Business dynamism in the Netherlands
and the role of high-tech startups
Tim Oudenampsen1
Faculty of Economics and Business, University of Groningen Master Thesis Economics
Supervised by: prof. dr. K.F. Roszbach Abstract
Using data on all limited liability companies in the Netherlands over the period 2006-2018, I show that business dynamism in the Netherlands is in slight decline. This decline is related to a decline in entry rates and an increase in exit rates during and short after the great recession. Homogeneity in employment growth among firms increases in this same period. I highlight the role high-tech startups play in this decrease in dynamism and show that among high-tech startups the reaction to the crisis is similar but more pronounced compared with the average firm. Furthermore, I examine the difference in impact between firm size and age on weighted net employment growth. I find some evidence in support of the perception that young firms disproportionately contribute to job reallocation and net job creation. However, this is only true given these firms survive, hence only when considering the negative impact of firm exit. In contrast to other research, I find that size does not significantly matter for job creation, but small firms do contribute more to job reallocation.
Course code: EBM877A20
Keywords: Business dynamism; Growth; Entrepreneurship; Job creation; High-growth firms JEL codes: E32, J62, L25, L26
Date: 09-01-2020
1 Student number: s2752042
2 1. Introduction
High business dynamism implies that resources are reallocated away from less productive firms towards high productive firms, hence contributing to productivity growth. This process of “creative destruction” causes unproductive incumbents to be pushed out of the market. Business dynamism varies substantially across countries (Criscuolo et al., 2014), but a vast amount of literature (Decker et al., 2018; Bijnens & Konings, 2018; among others) suggests a decline in this dynamism is present in developed economies. Heterogeneity among countries on this matter exists, and Bravo-Biasco et al. (2016) show that U.S. firms perform better in terms of dynamism compared with European firms. The underlying drivers affecting business dynamism are still debated about, and this started a series of investigations to the possible reasons for this decline in developed economies. In this paper I contribute to this debate by offering insights into the Dutch economy on this declining dynamism.
Over the years, the Dutch economy changed substantially, and these changes might have contributed to the trends we see today, shifting the balance away from competition towards the market leaders. Less competition and fewer firm entries decline the dynamism of an economy. Mainly important in this decline are small and young firms, as they have historically been an important contributor to employment growth (Decker et al., 2018). Several empirical observations have been identified by the literature which might be related to this decline in dynamism. My aim in this thesis is to summarize the literature on the possible causes of this decrease in business dynamism occurring in the developed economies. Furthermore, I closely examine the situation in the Dutch economy from 2006 until 2018 using data from Statistics Netherlands, to help identify the factors that could have driven the trend.
This paper contributes to the literature, since the way business dynamism has evolved in the Netherlands still is unclear. The emphasis in this paper lies on the difference in contribution firms make to economic growth and business dynamism, herefore I divide firms in age and size classes. Moreover, no research before explored the actual contribution high-tech startups make in job creation or growth in the Netherlands2. In this research I follow related literature (Bos &
Stam, 2014) and consider high-tech startups to be firms that are innovative by design, tech-based or tech-enabled and strive for rapid growth.
The Dutch government has taken several actions over the years to improve the startup landscape and counteract a decline in (startup) entry rates and business dynamism. In an article by the Dutch government (Rijksoverheid, 2019) they acknowledge recent research (e.g. Haltiwanger et al., 2013) that startups are the most important drivers of growth and an important contributor of jobs, which was the reason for a significant investment in this sector. However there does not seem to be an evincive paper on its impact. Therefore, I empirically test the difference in contribution to employment growth by firms of certain age or size.
3
This paper is structured in the following way. In Section 2 I give an overview of the related literature. I present the goal of the thesis in section 3 and introduce the question and hypotheses central to this study. In Section 4 I discuss the available data and clarify some essential concepts to comprehend this research. After this I lay out the empirical strategy in Section 5. Next, in Section 6, I display the main results of the empirical model. Finally, in Section 7 I answer the question central to this study, present a discussion around the methods used and conclude the paper.
2. Literature overview
As noted in the introduction, there have been several changes in the economy over the years. However, which of those changes contributed to the trend of declining business dynamism in the developed economies is long debated and not yet identified. Though several possible channels have been put forward. In this section I indicate the importance of business dynamism and highlight a few possible drivers of this declining trend of dynamism that prevails in the literature.
2.1 Business dynamism
Bartelsman et al. (2004) show that in most developed economies there is evidence that firms undergo significant changes over time, both through the process of firm exit and entry and through resource reallocation between existing firms. The level of efficiency with which an economy deals with these changes is important not only for productivity, but also for dynamics in unemployment and the labour market. The significant amount of churning of firms (that is the process of entry and exit) along with high reallocation rates of labour across surviving firms implies high search and adjustment costs (Bartelsman et al., 2004).
4
In the matter of linking entrepreneurship to employment growth and economic growth, it is important to distinguish between the different roles of the entrepreneur in the economy. Following the description by Wennekers & Thurik (1999), entrepreneurs can start a business, whether there is anything innovative about it or not. However, true innovative entrepreneurs turn ideas and inventions into economically viable products, which might, in doing so, lead to the creation of a firm. Innovation is the process of investing in research and development (R&D), where successful innovators become new producers and stimulate growth because of higher efficiency (De Ridder, 2019). Especially this group of true innovative entrepreneurs is more alert to opportunities than managers of incumbent firms, and in pursuing these opportunities they increase the competitive pressure (Spencer & Kirchhoff, 2006). In this paper I focus on the growth of high-tech startups, which is the group of firms associated with innovative entrepreneurship.
Entrants have high heterogeneity in initial size, reflecting both their own perceived ability as well as their ease to find initial funding. Note that because of high uncertainty, even very successful (ex-post) entrants tend to start small. This is also one of the explanations to why young and small surviving firms show rapid growth (Bartelsman et al., 2004). In their next phase, successful firms accumulate assets and experience, which leads to further strengthening and growth.
2.2 Declining dynamism
As argued above, business dynamism is important in economics, but the literature is inconclusive about the drivers. However, there are several potential drivers put forward in the literature. First, market concentration has risen in the past decades as documented by Autor et al. (2017) for the U.S. (Bajgar et al., 2019 show similar patterns for Europe as well). This higher market concentration however does not necessarily imply higher firm market power and lower competition (Autor et al., 2017).
Second, there has been a decline in the share of young firms in economic activity, and a decline in firm entry rates in the U.S. (Autor et al., 2017) and Belgium (Bijnens & Konings, 2018). Additionally, Autor et al. (2017) show that gross job reallocation rate has been falling, indicating less dynamism, which they attribute to a lower responsiveness to idiosyncratic productivity shocks by (young) firms. Furthermore, this second trend is harmful for the economy as fast-growing young firms disproportionally contribute to job creation (Bravo-Biosca et al., 2016; Haltiwanger et al., 2013).
