Scand. J. of Economics 110(1), 1–21, 2008 DOI: 10.1111/j.1467-9442.2008.00522.x
When Do Employees Leave Their Job for Entrepreneurship?∗ Ari Hyytinen University of Jyv¨askyl¨a, FIN-40014 Jyv¨askyl¨a, Finland
[email protected]
Mika Maliranta Research Institute of the Finnish Economy (ETLA), FIN-00120 Helsinki, Finland
[email protected]
Abstract Existing firms are argued to be an important source of new entrepreneurs. Yet, relatively little is known about the characteristics of firms that breed new entrepreneurs. We use a large linked employee–employer dataset to trace and characterize the types of firms which generate new entrepreneurs in Finland. We find that such transitions are rare and that smaller firms spawn new entrepreneurs more frequently than larger firms. We also find that firms’ R&D intensity and, to a lesser extent, their productivity are negatively related to the probability that employees transit into entrepreneurship. These results are robust to controlling for a number of employee and employer attributes. Keywords: Entrepreneurship; occupation choice; mobility JEL classification: G14; G31; G32
I. Introduction Existing firms are argued to be an important source of new entrepreneurs, especially in the U.S. 1 Yet, little is known about the firms that breed (or ∗
We are indebted to two anonymous referees for their insightful comments. We also thank Petri B¨ockerman as well as conference/seminar participants at the 33rd Conference of the European Association for Research in Industrial Economics (EARIE 2006), the Government Institute for Economic Research (VATT) in Helsinki, the ONS Analysis of Enterprise Microdata Conference (CAED 2005) in Cardiff and the XXII Meeting of Finnish Economists in Jyv¨askyl¨a (2005) for useful suggestions and discussions. This paper grew out of a research project on entrepreneurship at ETLA/Etlatieto, funded by the National Technology Agency of Finland, Tekes (project 579/31/03 and 10582/25/04). Maliranta gratefully acknowledges funding within the Research Programme for Advanced Technology Policy (ProAct) by Tekes. The views expressed are those of the authors and do not necessarily reflect the views of the institutions/organizations with which the authors are affiliated. 1 Bhide (1994) reports that 71% of the founders of the firms on the 1989 Inc. 500 list of the fastest growing firms in the U.S. replicated, or modified at least to some extent, a business idea/model that they had come across in their previous employment. See also Agarwal, Echambadi, Franco and Sarkar (2004), Gompers, Lerner and Scharfstein (2005), Klepper and Sleeper (2005) and Hellmann (2007). C The editors of the Scandinavian Journal of Economics 2008. Published by Blackwell Publishing, 9600 Garsington Road, Oxford, OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA.
2 A. Hyytinen and M. Maliranta
“spawn”) new entrepreneurs. From what kind of firms do employees quit to become entrepreneurs? Do certain corporate attributes increase labor mobility and, in particular, the likelihood of an employee leaving her job for entrepreneurship? In this paper we take advantage of a unique and large linked employee– employer dataset from Finland to address these questions. Besides having around 1.4 million person-year observations (0.41 million private sector employees over five years), the strength of this representative dataset is that it allows us (i) to measure entrepreneurial spawning (employee spin-offs) per number of employees in the parent firm across industries, (ii) to trace and characterize the types of firms from which each new entrepreneur comes and (iii) to contrast the transitions into entrepreneurship with other forms of labor market mobility. 2 This means that our measure(s) of entrepreneurial spawning and unit of analysis differ from those used in prior studies, such as Agarwal et al. (2004), Gompers et al. (2005) and Klepper and Sleeper (2005). Unlike Agarwal et al., who study the diskdrive industry, or Klepper and Sleeper, who focus on the laser industry, our analysis of entrepreneurial spawning covers a representative sample of industries from the business sector. Moreover, to the extent that earlier studies investigate entrepreneurial spawning, they use spin-off companies as their unit of observation. This is somewhat problematic, not least because what is observed in these studies are positive outcomes (i.e., employees who leave their job for entrepreneurship), but not negative outcomes (i.e., employees who stay), nor other labor market outcomes (i.e., employees who switch to a new job). Our analysis also differs from Gompers et al., who look at a (non-random) sample of venture-capital-backed start-ups and focus on the number of spin-offs per parent firm, as opposed to per employee. The available literature identifies a couple of prominent firm attributes that are likely to have an effect on the rate at which an established company breeds new entrepreneurs. The first attribute is firm size. Size matters, because especially smaller incumbents can serve as hatcheries where entrepreneurial learning takes place. There are many opportunities for such learning in the sense that employees of smaller firms often work alongside the firms’ manager-founder(s), thereby allowing them to observe how small businesses are run; see e.g. Gompers et al. (2005) and the references therein. A contrasting view of the effect of firm size is that employees are pushed from large, bureaucratic firms into entrepreneurship due to the reluctance of such firms to develop their employees’ entrepreneurial ideas
2
See also Maliranta and Nurmi (2004), who describe in detail the Finnish employee–employer dataset and how it can be used to trace transitions into entrepreneurship/self-employment.
