or Dutch firms. Therefore, comparisons based on commercial databases, which typically only have good data for firms over 50 employees (or just for public companies), not only miss a large share of the economy, but also of a different magnitude in each country.

For each firm we compute the average annualised growth rate ¹⁸ and allocate the firm into one of the 11 growth categories. ¹⁹ The data is then collapsed into 11 cells that summarize the growth distribution, containing the number of firms for which the growth rate falls within the interval (see Table 3 for summary statistics). In addition, for each cell the initial and final number of employees (and turnover, whenever feasible) are also computed. Overall statistics on the number of employer-firms surviving from time t-i to t relative to the total number of employer-firms at time t-i are also produced.

Growth distributions are constructed at the aggregate country level but also broken down for up to 51 sectors, 10 size classes and 10 age intervals. ²⁰ The sector breakdown is made at the 2-digit ISIC sector level, although data for some 3-digit subsectors is also collected. Size classes are based on the firm's initial number of employees. Age data is only available for a subset of countries. Age is defined as the age of the firm's oldest establishment at the beginning of the period, and the birth of that establishment corresponds to the first date when it recruited employees. Multiple breakdowns combining some of these categories are also available, although more aggregated to avoid disclosure problems. Any cell containing a number of firms below the confidentiality threshold established by each national statistical office is blanked out and codified as missing to avoid the release of legally-protected confidential information.

2.2 Data limitations

While business registers provide some of the most accurate firm-level data available, there are limitations. ²¹ The size threshold for payroll taxes or VAT registration can vary across countries. In practice, harmonization efforts imply that differences in the coverage of business registers (if any) are likely to be restricted to very small units. Therefore, they capture almost all the (legitimate) private sector activity in the country. The trade-off for the extensive coverage is the limited information they record, so we measure growth in employment and, whenever feasible, in turnover, but not in value added.

Measuring growth rates requires tracking firms over time. Business registers usually assign to each enterprise a unique identifier that is constant over time regardless of changes in name or ownership. While the quality of longitudinal databases has improved significantly, this is not always the case, which may result in “false” entries and exits. For instance, only some business registers are able to maintain before-and-after linkages for firms involved in mergers, acquisitions or spinouts. Therefore, to maximize participation when collecting the data an acquired firm is coded as an exit, ²² a spinout as a contraction in size and a new merged entity as the entry of a new firm. ²³ As a result, we do not distinguish between organic growth and growth by acquisitions. Both strategies are commonly used by firms to expand and, as Jovanovic and Rousseau (2002) point out, both can result in economy-wide productivity improvements (consider for instance an above-average productivity firm that expands by opening new plants - forcing others to close - vs. a similar firm that acquires existing plants and transfers its superior practices to them).

However, including acquisitions when measuring growth implies that the measures of job creation discussed throughout the paper capture all jobs gained by surviving firms, regardless of whether these jobs are actually new or they did already exist in acquired firms. Similarly, job destruction measures all jobs lost, regardless of whether they continue to exist in a divested subsidiary. The alternative approach to measure job flows, using establishment-level data, also has caveats, since employees reallocated across different plants within firms are also counted as job creation and destruction. Matched firm-establishment datasets are necessary to get accurate measures of job creation and destruction, but unfortunately they are not available across all the participating countries (see Haltiwanger, Jarmin, & Miranda, 2010, for an example). ²⁴

Determining the boundaries of the firm is another concern. The administrative or legal definition of an enterprise (or establishment) used by business registers does not necessarily coincide with the economic definition of the firm (which itself is often diffuse too). For instance, a new subsidiary of a larger firm is generally coded as new entering firm. ²⁵ Shifting of activities from one plant to another is treated differently if the plants belong to the same subsidiary or to two different subsidiaries of the same firm. Outsourcing to an external provider decreases employment growth (but not turnover growth). Employment outside the home country is not measured in business registers, so FDI or offshoring are not properly captured. ²⁶ However, these concerns should not be over-emphasized, since the boundaries of the firm are relatively clear for the majority of firms. After all, as Bartelsman, Scarpetta and Schivardi (2003) point out, the average number of plants per firm is 1.2 in the United States and 1.1 in Finland, despite the large difference in country size. ²⁷

A final note of caution is needed. Our measures of firm growth are substantially more accurate than the data on job creation. Specifically, job creation measures in this paper are subject to three weaknesses: (i) they ignore job creation and destruction via entry and exit, which account for 15-20% of gross job creation and destruction in the US (Haltiwanger, 2012); (ii) we cannot control for reversion to mean effects (Davis, Haltiwanger, & Schuh, 1996); and (iii) mergers and acquisitions are not accounted for. Therefore, we focus most of the discussion in the paper on the dynamics of firm growth, rather than answering the question of who creates jobs. This later question is better addressed by two other cross-country projects that began after the data collection for this one had been completed: the DYNEMP project led by the OECD in collaboration with national statistical agencies (Calvino, Criscuolo, & Menon, 2015), and the international cohort comparison study (Anyadike-Danes, et al., 2015).

2.3 Building distributions

We use histograms to plot the firm growth distribution, but also the density and cumulative distribution functions (cdf). To obtain the percentiles of the growth distribution, we start from the 11 growth intervals. The 10 finite interval endpoints provide the skeleton of the cumulative distribution function, so we interpolate between them to derive the full distribution. ²⁸ We assume the cdf to be continuous and only compute the cdf when the number of firms in a cell is above 50. We do not use a linear interpolation but instead try to control for the common curvature of the growth cdf using the Laplace distribution, which according to other studies calibrations' fits the empirical distribution of firm growth closely (Bottazzi & Secchi, 2006; Coad, 2007).