5
the developed economies. They show that the current ultra-low interest rates have a negative impact on industry competition. The traditional way low interest rates impact the economy is via increased incentives to invest, as future cash flows are valued higher ceteris paribus (Liu et al., 2019). However, they show that there is a second effect, the “strategic effect”, which triggers industry leaders to invest more aggressively. This second effect is driven by the fact that the lower interest rates increase the strategic incentive to invest for the leaders. This is not a symmetric argument for the followers, as their incentive to invest is driven by their potential to become market leader, which is endogenously diminished by the excessive investments made by the leaders in the market (Liu et al., 2019). Hence the fall in interest rates in the last decade might be the cause of higher market concentration and a wider productivity gap, which in turn decreases dynamism.
Furthermore, some argue that this third observation might be driven by the distortion in the knowledge flow between firms (De Ridder, 2019; Akcigit & Ates, 2019). In fact, De Ridder (2019) shows that all three above-mentioned regularities can be explained by a shift in the way firms produce, as more intangible inputs are used. Intangible inputs, like software and information technology, are scalable, implying they can be duplicated at near zero costs. This shifts production costs towards higher fixed costs in the form of R&D, while firms face lower variable costs. This may in turn lead to a “winner takes (almost) all” situation, as the highest quality versions of production technology result in the lowest costs. Products of lower quality, but produced with better technology, combined with economies of scale and network externalities, can in this circumstance outcompete higher quality products based on costs (De Ridder, 2019). The increased share of large firms may be a consequence of this rise in returns to scale. Economies of scale impose barriers to entry for new firms, since they must operate in a low-cost environment. Even in the presence of governments stimulating (startup) entry, it is difficult to match the cost-minimizing advantages that scale-economics brings to the larger firms. Size and scale may improve efficiency, but also create an uneven playing field, suppressing dynamism, and in the long run this may decrease efficiency again.
6
support in market experimentation to show little short-term productivity growth and lower churning, whereas in the long run this shifts to a higher contribution in productivity and dynamism. All in all, different distortions act simultaneously in a country. Where some policies show a preferential treatment for incumbents and impose high barriers to entry, others increase the likelihood of exits by incumbents in the long run. This also shows caution should be taken in using productivity gaps between entering and exiting firms as the key factor in analysis on business dynamics.
3. Thesis goal
In Section 2 I clarify the importance of business dynamism and discuss the reason to include high-tech startups as a separate group to investigate. In this section, I address the main goal of this paper and formalize the central question and hypotheses of this study.
3.1 Research question
Following the hypotheses of related literature that the positive job growth contribution of firms is driven by a small number of high-growth firms (Bijnens & Konings, 2018) and that small (young) high-growth firms have historically been an important contributor to productivity growth (Decker et al., 2018) and job creation (Bravo-Biasco et al., 2016; Haltiwanger et al., 2013), I argue how these hypotheses relate to the Dutch economy. However, the focus of this research primarily lies on the way business dynamism evolves in the Netherlands and how this determines net job creation. Following Haltiwanger et al. (2013), I track the role of firm size, age and the firms’ growth path in this trend of changing business dynamism. I also shed light on the debate around the importance of startups in this process and how significant this seemingly important group of companies for the Dutch economy is.
Hence the question I answer in this paper is: how does business dynamism evolve in the Netherlands over the period 2006-2018, and how does the impact of high-tech startups on this dynamism differ from the average firm?
3.2 Hypotheses
The first hypothesis is related to the survival probabilities of firms. I expect the survival probability, that is the chance of staying in business until the next calendar year, to decrease for firms of the same age after the great recession. I expect this to encounter both within the sample of all firms and startups. Hence the first hypothesis is:
7
Moreover, related to the expected decline in survival probabilities, I expect a decrease in relative entry and exit rates. Related to the declining trend in business dynamism in the world as shown by Criscuolo et al. (2014), I expect that these rates are in decline for the average Dutch firm. Research by Bijnens & Konings (2018) finds that high-tech sectors disproportionately react to crises compared with the average firm. I therefore expect the dynamism among high-tech startups to deviate from the average firm in the sense that the startups’ entry and exit rates react more severely. Hence the second hypothesis is:
H2: Entry rates decline and exit rates increase in the Netherlands over the period 2006- 2018 for the average Dutch firm, and for high-tech startups these trends are stronger. As already mentioned in the literature overview section, young and small firms tend to disproportionately contribute to net job creation. In line with this finding, I expect young and small firms to contribute disproportionately to net job creation, and hence the third hypothesis is:
H3: Young and small Dutch firms are disproportionate net job creators over the period 2006-2018.
In addition, I expect only a small part of firms to grow fast, hence this is expected to show up in the skewness of growth rates3. In line with Calvino et al. (2016), skewness is expected to
show that the median firm experiences little growth, while a small portion (the fastest growing 10 percent) of the firms show high growth in employment, which are coined the “superstar” firms in literature (Autor et al., 2017). I expect that the existence of “superstar firms” is even more evident in for high-tech startups as compared with the average firm. Hence the fourth hypothesis is as follows:
H4: Over the period 2006-2018, the distribution of net employment growth is skewed to the right considering the average Dutch firm and even more heavily skewed to the right in case of high-tech startups.
As hypothesized above, startups are expected to show higher growth compared with the average firm. Within this group of firms however, I expect that there are significant differences in weighted net employment growth if one investigates it on an age class level or on a size class level. Both in terms of all firms and startups, I expect weighted net employment growth to fall with firm size and firm age. Hence I expect the following to hold:
3 With skewness I measure the symmetry of net employment growth distribution. The fastest growing 10
8
H5: In the Netherlands, over the period 2006-2018, weighted net employment growth of all firms and startups in the Netherlands is inversely related to firm size and firm age. Lastly, startups and young firms tend to have much higher exit rates compared with more mature firms, and thus are expected to destroy more jobs. Hence the related sixth and last hypothesis is:
H6: The rate of job destruction decreases with age.
4. Data
In answering the main question stated above, I conduct an empirical analysis on data I gather primarily from two sources: Statistics Netherlands and Techleap. In this data section I first describe the sources of the data. In the second part of this data section, I present the type of data and the steps I take in constructing the main dataset. In the third part, I explain several definitions which are important for this research but may need some clarification. Next to these definitions, I also explain the construction of the main variables. For a thorough stepwise explanation of the construction of the “master file”, see appendix A.1. Finally, in the fourth and last subsection I introduce descriptive statistics.
4.1 Data sources
In analysing business dynamism in the Netherlands, I exploit large microdata sets from Statistics Netherlands. The data stems from the general business register (for simplicity abbreviated as GBR) of Statistics Netherlands, which comprises companies and institutions with their identification- and structure data, established and registered in units suitable for statistical research. Due to changes in the definitions in the GBR, I use data for the period 2006 to 2018.
9
regarding the obligation to declare VAT. The number of EP in the GBR is the sum of the employees and non-employees converted to FTE.
I analyse the data from Statistics Netherlands over the years 2006 up until 2018, where I limit the analysis to this period because of data restrictions. In this time frame I consistently measure the age and size of all firms and corresponding establishments. For firms started before 2006, the age is unknown and hence these firms are left out.
Appendix A.1 describes in detail how I merge all the available data in the GBR. I use a firm level analysis. My final sample consists of 1.479.454 observations on a total of 449.603 firms4.
In the analysis on high-tech startups, data is made available by Techleap, a public-private partnership in which both the startup community and the government contribute and help to build a solid startup landscape. With the help of Dealroom, a data analyst on business information in Europe, the startups are identified. This dataset includes only high-tech startups, which are gathered according to a selection criterion of Techleap.
