The explanatory variables - Krannert School of Management

Entrepreneurs’ Human Capital and the Start-up Size of New Technology-Based Firms* Massimo G. Colomboa Marco Delmastrob Luca Grillia a

Politecnico di Milano, Department of Economics, Management and Industrial Engineering

b

Autorità Garante della Concorrenza e del Mercato

Jel Classification: L11; M13 Keywords: New technology-based firms; Firm start-up size; Human capital.

Abstract This paper investigates the determinants of the start-up size of new technology-based firms. While previous empirical studies generally focussed on industry-specific variables, we draw attention to the characteristics of founders, notably their human capital. In the empirical section we consider a sample of 391 young Italian firms operating in high-tech industries in both manufacturing and services. The econometric estimates confirm the explanatory power of the industry-specific effects highlighted by previous work. In addition, they indicate that the human capital of founders figures prominently in explaining firms’ start-up size. Furthermore, the specific component of human capital associated with industry-specific professional knowledge and managerial and entrepreneurial experiences is found to have a greater positive impact on initial firm size than the generic component, proxied by education and general (i.e. non industry-specific) working experience.

(*) We gratefully acknowledge the support of MIUR 2000 and 2002 funds and a grant from CNR (C00E3AF). We are indebted to Thomas Åstebro, Mario Calderini, Xavier Castaner, Bernard Garrette, Steve Klepper, Josè Mata, Luigi Orsenigo, Bertrand Quelin, Kenneth Simons, Peter Thompson, Enrico Santarelli, Marco Vivarelli, participants in the 29th EARIE Conference, the 12th AiIG conference, and seminars held at Groupe HEC, Università di Bologna, Università di Pavia and Università di Torino, and an anonymous referee for helpful comments. While the paper is the result of the joint work of the authors, Massimo G. Colombo has written sections 1 and 2, Marco Delmastro sections 3 and 4.2, and Luca Grilli sections 4.1, 5 and 6. Correspondence: Massimo G. Colombo, Politecnico di Milano, Department of Economics, Management and Industrial Engineering, P.za Leonardo da Vinci, 32, 20133 Milan (ITALY), [email protected]

1. Introduction Since the early ‘80s, a rich stream of empirical literature has analysed the determinants of new firm creation and the post-entry performances of new firms (for a survey see Geroski 1995, Sutton 1997, Caves 1998). Such studies have established several interesting “stylised facts”. First, although new firms are very numerous, they generally are much smaller than incumbents (Cable and Schwalbach 1991). Second, in the years that immediately follow foundation mortality rates are very high among newly born firms; however, they decline with start-up size. In other words, the higher the initial size of a new firm, the higher the probability of survival, all else equal (Evans 1987a and 1987b, Dunne et al. 1988 and 1989, Philiphs and Kirchhof 1989, Audretsch 1991, Mata and Portugal 1994, Audretsch and Mahmood 1994 and 1995, Mata et al. 1995, Audretsch 1995b, Cabral and Mata 2003). Third, Gibrat’s law claiming that firms’ growth rates are independent of firm size, has been found not to hold for young firms. Studies relating to different countries and industries have shown that smaller new firms exhibit significantly higher growth rates than their relatively larger counterparts (see Evans 1987a and 1987b, Dunne et al. 1988 and 1989, Hart and Oulton 1996). This result is generally interpreted as a consequence of the need to eliminate as rapidly as possible the cost disadvantage accruing from operating at sub-optimal scale. The fact that the survival prospects of new firms are generally found to be lower and the growth rates of new surviving firms to be greater in industries where there are substantial economies of scale lends support to such view (see for instance Audretsch and Mahmood 1994, Audretsch 1995b. For a different view see Mata and Portugal 1994).1 If a larger start-up size positively affects the likelihood of survival of new firms, and if surviving new firms that started operations at smaller scale struggle to catch up, the question arises why there are firms with small initial size. Unfortunately, the analysis of the determinants of the size of new firms has so far remained rather undeveloped. A few empirical studies have tried to relate the initial scale of firms to specific characteristics of the industry in which they are going to operate (Mata 1996, Mata and Machado 1996, Görg et al. 2000). Such studies show that start-up size increases with the minimum efficient scale (MES) of the industry, the cost disadvantage of operating at sub-optimal scale, and industry growth, while it diminishes with the entity of sunk costs, inversely measured by the easiness of entry into and exit from the industry. The impact of market size is more controversial, being positive but weakly significant in Mata (1996) and Mata and Machado (1996) and prevalently 1

As regards Italy, Audretsch et al. (1999) estimate models for survival and growth of firms born in January 1987 and tracked up to January 1993. They find virtually no evidence of a positive relation between firm’s initial size and survival (similarly insignificant results are obtained by Wagner 1994 for Germany). However, they do find that in most industries smaller new firms grow at higher rates than larger ones. 1

negative in Görg et al. (2000). Note that both Mata and Machado (1996) and Görg et al. (2000) acknowledge that there is size heterogeneity among new firms in a given industry; however, the sources of heterogeneity generally remain unobserved due to lack of proper data at firm level. A different stream of the economic literature that has analysed the entrepreneurial choices of individuals (see Evans and Jovanovic 1989, Evans and Leighton 1989, Holtz-Eakin et al. 1994a, Lindh and Ohlsson 1996) has shown that both personal characteristics such as age, education and working experience, and financial conditions play a key role in shaping the decision to become an entrepreneur. In spite of the fact that such factors are very likely to influence also the size of a new firm, the evidence so far provided on this issue is rather scarce. Mata (1996) considers some covariates reflecting the human capital of new firms’ founders, namely their age as a proxy for working experience, and education. His estimates of a sample selection model highlight a positive and statistically significant effect of the two above mentioned variables on the size of new Portuguese firms, measured by the log of employment: older and more educated people set up larger businesses. Holtz-Eakin et al. (1994b) consider a group of people in the US who received inheritances and show that for individuals who started a new company, the amount of capital invested in the new firm increases with an increase of the size of the inheritance; such a finding suggests that liquidity constraints influence start-up size. Ǻstebro and Bernhardt’s (1999) study is the most similar to the present work. They examine the determinants of start-up capital for 986 US firms created in 1987 by 1,194 individuals. The amount of capital initially invested in a firm turns out to increase with an increase of the human capital of the founding team, proxied by years of working experience and managerial and entrepreneurial competencies. In addition, individuals with greater predicted household income are found to start larger enterprises. In turn, with all else equal, household income is positively related to the human capital of individuals. Such results indicate that entrepreneurs’ human capital has both direct and indirect positive effects on firms’ start-up capital, with the indirect effect arising from relaxation of financial constraints. They also suggest that entrepreneurs indeed are financially constrained.2 In this paper we adhere to and extend this approach. Following Mata (1996) and Ǻstebro and Bernhardt (1999), rather than focussing solely on industry characteristics we examine the effects of the human capital of founders on the initial size of new technology-based firms (NTBFs). Initial size is measured by the log of the number of employees after twelve months from the date on 2

In the household income equation, the dummies capturing educational attainments and years of working experience exhibit the greatest explanatory power. In contrast, in the start-up capital equation, it is variables reflecting individuals’ prior managerial and entrepreneurial experiences that are most significant, together with founders’ predicted household