C The editors of the Scandinavian Journal of Economics 2008.
When do employees leave their job for entrepreneurship? 3
further within the firm. 3 Larger firms may also breed entrepreneurs because of peer effects, i.e., their employees may be exposed to an intense flow of entrepreneurial ideas that originate from co-workers. The second firm attribute that is likely to have an effect on a firm’s propensity to breed new entrepreneurs is its innovativeness. The likelihood that an employee learns about new technologies, product innovations and new forms of organizing production is more likely, the greater the R&D intensity and innovativeness of her current employer. Albeit employees may pay for transferable knowledge (human capital) through lower wages, as shown by Pakes and Nitzan (1983) and Møen (2005), the incentive to make commercial use of an innovative employer’s knowledge by quitting and starting up a rival may remain; cf. Arrow (1962). While this view suggests a direct relation between a firm’s innovativeness and its likelihood of spawning new entrepreneurs, there are confounding effects. Kim and Marschke (2005) argue, for example, that as the risk of employees departing a firm (to join or start a competitor), increases, so do the efforts of the firm to prevent unauthorized use of its knowledge stock through, e.g., patenting. Other protective measures are the use of tailored employment contracts for research personnel to govern employee mobility, as in Pakes and Nitzan (1983), and enhancement of a firm’s ability to capitalize its employees’ ideas within the firm, as in Gromb and Scharfstein (2003) and Hellmann (2007). 4 These firm-level protective measures may mitigate the positive relation between a firm’s innovativeness and its likelihood of spawning new entrepreneurs, or even lead to an inverse relation. By investigating occupational choices in the Finnish labor market from 1997–2001, this study explores the effect of firm size and innovativeness on the likelihood that an employee leaves her job for entrepreneurship. We find, in particular, that smaller firms spawn new entrepreneurs more frequently than larger firms. The result is robust to controlling for the minimum firm size in the estimating sample, firm productivity and R&D intensity, industry, and a number of other employer and employee attributes. This finding does not depend on the employees of smaller firms being generically more 3
The organizational capacity of larger—and presumably more bureaucratic and hierarchical— firms to respond to entrepreneurial ideas (and change more generally) may, for example, be limited; cf. Henderson (1993). Such organizations usually process soft information about new business ideas rigidly; cf. Stein (2002) and Berger, Miller, Petersen, Rajan and Stein (2005). They also have internal capital markets that may disproportionately favor established lines of business; cf. Scharfstein and Stein (2000). Gompers et al. (2005) call this the “Xerox view” of entrepreneurial spawning. They found Xerox to be one of the most prominent examples of large bureaucratic firms whose top executives were reluctant to fund its employees’ entrepreneurial ideas in the 1960s and 1970s. 4 A firm’s ability to capitalize its employees’ ideas reflects at least partly its willingness to allow for “intrapreneurship”, i.e., within-firm entrepreneurship. C The editors of the Scandinavian Journal of Economics 2008.