In constructing their dataset, Techleap and Dealroom try to filter high-tech startups in their first years of existence, based on their criteria. This early identification helps to counter most concerns regarding endogeneity, as late firm identification automatically leads to higher probabilities of higher growth rates in the future. Among the variables are data on their entry/exit dates as well as their employment levels. In section 4.3.1 I explain the construction of the startup database in more detail, which includes more information on the selection process. The total amount of observations in my sample consisting of only high-tech startups is 13.690, which represents a total of 4.166 firms.
4.2 Firm level analysis
In this part I clarify why I use firm analysis as opposed to establishment analysis. It is important to note that one can perform the analysis in this paper both on the establishment level as well as on the firm level. However, the focus in this paper is on business dynamics, and in that case firm level analysis is preferred.
Only using establishment data poses some problems, as opening new stores for example shows up as new entry in that particular year, although its age and size should be based on the size and age of the parent firm. Hence, the age and size of the establishments may not provide the right information on the relevant parent firm’s age and size. I take a large retail chain or supermarket chain as an example: their new establishments could lead to overestimated results in entry and exit dynamics. In this paper I refer to an establishment when it serves as a specific
4 Some firms have multiple observations per year, as there are multiple establishments part of the same
10
location where business activity takes place, while a (parent)firm captures all establishments under the common operational control of this firm.
According to Haltiwanger et al. (2013), the most important drawback of firm level analysis as opposed to establishment level analysis is the fact that net employment change may simply reflect firm structure changes (for instance M&A activity). Thanks to the high quality of the data from Statistics Netherlands, I abstract from this. In the data from Statistics Netherlands, I construct a distinctive observation for every merger, restructuring or acquisition. In this way, I minimize the impact of new establishments, because the data from Statistics Netherlands provides proper establishment linkage to the parent firm. When a new establishment identifier appears in the GBR data, I assign this establishment an age based upon the age of the oldest establishment of the parent firm. Only if this establishment creates a parent firm on its own I consider it an entrant. In my analysis I consider a firm to exit only when it stops its business completely, or because it is acquired by another firm. In both cases the firm is no longer a parent firm, and hence I delete it from my dataset. Thus, the firm age and size measures are robust to ownership changes. Furthermore, also in case of startups in the Techleap data, the data is robust to ownership changes.
Thus, using firm level analysis, job creation equals employment gains from new and expanding firms, and job destruction equals the job losses from closing and contracting firms. Note that firm level data measures are lower in terms of job creation and job destruction, as the establishment level data measures include establishment reallocation as new entries/exits.
4.3 Definitions
Startups are at the centre of this paper, and therefore a clear definition of the concept is necessary. What is for instance the difference between an SME and a startup, or the relation between entrepreneurship and startups? The answers to such questions remain fuzzy, hence a thorough understanding of these concepts is vital for this project. Next to the understanding of what high-tech startups are, it is important to distinguish which firms I consider in the analysis.
4.3.1 Entrepreneurship and startups
11
As explained above, both individuals and corporations can conduct entrepreneurial activities. The crucial differences between innovative entrepreneurs and SME-owners is innovation (Zahra et al., 2011). Startups are the businesses that pursue opportunities and strive to be innovators, and in doing so seek for rapid growth, whereas SMEs do not have such ambitions. Techleap, in constructing its database, distinguishes between several characteristics of startups. These companies share most (not necessarily all) of the following characteristics; (i) innovative by design, that is the product and/or business model are innovative, (ii) based or tech-enabled companies, (iii) rapidly scaling or scalable businesses, and (iv) global ambitions. Thus, the companies I define as startups in this report are firms that are innovative by nature, are the firms with business models and technologies and are embedded in internet verticals (that is an firm with a business model that is closely linked to the internet like telecom, fintech or transport). Next to the criteria above, Techleap excludes a large group from of firms having less than 1 FTE. In addition, firms being older than 10 years old and service providers are excluded. However, given the above-mentioned criteria, I recognize that having very strict definitions is tempting but often creates many nonsensical results. Global ambitions is one of the criteria for startups, but this seems rather vague. Its primary purpose is to filter firms that explicitly do not have global ambitions. Thus, these definitions are not to be strictly followed, as I make several exceptions. Take the age of the firms: most research take an upper limit of 10 years old for startups, but some companies who exceed this still are relevant for the analysis of Dutch startups (for instance Adyen), and hence I include those. Furthermore, the growth of a firm is often used as characteristic for distinguishing startups. However, the difficulty with growth metrics is the fact that this metric might not be available for a large part of young firms. In fact, it can even be misleading to filter only high growth firms, since the development of a firm happens in different stages of growth and looking to narrowly at growth might cause a bias in the results (Bravo-Biosca et al., 2016).
For the detailed stepwise process of selecting startups and merging the company-identifiers with the data of Statistics Netherlands see Appendix A.2. After the process of cleaning and merging the data, the number of companies that represent high-tech startups in the Netherlands is only 4166, spread over the whole sample period. This group of companies is a rather small part of the Dutch business environment, but the impact some startups have still is significant. In Section 4.4 I give an overview of descriptive statistics on startups.
4.3.2 Small, large, mature
12
Above I define some broad size and age classes, but the construction of the variables firm-age and firm-size needs some further clarification. The firm size measure I use is relatively straightforward. I comprise firm size by the aggregate of employment of all establishments belonging to the same parent firm. Size is measured as the average size of the firm in year t-1 and year t. New firms are assigned an average size equal to their size in year t.
Firm age is a bit more challenging to construct. Since I analyse the data based on the firm level, I assign a parent firm an age based upon the age of the oldest establishment in the firm. I base a new establishment’s age on the age of the oldest establishment belonging to the same parent firm. Note that a firm birth only appears in the data when an establishment instantaneously creates its own parent firm. A firm death occurs when a firm stops existingand thus with it all its establishments. Furthermore, note that in case of an acquisition, the acquiring firm determines the age of the acquired firm, even if the acquiring firm is a younger firm.
4.3.3 Firms
In my analysis, I differentiate between two types of firms in the Netherlands. As the focus of this paper is on business dynamism and employment growth, it is useful to only consider non-financial limited liability firms. First, the reason to only consider limited liability firms is essential. The number of employees with flexible contracts has risen substantially in the last two decades. A study by the Bureau for Economic Policy Analysis (Hoekstra et al., 2016) in the Netherlands finds that flexible contracts and the number of self-employed persons has risen with approximately 10 percent from 2003 until 2015. Thus, in case I include freelancers and self-employed persons, the dynamism in the country is overestimated because of this shift in structure. Furthermore, in all cases, high-tech startups start as limited liability company, hence leaving out the non-limited liability firms improves the comparison of high-tech startups versus the average firm. Second, I study only non-financial firms, because including financial firms skews the results. A large part of company owners set-up a financial holding next to their regular company. Financial holdings might have a significant impact on churning, as these are set-up by company owners or large corporations for pure administrative or financial reasons. Thus, this would significantly overstate the entry dynamics, as a limited liability company is often accompanied with the establishment of a financial holding in the Netherlands. Important to note is when I mention “firms” or “all firms” in the subsequent sections of this paper, I refer to the all non-financial limited liability firms.