The Laplace distribution is part of the same family as the normal distribution but exhibits flatter tails (i.e., extreme occurrences are more likely) and has a symmetric tent-shape when plotted in log-scale. The Laplace density function and cdf are:

f(x|µ, b) = 1/2b exp(-|x-µ|/b)
F(x) = 0.5 [1 + sgn(x – µ) (1 – exp(-|x-µ|/b))] (1)

Although the cdf is not differentiable, it can conveniently be solved in a closed form, ²⁹ so we build from this expression to interpolate the cdf. We do not impose that the firm growth distribution follows a Laplace distribution. Instead, we only assume that the Laplace distribution is a good local approximation between interval endpoints for the actual growth distribution, and calculate for each growth interval the two parameters that define it (µ and b). For each growth interval we know: (1) the lower bound rate of growth, (2) the share of firms with smaller growth rates than the lower bound, (3) the upper bound rate of growth and (4) the share of firms with growth rates higher than the upper bound, so we end up with two equations and two unknowns from which we can solve for µ and b. We then obtain an approximation of the cdf for each growth interval and compute the growth rate for each percentile of the distribution of firm growth.

The calculation is different for the top and bottom growth interval (below -20% and above 20%), since they are not bounded. No firm can contract by more than 100%, so we impose this lower bound for the bottom interval and proceed as with all the other growth intervals. For the top growth interval we use the share of firms that fall into the category and their average growth rate ³⁰ to solve for µ and b and obtain the cdf for the top end of the distribution. However, this approximation is very sensitive to the underlying distributional assumptions, so the percentiles for the extremes of the distribution need to be interpreted with care. Table 4 presents the summary statistics for the percentiles resulting from this exercise, which we discuss in more detail in the section that follows.

3 Stylized facts on firm growth and employment dynamics

We first discuss a series of stylized facts that emerge from the analysis of the database and that hold across countries, sectors and time periods. For brevity, the figures in this section focus on the cross-country average and present only the data collected for the first period (2002-05). In order to avoid the cross-country average being driven by extreme values, the highest/lowest values are replaced by the second highest/lowest values when computing the average. The appendix includes country-by-country data for all the indicators discussed here, as well as data for the second period whenever available. We measure growth in terms of employment, but detailed data on turnover growth for some countries is also provided in the appendix (and it is also briefly discussed at the end of this section).

In order to examine the robustness of some of these facts across sectors and countries, Table 5 presents the summary statistics at the industry-country pair level, ³¹ and Table 6 reports the proportion of cells for which they are satisfied across the different breakdowns available, whether by country, industry, period, size, age, or, whenever feasible, growth measure. ³² The appendix provides additional figures and tables that confirm that the facts discussed here hold across most (if not all) countries and sectors.

Heterogeneity in firm growth rates, but regularity in the distribution

Figure 1 presents the distribution of firm growth across different sizes and countries. Each column indicates the share of firms with average annual employment growth rates over a three year period falling within that growth interval (with the range covering 11 intervals from less than -20 per cent to more than +20 per cent employment growth per annum).

The growth distribution for all firms with 1 or more employees is relatively polarized. Firms either remain static or experience fast growth or decline. This is due to a combination of two factors. First, micro firms constitute a disproportionate share of the business stock (80% of surviving firms), even if they only account for less than 20% of employment (Table 2). Therefore, they drive the aggregate growth distribution, given that by construction percentiles are not “employment-weighted” (in contrast to the average growth measures used in this paper). Second, the interpretation of growth rates is very sensitive to the initial size of the firm, particularly when measured in employment. A firm with one employee achieves a 100% growth rate by hiring another worker (“a doubling of their workforce”), while a firm with a 100 employees which incorporates another 10 employees has instead 10% growth rate. Because of this, unless otherwise stated, in the rest of the paper we focus on firms with 10 or more employees when discussing the distribution of firm growth rates.

Looking only at firms with 10 or more employees, the growth distribution becomes more evenly distributed, with a larger mass in the middle than in the extremes. It is unimodal, roughly symmetric and has fat tails. Therefore, extreme bursts of growth or decline occur quite regularly. The mass is higher at the bottom end of the distribution than at the top end, which implies that there are more high-decline firms than high-growth firms. This high level of growth and contraction leads to very high job reallocation rates across surviving incumbents, with the share of jobs created or destroyed by incumbents over a three year period being around 30% on average (Table 3). Going down one level to look at sectors confirms that substantial job reallocation also emerges within industries (Table 5).

The growth distribution looks very similar across all the different participating countries, with the exception of the UK (Figure 1.b). ³³ The median firm with 10 or more employees experiences zero or slightly negative growth in most countries and industries (Tables 4 & 5), and therefore existing 10+ firms are more likely to shrink than grow. However, this does not hold across all countries and sectors, even if it is still quite rare that the median firm grows by more than 1% pa (Table 5). ³⁴

Figure 2 shows that a comparable degree of heterogeneity in growth rates exists within sectors and size classes (as well as within narrowly defined age classes, as seen in the appendix). That is, the shape of the distribution is remarkably similar across sectors, sizes and ages. Therefore, firm growth heterogeneity is not the result of aggregation across different subsets of firms. This contrasts with firm size heterogeneity (and to a lesser extent, productivity heterogeneity), which falls substantially when looking within more disaggregated sectors (e.g., Coad, 2007).

Limited up-or-out dynamics

Standard learning and selection models (Jovanovic, 1982; Hopenhayn, 1992) predict a relationship between an age, survival and growth. Entrepreneurs enter the market, learn about their type and, depending on it, either expand or contract and exit. This up-or-out dynamic has been documented empirically in the literature (e.g., Haltiwanger, Jarmin, & Miranda, 2010) and it is confirmed as well by the top panel in Figure 3.

Young firms have lower survival rates than more mature firms. But conditional on survival, they experience higher growth on average. On the contrary, more mature firms have negative growth and destroy jobs on average (even if there is some significant heterogeneity across countries). This up-or-out dynamic is particularly strong for very young firms (1-2 years). They are 25% more likely to exit than the average firm, but conditional on surviving, they grow 3.5 times faster.

Similar patterns emerge when looking at median instead of average growth. Surviving young firms are more likely to grow than shrink, while older firms are more likely to shrink than grow (even if this is not always the case). As a result, average firm size increases as the cohort matures.

Up-or-out dynamics are clear when looking at averages, yet they appear much more limited when looking at the full distribution of growth. The last two plots in Figure 3 show that there is a large share of firms which neither expand nor contract over a three year period.