2

which the firm was incorporated.3 We take advantage of a quite detailed description of the human capital of founding teams; in particular, we are able to separate founders according to whether their previous work experience was related to the business the new firm is in or not. This is an important distinction. In fact, the personal wealth an entrepreneur may have access to is likely to increase with the years of work experience, but to be independent of its industry-specific nature. On the contrary, the productivity of human capital in the entrepreneurial job is likely to be greater for founders with related rather than unrelated working experience. Therefore, consideration of the nature of founders’ work experience helps shed new light into the reasons why new firms start operations at small scale. The empirical analysis is based on a sample composed of 391 Italian firms that were established in the ‘80s and ‘90s and operate in high-tech manufacturing and service industries. The NTBF sector offers an ideal testbed of theoretical hypotheses on the determinants of start-up size. First, newcomers allegedly play a fundamental role for static and dynamic efficiency (see Audretsch 1995a). Second, founders’ competencies are regarded as a key source of competitive advantage for new firms (see Cooper and Bruno 1977). Third, capital market imperfections are likely to be magnified for NTBFs (see Carpenter and Petersen 2002). In addition, while focussing on NTBFs, we are better able to control for the influence exerted on start-up size by environmental factors. The paper proceeds as follows. In next section we build on the literature on entrepreneurship to develop an empirical model of firms’ start-up size that takes into due account the influence of the human capital of founders. In Section 3 we present the data set. Section 4 is devoted to the specification of the econometric model and the description of the dependent and explanatory variables. In Section 5 we illustrate the results of the estimates. Summarising remarks in Section 6 conclude the paper.

2. The empirical model The aim of this section is to specify an empirical model highlighting the influence exerted on the initial size of NTBFs by the human capital of founders. For this purpose, we follow Becker (1975) in distinguishing between generic and specific human capital. Generic human capital relates to the general knowledge acquired by entrepreneurs through both formal education and professional

income. This means that the individuals who are most likely to switch into self-employment because of their entrepreneurial talent are also those who are most likely to be financially constrained. 3

Usually, owners of NTBFs (or at least a subset of them) have operating (i.e. managerial) roles in the firms they established. For this reason, as will be explained later in greater detail (see Section 4.2.1), we consider two measures of the number of employees of a new firm; the first one coincides with the number of salaried personnel, the second one also includes founders (i.e. individuals who provided equity capital to establish a new firm and had operating positions in the new firm). 3

experience.4 Specific human capital consists of the capabilities of individuals that can directly be applied to the entrepreneurial job in the newly created firm; it is very much related to the industryspecific skills that founders learned in the organization by which they were formerly employed and to the “leadership experience” gained either through a managerial position in another firm or in prior self-employment episodes (Cooper 1985, Preisendörfer and Voss 1990, Brüderl et al. 1992, Brüderl and Preisendörfer 2000). In the following, we will build on previous work (Evans and Jovanovic 1989, Cressy 1996, Xu 1998, Åstebro and Bernhardt 1999) that modelled an individual's choice of switching between the state of being a salaried worker and that of being an entrepreneur. We assume that the personal wealth Wi an individual has access to is an increasing concave function of her (total) human capital

ψi: Wi=ψiγ. Let Ii and Πi indicate income of the entrepreneur and earnings of the newly started firm, respectively: Ii=Πi+c(Wi-k). k denotes the amount of capital the entrepreneur invested in the new venture and c is the opportunity cost of capital. Firm’s earnings are assumed to depend on both the initial investment k and the entrepreneurial ability of the founder. In turn, it seems plausible to assume that this latter is positively related to the specific component ϕi of human capital.5 Hence:

Πi=kα ϕiδ. Let us indicate with k* the amount of initial capital that maximizes founder's own earnings. One obtains: If the founder is not financially constrained (that is, either k*≤Wi or there is a frictionless capital market), she will indeed start operations at k*. Let us further assume that capital and labor are complements (the discussion of this issue is postponed to Section 4.2.1). According to the above expression the start-up size of a new firm measured by the employment level, will increase with the specific human capital of the founder. Nevertheless, obtaining access to external capital may be difficult due to the existence of market imperfections, a situation which frequently occurs to NTBFs (see Carpenter and Petersen 2002 for an in-depth analysis of this issue). Under such circumstances, the "financing hierarchy" hypothesis (see Fazzari et al. 1988) suggests that an entrepreneur will look for external financial 4

In previous empirical works (see for instance Bates 1990, Stuart and Abetti 1990, Brüderl et al. 1992, Storey 1994, Westhead and Cowling 1995, Almus 2000, Brüderl and Preisendörfer 2000) knowledge acquired by entrepreneurs through education is captured by education attainments such as graduation and achievement of a Ph.D. degree, or years of schooling; professional knowledge is generally mirrored by the years of working experience before establishing the new firm or simply by entrepreneurs’ age. 5

The available empirical evidence generally lends support to such contention. In fact previous studies show that both the survival likelihood of new firms and the growth rate of surviving new firms are positively related to the specific component of the human capital of founders; in contrast, results relating to the generic component are far less robust (see for instance Cooper and Bruno 1977, Cooper 1985, Dunkelberg et al. 1987, Bates 1990, Brüderl et al. 1992, Chandler and Jansen 1992, Siegel et al. 1993, Westhead and Cowling 1995, Brüderl and Preisendörfer 2000. For a survey see Storey 1994). 4

resources only if her own financial endowment Wi is not enough to finance the new venture at the desired scale; this occurs if k*>ψiγ. Given entrepreneur’s specific human capital, the higher his total human capital, the lower the likelihood that financial constraints be binding. Total human capital includes both the specific and the generic components; therefore one expects also the generic component of human capital to have a positive effect on start-up size - even though a smaller one than the specific component, in so far as it alleviates possibly binding financial constraints. In previous work (see again Evans and Jovanovic 1989, Cressy 1996, Xu 1998, Åstebro and Bernhardt 1999), in determining optimal start-up size the assumption was made that individuals know with certainty their entrepreneurial talent before starting a new firm;6 that is, entrepreneurial talent is a deterministic variable. This may not be the case. Actually, it is more realistic to assume that new firms’ founders have beliefs concerning their entrepreneurial ability that are surrounded by uncertainty (see Jovanovic 1982). While there is evidence that individuals going into selfemployment often are overoptimistic (see Camerer and Lovallo 1999, Arabsheibani et al. 2000), they will rationally be more optimistic about the prospects of the new venture the better their specific human capital, as this latter characteristic is a reasonably accurate predictor of entrepreneurial success (see again footnote 5). In particular, the higher the specific human capital of founders, the lower the likelihood they will assign to failure of the new firm. Furthermore, the creation of a new firm always involves the commitment of investments a portion of which is unrecoverable in case of failure; in other words, there are sunk costs in building new capacity from scrap. Real option theory (see Pindyck 1991, Dixit and Pindyck 1994) contends that there is an opportunity cost of making an irreversible investment expenditure due to the lost option value of waiting for new information to arrive; such cost increases with the uncertainty of the future returns the investment will generate. A firm’s capacity choice is optimal when the present value of the expected cash flow generated by a marginal unit of capacity equals the full cost of that unit, including the opportunity cost of exercising the option to buy the unit. Therefore, when there is considerable uncertainty, firms will limit the amount of unrecoverable investments so as to avoid the risk of incurring losses if unpredicted contingencies occur (Pindyck, 1988 and 1993). Accordingly, new firms will optimally start operations at a small scale and expand if circumstances prove to be favourable (Cabral 1995). The appeal of a smaller initial scale increases with an increase in the probability entrepreneurs assign to failure. It follows that with everything else equal,