4.4 Descriptive statistics
In this subsection I illustrate the points mentioned above with the aid of tables and figures with descriptive statistics. The definition of age and size I discuss in Section 4.3.2, and I now present the statistics of these firm age and size classes.
13 Table 1: Distribution of firms
All firms Startups
Size class No. % No. %
0-1 1024672 69,26 5951 43,47 2-4 244216 16,51 3322 24,27 5-9 106246 7,18 2145 15,67 10-19 58993 3,99 1269 9,27 20-49 31048 2,10 653 4,77 50-199 12079 0,82 301 2,20 200-499 1645 0,11 37 0,27 500+ 555 0,04 12 0,09 Total 1479454 100,00 13690 100,00
All firms Startups
Age class No. % No. %
0 389017 26,29 3412 24,92 1 293923 19,87 2731 19,95 2 201199 13,60 2015 14,72 3 150935 10,20 1492 10,90 4-5 202212 13,67 1985 14,50 6-7 123795 8,37 1097 8,01 8-9 74452 5,03 599 4,38 10+ 43921 2,97 359 2,62 Total 1479454 100,00 13690 100,00
In this table I present the distribution of firm observations per size and age class. This is the distribution of firm observations in the Netherlands over the whole sample period 2006-2018. The total amount of observations is captured under No., while the share of firm observations from the total in an age or size class is captured by %. One firm has multiple observations in the dataset if it is in business for more than 1 year, and hence the amount of observations exceeds the number of actual firms.
Source: Statistics Netherlands – general business register & Techleap database
14
Next to the distribution in age and size classes, another important statistic is the survival rate of firms. What is the probability that a firm survives until the next year? In Table 2 I shed light on this measure, as I show the probability a firm survives until the next year given its year of entry. In Table 2a I show the survival chart for all firms, and in Table 2b I show the survival chances for startups of the same year cohort.
In Table 2a I show that, considering the average firm in the Netherlands, the probability of surviving up to an age of 1 fluctuates around or slightly above 80 percent. The probability of surviving 1 year is at a small peak in 2007, from where it declines until 2011. Thus, during and shortly after the great recession of 2007-2009, these survival rates drop. This is not only evident for the survival until the age of 1, but also for higher ages of firms. What furthermore is evident from Table 2a, is the fact that only approximately 50 percent of all firms in the Netherlands with an entry year after 2005 survives until the age of 4.
If I compare the statistics in Table 2a with that in Table 2b, the most striking outcome is the difference in magnitude of the survival rates, as they are higher for every age-cohort considering high-tech startups. Considering all firms, the probability of survival up to an age of 1 lies around 80 percent, however, this is around 90 percent in most cohorts of high-tech startups. Similar to the the average firm, survival probabilities of startups decrease, with the most prominent drop in the first year right after the great recession. After 2010 there is more fluctuation in this first year survival probability. Next to the decline in the probability of first year survival, I also detect a decline in survival probabilities to higher ages in the years after the recession. Thus, in case of the average Dutch firm and high-tech startups, the survival probabilities decrease since the occurrence of the great recession. This last statement is in line with the first hypothesis of declining survival probabilities after the great recession, both in case of all firms and startups. Thus, firms entering in post-crisis years do experience a lower probability of remaining in business.
In Table 1 I illustrate the trend of higher probability of exit, however I fail to show whether the relative exit share within the same year cohort is in decline too. Plotting the exit share together with the relative entry share helps in demonstrating the evolution of dynamism over the period 2006-2018. Therefore, In Figure 1 below, I illustrate the relative shares of entry and exit by year. First in Figure 1a I illustrate the average firm entry and exit shares, and in Figure 1b I illustrate these shares for the subset of startups.
15 Table 2: Survival chart
A) All firms Year of entry Year 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2006 1,00 2007 0,83 1,00 2008 0,70 0,89 1,00 2009 0,60 0,69 0,86 1,00 2010 0,49 0,57 0,70 0,84 1,00 2011 0,42 0,48 0,58 0,64 0,79 1,00 2012 0,38 0,43 0,51 0,55 0,62 0,78 1,00 2013 0,34 0,39 0,45 0,48 0,52 0,59 0,81 1,00 2014 0,32 0,35 0,40 0,42 0,45 0,49 0,60 0,82 1,00 2015 0,29 0,32 0,36 0,37 0,39 0,40 0,48 0,58 0,80 1,00 2016 0,27 0,29 0,33 0,34 0,35 0,36 0,41 0,48 0,58 0,82 1,00 2017 0,24 0,27 0,30 0,32 0,32 0,33 0,36 0,39 0,41 0,54 0,83 1,00 2018 0,23 0,26 0,29 0,29 0,30 0,30 0,34 0,37 0,41 0,52 0,60 0,87 1,00 B) Startups Year of entry Year 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2006 1,00 2007 0,98 1,00 2008 0,97 0,96 1,00 2009 0,96 0,91 0,99 1,00 2010 0,87 0,89 0,91 0,98 1,00 2011 0,85 0,80 0,86 0,89 0,85 1,00 2012 0,78 0,74 0,80 0,82 0,71 0,92 1,00 2013 0,68 0,69 0,72 0,76 0,65 0,84 0,86 1,00 2014 0,67 0,68 0,67 0,73 0,61 0,77 0,74 0,91 1,00 2015 0,63 0,62 0,66 0,66 0,58 0,68 0,65 0,78 0,87 1,00 2016 0,59 0,58 0,62 0,59 0,49 0,67 0,64 0,73 0,74 0,93 1,00 2017 0,47 0,53 0,58 0,49 0,43 0,59 0,53 0,59 0,59 0,68 0,80 1,00 2018 0,48 0,51 0,54 0,46 0,43 0,55 0,52 0,58 0,52 0,61 0,69 0,83 1,00 This table shows the probability of surviving until the next year for different cohorts of firms in the
Netherlands. Table 2a captures the survival rates of all firms, Table 2b demonstrates these rates
considering startups. The year of entry indicates in what year the company started its business. The cells give a percentage chance of surviving until the year stated in the left column. As an emample, for a company entered in year 2007, the percentage chance of surviving until 2007 is 100 percent, while the chance is 89 percent to survive until 2008.
16 Figure 1: Relative entry and exit rates
A) All firms
B) Startups
This figure presents the relative entry and exit share by year for the whole sample of firms and for startups only in the period 2006-2018. It plots the number of firms entering or exiting the market as a percentage of total firms. Figure 1A graphs these rates for all firms, Figure 1B graphs these rates using only startups.