4 A. Hyytinen and M. Maliranta
mobile since, in our data, firm size is not similarly (i.e., inversely) related to the likelihood that a private-sector employee switches to a new job. We also document that the relation between a firm’s innovativeness and its likelihood of spawning new entrepreneurs is inverse. Both the R&D intensity of a firm and, to a lesser extent, its productivity are negatively related to the probability that one of its employees transits into entrepreneurship. The inverse relation is not due to the employees of less innovative firms being intrinsically more mobile, since our proxies for firm innovativeness have either no or a non-linear effect on the likelihood of interfirm switches. In sum, our findings allow us to reject the view that employees are pushed from larger, bureaucratic firms into entrepreneurship and to question the (often cited) conjecture that especially the most innovative firms are at risk of losing good ideas and inventions because their employees are more prone to quit and quickly establish new firms. The rest of the paper is organized as follows. In the next section we outline a framework for our empirical analysis. In Section III we discuss the data. In Section IV we report the results of our empirical analysis. Section V contains a brief summary.
II. Empirical Framework Consider an employee, labeled e, who faces a choice among staying in her current job (Stay), switching to a new job (Switch), becoming an entrepreneur (Selfemp) and transiting into unemployment (Unemp). The utility she obtains from alternative j ∈ {Stay, Switch, Selfemp, Unemp} is Uej , and she chooses alternative i if and only if Uei ≥ Uej for all j = i. We assume that Uej ≡ Vej + ε ej , where Vej is the observed part and where the unobserved part, ε ej , is i.i.d. and of type I extreme value. The observed part is linear, i.e., Vej = xe β j + α j , where xe refers to a vector of employees’ characteristics and their current employer’s attributes. With these assumptions, the multinomial logit (MNL) choice probability due to McFadden (1974) is: exp xe βi + αi (1) Pr(e’s choice = i) ≡ Pei = 4 , j=1 exp x e β j + α j for e = 1, . . . , N. As the attributes of the alternatives are not observed, the parameters of this model are unidentified unless the parameter vector of one of the alternatives is normalized. For estimation we set β Stay = αStay = 0, in which case the remaining coefficients measure the change relative to the employees who stay in their current job. Under this normalization, αSelfemp corresponds, for example, to the average effect of omitted factors on C The editors of the Scandinavian Journal of Economics 2008.
When do employees leave their job for entrepreneurship? 5
the utility of becoming self-employed relative to staying in one’s current job. We estimated the model by the method of maximum likelihood and report the marginal effects (evaluated at the means). The marginal effects measure the impacts of infinitesimal changes in the continuous variables and discrete changes in the dummy variables. To allow for within-firm correlation in employees’ propensity to leave their firm, we use standard errors that are clustered at the level of plants in which the employees work. A well-known weakness of the MNL model is that it assumes that the ratio of Pei to Pek does not depend on alternatives other than i and k (i.e., that the assumption of independence from irrelevant alternatives, IIA, holds). This assumption is less of a problem when the alternatives are reasonably distinct and dissimilar; see e.g. Amemiya (1981). Our view is that this is the case here. It also turns out that the data are consistent with this view in that Hausman IIA tests do not reject the assumption (see robustness tests for details).
III. Data Description of Data Source and Transitions The data used in this paper represent a random sample from the Finnish Longitudinal Employer–Employee Data (FLEED) of Statistics Finland. The FLEED data are constructed by linking various administrative registers, such as Employment Statistics, Business Register, Financial Statements Statistics and the R&D survey of Statistics Finland. The basic unit in this dataset is an individual who belongs to the working population of Finland and who—if organizationally employed—can be linked to the company and plant in which she works. The original FLEED data have three characteristics that are particularly important for this study: the data follow over time basically the entire working population of Finland, include a wealth of information about the individuals and their occupations, and make it possible to trace individuals’ labor market transitions. The sample available for this study thus allows us to measure entrepreneurial spawning (employee spin-offs) per number of employees in the parent firm, to study spawning across a representative set of industries, to characterize the types of firms that spawn entrepreneurs and to study all forms of labor market mobility. The sample covers the years 1997–2001 and consists of roughly every third individual in the original FLEED data. Our analysis focuses on the labor market behavior of business sector employees, which gives us about 1.356 million person-year observations that consist of 409,277 individuals C The editors of the Scandinavian Journal of Economics 2008.