6

Even if individuals are not aware of their entrepreneurial talent, financiers might tell the difference between them. In particular, if financiers are able to distinguish between good and bad entrepreneurial projects according to the specific human capital of entrepreneurs, individuals with greater specific human capital will again establish larger firms. Nonetheless, under such circumstances the generic component of human capital will have no effect on initial size, as there would be no imperfections in capital markets. 5

individuals with greater specific human capital and greater confidence in their own entrepreneurial talent will start a new firm at a relatively larger scale of operations.7 To sum up, the arguments illustrated in the present section suggest that firm start-up size will be: •

positively related to the specific human capital of founders;

•

positively related to the generic human capital of founders, with the impact of generic human capital being smaller than that of specific human capital.

3. The data In this paper we consider a sample composed of 391 Italian NTBFs. Sample firms were established in 1980 or later, were independent at start-up time (i.e. they were not controlled by another business organization even though other organizations may have held minority shareholdings in the new firms) and operated in the following high-tech sectors, in manufacturing and services: computers, electronic components, telecommunication equipment, optical, medical and electronic instruments, biotechnology & pharmaceuticals, aerospace, multimedia content, software, Internet services (e-commerce, ISP, web-related services), and telecommunication services. The sample of NTBFs was extracted from the RITA (Research on Entrepreneurship in Advanced Technologies) database, developed at Politecnico di Milano. The RITA database was created in 1999, and was updated and extended in 2001; it contains detailed information on more than 400 Italian NTBFs and more than 1,000 of their founders. The development of the database went through a series of steps. Firstly, Italian firms that complied with the above mentioned criteria relating to age and sector of operations were identified. For the construction of the target “universe” a number of sources were used. These included lists provided by national industry associations, online and off-line commercial firm directories, and lists of participants in industry trades and expositions. Information provided by the national financial press, specialized magazines, other sectoral studies, and regional Chambers of Commerce was also considered. Altogether, around 2,000 firms were selected for potential inclusion in the database. For each firm, a contact person (i.e. one of the owner-managers) was also identified. Unfortunately, data provided by official

7

Of course, uncertainty may arise for different reasons than founders’ untested entrepreneurial ability. If there are sunk capacity costs, greater uncertainty will lead to optimal choice of a smaller start-up size, independently of its sources. Previous studies have shown that start-up size increases with a reduction of sunk costs; a reduction of uncertainty due for instance to market growth has a similarly positive effect (see again Mata 1996, Mata and Machado 1996, Görg et al. 2000). 6

national statistics do not allow to obtain a reliable description of the universe of Italian NTBFs.8 Note also that for obvious reasons, the selection procedure led to the oversampling of growthoriented firms. Second, a questionnaire was sent to the target firms either by fax or by e-mail. The first section of the questionnaire contains detailed information on the human capital characteristics of entrepreneurs such as age, education, and prior working experience. The second section comprises further questions concerning the characteristics of the firms at start-up time, including initial size (for a precise definition of start-up size see Section 4.2.1.). Lastly, answers to the questionnaire were checked by educated personnel; when it was deemed necessary, phone or faceto-face follow-up interviews were made with firms' owner-managers. This final step was crucial in order to obtain missing data and ensure that answers were reliable. The sample used in the present work consists of 391 NTBFs for which we were able to create a complete data set (see Section 4.2). The only exception concerns data on previous entrepreneurial experiences of firms’ founders that were available only for a sub-sample of 260 firms. The sample consists of 19 firms in the biotechnology and pharmaceutical industry (4.8% of the sample), 23 firms in the multimedia content sector (5.9%), 112 software houses (28.6%), 156 Internet and telecommunication service firms (39.9%), while the remaining 81 firms (20.7%) operate in the following manufacturing sectors: telecommunication equipment, electronic components, computers, optical, medical and electronic instruments, and aerospace. The sample is quite large and as will be shown in Section 4.2, it exhibits considerable heterogeneity as to the relevant variables. Therefore, it provides a reasonably adequate testbed of the theoretical hypotheses we are concerned with in this work. Of course, there is no presumption here to have a random sample. From one side, as was mentioned above, absent reliable official statistics, it is very difficult to identify unambiguously the universe of Italian NTBFs. From the other side, the sample was drawn in 1999; so only firms having survived up to the survey date were included. In principle, attrition may generate a sample selection bias that is difficult to control; in fact, the empirical literature generally documents that failure rates of new firms decrease with both start-up size and the human capital of founders (see footnote 5 and the literature mentioned in the introductory section). Nevertheless, Audretsch et al. (1999) have highlighted that the initial size of Italian firms does not significantly affect their subsequent survival rates, while Del Monte and Scalera (2001) have detected a negative relationship between initial capital and life duration of small Italian firms created within a public start-up programme. In accordance with such results, in our sample there is almost no correlation between the start-up size of firms and their age at survey 8

The main problem is that in Italy, most individuals who are defined as “self-employed” by official statistics actually are salaried workers with atypical employment contracts. Unfortunately, on the basis of official data such individuals cannot be distinguished from entrepreneurs who created a new firm. 7

date: the value of the Pearson correlation index is 0.0186 and it is not statistically different from zero at conventional confidence levels.

4. The econometric models 4.1. The specification of the models We investigate the determinants of start-up size through econometric estimates of a series of models relating firms’ initial scale measured in logarithm to variables reflecting the human capital of founders and a set of control variables ( xi ).We can express the basic model as:

y i = xi' β + ε i ;

(1)

β is the vector of parameters to be estimated; the disturbances ε i are assumed to be N(0, σ ε 2 ). Note that OLS estimates are likely to be biased because of the truncated nature of the sample; in fact, we only observe those firms that were actually founded (Mata 1996). We do not have any information on individuals who possibly wanted to become entrepreneurs but were not able to do so and chose to remain workers or eventually became unemployed. Following Maddala (1986), we can take this factor into account using a latent variable framework. The model can then be defined as follows:

y i* = xi' β + ε i ,

(2)

where y i* is the log of the potential level of start-up size. The unobservable initial size of firms that were not founded is naturally smaller than one. Therefore, the observed data y i are such that:

y i = y i* if y i* ≥ 0 , while y i is not observed otherwise. The likelihood function for the model is given by: L = ∏ Φ ( xi' β / σ ε ) −1 σ ε−1φ [( y i − xi' β ) / σ ε ] ,

(3)

yi ≥ 0

and the expected value for the log of observed start-up size is: E[ y i | y i* ≥ 0] = xi' β + σ ε φ [ xi' β / σ ε ] / Φ[ xi' β / σ ε ] .