Source: Statistics Netherlands – general business register & Techleap Database
0 .05 .1 .15 .2 .25 2006 2008 2010 2012 2014 2016 2018 year
95% Confidence interval 95% Confidence interval
% Exit % Entry % 0 .05 .1 .15 .2 .25 .3 .35 .4 2006 2008 2010 2012 2014 2016 2018 year
95% Confidence interval 95% Confidence interval
17
similar trend in exit rates, only they find that in Belgium the exit rates start to increase a year later, from 2011 onwards. One possible explanation for this disparity might be the higher level of international trade in the Netherlands, initiating a sooner downturn. In the entry and exit dynamics for high-tech startups I detect a slightly different pattern. As I show in Figure 1b, the relative entry share declines from 2006 until 2009, from where it starts to increase again, but from 2012 onwards the entry rates are in steady decline again. Where the entry rates of the average firm revers back to the pre-crisis level, the entry rates of high-tech startups are significantly lower in 2018 as compared with 2006. The higher responsiveness to recessions from high-tech firms is not completely surprising, as this is also documented by Decker et al. (2018) for U.S. firms in the high-tech sector. Moreover, the relative exit share of high-tech startups moves in a different fashion as the relative exit share of the average firm as shown in Figure 1a. In Figure 1b I demonstrate that the exit share of high-tech startups declines in the last year prior to the crisis, stays relative stable within the crisis, but jumps up to a significantly higher level in 2010. From 2010 onwards, the relative exit share of high-tech startups then remains relatively stable at this higher level. Thus, where the exit rates of the average firm show 2 peaks but almost completely reverts to its pre-crisis level, the exit rates of startups remain higher over a longer period after the great recession. Nevertheless, the change in exit rates within the group of startups from 2006 until 2018 is not significant, as the 95 percent confidence interval in 2018 overlaps with the 2006 exit rate.
All in all, considering the whole sample period, the entry rate of the average Dutch firm shows a decline in reaction to the crisis, but the 2018 value does not significantly deviate from its 2006 value. This contradicts part of my second hypothesis, since I expect a decline in those rates for the average Dutch firm. Though, in line with the second hypothesis is the entry dynamics among startups. The entry rate of startups is significantly lower in 2018 compared to 2006, and hence the effect of the recession is stronger compared with the average Dutch firm. Now for the exit rates, both in case of the average firm and startups, I observe it is in contradiction with the hypothesis of an increase in exit rates between 2006 and 2018. Although in both cases the exit rates increase during this period, they revert to the pre-crisis level of 2006. In the remainder of this paper I illustrate how entry and exit dynamics relate to the different size and age classes and their contribution to net job creation. However, the dynamics I present in Figure 1 together with Table 2 already hint at a decrease in overall business dynamism. This decline in business dynamism is mostly apparent in the first few years after the crisis with higher probability of exit and increasing exit rates together with a lower level of entry. For high-tech startups I display that this decrease in business dynamism continues to exist for a longer period after the great recession.
18 Table 3: Net job creation 2018
A) All firms
Size class
Age class Small Medium Large Total
<1 19743 9314 2887 31944 1-4 11082 10943 9448 31473 5-9 808 5258 6564 12630 10+ 29 2301 5642 7972 Total 31662 27816 24541 84019 B) Startups Size class
Age class Small Medium Large Total
<1 692 265 10 967
1-4 4261 1181 299 5741
5-9 1310 899 424 2633
10+ 719 315 100 1134
Total 6982 2660 833 10475
This table shows the total amount of net job creation within a particular age or size class in 2018. Table 3A presents the net jobs created when observing all firms in the Netherlands. Table 3B shows the total amount of net job creation for startups only. Net job creation is measured as the difference between gross job creation and gross job destruction at a firm level.
Source: Statistics Netherlands- general business register & Techleap database
Table 3 relates to figure 1 on entry and exit dynamics, as entry of firms creates jobs, whereas firm exit destroys jobs. I split the firms into either small (<10), medium (10-49) or large (50+) in terms of average employment. The age classes span firm births (age <1), young firms (age 1 - 4) or mature firms split up into two categories (age 5-9 and 10+). Both Table 3a and 3b yield several interesting facts. The first thing to notice is that the total amount of net new job creation in 2018 equals 84.019. Moreover, what is interesting is the amount of net job creation when comparing the different age and size classes. I observe, both considering all firms and high-tech startups, that all age and size categories are net job creators. Considering all firms, the group with the most net jobs created in 2018 is firm births. Important to note about this figure is the fact that firm births do not destroy jobs, hence their net job creation equals their gross job creation. Furthermore, what is striking is the fact that the largest share of net jobs created (approximately 75 percent) comes from firms aged 0 to 4, while firms aged 10+ are merely job creators. The larger an old firm gets, the higher is their contribution to net job creation.
19
show high growth in their early stage. Thus, while net job growth is dominated by the young firms considering the average Dutch firm, it is mostly dominated by the younger small firms in case of startups.
Although I extract some useful facts from Table 3, it is not clear how this relates to the relative size of the group of firms. From this, I am unable to answer whether young or small firms are disproportionate net job creators, as expected in my third hypothesis. This is the basis of figure 2, as I concentrate on the accumulated employment shares and the relative shares of job creation and destruction. In Figure 2a I illustrate these shares by age class, and in Figure 2b I highlight the shares by size class. I use the same age and size classes as in Table 3. I focus on the interaction of age and size classes later in this paper, when I turn to the empirical section. First, in figure 2a I capture the relative shares of employment, gross job creation, gross job destruction and net job creation by age class. Notice that job creation and job destruction appear on the establishment level, but I classify it by the characteristics of the parent firm that owns them. In both figures 2a and 2b I solely focus on the average Dutch firm, not on high-tech startups specifically. One should note that if the contribution to employment is smaller than the contribution to gross jobs created, I consider the group of firms to create jobs disproportionately. The same holds for the job destruction statistic or the net job creation statistic. Hence a disproportionate net job creator is a firm which has a higher share in net job creation than their relative contribution to overall employment.
20 Figure 2: Employment and net job creation
A) shares of employment and (net)job- creation and destruction by age-class
B) shares of employment and (net)job-creation and destruction by size-class
Both figure 2a and 2b demonstrate the shares of employment, job creation, job destruction and net job creation. The blue bars represent the share of employment as a percentage of total
employment. Green and Red respectively represent the share of job creation and destruction as percentage of total job creation and destruction. In the orange bar I present the relative shares of net job creation, that is the difference between job creation and destruction.
Source: Statistics Netherlands – general business register & Techleap database
0 .2 .4 .6 .8 <1 1-4 5-9 10+
Employment Job creation
Job destruction Net job creation
% Age class 0 .1 .2 .3 .4 .5
Small Medium Large
Employment Job creation
Job destruction Net job creation
Size class
21
exit in the Results Section even more.
Second, in figure 2b I illustrate the same relative shares as in figure 2a, only now I divide the firms by size class. Notice that both small and medium size firms are disproportionate net job creators, as their share in net job creation exceeds their share in employment. Furthermore, related to this first observation is the relative low share in job creation by large firms. Most jobs in the economy are situated at large firms, but their share in both job creation and job destruction falls short of their employment share. The opposite is true for small and medium size firms, which illustrates the importance of these firms in job reallocation. All in all, the age and size classes I consider disproportionate net job creators are firm births, and small and medium sized firms. Thus indeed, as expected in the third hypothesis in this paper, small firms are disproportionate net job creators. My findings are in line with related literature (Bravo-Biasco et al., 2016; Haltiwanger et al., 2013; among others) stating the importance of small and young firms in the creation of jobs and the reallocation of labour. In terms of reallocation, it appears age matters more, as young firms have high levels of job creation and destruction. However, in terms of actual job creation, it appears size matters more, as job creation mostly happens within the small and medium size firms
As startups tend to be small and young, they might contribute to the fact that I consider these smaller (and to a lesser extent younger) firms job creators. In the empirical section I distinguish between the contribution of startups and young/small firms to weighted net job creation, and hence investigate whether they indeed contribute in improving the outlook of small and young firms.