Whether young or old, ca. 40% of surviving firms are static, with an average annual employment growth rate between -1 and 1%. If a wider definition is used (-5,5%), ca. half of firms remain stable over the period. Therefore, not only do a majority of young firms survive over the three year period, but conditional on survival, a majority of them do not grow in most countries either. ³⁵ As a result, the size/employment distribution among older firms, even it shifts right, continues to be dominated by SMEs, which account for over 50% of employment and 90% of firms.

Limited up-or-out dynamics are not surprising given that a majority of entrepreneurs set up lifestyle business without ambitions to grow. Yet as Hurst & Pugsley (2011) argues, it raises some challenges for many of the standard models of entrepreneurship and firm growth, which often ignore that the most common motivation for entrepreneurs is becoming their own boss rather than growing the next 1 billion dollars business.

The dispersion of the growth distribution falls with size and age

Since Gibrat's law a large literature has examined the relationship between firm size and growth. Relatively less attention has been given to the growth-age link, and the combination of both dimensions. Figure 4 explores this interaction by plotting the firm growth density function by size and age. The density function narrows both with age and size unconditionally, but it is more responsive to changes in age than in size.

We can also look at how the density function varies with age and size conditioning on each other. Holding age constant, the density function narrows with size, although there is heterogeneity in the effect across different age classes. The dispersion falls much more for young firms that for firms with 11+ years, for which differences are more negligible. Keeping instead size constant, the distribution narrows with age. But there is also heterogeneity, since age matters relatively less for larger firms.

In order to appreciate the differences better, Figure 5 presents the main percentiles of the growth distribution, from the 99th percentile to the 10th percentile. The dispersion of growth rates falls with size. It is clear for the extreme top and bottom percentiles, but it is also the case for percentiles closer to the middle (even if more difficult to appreciate). Smaller firms grow and shrink faster than larger firms, and controlling for age does not make a big difference either. The variance of growth rates narrows with size even when we condition on age.

Figure 5.b considers instead the impact of firm age. Consistent with a learning and selection model, it shows higher dispersion of growth rates for younger firms than for more mature firms, as long as they have 10 or more employees, and this is the case regardless of whether we control for size or not. In case of micro firms, the picture is more nuanced. The distribution shows the same narrowing up to percentile 95th, but it does not arise for the two highest percentiles (99th, 98th). Finally, and not surprisingly, both the interquartile range and the p90-p10 range fall with size and age, unconditionally or conditioning on each other.

Table 6 examines the robustness of this stylized fact (as well as others) across the different sectors, countries and time periods available. Specifically, the first two columns report the percentage of country-period pairs for which each statement is satisfied (with the first one considering all firms and the second one only firms with 10+ employees). Columns 3 and 4 add a sector dimension, and therefore show the percentage of country-period-industry combinations for which the statement is satisfied. Finally, columns 5 and 6 consider as well size and age breakdowns, and therefore show how robust the statement is after conditioning for size and/or age. The total number of cells (or combinations) is reported in parenthesis. ³⁶

In over 95% of the country-industry pairs considered the interquartile range and the p90-p10 range are higher for young firms (1-5 years) than for old firms (11+ years), and this holds as well when conditioning on size class. The firm growth distribution also narrows as firms grow in size, but this fact is more robust at the extremes than closer to the middle of the distribution. In other words, the p90-p10 range is higher for SMEs than for large firms (250+) in 80-100% of the industry-country pairs considered, while for the interquartile range it holds “only” in 70-80% of cells. The robustness falls slightly when conditioning on age, but if it still holds for large majority of age groups across countries and sectors. ³⁷

What about the centre of the distribution? The median and the average employment growth rate falls with size and age, conditionally and unconditionally (as can be seen in Figure 5 and Table 6). Therefore, as firms become larger and/or older the distribution shifts downwards.

Hence, in contradiction with the strong version of Gibrat's law, ³⁸ size impacts the shape of the growth distribution, and the same picture emerges if we look at shares rather than percentiles. ³⁹ In summary, the average, the median and variance of the growth distribution fall both with size and age, conditionally and unconditionally, and this narrowing emerges at both extremes of the distribution.

Both age and size matter for job creation, but not equally

There is an on-going debate among policy makers, business groups and researchers about which are the firms that matter most for job creation, with answers ranging from the young, the small or the high- growth (see for instance Haltiwanger, Jarmin, & Miranda, 2010; Bravo-Biosca, 2011; Calvino, Criscuolo, & Menon, 2015; for an overview of the different positions).

The top panel in Figure 6 shows that gross job creation falls both with size and age. This negative relationship remains when controlling for each other. Even controlling for age, SMEs have higher job creation rates than large firms, also when they are old (Table 6). In contrast, there is no clear relationship between size and job destruction. Conditional on survival, small and large firms have a similar job destruction rate, particularly when not controlling for age. Comparing firms by their age, we find that job destruction is higher for younger firms than older firms, but not when looking at micro firms. Job destruction rates for firms with less than 10 employees are independent of firm age, with suggests that downward adjustment for small young firms happens through the extensive margin rather than the intensive margin, consistently with the literature showing that job destruction by exit is much larger for young firms (Haltiwanger, Jarmin, & Miranda, 2010).

Figure 6 also shows the substantial job reallocation process among surviving incumbents, with a large gap between net and job gross creation. As discussed earlier, around 30% of jobs are created or destroyed in each period, and this share is significantly larger for young and/or small firms. The last chart in Figure 6 considers an alternative measure of job churning, excess job reallocation. Measured as the sum of job creation and destruction less the absolute value of net change, it captures the total amount of churning jobs over and above that due to net changes (Davis, Haltiwanger, & Schuh, 1996). ⁴⁰ Similar patterns emerge. It falls substantially with size, particularly as firms get larger, as well as with age, and this holds as well when conditioning on each other (excluding micro firms).