(4)

The truncated regression model assumes that the same set of parameters ( β ) and variables ( xi ) that determine the potential level of employment of firms at start-up ( y i* ), also determine whether firms are created or not: in fact, the firm is in the market if y i* ≥ 0 , it is never born otherwise. A more general approach is the one adopted by Mata (1996), which following Bloom and Killingsworth (1985), considers a sample selection model with incidental truncation. In this framework, we have: y i = xi' β y + ε Yi ,

(5a) 8

s i = z i' β s + ε S ,

(5b)

i

where (5a) and (5b) are respectively the regression equation and the threshold or selection equation. The latent variable si represents the difference between the income entrepreneurs expect to get from the new venture and the wage they command in the labour market. Only people for whom si >0 become entrepreneurs. So, the start-up size y i is only observed if si >0. Then ε Yi and ε Si are

bivariate normal mean-zero random variables, uncorrelated with the x and z, with variances given by σ YY and σ SS respectively, and with covariance σ SY . Notwithstanding the fact that si is not observed and therefore equation (5b) cannot be estimated by itself, Bloom and Killingsworth (1985) show that it is possible to obtain unbiased estimates of the parameters of both equations (as long as identification requisites are met) using the information that the observation of y i is feasible only as long as s i > 0 . The likelihood for a sample of observations y i conditional on a set of regressors xi and given that they all are in the truncated sample is:

φ( L = ∏[ yi ≥ 0

εY

i

1 2 YY 1 2 YY

σ σ

  '  [− z i β s + (σ SY ε Yi / σ SY ]  ) 1 − Φ  1 [σ (1 − (σ 2 / σ σ ))] 2  SY YY SS  SS ] ' −z β [1 − Φ ( i1 s ]

     .

(6)

σ SS2

The expected value of y i will now be given by: 1

1

1

E[ y i | si > 0] = xi' β y + [σ SY / σ SS2 ][φ ( z i' β s / σ SS2 ) / Φ ( z i' β s / σ SS2 )] .

(7)

In order to make the model identifiable, σ SS is set equal to unity without loss of generality, and σ YY and σ SY must be different from zero. Following Mata (1996), we include as regressors in the selection equation (5b) those covariates that refer to the human capital of founders.9

9

Muthen and Jöreskog (1983) and Maddala (1986) have questioned the ability of the model to catch the “selection” decision since estimates of the parameters of the selectivity portion of the model are derived only through functional form and therefore are often unreliable. For this reason, they are not reported in the paper; they are available from the authors upon request. Note also that if we admit that regressors affect not only entry but also the probability of survival in the following years, both the truncated and the sample selection model with incidental truncation can also be viewed as attempts to account for the non-randomly sampling of our data. 9

4.2. The variables of the econometric models

4.2.1. Start-up size

In this work, start-up size is defined as the number of firms’ employees measured twelve months after the date on which the new firm was incorporated. The number of firms’ employees is operationalised alternatively as the number of salaried personnel or the sum of the number of founders (i.e. individuals who were shareholders of and took managerial positions in the newly

created firm) and the number of salaried personnel. In most previous works that analysed the determinants of start-up size, size is measured by the firm’s employment (see Mata 1996, Mata and Machado 1996, Görg et al. 2000. See also Audretsch et al. 1999). Åstebro and Bernhardt (1999) use start-up capital. In our sample, we had no precise information relating to this latter variable. However, as the new firms we consider are in high-tech industries, the amount of total capital is likely to be closely correlated with the employment level. In fact, in the early years of firms’ life costs mainly relate to R&D and new product and service development, and there rarely are sizable investments in physical production assets. So labour and capital are likely to be complements rather than substitutes. Note also that NTBFs are rarely profitable in the early period of their life. Hence, the ability to pay salaries is constrained by the financial resources firms may have access to. As a corollary, binding financial constraints may induce founders to hire a lower number of salaried employees than the optimal one. Furthermore, the inclusion of the number of founders in one of the measures of start-up size was based on evidence that founders often account for a sizable portion of a new firm’s workforce. In our sample, the mean number of founders was 2.80, while the mean number of salaried personnel at start-up time was 4.42. Lastly, in previous work there generally is no clear indication as to the exact date on which initial size is measured. The reason may be that it is difficult to define unambiguously when start-up actually occurs. The criterion adopted here has two advantages. First, when a new firm is created time is needed to hire personnel and organize operations. Twelve months seem to be a reasonable period to allow a new firm to reach the size founders had in mind when they established the firm. Second, we are confident that respondents provided information on firm size at exactly the same point in time in the life of the new firms. 4.2.2. The explanatory variables

The explanatory variables are illustrated in Table 1. They can be subdivided in two groups. The first group encompasses variables aimed at analysing the role played by founders’ generic and specific human capital, as captured by education and professional experience. On the 10

basis of both the predictions of our empirical model and previous findings (see Mata 1996, Ǻstebro and Bernhardt 1999), such variables should have a positive impact on start-up size. From one side, individuals with greater specific human capital should perform better as entrepreneurs and be more confident of their entrepreneurial ability; hence the desired initial size should be larger. This effect is captured by three variables. Specworkexp measures the years of professional experience of founders in the same sector of activity of the new firm. This is a key component of specific human capital. In fact, entrepreneurial competence is likely to be higher and uncertainty about firm’s post-entry performance lower if founders have deep direct knowledge of the market, technological, and competitive environment in which the new firm operates (Agarwal and Audretsch 2001). Furthermore, we have proxied entrepreneurial ability with two additional variables. DManager and DEntrepreneur equal 1 if prior to the establishment of the new firm, one or more founders had a managerial position in a medium or large company (i.e. number of employees greater than 10010) and self-employment experience, respectively. From the other side, the personal wealth of entrepreneurs inclusive of funds provided by family members and friends, reportedly increases with human capital, independently of its specific or generic nature (Ǻstebro and Bernhardt 1999). If there are imperfections in capital markets and individuals are financially constrained, greater personal wealth should help them relax such constraints and achieve the desired initial firm size. Previous studies concerned with entrepreneurship have provided evidence consistent with the view that new firms suffer from financial constraints. For instance, both cross-sectoral (Meyer 1990, Blanchflower and Oswald 1998) and longitudinal (Evans and Jovanovic 1989, Evans and Leighton 1989, Black et al. 1996) studies have shown that the likelihood of being self-employed increases with individuals' net worth. Lindh and Ohlsson (1996), using Swedish microdata, have shown that the probability of being selfemployed increases when individuals receive windfall gains in the form of lottery winnings and inheritances. Similarly, Holtz-Eakin et al. (1994a) have analysed reception of an inheritance. Their results indicate that the likelihood of establishing a new enterprise and the initial capital committed to the enterprise by the founder significantly increase with the size of the inheritance and that such effect is more pronounced for low net-worth individuals. In addition, if one focuses attention on entrepreneurs that received an inheritance, the greater the inherited amount the greater the likelihood of survival and the growth rate of the new venture (Holtz-Eakin et al. 1994b). Åstebro and Bernhardt (1999) have shown that the predicted household income of US entrepreneurs positively affects the amount of capital committed to a new venture. Lastly, the analysis of the 10