6 A. Hyytinen and M. Maliranta
during a five-year period. 5 Identifying the (organizationally) employed and unemployed from this sample is relatively straightforward, but identifying entrepreneurs is less so. Individuals are identified as self-employed if they are insured on the basis of the Self-employed Persons’ Pension Act (YEL). Finnish law mandates such insurance for those who are partners in a general partnership, assume the role of a general (i.e., responsible) partner in a limited partnership or own 50% or more of the stocks in a limited liability company. The law only applies to the self-employed who reside in Finland and are aged 18–64, requires that self-employment has continued for at least four months, and stipulates that for the insurance requirement to apply, earnings should not fall short of a certain low threshold (5,504.14 euros in 2004). This register-based measure does not differentiate between entrepreneurship and self-employment, which implies that for a large part of our analysis, we use the terms “entrepreneurs” and “self-employed” interchangeably. However, in connection with the robustness tests, we show that the main results of this paper hold in sub-samples in which it is unlikely that the transitions reflect “bread-and-butter” self-employment. Table 1 gives an overview of the data and particularly of the transitions we observe: from salary work into a new job (Switch), self-employment (Selfemp) and unemployment (Unemp). The table also reports the share of immobile employees who do not leave their current job (Stay). As this transition matrix shows, around 20% of those who are employed in year t transit into a new plant during year t + 1. About 15% of employees switch Table 1. Transition matrix, 1997–2001 Status next year Year 1997 1998 1999 2000 2001 Total
Stay
Switch
Selfemp
Unemp
Total
208,113 81.62 216,778 80.52 215,509 80.56 222,274 79.85 229,580 80.44
34,839 13.66 38,581 14.33 40,518 15.15 41,671 14.97 40,708 14.26
1,646 0.65 1,870 0.69 1,883 0.70 1,909 0.69 2,031 0.71
10,394 4.08 11,987 4.45 9,599 3.59 12,522 4.50 13,100 4.59
254,992 100.00 269,216 100.00 267,509 100.00 278,376 100.00 285,419 100.00
1,092,254 80.58
196,317 14.48
9,339 0.69
57,602 4.25
1,355,512 100.00
5
Employees who work in plants or firms that are closed down are not included in the basic estimating sample. For them, staying in one’s current job is obviously not possible.
C The editors of the Scandinavian Journal of Economics 2008.
When do employees leave their job for entrepreneurship? 7
annually to a new job and around 4%–5% transit into unemployment. Only 0.7% of the employed transit into entrepreneurship, which makes these transitions a relatively rare labor market event. The transition matrix also shows that the shares of individuals who make a move on the labor market have remained constant over time. The stability of these aggregate shares fits well with the finding that the determinants of individual-level transitions have also been stable. 6 We therefore used the pooled sample from 1997–2001 as our base sample in what follows.
Definitions of Conditioning Variables Firm Size and Innovativeness. We are interested in how the size and innovativeness of firms for whom an employee worked at the end of year t affect the likelihood of her transiting in year t + 1. Our measure of firm size is the number of employees the firm (plant) had at the end of year t. This variable consists of seven employment categories (0–4, 5–9, 10–19, 20–49, 50–99, 100–299 and >300) to which each firm in our data has been assigned. 7 Measuring firm innovativeness is rarely straightforward, and our study is no exception in this regard. Our primary measure of innovativeness is R&D-dummy that equals one if the ratio of R&D expenditures to turnover exceeds 3.5%, and zero otherwise. 8 It is our primary measure, because the current R&D efforts of a firm are likely to mirror investments in
6
That is, we have not found any evidence of (structural) breaks in the determinants across the sample years. 7 It is worth mentioning that our results are robust to using a quasi-continuous size measure that can be obtained by reverse-engineering the categorical size variable. 8 Unfortunately, we do not have a continuous R&D variable. The reason is that, due to certain data confidentiality concerns of Statistics Finland, the data include only a categorical variable for R&D intensity (R&D per turnover). We use an R&D dummy even though the original variable has more than two categories for some firms. However, the frame of the underlying R&D survey, which generates the R&D data, is such that for many of the smaller firms in our estimating sample, it can only be determined whether the ratio of R&D to sales exceeds or falls short of 3.5%. This threshold is used for three reasons. First, due to the design of the survey, we believe that it minimizes measurement error. For this we interpret the design of the R&D survey to be such that the firms that are not covered by the survey are low R&D firms (i.e., their R&D per turnover