This section previews the dynamics on businesses in the Netherlands, but not examines the evolution of net job growth within certain age or size classes. I illustrated the fact that survival probabilities of firms in the Netherlands are in decline since the great recession. The relative entry and exit rates however only realize a short spike, but no long-term reaction. Although, in the case of startups, the entry rate declines significantly after 2012. In terms of contribution to net job creation, the small and medium size firms together with firm births are the largest contributors to net job creation in relation to their share in employment.
In the next section I focus on the weighted net job creation among different age and size classes. I introduce a saturated regression model and explain how I estimate the contribution of different groups of firms in the weighted net employment growth and business dynamism.
5. Empirical strategy
22
growth rate of firms, and examine how this relates to the fourth hypothesis in this paper. For this matter, I examine the evolution of skewness of the net employment growth rate distribution. After this, in the second part of this section, I shift the focus to a regression model in measuring the relative contribution of classes of firms on the total employment, and how the firm classes relate to each other in terms of exit dynamics and growth. I discuss the results of this empirical strategy in Section 6.
5.1 Skewness of employment growth
I measure skewness using firm employment growth rates. For the firm employment growth rate, I use a weighted net employment growth measure introduced by Davis, Haltiwanger and Schuh (Davis et al., 1996) abbreviated by DHS. This growth metric compares the absolute growth of a firm between period t and t-1 (Eit – Eit-1) with the average employment of the firm
in the two periods (Xit,). Hence DHS growth is a weighted net employment growth rate, and I
capture this by the following expressions5:
Git = (Eit – Eit-1) / Xit = Cit – Dit,
with Xit = (Eit + Eit-1)/2
where Git measures the change in weighted net employment, Eit is the employment level of firm
i in period t, and Cit and Dit represent respectively job creation and job destruction by firm i in
period t. Net job creation is the most relevant measure in this analysis, however it is insightful to look at gross job creation and destruction separately as well. Net job creation of five percent can be the result of six percent gross job creation and one percent gross job destruction, or 20 percent gross job creation and 15 percent gross job destruction. The difference is the amount of jobs reallocated, which is an important issue in macroeconomics. Job creation of all firms at time t equals the sum of the change in employment level of the subset of firms that expand, hence
Cit = ∑𝑓𝑓∈𝑆𝑆+(𝑋𝑋𝑖𝑖𝑖𝑖 − 𝑋𝑋𝑖𝑖𝑖𝑖−1)= ∑𝑓𝑓∈𝑆𝑆+ ∆𝑋𝑋𝑖𝑖𝑖𝑖,
where ∆𝑋𝑋𝑖𝑖𝑖𝑖 denotes the first-difference operator of employment, hence the change in
employment from year t-1 to t for firm i. The 𝑓𝑓 ∈ 𝑆𝑆+ sign reflects the fact that this summation
only holds for firms that expand employment levels.
Job destruction by all firms at time t equals the sum of the changes in employment level of the subset of firms that contract, hence
Dit = ∑𝑓𝑓∈𝑆𝑆− ∆𝑋𝑋𝑖𝑖𝑖𝑖,
where the 𝑆𝑆− sign denotes the subset of firms (𝑓𝑓) that contract in terms of employment levels,
and ∆𝑋𝑋𝑖𝑖𝑖𝑖 again denotes the first-difference operator of employment. Considering job
5 In the rest of the paper, I use the terms DHS growth, weighted net employment growth rate, or employment
23
destruction, I check the share of job destruction accounted for by firms of different age and size classes.
This DHS growth measure I utilize in the analysis is a second order Taylor approximation to the log first differences. This is useful in this case as the growth rate of employment usually is much smaller than 1, hence using the advantage of the Taylor approximation I can rewrite the growth rate of Git in the functional form as I state above. It is standard in firm dynamics analysis
because of its two useful properties: (i) it accounts for both entry and exit rates, and (ii) it is symmetrical around zero and bounded by -2 and +2, where firms on the -2 end represent exits and on the +2 end represent entrants. By construction this measure cannot exceed 2 or -2, as the denominator cannot be more than two times smaller as the numerator. I divide the sum of total employment change by 2, hence this is never two times smaller than the actual change in employment from year t-1 to year t. Another important but vastly overlooked benefit from using DHS growth as growth metric is pointed out by Neumark et al. (2011), as they state that classifying a firm by start-year size biases measurements. Firms who just experienced a negative shock to their employment appear to be small in time t and will be mistakenly classified as smaller. Hence, this implies this “smaller” firm will grow faster next year because of regression to the mean effects. The reverse can be said about using end-year size classification, as this yields a positive size bias. I avoid this measurement error since I measure employment growth according to the DHS method. Exploiting this average size measure, I only classify a firm as small if it stays small for two consecutive years.
Accordingly, by using the above explained growth metric, I generate skewness statistics and test the fourth hypothesis in this paper. I calculate skewness by calculating the difference between the 90th and 50th percentile of employment growth and from this I subtract the 50-10
employment growth differential. In Figure 3 I depict the mean skewness for all firms and for startups over the period 2006-2018, together with the 95 percent confidence interval.
A positive skewness indicates high employment growth among the highest percentiles of growth, while a negative skewness suggests a relatively low employment growth among the slowest growing firms. In Figure 3 I observe a negative skewness throughout the whole sample period, both for all firms and startups. This contradicts my fourth hypothesis related to the existence of “superstar firms”, as a negative skewness means the employment growth difference between the 50th percentile and the 10th percentile is higher than the difference in
employment growth between the 90th percentile and the 50th percentile. Or in other words, the
24 Figure 3: Skewness
In this figure I demonstrate the skewness statistics with 95 percent confidence interval for the whole sample (all firms) and for startups only in the period 2007-2018. Skewness refers to the difference between the growth differentials 90th – 50th and 50th – 10th . The connected
red-line represents the mean skewness level, whereas the light-blue area depicts the confidence interval of 95 percent.
Source: Statistics Netherlands – general business register & Techleap database
Altogether, skewness converges towards zero after the crisis, which signifies Dutch firms become more homogeneous in terms of employment growth. This trend is in line with the decrease in skewness and the decreasing proportion of high-growth firms in the U.S. as documented by Decker et al. (2016). More homogeneous growth also hints at less creative destruction and less reallocation. In turn, declining dynamism can be derived from higher homogeneity, less reallocation of resources and lower entry and exit rates. Hence again, related to the same observation in Section 4.4., I conclude that business dynamism seems to be in decline.
In the next subsection I turn to a regression model and estimate the relationship between DHS growth and the different age and size classes a firm can fall in. I furthermore examine the relationship between job destruction and firm age or size. The results of these estimations I present in Section 6. -1.5 -1 -.5 0 2006 2008 2010 2012 2014 2016 2018 2006 2008 2010 2012 2014 2016 2018
All firms Startups
95% Confidence interval Skewness
year
25 5.2 A fully saturated dummy variable model for net growth
To further analyse business dynamism, I shift the focus to regression analysis and try to find a causal relation between dynamism in the Dutch economy and net job creation and growth. Net employment growth of firms is influenced by firms’ size, entry and exit, and their age. In this analysis I emphasize the role of these determinants individually, as well as together, on the growth path of firms. According to Haltiwanger et al. (2013) the startup process in the U.S. is a key determinant in net job creation. They stress that, although the ongoing dynamics of mature firms is important, the entry process and post-entry dynamics of young and small (startups) really make a difference. The purpose of this and the next section is to examine how this relates to the Dutch economy.