In small family-owned Italian companies decision authority is often centralised in the owner-manager’s hands (see Colombo and Delmastro 1999), while salaried managers are assigned execution tasks. So entrepreneurial learning associated with such managerial positions generally is fairly limited. 11

evolution over time of the size distribution of Portuguese firms performed by Cabral and Mata (2003) indicates the presence of binding financial constraints that prevent new firms from attaining their optimal initial size. Nonetheless, the view that new firms face tight financial constraints is not unanimously shared in the literature. In particular, it has been argued that the fact that individuals’ assets are positively correlated with firm creation and post-entry performance may be the effect of a spurious correlation: if assets and human capital are correlated, failure to include into econometric models a proper specification for founders' human capital may lead to the erroneous detection of a capital market imperfection (see Cressy 1996).11 If there are binding financial constraints, we expect both the specific and generic component of founders’ human capital to have a positive effect on start-up size. As was indicated above, the specific component also captures founders’ greater entrepreneurial ability and self-confidence, while the generic component does not, being simply a proxy for individuals’ wealth; so we predict a greater positive effect on firms’ initial size of variables that reflect the former component than those associated with the latter one. Generic human capital variables include the level of education measured by the mean number of years of education of founders (Education) and the years of professional experience in other sectors than the one of the new firm (Genworkexp). As concerns graduate and post-graduate education, we also distinguish between economic and law studies (Ecoeducation) and technical and scientific studies (Techeducation).12 In addition, in order to facilitate comparison with previous works, we consider Workexp given by the number of years of working experience of a firm’s founders independently of the sector of activity. The second group includes control variables. A new-born firm may receive valuable tangible and/or intangible resources from a “mother” company (e.g. complementary technologies, access to distribution channels, after-sale services, support to entry into international markets, financing). Such situation indicated by the dummy variable DMother company, is likely to result in greater start-up size.

Second,

a number of sample firms were located in technology incubators

(DIncubated). Such location often involves a physical constraint; in fact the limited space at 11

In addition, Levenson and Willard (2000) have documented that the extent of credit rationing in the US is fairly limited: according to their estimates only 2.14% of firms did not get the funding for which they applied, while an additional 4.22% were discouraged from applying because of the expectation of denial. However, constrained firms proved to be smaller and younger than unconstrained ones. Even in the absence of any correlation between individuals' net worth and human capital, the positive relation between net worth and the likelihood of self-employment may be explained by the lower risk aversion of richer individuals (see Cressy 2000). 12

Ecoeducation measures years spent for the attainment of degrees in economics, law, management, and political sciences, while Techeducation reflects years spent for obtaining degrees in engineering, physics, biology, chemistry, medicine, pharmaceutics, and computer science. In order to properly judge the effective level of competencies of founders, we consider the minimum length of time necessary to attain a certain degree. In order to attain an Italian graduate degree in economics, law, management, political sciences and most scientific degrees four years of studies are requested, while five years is the minimum time for a degree in engineering. Master and Ph.D. programmes require one and three additional years respectively, independently of the specific field. 12

disposal of a firm in an Italian incubator may impede achievement of the desired size (see Colombo and Delmastro 2002 for a description of Italian technology incubators). Therefore, we expect a negative coefficient for this variable. Rreal is the real interest rate in the year of firm’s foundation (see Banca d’Italia 2001). As greater cost of capital negatively affects investments, we predict a negative impact of such variable on start-up size. The variable Infrastructure reflects the level of infrastructure development in the county of firm’s location. It is provided by Centro Studi della Confindustria (1991) and it is calculated as the average of the following indexes: per capita value added, share of manufacturing out of total value added, employment index, per capita bank deposits, automobile-population ratio, and consumption of electric power per head. The Italian benchmark value is 100, while the value of the variable ranges for sample firms from 44 to 175. Since the average value of Infrastructure for sample firms is 115, this shows that high-tech start-ups are usually located in more developed regions. Location in an area with efficient infrastructure may make founders more confident on the future prospects of the firm and convince them to start with a greater size. Other covariates in this group reflect specific characteristics of the industries (at the three digit NACE-CLIO classification) in which start-ups operate; most of them have been considered by previous studies of the determinants of firms’ start-up size. The minimum efficient scale (Mes) is computed as the log of average employment of firms, while Suboptimal is the proportion of employment in firms operating at less than minimum efficient scale (see Görg et al. 2000).13 This latter variable inversely proxies the cost disadvantage experienced by firms that operate at suboptimal size. In accordance with the results of previous studies (see Mata 1996, Mata and Machado 1996, Görg et al. 2000), we expect a positive and a negative impact of such variables on start-up size. In order to create a proxy for industry uncertainty, we had recourse to the database on European initial public offerings (IPO) that was jointly developed by Politecnico di Milano and Tilburg University. This database includes data on 482 IPOs that occurred between 1996 and 2001 in five European new stock markets (Neuer Markt, Nuovo Mercato, Nouveau Marchè, Euro NM, Nmax).14 Uncertainty measures the industry average of the normalized standard deviation of the

market price of newly listed firms in the 50 days following the IPO; its predicted sign is negative. In fact, great variability of post-IPO stock prices in an industry signals great ex-ante uncertainty on new firms’ performance. Under such circumstances, founders of new firms will have incentives to

13

Data sources are the 1981, 1991 and 1996 ISTAT Census. Due to lack of data relating to the Internet sector, the minimum efficient scale in this sector has been assumed to be the same as in the software sector. Alternative measures of Mes such as the one proposed by Caves et al. (1975) and Lyons (1980) could not be computed because of lack of data. 14

Data on IPOs have been collected primarily through IPOs brochures and companies web sites, while data on market prices have been obtained from the Datastream database and the web sites of the above cited new markets. For further details see Giudici and Roosenboom (2002). 13

limit the initial commitment of resources so as to avoid sunk costs (Cabral 1995); with everything else being equal, we expect smaller initial firm size in such industries. Lastly, in some specifications the above mentioned industry-specific variables were replaced by industry dummies; this allowed adoption of a more detailed industry classification (see Table 1). Industry dummies aim to take also into account other industry-specific effects (e.g. the existence of different business opportunities in different sectors) which may influence firms’ start-up size.15 Table 2 illustrates descriptive statistics of the dependent and explanatory variables. The mean value of firms’ initial size measured by the number of salaried employees is 4.42; if one adds the number of founders, mean initial size becomes 6.7 workers. There obviously are remarkable differences across sectors. In particular, the biotechnology and pharmaceutical industry presents the greatest number of total employees per firm (12.5 on average including founders), while in the multimedia content sector firms usually start operations at the smallest scale (4.6). In Table 3 we illustrate the correlation matrix of explanatory variables. Correlation across variables generally is low, suggesting the absence of any relevant problem of multicollinearity. However, there appear to be nonnegligible differences across industries in the human capital characteristics of founding teams. Note for instance the relatively large values of the correlation index between the dummy variable relating to the biotech & pharmaceutical sector from one side and Education, Techeducation and Specworkexp from the other side (0.185, 0.212 and 0.113, respectively). Quite

unsurprisingly, correlations between industry dummies and other industry-specific variables (i.e. Mes, Suboptimal and Uncertainty) generally are quite high.