An alternative approach to address the same question is to look at the share of employment, job creation and destruction accounted by each group of firms (Figure 7). Recall that these measures do not control for M&As and only capture surviving firms, and so exclude job reallocation due to entry and exit. ⁴¹ In summary, young SMEs account for a larger share of job creation and destruction than their employment share, and make an overall positive contribution to employment (in fact, they have the highest job creation rate and average employment growth rate of all groups of firms). Large old firms account for a lower share of job creation and destruction than their employment share, and destroy jobs on average. Both of these facts are robust across countries and sectors (Table 6). What about old SMEs? The answer is it depends. They consistently have smaller job creation rates than young SMEs, but also higher than large old firms. However, no clear pattern emerges with regard to their net contribution to employment growth, which can be positive or negative depending on the country and sector considered (Table 6).

Job creation and destruction is concentrated among a minority of firms

Figure 8 plots the distribution of job creation and destruction across firms with different growth rates. While most firms experience small changes in employment, most change in employment is the result of large changes in employment in a small number of firms. Firms in the top growth interval, which corresponds to the share of high-growth firms according to the OECD definition, ⁴² only represent 4.5% of surviving firms with 10 or more employees, yet they account for ca. 40% of all jobs created by all firms with ten or more employees (even if there are some important differences across countries). Job destruction is also concentrated. Less than 10% of firms decline by more than 20% a year on average over the period, yet they account for 45% of jobs lost by surviving firms with 10 or more employees.

Firm growth and contraction are correlated, controlling for aggregate shocks

Figure 9 examines the correlation between the top and bottom percentiles of the growth distribution, net of shocks that shift the full distribution up or down (i.e., subtracting the median). Each dot represents a country-industry pair (2-digit sector) in the first period, and we report both the unconditional correlation and the conditional correlation after controlling for industry and country fixed effects.

A clear relationship emerges. The faster firms grow at the top of the distribution (percentile 90th), the faster firms shrink at the bottom (percentile 10th). Similar, albeit weaker, findings emerge when looking at percentile 75th and 25th respectively.

Results change however if the median is not subtracted to control for aggregate shocks. Shocks typically shift the full distribution upwards or downwards, which creates a strong negative correlation between the share of growing firms and shrinking firms (see appendix). As a result, the negative correlation between the 90th and 10th percentiles becomes weaker, while in the case of the 75th-25th correlation, the sign is reversed. This is consistent with intermediate quartiles (e.g., p75-p25) being more closely correlated with the median than percentiles at the extremes of the distribution.

Some other results

Looking across different sectors, we find that the shape of the growth distribution is broadly similar and these facts continue hold. There are however some differences as well. Services industries display a much more dynamic growth distribution than manufacturing, which suggests that part of the literature that has looked only at the manufacturing sector may underestimate the degree of churn in the economy (in particular given its relative small share relative to services).

Table 7 also reports how robust some of the stylized facts discussed above are if instead of considering employment growth we look at real turnover growth. The distribution of turnover growth also narrows with age (and shifts down too, even if not as strongly), conditionally and unconditionally. But is no longer true that SMEs grow faster than large firms (whether looking at the average or the median), and the turnover growth distribution narrows less strongly with size in comparison the employment growth distribution. This would be consistent with large firms experiencing faster productivity growth (but also more volatile).

Comparing the employment and turnover distribution (bottom panel in Table 8) shows that average turnover growth is consistently higher than employment growth (but not so median growth). Similarly, the dispersion of the growth distribution is consistently higher for turnover than employment. Putting these findings and others in earlier sections together suggest that, as investment, employment adjustment is also lumpy, not fluctuating as much as turnover. Some implications for productivity follow, since increases in employment at firm level will be associated with a fall in their productivity levels as a result. ⁴³

Finally, several other stylized facts are visible in the data presented in the appendix. Even if not discussed here, we briefly list some of them. We find that: (a) survival rates are more dependent on age than size; (b) the firm size distribution is more skewed to the left for young than old firms, and so average size increases with age. Looking at the growth distribution by country-industry pair, we also find: (a) little or no correlation between aggregate employment growth and most proxies for the dynamism of the growth distribution, such as the share of static firms, interquartile range, p90-p10 range or excess job reallocation; (b) little or no correlation between average employment growth for young firms and the minimum efficient scale (proxied by average firm size in the sector); (c) substantial negative correlation between the dynamism of the growth distribution and survival rates; (d) little correlation between employment growth of the top 5% of firms and median growth, while the correlation with the median is stronger for the rest of percentiles.

4 Cross-country differences on firm growth and employment dynamics

The prior section discussed several stylized facts that hold across countries. Next we briefly consider some of the differences that emerge, which are substantial across several of the metrics collected. It is however difficult to tease out in a definite way what accounts for them. While structural institutional factors certainly play an important role (see Bravo-Biosca, Criscuolo, & Menon, 2016 for evidence on the impact of labour regulation and financial institutions), some of the differences observed may be the result of measurement issues and differential business cycles.

Specifically, as discussed in Section 2.2, while business registers are the most comprehensive and internationally comparable source for firm-level data, they still have some limitations. In addition, the limited time coverage of the database does not allow controlling for potential business cycles effects. Nevertheless, during the period considered here most participating countries experienced moderate economic growth, and the same patterns emerge in the later period (2004-07/2005-08) in the subset of countries for which data is available (see appendix).

In light of this, this section focuses on the relative performance of different groups of companies within and across countries, rather than on average comparisons, under the assumption that the former will be less dependent on the business cycle than the latter, and therefore better capture more structural differences. See appendix for more cross-country comparisons on the wider range of indicators available.

Figure 1 displays the firm growth distribution for all participating countries, yet it is difficult to observe clearly the existing cross-country differences. Because of this, Figure 10 plots the growth distribution relative to the US. Specifically, each bar indicates how much higher/lower in percentage terms the share of firms with a growth rate falling within that interval is in that country relative to the US.

A clear pattern emerges: Firms in the US grow and shrink more rapidly than in European countries, which have a much larger share of static firms, for which employment does not vary much (up or down). This pattern holds for a majority of countries, sectors and sizes classes (Figure 11). A similar pattern emerges as well when comparing the growth distribution for young firms between the US and the European countries for which we have data (see appendix).