In the regression analysis I use a panel data regression with a fully saturated dummy variable model. These types of regressions use discrete explanatory variables, which is true for the explanatory variables I use in this research. I include a separate parameter for every value the explanatory variables can take. Hence in this case I include a single dummy for every age and size class a firm can be in. My model takes on two explanatory variables and as such is only saturated when I include dummies for all age and size classes, as well as their product, and a constant. The coefficients on the dummies in this case are the main effect, whereas their product is the interaction term. The regressions I adopt is an one-way dummy variable model in firm age or firm size and an interaction term. Note that if I regress the dependent variable ‘weighted net employment growth’ for every value of age and size, I still estimate the conditional expectation function as in the case of including these variables as continuous explanatory variables. Moreover, this avoids issues in dealing with a limited dependent variable such as the DHS growth rate in this research (which is bounded by -2 and +2, as explained in Section 5.1). One should note that the saturated dummy variable model gives many estimates, because it includes all main effects and possible interactions. I explicitly focus on (i) the partial effect (or average mean effect) of age classes on net employment growth with and without size controls, and (ii) the partial effect of size classes on net employment growth with and without age controls.
Using a saturated dummy variable specification is an extension to a standard regression model where one includes an interaction term. The standard form regression with interaction term, as I utilize in this paper is:
git = α + β1 sit + β2 ait + β3 sit * ait + εit, (1)
where, in the first regression, git represents the DHS growth of firm i in period t, sit is the size
of firm i in period t, ait is the age of firm i in period t, sit* ait is an interaction term, and α is a
26
the job destruction rate by exit. Hence, with this as dependent variable, I capture the relation between firm size or age and job destruction by exit.
Following Haltiwanger et al. (2013), in constructing the one-way saturated model, I generate and include 8 age- and size-classes. In this case “one-way models” refer to models creating a single dummy variable for all age or size classes, keeping the other factor constant. I test for partial associations, hence either holding the age distribution constant while regressing size on net employment growth, or vice versa. I first test for partial associations of firm size holding age constant, and thus include 8 size classes. I estimate the following regression, where si = 0,
1, 2, … ,8:
𝑦𝑦𝑖𝑖𝑖𝑖 = 𝛽𝛽0+ ∑7𝑘𝑘=1𝛽𝛽𝑘𝑘𝑑𝑑𝑘𝑘𝑖𝑖𝑖𝑖 + 𝛽𝛽2 𝑎𝑎𝑖𝑖𝑖𝑖+ ∑7𝑘𝑘=1𝛽𝛽3(𝑑𝑑𝑘𝑘𝑖𝑖𝑖𝑖𝑎𝑎𝑖𝑖𝑖𝑖) + 𝜀𝜀𝑖𝑖𝑖𝑖, (2)
where 𝑦𝑦𝑖𝑖𝑖𝑖 is DHS growth, 𝛽𝛽0 is a constant, and the second right-hand side termrepresents 7
main effects of the size category on net employment growth. In this second right-hand side term, dsi = 1[si = k] is a dummy variable indicating size class k for firm i. In equation 2 I
furthermore include one main effect of age, and 7 size-age interactions which I capture with the fourth right-hand side term in the equation. The interaction terms, 𝛽𝛽3, indicates how each
of the size effects differ by age. Note that I perform this analysis twice, using all firms and startups only. Also note that in the above regression I generate results that represent the difference relative to a baseline group. This is the reason why I only interact seven size classes with age, as I use the biggest size class as baseline group, thus omit it. Hence, in the Results section I highlight the effect of size on net employment growth relative to the group of firms in the largest size class. The reason I chose the largest size class is mainly because I focus my analysis on small and young firms, thus omitting the smallest firms would complicate the interpretation of the results.
In the second estimation, I test for partial associations of firm age holding size constant, and I include 8 age classes. This leads to the following regression, where in this case I use ai = 0, 1,
…, 8.
𝑦𝑦𝑖𝑖𝑖𝑖 = 𝛽𝛽0+ ∑7𝑗𝑗=1𝛽𝛽𝑗𝑗𝑥𝑥𝑗𝑗𝑖𝑖𝑖𝑖 + 𝛽𝛽2 𝑠𝑠𝑖𝑖𝑖𝑖+ ∑7𝑗𝑗=1𝛽𝛽3(𝑥𝑥𝑗𝑗𝑖𝑖𝑖𝑖𝑎𝑎𝑖𝑖𝑖𝑖) + 𝜀𝜀𝑖𝑖𝑖𝑖, (3)
where 𝑦𝑦𝑖𝑖𝑖𝑖 is DHS growth, 𝛽𝛽0 is a constant, and the second right-hand side termrepresents 7
main effects of the age category on net employment growth. In this second right-hand side term, dai = 1[ai = j] is a dummy variable indicating age class j for firm i. In the third equation I
furthermore include the main effect of size, and 7 size-age interactions which I capture with the fourth term. Here the interaction terms, 𝛽𝛽3, indicates how each of the age effects differ by
27
Next to the analysis on the impact of size and age classes on net employment growth, I also conduct a regression on the impact of age or size classes on job destruction by firm exit. Job destruction by firm exit is measured as the total amount of employment within a firm that exits the market. I capture this in equation 1, and I present the results from this regression in the Results section in Figures 5 and 7.
6. Results
This section covers the results from the empirical analysis. In section 6.1 I first elaborate upon the estimation results from equation 2 together with the dynamics of job destruction. In the next subsection I demonstrate the link between DHS growth and firm age as I capture with equation 3. Furthermore, I elaborate on how job destruction by exit differs by age class. In the third subsection, I present a robustness check.
6.1 Net employment growth and firm size
I present the regression results from equations 2 and 3 in Table 4a for all firms, and in Table 4b for startups. Given the different age and size classes, many coefficients are present in Table 4. In interpreting the table, and the figures based on this table, it is important to notice that the estimated coefficients represent the difference in relation to an omitted size or age class. Note that the amount of observations is very high, hence the standard errors are extremely small. Because of the large number of coefficients in Table 4a and 4b, it is easier to illustrate the most important findings in figures. First, I plot the results of equation 2 in Figure 4a and 4b, and hence plot the relation between DHS growth and firm size classes, both with and without controlling for age. In Figure 4a I present the regression estimates based on the whole sample, whereas in Figure 4b I plot the estimates using only startups.
28 Table 4a: Fully Saturated Net Employment Growth Regression - All Firms
Parameter Age Size Age + Size
controls Size + Age controls
Age 0 # Average employment 0.00046*** 0.00086***
(0.000) (0.000)
Age 1 # Average employment -0.00166*** -0.0025***
(0.000) (0.000)
Age 2 # Average employment -0.00186*** 0.00003
(0.000) (0.000)
Age 3 # Average employment -0.00150*** 0.00009
(0.000) (0.000)
Age 4-5 # Average employment -0.00067*** 0.00046***
(0.000) (0.000)
Age 6-7 # Average employment -0.00034*** 0.00014
(0.000) (0.000)
Age 8-9 # Average employment 0.00001 0.00010
(0.000) (0.000)
Age 10+ # Average employment 0 0
(.) (.)