5. Empirical Results Results of the econometric analysis are reported in Tables 4 and 5. In all tables, in models I and II start-up size is calculated as the number of salaried personnel, while in models III and IV we add the number of founders. For the sake of synthesis, we only report the results of truncated regression models. The results of OLS and sample selection models can be found in the Appendix (Tables A1 and A2).16 Most estimated coefficients of the explanatory variables have the predicted sign. In 15

Industry-specific business opportunities are more easily discovered by people who already operate in the industry. Accordingly, in our sample there is a considerable number of founders who established a new firm in the same sector of the firm by which they were formerly employed. Such individuals account for 40% of the total number of founders. In some industries such percentage is substantially higher: in particular, in the biotech & pharmaceutical industry it is 55%. This may be a source of unobserved heterogeneity. For instance, suppose that for reasons independent of the human capital characteristics of founders, biotech & pharmaceutical new firms require more employees than firms in other industries. Suppose further that in this industry firms employ individuals with greater human capital. Then failure to control for industry-specific effects may lead to biased estimates. We are grateful to an anonymous referee for raising this point. Of course, when it was not possible to rule out the non-identification issue (i.e both σYY and σSY were not statistically different from zero at 95%), estimates of the sample selection model are not reported. 16

14

addition, no substantial differences emerge in the results across different regression approaches and different specifications of the dependent variable. First of all, in Table 4 we consider founders’ education and working experience. More qualified individuals are more likely to perform better as entrepreneurs and to be more confident about the future performance of the new venture; so their desired start-up size should be greater. In addition, as the personal wealth entrepreneurs have access to generally increases with their education and working experience, the financial constraints that may inhibit achievement of the desired size will be eased. In accordance with the above arguments, the working experience gained by founders in previous jobs captured by Workexp, has a positive statistically significant (at 99%) effect on initial size. The coefficient of Education also is positive, but it is statistically significant (at 95%) only when industry-specific control variables are replaced by industry dummies.17 In this paper we are mainly interested in understanding the relative explanatory power of the “wealth effect” and the “entrepreneurial ability effect” of founders’ human capital. For this purpose, let us turn to Table 5. In these models Workexp was replaced by two additional covariates, Specworkexp and Genworkexp. The former variable measures the years of professional experience

gained by founders in the sector of the new firm, while the latter reflects professional experience unrelated to the activity of the new firm. Both variables similarly capture the wealth effect. Nevertheless, Specworkexp more directly reflects the specific component of founders’ human capital; it reveals superior competencies arising from better knowledge of the target industry, and consequently also higher level of confidence about firm’s prospects. Such factors are expected to lead to greater initial size, with the positive effect of Specworkexp being greater than that of Genworkexp. This argument is confirmed by the results of the estimates. While both Specworkexp

and Genworkexp have positive coefficients, significant at conventional confidence levels (with one exception), the coefficient of the former variable always is greater than the one of the latter: Wald tests show that in all specifications but one the difference between the two coefficients is statistically significant at conventional confidence levels.18 In the same way it has to be interpreted the positive coefficients significant at conventional confidence levels of DManager and 17

This result partially differs from those of previous works. A possible reason is that among Italian high-tech entrepreneurs, education may be a poor proxy of both financial wealth and entrepreneurial talent. Åstebro and Bernhardt (1999) using US Census of Population 1990 data, show that household income is closely associated with the educational attainments of entrepreneurs, and positively influences the amount of capital committed to a new venture. Nevertheless, they also find that after controlling for predicted household income, education does not have any additional direct effect on start-up capital. These results support the view that education is not associated with entrepreneurial ability (see also the references mentioned in footnote 5). On the contrary, founders’ education is found by Cabral and Mata (2003) to positively affect the size of Portuguese firms not only at entry (see Mata 1996 for a similar result) but also afterwards. Such evidence is interpreted by the authors as witnessing the greater efficiency of firms created by more educated people.

15

DEntrepreneur (in models II and IV). The fact that within the team of founders one or more of them

previously had a managerial position in a medium or large company or were involved in previous entrepreneurial episodes is not only to be associated with the availability of greater personal funds but also signals the quality of the managerial competencies on which the new venture relies. Considering model II (IV) in Table 5a, when such variables are set at 1 while all other dummies equal null and the remaining variables are evaluated at mean value, the estimated start-up size, measured by the number of salaried employees (by the sum of the number of founders and salaried employees), increases by 19% (35%) and 14% (19%), respectively. Note that when start-up size also includes the number of founders (see models III and IV), the coefficient of Genworkexp, though positive, is no longer significant. This result confirms the role of financing constraints. In fact, young, inexperienced founders generally lack personal capital and so it is difficult for them to hire a number of salaried personnel corresponding to the desired initial size: accordingly, with all else equal, the number of a new firm’s salaried personnel increases with Genworkexp (see the estimates of models I and II). Such individuals have greater incentives to find partners to jointly set up a new firm so as to escape financing constraints. In accordance with this argument, in our sample the Pearson correlation index between Genworkexp and the number of founders is equal to -0.1256 (p value < 0.05). If one considers Workexp, which reflects founders’ working experience irrespective of its nature, the negative correlation is even stronger (-0.165, p value < 0.01). In order to gain further insights into the role played by education, in the models presented in Table 5 we distinguish between economic/law education and scientific/technical one, at graduate and post-graduate level. In all specifications, Ecoeducation has a positive and strongly significant effect on the scale of new-born firms, while the coefficient of Techeducation is insignificant. These results support the view that scientific and technical education does not reflect the specific technical competencies of high-tech entrepreneurs, which are instead mainly connected with professional experience. They may also be indicative of the greater wealth of individuals who got a degree in economics, law, management, and political sciences (e.g. an MBA degree. For a similar result see Åstebro and Bernhardt 1999). Let us now briefly consider the effect of control variables. First of all, the regressions show that firms’ initial size positively depends on help received by a “mother" company, captured by DMother company. Infrastructure also has a positive coefficient, significant at conventional

confidence levels in all specifications but one; this points to the role played by the local endowment