Figure 12 considers two other metrics that also capture the cross-country differences in business dynamism. The left figure displays excess job reallocation for all firms with one or more employees, which (partially) controls for business cycles effects by taking out job churn due to economy-wide net employment changes. This is higher in the US than in all continental Europe countries, with the surprising exception of Greece, suggesting that there is a much more active process of resource reallocation across incumbent firms in the US. The right figure plots the percentiles of the growth distribution and the interquartile and p90-p10 ranges by country, sorted according to the interquartile range. There are also sizeable differences, both when looking at the percentiles at the extremes of the distribution and when looking at the interquartile and p90-p10 ranges, with the US ranking higher than most European countries.

One of the potential effects of a less dynamic growth distribution is that it becomes more difficult to challenge incumbents. Figure 13 looks at the differences in performance between SMEs vs. large firms and young firms vs. old firms across countries. Specifically, the two measures considered are the average employment growth rate and job creation as share of initial jobs (in other words, net and gross job creation rates). Each bar corresponds to the difference in percentage points between the rate for SMEs/young firms and the rate for large/old firms. Again, the differences are substantial, with countries like the US displaying a much larger gap than most European countries in the sample.

One possible interpretation is that a larger gap is a signal that the country's institutional framework makes it relatively easier for younger and smaller firms to challenge incumbents. However, this is not the only possibility. In some circumstances a poor institutional background can also lead to large gaps in the growth rates of younger and smaller firms relative to large firms, which could help explain the position of countries such as Italy and Greece in Figure 13. For instance, Arellano, Bai, & Zhang (2009) show that small firms grow disproportionally faster than larger firms in less financially developed countries, because limited access to external finance constraints their growth to what their current cashflows can fund.

In the appendix we explore in a little bit more detail whether there are cross-country differences on who creates jobs. As discussed in Section 2.2, given the limitations in our data (particularly the fact that it only captures surviving firms and not entry and exit), these results need to be interpreted with care. The data shows that on aggregate surviving firms create jobs in some countries but not in others, even if looking only at economies that are growing. Specifically, we find that surviving firms' job creation is larger than job destruction in most European countries with positive aggregate growth, in contrast to the US, where net job creation is negative (despite also growing during the period). These differences are mostly driven by incumbents with more than 250 employees, that contract significantly in the US but not across Europe. A back-of-the-envelope calculation would suggest that start-ups make a more important contribution to job creation in the US, although more recent U.S. Census data suggest that there has been a substantial decline in the contribution of new firms to job creation (Pugsley & Şahin, 2015; Decker, Haltiwanger, Jarmin, & Miranda, 2016), and this decline has occurred in many other OECD countries (Calvino, Criscuolo, & Menon, 2015).

Overall, there are at least three takeaways from these cross-country comparisons.

First, much of the policy debate in Europe around business growth has been framed around its lacklustre performance at generating high-growth firms that become global champions. This evidence clearly shows that differences go beyond that. The US has more high-growth firms than Europe, but this is only one part of the picture. European countries have a less dynamic firm growth distribution overall, with slower growth and slower contraction, and unless this is recognised, we are likely to draw the wrong policy conclusions.

Second, it is important to understand the institutional drivers behind these differences, and what policy levers exist to address them. While there is a body of work looking at some these questions, there are still many unanswered questions.

Finally, much of the literature on firm dynamics has been developed using US datasets. This data shows that it cannot be taken for granted that the same results that emerge in the US will hold across other countries, and therefore suggests the need for more replication of studies done in one country with other countries' datasets.

5 Firm growth dynamics and productivity growth

We next examine whether differences in the distribution of firm growth impact productivity growth. There are several channels by which a more dynamic firm growth distribution could potentially lead to faster productivity growth. From a pure accounting perspective, it speeds up the reallocation of labour and capital, most likely from unproductive incumbents towards innovative firms that have successfully developed superior practices. ⁴⁴ It may also have an additional indirect effect, increasing competitive pressures which force firms to improve their internal practices or else shrink and exit. ⁴⁵ A more dynamic firm growth distribution may as well signal an environment in which firms are willing to experiment and put new ideas into practice, while being able to backtrack and shrink without major consequences if they do not succeed. ⁴⁶ Hence there is also a dynamic effect, since knowing that it will be easy to scale up tomorrow if an invention is successful increases the incentives to invest in innovation today. On the contrary, a large share of static firms may signal instead an unwillingness to take risks to innovate, since trying a new business model, exploring a new technology or launching a new product often requires a firm to expand its capabilities, even if only temporarily and with uncertain success (Saint-Paul, 1997; Bartelsman, Perotti, & Scarpetta, 2008).

These effects are however not unambiguous. High levels of resource reallocation may not lead to higher productivity if, for instance, it is directed towards the more unproductive firms, either because they have better access to finance, they are well-connected, their managers are prone to mistakes ⁴⁷ or care more about empire-building than improving performance, or when a speculative bubble distorts the allocation process. Even when this is not the case, resource reallocation also creates significant adjustment costs for firms and workers. Firms may lose the intangible capital embedded in their workers, demotivate them to take risks ⁴⁸ and face disruption in their organization, resulting in lower productivity. Employees are likely to lose firm-specific human capital and face significant uncertainty (Hall R. E., 1995). Moreover, a more dynamic growth distribution can increase frictional unemployment due to job search and matching frictions, particularly in poorly functioning labour markets with high unemployment rates (Mortensen & Pissarides, 1994). Finally, too much competition can also reduce incentives for experimentation (Aghion, Bloom, Blundell, Griffith, & Howitt, 2005).

Which of these effects dominate may depend on the position of the country relative to the world technology frontier. As countries get closer to the frontier, both experimentation and selection become more important (Acemoglu, Aghion, & Zilibotti, 2006). First, at the frontier innovation replaces imitation as the main driver of productivity growth, so experimentation is more important. Second, the reallocation of resources towards more productive firms may also play a more significant role at the frontier, since only a fraction of the firms attempting to innovate succeed, while many others fail to improve their productivity. In contrast, imitation is less uncertain and does not require as much skill, so far from the frontier within-firm productivity improvements across the board are more feasible (Acemoglu, Aghion, & Zilibotti, 2006). Finally, the effect of competition on innovation and productivity growth is also stronger at the frontier (Aghion, Bloom, Blundell, Griffith, & Howitt, 2005; Aghion, Blundell, Griffith, Howitt, & Prantl, 2009). Altogether, these different channels suggest that the impact of a more dynamic growth distribution should be stronger the closer the country is to the technology frontier.