Size 0-1 # Company age 0.15524*** -0.01927
(0.009) (0.014)
Size 2-4 # Company age 0.15206*** -0.09432***
(0.009) (0.014)
Size 5-9 # Company age 0.12804*** -0.07339***
(0.009) (0.014)
Size 10-19 # Company age 0.09580*** -0.06905***
(0.009) (0.014)
Size 20-49 # Company age 0.06326*** -0.05925***
(0.009) (0.014)
Size 50-199 # Company age 0.02998*** -0.05941***
(0.009) (0.014)
Size 200-499 # Company age 0.00481 -0.04955***
(0.010) (0.016)
Size 500+ # Company age 0 0
(.) (.)
R^2 0.060 0.065 0.160 0.077
N 1479454 1479454 1479454 1479454
Standard errors in brackets. P-values: * p < 0.05 ** p < 0.01 *** p < 0.001. N = observations
This table displays the regression results from executing equations 2 and 3. Note that the regression is performed using all firms, in the whole sample period 2006-2018. The column “Age” displays the coefficients and standard errors of estimating equation 3. The column “Age with controls” captures the same, but with control variable average employment. Column “size” captures the coefficients and standard errors of equation 2, and “size with controls” gives the coefficients and standard errors of equation 2 with control variable age. What is shown in the table is the estimated coefficient of the
interaction term (interaction effect). I surpress the main effects of age and size class and the constant. The coefficients and standard errors of the largest or oldest size classes are zero because it concerns the omitted group.
29 Table 4b: Fully Saturated Net Employment Growth Regression - Startups
Parameters Age Size Age + Size Controls Size + Age
controls
Age 0 # Average employment 0.01068* 0.00705*
(0.001) (0.001)
Age 1 # Average employment -0.00261* 0.00119
(0.001) (0.001)
Age 2 # Average employment -0.00240* 0.00045
(0.001) (0.001)
Age 3 # Average employment -0.00121*** 0.00020
(0.001) (0.001)
Age 4-5 # Average employment 0.00011 0.00037
(0.001) (0.001)
Age 6-7 # Average employment 0.00025 0.00013
(0.000) (0.000)
Age 8-9 # Average employment 0.00000 0.00002
(0.000) (0.000)
Age 10+ # Average employment 0 0
(.) (.)
Size 0-1 # Company age 0.00068 -0.14110
(0.052) (0.101)
Size 2-4 # Company age 0.05038 -0.19309***
(0.051) (0.101)
Size 5-9 # Company age 0.04378 -0.14027
(0.051) (0.101)
Size 10-19 # Company age 0.03282 -0.10535
(0.051) (0.101)
Size 20-49 # Company age 0.03732 -0.10110
(0.051) (0.101)
Size 50-199 # Company age 0.01979 -0.08866
(0.051) (0.102)
Size 200-499 # Company age 0.01257 -0.12008
(0.059) (0.109)
Size 500+ # Company age 0 0
(.) (.)
R^2 0.120 0.108 0.196 0.129
N 13,690 13,690 13,690 13,690
Standard errors in brackets. P-values: * p < 0.05 ** p < 0.01 *** p < 0.001. N = observations
This table displays the regression results from executing equations 2 and 3. Note that the regression is performed using startups only, in the whole sample period 2006-2018. The column “Age” displays the
coefficients and standard errors of estimating equation 3. The column “Age with controls” captures the same, but with control variable average employment. Column “size” captures the coefficients and standard errors of equation 2, and “size with controls” gives the coefficients and standard errors of equation 2 with control variable age. What is shown in the table is the estimated coefficient of the interaction term (interaction effect). The main effects and constant are suppressed. The coefficients and standard errors of the largest or oldest size classes are zero because it concerns the omitted group.
30 Figure 4: DHS employment by size class
A) All firms
B) Startups
This figure captures the regression output from equation 2 using all firms and startups in the period 2006-2018. The four lines depict the relation between a particular size class and the DHS
employment growth rate. In both figures, the blue lines depict the estimation coefficients and the corresponding standard errors from equation 2 with size controls. The red lines depict the
coefficients and standard errors from equation 2 without size controls. The points are estimated as the percentage difference to the omitted group, which in this case is the group of companies with a size that exceeds 500 average employees.
Source: Statistics Netherlands – general business register & Techleap database
-.2 -.1 0 .1 .2 0-1 2-4 5-9 10-19 20-49 50-199 200-499
Size + Age controls Size
Firm age class
% -.4 -.2 0 .2 .4 0-1 2-4 5-9 10-19 20-49 50-199 200-499
Size + Age controls Size
%
31
small firms. However, when controlling for firm age the effect changes drastically. Instead of an inverse relation, I now find a positive relation between firm size and employment growth, except for the smallest size class. Hence, once I control for firm age, I find no evidence for higher growth at smaller firms. In fact, there seems to be a rather homogenous effect of firm size on net employment growth. Both findings, with and without considering age controls, are roughly in line with the findings of Haltiwanger et al. (2017) for the U.S, only they find a stronger positive relation when estimating equation 2 with age controls. These findings highlight the importance of age controls when investigating the impact of size on employment growth.
Now considering Dutch startups, there is no clear relation between firm size and net employment growth. There is no significant difference in growth between firm size classes. The results from estimating equation 2 only suggest a slight significant difference between firms with 2 to 4 employees and the largest firm class when estimating with age controls. Hence, for the average Dutch firm, size seems to matter, as the growth rate increases with size once we control for age. Within the group of startups, there does not seems to be a strong correlation between size and growth. These two facts contradict the fifth hypothesis in this paper, in which I declare my expectation that firm size is inversely related to employment growth both considering all firms and startups. Thus, policy aimed at smaller firms motivated by a higher employment growth among small firms seem to ignore the crucial role of firm age. It should include a thorough analysis on the role of age and the way this interaction has an impact on business dynamism in general to effectively target higher job creation.
As noted above, small firms in the economy tend to have a lower net employment growth rate when controlling for firm age. This is partly due to the exit of firms, as this brings down the weighted net employment growth rate drastically, and naturally small firms tend to have higher exit rates. I explore this further in figure 5 below, in which I plot the share of job destruction by exit per size class.
This figure is interpretable as an employment-weighted firm exit rate. What is immediately evident is that firm exit rates fall with size, both in the case of all firms and startups. This is in line with the expectations of hypothesis 6, namely that small firms have higher exit rates. The share of job destruction by exit considering the average firm is approximately 3 percent for firms in the smallest size class, whereas it is slightly lower than 2 percent for firms in the largest size class. Also in case of high-tech startups this share of job destruction by exit is declining by size, where the largest 3 size classes even have zero job destruction rate of zero. These last two findings also partly explain the trend in figure 4 why small firms experience lower net employment growth rates when controlling for age and highlights the importance of job destruction by firm exit in net employment growth figures.