18

A F-test run on the OLS estimates and a Wald test run on the sample selection estimates generally confirm this result (see the Appendix). 16

of infrastructure in influencing the expected performance of newly born firms, thus leading entrepreneurs to start new ventures at a larger scale. Location in a technological incubator and the opportunity cost of capital, measured by the real interest rate at the time a firm was created, turn out to have the predicted negative effect, but statistical significance is weak. As to industry-specific variables, in accordance with previous works (Mata 1996, Mata and Machado 1996, Görg et al. 2000) relatively larger firms enter into markets characterised by substantial economies of scale; Mes has a positive coefficient statistically significant at conventional confidence levels, while the coefficient of Suboptimal, though negative as was predicted, is not significant. Moreover, all other things equal, greater industry uncertainty while deterring unrecoverable investments, results in lower mean initial size, due to the desire of entrepreneurs to avoid sunk costs; however, statistical significance again is weak. In the specifications illustrated in Table 5b, industry-specific variables are replaced by industry dummies, with the baseline in the estimates being represented by the computer industry; the aim is to better control for unobserved heterogeneity across sectors. A LR test confirms the explanatory power of industry effects: in all models the industry dummies are jointly significant at 99%. The results previously illustrated basically remain unchanged. In particular, with the exceptions mentioned above, both Specworkexp and Genworkexp have positive statistically significant coefficients and the null hypothesis that the values of the coefficients of the two variables be equal is rejected by a Wald test at conventional confidence levels. In addition, the coefficients of DManager and DEntrepreneur are positive and significant at conventional confidence levels. Lastly, since in the empirical analysis we use survey-based data, this may raise the concern that the longer the time elapsed since firms’ foundation, the less reliable the data relating to the start-up size of firms and the characteristics of firms’ founders. For this purpose, we firstly run heteroskedasticity tests. In particular, we tested the null hypothesis of homoskedasticity of the error terms against the presence of multiplicative heteroskedasticity caused by the log of the age of the firms at survey date. We find that the null hypothesis is accepted by LR tests in all equations of Tables 5a and 5b (results are available from the authors upon request).19 Second, we focused on a subsample composed of 203 firms that were established in 1995 or later. For these firms a rather short time period has elapsed since foundation: so information provided by firms’ owner-managers is likely to be more accurate. Of course, the shortcoming is that the number of observations decrease quite substantially. We rerun the econometric models for this subsample. Results turned

19

The restriction imposed on the model and tested by the LR tests is var (εi) = exp (γ’ Agei) with γ equal to zero as null hypothesis. See Godfrey (1988) for more details on tests for heteroskedasticity in limited dependent variable models. 17

out to be quite close to those relating to the entire sample (again they are available from the authors upon request).20

6. Concluding remarks The aim of this paper was to extend our understanding of the determinants of firms’ start-up size. The decision as to the initial scale of operations is an important one. As is well documented in the literature, in the early years following entry start-up size positively affects the probability of survival. In addition, surviving new firms that started operations at sub-optimal scale struggle to grow so as to rapidly eliminate the disadvantage accruing from small size. Nevertheless, the analysis of the factors that influence the initial size of firms is quite undeveloped. The few empirical studies on this topic primarily focus on industry characteristics, such as the presence of economies of scale and environmental uncertainty; because of lack of proper data, they generally are unable to explain the observed heterogeneity among new entrants in a given industry. Therefore the question why firms enter into the same market with different sizes so far remains largely unexplored. This paper directly addresses this issue; while controlling for industry-specific and other contextual factors, it draws attention to the influence exerted on start-up size by founders’ human capital. In particular, we aim to disentangle the “entrepreneurial ability” and “wealth” effects of human capital. For this purpose, we consider a sample composed of 391 Italian firms that operate in hightech industries, in both manufacturing and services, were created in 1980 or later, and were independent at start-up time. We estimate different econometric models (OLS, truncated, sample selection models) relating firms’ initial size proxied alternatively by the number of salaried employees and the sum of the number of founders and salaried employees, to a series of covariates. The key findings can be summarised as follows. First, the human capital of entrepreneurs measured by several indicators of educational attainments and working experience, turns out to have a crucial influence on start-up size. The effect of human capital is twofold. On the one hand, founders with greater entrepreneurial talent and greater confidence in the prospects of the new venture start operations at greater scale, all being equal. On the other hand, more educated, better qualified, and probably wealthier individuals suffer to a lesser extent from financial constraints associated with imperfections in capital markets that otherwise hinder achievement of the “optimal” start-up size. In accordance with this view, all human capital variables generally have a positive impact on firms’ initial size. However variables 20

As to human capital variables, the main difference is that variables capturing education have a more positive effect on initial size, while Genworkexp looses its explanatory power. 18

that reflect the specific component of human capital (i.e. years of working experience of founders in the same sector of the new firm and variables indicating their managerial and entrepreneurial experiences) and thus capture both the “wealth” and the “entrepreneurial ability” effects of human capital, exhibit greater explanatory power than those that only reflect the generic component (i.e. notably working experience in other sectors of activity). These results are interesting in their own right as they confirm the view that the existence of both firm-specific persistent shocks and financial constraints is a key driver of the dynamics of young firms (see Cooley and Quadrini 2001). They also have important policy implications. Some authors (see for instance Holtz-Eakin 2000, Santarelli and Vivarelli 2002) question the rationale for public support of new firms. Actually, failure rates are especially high among such firms. Therefore, public support may distort and delay the competitive selection process, subsidising inefficiencies. This is especially worrisome if firms are not financially constrained. In this paper we have shown that the start-up size of Italian NTBFs increases with the industry-specific and managerial skills of founders. As there is a positive relation between firms’ initial size and the probability of survival, the likelihood that the firms established by such highly qualified individuals be able to stay in business is greater. In addition, the initial size of firms also increases with the level of education and the generic working experience of founders, two variables that generally indicate availability of grater personal wealth to finance the new firm. So this evidence possibly suggest that Italian NTBFs indeed are financially constrained; survey based evidence on Italian high-tech start-ups supports such view (Giudici and Paleari 2000. See also Colombo and Grilli 2004). While we agree with the view that indiscriminate public support to the NTBF sector is both unfeasible and inefficient, this does not mean that there is no need for public support. In particular, the evidence provided in this study argues in favour of public interventions that stimulate the establishment of new firms by individuals endowed with high level of specific human capital and facilitate the provision of seed and start-up capital to those ventures.

19

References Agarwal, R., Audretsch, D.B., 2001. Does entry size matter? The impact of the life cycle and technology on firm survival. Journal of Industrial Economics 49, 21- 43. Almus, M., 2000. What characterizes a fast growing firm?. ZEW Discussion Paper 00-64. Arabsheibani, G., de Meza, D., Maloney, J., Pearson, B., 2000. And a vision appeared unto them of a great profit: Evidence of self-deception among the self-employed. Economic Letters 67, 35-41. Ǻstebro, T., Bernhardt, I., 1999. The winner’s curse of human capital. Working Paper CES 99-5, Center for Economic Studies, U.S. Department of Commerce. Audretsch, D.B., 1991. New-firm survival and the technological regime. Review of Economics and Statistics 72, 441-450. Audretsch, D.B., 1995a. Innovation and industry evolution. Cambridge, Mass: MIT Press. Audretsch, D.B., 1995b. Innovation, growth, and survival. International Journal of Industrial Organization 13, 441-457. Audretsch, D.B., Mahmood, T., 1994. Firm selection and industry evolution: The post-entry performance of new firms. Journal of Evolutionary Economics 4, 243-260. Audretsch, D.B., Mahmood, T., 1995. New firm survival: New results using a hazard function. Review of Economics and Statistics 77, 97-103. Audretsch, D.B., Santarelli, E., Vivarelli, M., 1999. Start-up size and industrial dynamics: Some evidence from Italian manufacturing. International Journal of Industrial Organization 17, 965-983. Banca d’Italia, 2001. Relazione generale sulla situazione economica del Paese e bollettino economico. (In Italian). Bates, T., 1990. Entrepreneur human capital inputs and small business longevity. Review of Economics and Statistics 72, 551-559. Becker, G.S., 1975. Human capital. New York: National Bureau of Economic Research. Black, J., de Meza, D., Jeffreys, D., 1996. House pricing, the supply of collateral and the enterprise economy. Economic Journal 106, 60-75. Blanchflower, D., Oswald, A., 1998. What makes an entrepreneur. Journal of Labor Economics 16, 26-60. Bloom, D., Killingsworth, M.,1985. Correcting for truncation bias caused by a latent truncation variable. Journal of Econometrics 27, 131-135. Brüderl, J., Preisendörfer, P., 2000. Fast growing businesses: Empirical evidence from a German study. International Journal of Sociology 30, 45-70. Brüderl, J., Preisendörfer, P., Ziegler, R., 1992. Survival chances of newly founded business organizations. American Sociological Review 72, 227-242. Cable, J., Schwalbach, J., 1991. International comparisons of entry and exit. In Geroski, P.A., Schwalbach, J. (Eds.), Entry and market contestability: An international comparison. Oxford: Blackwell, 257-281. Cabral, L., 1995. Sunk costs, firm size and firm growth. Journal of Industrial Economics 43, 161172. 20