We use standard OLS to test whether a more dynamic growth distribution is associated with faster productivity growth, as the literature suggests it should. Firm growth dynamics are clearly endogenous, so the results need to be interpreted with this in mind. An alternative approach would have been to use standard productivity growth decompositions (e.g., Bartelsman, Haltiwanger, & Scarpetta, 2005), which compute the share of productivity growth accounted by within-firm improvements, the entry and exit of firms and the reallocation of resources across continuing firms. They however require firm-level productivity data and, in addition, do not capture the indirect effect that a more dynamic firm growth distribution may have on within-firm productivity growth arising, for instance, from stronger competition.

The baseline specification regresses annual total factor productivity growth for industry j in country i in 2002-2005 on the share of static firms, which is a measure of the (lack of) dynamism of the firm growth distribution. The share of static firms is defined as the share of all surviving firms with 10 or more employees ⁴⁹ with annual average employment growth between -1% and 1% (>1% are growing firms, <-1% are shrinking firms). We include country (µ) and industry (τ) fixed effects in the regression to mitigate omitted variables concerns, and control for the distance to the frontier to account for potential convergence effects. Industry-country specific shocks can shift as well the distribution of firm growth, so we control for employment growth at the industry-country pair level.

TFPgrowthij = βōshareij + β₁employmentgrowthij + β2distancefrontierij + μ¡ + Tj + Eij (2)

We use EUKLEMS data for industry-level productivity measures. Annual TFP growth is value added-based and annual labour productivity growth (which we use as a robustness check) is defined as gross value added per hour worked by persons engaged (O'Mahony & Timmer, 2009). We exclude outliers defined as those industry-country pairs with TFP growth more than two standard deviations away from the industry or country mean. ⁵⁰ Employment growth corresponds to the average annual growth in the industry's number of employees over the period, also from EUKLEMS. Distance to frontier is defined as as -ln(TFPij/TFpleader(j)) at the beginning of the period, where TFpleader(j) corresponds to the highest TFP level for industry j across countries (as long as it is within 2 standard deviations from the mean for the industry). ⁵¹ TFP levels data is obtained from the GGDC Productivity Level Database (Inklaar & Timmer, 2008), which also builds on EUKLEMS, and specifically corresponds to value added-based (double deflated) multi-factor productivity.

Table 6 reports the results of this exercise. Each of the 144 observations corresponds to an industry-country pair, with 8 countries ⁵² and up to 22 industries included in the regressions. All regressions are estimated using OLS, with standard errors in parentheses clustered both at country and industry level. ⁵³ Column 1 reports the results of the baseline specification, with country and industry effects but without any additional control. The coefficient indicates that a 1pp increase in the share of static firms is associated with -0.187pp lower annual TFP growth, and it is significant at the 1% level. Controlling for convergence effects with the industry's distance to the technology frontier and for potential industry shocks with employment growth does not make a difference (Column 2). ⁵⁴

A low share of static firms could be driven both by a high share of growing and/or shrinking firms. Therefore, the results could just be picking up some positive correlation between growing industries and TFP growth not captured by average employment growth, with no relationship to selection processes. Column 3 replaces the share of static firms by the shares of growing and shrinking firms. While the coefficient for the growing share (0.251) is higher than for the shrinking share (0.171), they are both large and significant. Consequently, this supports the hypothesis that industries with a higher degree of selection, that is, with a higher share of growing and shrinking firms and fewer static ones, experience faster productivity growth.

Experimentation and selection may become more important the closer to the technology frontier an industry is. Columns 4-5 examine this hypothesis. The interactions between distance to frontier and the shares of static, growing and shrinking firms are significant and with the right sign. Far from the frontier, a large share of static firms is not associated with lower TFP growth, while at the frontier it is. The same finding arises when looking at the share of growing and shrinking firms. Thus, this evidence suggests that a dynamic firm growth distribution becomes more important for productivity growth the closer countries get to the technology frontier, as predicted by Acemoglu, Aghion and Zilibotti (2006).

A variety of methodological issues arise when estimating TFP, which could potentially affect comparisons across countries and industries. Therefore, for robustness we consider as well labour productivity growth as outcome variable, since it is subject to relatively fewer measurement issues. Columns 6-7 show that the same patterns emerge, even if with somewhat lower significance levels.

Summing up, these results suggest that a 5pp higher share of static firms is associated with 1pp lower annual productivity growth (both for TFP and labour productivity), and that this negative effect becomes stronger as countries converge to the technology frontier.

Is this a big or a small effect in economic terms? In the decade prior to the financial crisis, Europe's annual TFP growth lagged the US by 1pp on average (Ark, O'Mahony, & Timmer, 2008), while cross-country differences in the share of static firms average several percentage points. Alternatively, a one-standard deviation increase in the share of static firms is associated with 1.1pp lower annual TFP growth. Therefore, the magnitude of the coefficient and the implied correlation are non-negligible.

6 Final remarks

The new database that this paper introduces captures the heterogeneity of firm growth in a comparable way across different countries, providing a more detailed picture of firm growth dynamics than previously available. It thus provides new insights on how creative destruction (or, more precisely, resource reallocation) happens in different countries, as well as serves to test the robustness of some stylized facts in different settings.

Our results confirm that there is substantial heterogeneity in firm growth rates, but also that the growth distribution displays a very regular pattern across countries and sectors. We find that there are limited up or out dynamics, and that both age and size are negatively related to the mean and variance of the growth distribution. Nevertheless, we also find significant cross-country differences. European countries tend to display a much more static growth distribution, while in the US firms typically grow and shrink much more rapidly. There are also significant differences in the contribution that surviving incumbents make to employment growth, negative in the US but positive in many (even if not all) European countries. Finally, we explore whether this differences matter for productivity growth, and we find this to be the case, particularly as countries converge to the technology frontier.

References

• Acemoglu, D., Aghion, P., & Zilibotti, F. (2006). Distance to frontier, selection, and economic growth. Journal of the European Economic Association, 4(1), 37-74.
• Acharya, V., Baghai, P., & Subramanian, K. (2010). Labor Laws and Innovation. NBER Working Paper No. 13085.
• Aghion, P., & Howitt, P. (2006). Appropriate Growth Policy: A Unifying Framework. Journal of the European Economic Association, 4(2-3), 269-314.
• Aghion, P., Bloom, N., Blundell, R., Griffith, R., & Howitt, P. (2005). Competition and Innovation: An Inverted-U Relationship. The Quarterly Journal of Economics., 120(2), 701-728.
• Aghion, P., Blundell, R., Griffith, R., Howitt, P., & Prantl, S. (2009). The Effects of Entry on Incumbent Innovation and Productivity. Review of Economics and Statistics, 91(1), 20-32.
• Aghion, P., Fally, T., & Scarpetta, S. (2007). Credits Constraints as a Barrier to the Entry and Post-Entry Growth of Firms. Economic Policy, 22(52), 731-779.
• Aidis, R., Estrin, S., & Mickiewicz, T. (2007). Entrepreneurship in Emerging Markets: Which Institutions Matter? UCL SSEES Economics Working Paper No. 81.
• Alfaro, L., & Charlton, A. (2007). International Financial Integration and Entrepreneurial Firm Dynamics. NBER Working Paper No. 13118.
• Anyadike-Danes, M., Bjuggren, C.-M., Gottschalk, S., Hölzl, W., Johansson, D., Maliranta, M., & Myrann, A. (2015, April). An international cohort comparison of size effects on job growth. Small Business Economics, 44(4), 821-844.
• Arellano, C., Bai, Y., & Zhang, J. (2009). Firm Dynamics and Financial Development. NBER Working Paper No. 15193.
• Ark, B. v., O'Mahony, M., & Timmer, M. P. (2008). The Productivity Gap between Europe and the United States: Trends and Causes. Journal of Economic Perspectives, 22(1), 25-44.
• Arkolakis, C. (2011). A Unified Theory of Firm Selection and Growth. NBER Working Paper No. 17553.
• Baily, M. N., Hulten, C., Campbell, D., Bresnahan, T., & Caves, R. E. (1992). Productivity Dynamics in Manufacturing Plants. Brookings Papers on Economic Activity. Microeconomics, 187-267.
• Bartelsman, E. J., Gautier, A. P., & de Wind, J. (2010). Employment Protection, Technology Choice, and Worker Allocation. CEPR Discussion Paper 7806.
• Bartelsman, E. J., Haltiwanger, J. C., & Scarpetta, S. (2013). Cross-Country Differences in Productivity: The Role of Allocation Efficiency and Selection. American Economic Review, 103(1), 305-334.
• Bartelsman, E. J., Haltiwanger, J., & Scarpetta, S. (2004). Microeconomic Evidence on Creative Destruction in Industrial and Developing Countries. World Bank Policy Research Working Paper No. 3464.
• Bartelsman, E., Haltiwanger, J., & Scarpetta, S. (2005). Measuring and Analyzing Cross-country Differences in Firm Dynamics. NBER Conference on Research in Income and Wealth. Producer Dynamics: New Evidence from Micro Data.
• Bartelsman, E., Perotti, E., & Scarpetta, S. (2008). Barriers to Exit, Experimentation and Comparative Advantage. mimeo.
• Bartelsman, E., Scarpetta, S., & Schivardi, F. (2003). Comparative Analysis of Firm Demographics and Survival: Micro-Level Evidence for the OECD Countries. OECD Economics Department Working Papers No. 348.
• Bottazzi, G., & Secchi, A. (2006). Explaining the Distribution of Firm Growth Rates. The RAND Journal of Economics, 37(2), 235-256.
• Brandt, N. (2004). Business Dynamics in Europe. OECD Science, Technology and Industry Working Papers no. 2004/1.
• Bravo-Biosca, A. (2010). Growth Dynamics: Exploring business growth and contraction in Europe and the US. London: FORA-NESTA Research Report.
• Bravo-Biosca, A. (2011). A look at business growth and contraction in Europe. Nesta Working Paper No. 11/02.
• Bravo-Biosca, A., Criscuolo, C., & Menon, C. (2016, October). What drives the dynamics of business growth? Economic Policy, 31(88), 703-742.
• Cabral, L. M., & Mata, J. (2003). On the Evolution of the Firm Size Distribution: Facts and Theory. American Economic Review, 93(4), 1075-1090.
• Calvino, F., Criscuolo, C., & Menon, C. (2015). Cross-country evidence on start-up dynamics. OECD Science, Technology and Industry Policy Papers No. 2015/06.
• Cameron, C. A., Gelbach, J. B., & Miller, D. L. (2006). Robust Inference with Multi-way Clustering. NBER Technical Working Paper No. 327.
• Caves, R. E. (1998, December). Industrial Organization and New Findings on the Turnover and Mobility of Firms. Journal of Economic Literature, 36, 1947-1982.
• Coad, A. (2007). Firm Growth: A Survey. mimeo.
• Cooley, T. F., & Quadrini, V. (2001). Financial Markets and Firm Dynamics. American Economic Review, 91(5), 1286-1310.
• Cosh, A., & Hughes, A. (2007). British Enterprise: Thriving or Surviving? Centre for Business Research, Cambridge University.
• Davis, S. J., Haltiwanger, J. C., & Schuh, S. (1996). Job Creation and Destruction. Cambridge, MA: MIT Press.
• Davis, S. J., Haltiwanger, J., Jarmin, R., & Miranda, J. (2007). Volatility and Dispersion in Business Growth Rates: Publicly Traded versus Privately Held Firms. NBER Macroeconomics Annual 2006. 21. MIT Press.
• Decker, R. A., Haltiwanger, J., Jarmin, R. S., & Miranda, J. (2016, July). Where has all the skewness gone? The decline in high-growth (young) firms in the U.S. European Economic Review, 86, 4-23.
• Desai, M., Gompers, P., & Lerner, J. (2003). Institutions, Capital Constraints and Entrepreneurial Firm Dynamics: Evidence from Europe. NBER Working Paper No. 10165.
• Dinlersoz, E. M., & Yorukoglu, M. (2012). Information and Industry Dynamics. American Economic Review, 102(2), 884-913.
• Eurostat & OECD. (2007). Eurostat-OECD Manual on Business Demography Statistics.
• Fogel, K., Hawk, A., Morck, R., & Yeung, B. (2006). Institutional Obstacles to Entrepreneurship. In M. Casson, B. Yeung, A. Basu, & N. Wadeson (Eds.), Oxford Handbook of Entrepreneurship. Oxford University Press.
• Geroski, P., & Gugler, K. (2004). Corporate Growth Convergence in Europe. Oxford Economic Papers, 56(4), 597-620.
• Gibrat, R. (1931). Les Inégalités Économiques. Paris: Librairie du Recueil Sirey.
• Hall, B. H. (1987). The Relationship Between Firm Size and Firm Growth in the United States Manufacturing Sector. Journal of Industrial Economics, 35(4), 583-606.
• Hall, R. E. (1995). Lost Jobs. Brookings Paper on Economic Activity, 1, 221-256.
• Haltiwanger, J. (1997). Measuring and Analysing Aggregate Fluctuations: The Importance of Building from Microeconomic Evidence. Federal Reserve Bank of St. Louis Economic Review, 55-77.
• Haltiwanger, J. (2012). Job Creation and Firm Dynamics in the US. In J. Lerner, & S. Stern (Eds.), Innovation Policy and the Economy, Volume 12 (pp. p. 17 - 38). Cambridge: NBER.
• Haltiwanger, J., Jarmin, R. S., & Miranda, J. (2010). Who Creates Jobs? Small vs. Large vs. Young. NBER Working Paper No. 16300.
• Haltiwanger, J., Jarmin, R. S., Kulick, R., & Miranda, J. (2016). High Growth Young Firms: Contribution to Job, Output, and Productivity Growth. In J. Haltiwanger, E. Hurst, J. Miranda, & A. Schoar, Measuring Entrepreneurial Businesses: Current Knowledge and Challenges. University of Chicago Press. Forthcoming.
• Hart, P. E., & Oulton, N. (1996). Growth and Size of Firms. Economic Journal, 106(438), 1242-52.
• Henrekson, M., & Johansson, D. (2010). Gazelles as Job Creators: A Survey and Interpretation of the Evidence. Small Business Economics, 35, 227-244.
• Hoffmann, A. N., & Junge, M. (2006). Documenting Data on High-growth Firms and Entrepreneurs across 17 countries. mimeo.
• Holmes, T. J., & Stevens, J. J. (2010). An Alternative Theory of the Plant Size Distribution with an Application to Trade. NBER Working Paper No. 15957.
• Hopenhayn, H. A. (1992). Entry, Exit, and Firm Dynamics in Long Run Equilibrium. Econometrica, 60(5), 1127-1150.
• Hurst, E., & Pugsley, B. W. (2011). What do Small Businesses Do? Brookings Papers on Economic Activity, 43(2), 73-142.
• Ijiri, Y., & Simon, H. A. (1977). Skew distribution and the sizes of business firms. Amsterdam: North-Holland.
• Inklaar, R., & Timmer, M. P. (2008). GGDC Productivity Level Database: International comparisons of output, inputs and productivity at the industry level. GGDC Research Memorandum 104.
• Jovanovic, B. (1982). Selection and Evolution of Industry. Econometrica, 50(3), 649-670.
• Jovanovic, B., & Rousseau, P. L. (2002). Mergers as Reallocation. NBER Working Paper No. 9279.
• Klapper, L., Amit, R., Guillén, M. F., & Quesada, J. M. (2007). Entrepreneurship and Firm Formation across Countries. World Bank Policy Research Working Paper 4313.
• Klapper, L., Laeven, L., & Rajan, R. (2006). Entry regulation as a barrier to entrepreneurship. Journal of Financial Economics, 82, 591-629.
• Kogan, L., Papanikolaou, D., Seru, A., & Stoffman, N. (2012). Technological Innovation, Resource Allocation, and Growth. NBER Working Paper No. 17769.
• Mansfield, E. (1962). Entry, Gibrat's Law, Innovation and the Growth of Firms. American Economic Review, 52(5), 1023–51.
• Mortensen, D. T., & Pissarides, C. A. (1994). Job Creation and Job Destruction in the Theory of Unemployment. The Review of Economic Studies, 61(3), 397-415.
• Nelson, R. R., & Winter, S. G. (1978). Forces Generating and Limiting Concentration under Schumpeterian competition. The Bell Journal of Economics, 9(2), 524-548.
• O'Mahony, M., & Timmer, M. P. (2009). Output, Input and Productivity Measures at the Industry Level: the EU KLEMS Database. Economic Journal, 119(538), pp. F374-F403.
• OECD. (2008). Measuring Entrepreneurship: A Digest of Indicators.
• Penrose, E. E. (1959). The Theory of the Growth of the Firm. New York: Wiley.
• Petersen, M. A. (2009). Estimating Standard Errors in Finance Panel Data Sets: Comparing Approaches. The Review of Financial Studies, 22(1), 435-480.
• Pugsley, B., & Şahin, A. (2015). Grown-Up Business Cycles. Federal Reserve Bank of New York Staff Report No. 707.
• Saint-Paul, G. (1997). Is labour rigidity harming Europe's competitiveness? The effect of job protection on the pattern of trade and welfare. European Economic Review, 499-506.
• Scarpetta, S., Hemmings, P., Tressel, T., & Woo, J. (2002). The Role of Policy and Institutions for Productivity and Firm Dynamics: Evidence from Micro Data and Industry Data. OECD Economics Department Working Paper 342.
• Schumpeter, J. A. (1942). Capitalism, Socialism, and Democracy. New York: Harper and Brothers.
• Véron, N. (2008). The Demographics of Global Corporate Champions. Bruegel Working Paper no. 2008/03.