Cabral, L., Mata, J., 2003. On the evolution of the firm size distribution: Facts and theory. American Economic Review 93, 1075-1090. Camerer, C., Lovallo, D., 1999. Overconfidence and excess of entry. American Economic Review 89, 306-318. Carpenter, R.E, Petersen, B.C., 2002. Capital market imperfections, high-tech investment, and new equity financing. Economic Journal 112, F54-F72. Caves, R.E., 1998. Industrial organization and new findings on the turn-over and mobility of firms. Journal of Economic Literature 36, 1947-1982. Caves, R.E., Khalilzadeh-Shirazi, J., Porter, M.E., 1975. Scale economies in statistical analyses of market power. Review of Economics and Statistics 57, 133-140. Centro Studi Confindustria, 1991, Indicatori Economici Provinciali. Rome: SIPI, Collana Industria e Territorio (in Italian). Chandler, G.N., Jansen, E., 1992. The founder’s self-assessed competence and venture performance. Journal of Business Venturing 7, 223-236. CIRET, 2002. 1st RITA report on Italian new technology-based firms. Milan: Politecnico di Milano. Colombo, M.G., Delmastro M., 1999. Some stylized facts on organization and its evolution. Journal of Economic Behaviour & Organization 40, 252-274. Colombo, M.G., Delmastro, M., 2002. How effective are technology incubators? Evidence from Italy. Research Policy 31, 1103-1122. Colombo, M.G., Grilli, L., 2004, Funding gaps? Access to bank loans by high-tech start-ups, CIRET-Politecnico di Milano, Working paper n. 04.01. Cooley, T.F., Quadrini, V., 2001. Financial markets and firm dynamics. American Economic Review 91, 1286-1310. Cooper, A.C., 1985. The role of incubator organizations in the founding of growth-oriented firms. Journal of Business Venturing 1, 75-86. Cooper, A.C., Bruno, A.V., 1977. Success among high-technology firms. Business Horizons 20, 1622. Cressy, R., 1996. Are business start-ups debt-rationed? Economic Journal 106, 1253-1270. Cressy, R., 2000. Credit rationing or entrepreneurial risk aversion? An alternative explanation for the Evans and Jovanovic finding. Economics Letters 66, 235-240. Del Monte, A., Scalera, D., 2001. The life duration of small firms born within a start-up programme: Evidence from Italy, Regional Studies 35, 11-21. Dixit, A.K., Pindyck, R.S., 1994. Investment under uncertainty. Princeton, New Jersey: Princeton University Press. Dunkelberg, W.C., Cooper, A.C., Woo, C.Y., Dennis, W., 1987. New firm growth and performance. In: Churchill, N., Hornaday, J., Kirchhoff, B., Krasner, O., Vesper, K.H. (Eds.), Frontiers of entrepreneurship research. Wellesly, Mass.: Center for Entrepreneurial Studies, Babson College, 307-321. Dunne, T., Roberts, M.J., Samuelson, L., 1988. Pattern of firm entry and exit in U.S. manufacturing industries. Rand Journal of Economics 19, 495-515.

21

Dunne, T., Roberts, M.J., Samuelson, L., 1989. The growth and failure of U.S. manufacturing plants. Quarterly Journal of Economics 104, 671-698. Evans, D.S., 1987a. Test of alternative theories of firm growth. Journal of Political Economy 95, 657-674. Evans, D.S., 1987b. The relationship between firm growth, size and firm age: Estimates for 100 manufacturing industries. Journal of Industrial Economics 35, 567-582. Evans, D., Jovanovic, B., 1989. An estimated model of entrepreneurial choice under liquidity constraints. Journal of Political Economy 97, 808- 827. Evans, D., Leighton, L., 1989. Some empirical aspects of entrepreneurship. American Economic Review 79, 519- 535. Fazzari, S.M., Hubbard, R.G., Petersen, B.C., 1988. Financing constraints and corporate investment. Brooking Papers on Economic Activity 2, 141-206. Geroski, P.A., 1995. What do we know about entry?. International Journal of Industrial Organization 13, 421-440. Godfrey, L.G., 1988, Misspecification tests in econometrics, Cambridge: Cambridge University Press. Giudici, G., Paleari, S., 2000. The provision of finance to innovation: a survey conducted among Italian technology-based small firms. Small Business Economics 14, 37-53. Giudici, G., Roosenboom, P., 2002. Pricing initial public offerings on European ‘new’ stock markets. XIII Tor Vergata Financial Conference, Rome, Italy. Görg, H., Strobl, E., Ruane, F., 2000. Determinants of firm start-up size: An application of quantile regression for Ireland. Small Business Economics 14, 211-222. Hart, P.E., Oulton, N., 1996. Growth and size of firms. Economic Journal 106, 1242-1252. Holtz-Eakin, D., 2000. Public policy toward entrepreneurship. Small Business Economics 15, 283291. Holtz-Eakin, D., Joulfaian, D., Rosen, H.S., 1994a. Entrepreneurial decisions and liquidity constraints. Rand Journal of Economics 25, 334-347. Holtz-Eakin, D., Joulfaian, D., Rosen, H.S., 1994b. Sticking it out: entrepreneurial survival and liquidity constraints. Journal of Political Economy 102, 53-75. Jovanovich, B., 1982. Selection and evolution of industry. Econometrica 50, 649-670. Levenson, A.R., Willard, K.L., 2000. Do firms get the financing they want? Measuring credit rationing experienced by small businesses in the U.S.. Small Business Economics 14, 83-94. Lindh, T., Ohlsson, H., 1996. Self-employment and windfall gains: Evidence from the Swedish lottery. The Economic Journal 106, 1515-1526 Lyons, B., 1980. A new measure of minimum efficient plant size. Economica 47, 19-34. Maddala, G.S., 1986. Limited-dependent and qualitative variables in econometrics. Cambridge: Cambridge University Press. Mata, J., 1996. Market, entrepreneurs and the size of new firms. Economic Letters 52, 89-94. Mata, J., Machado, J.A.F., 1996. Firm start-up size: A conditional quantile approach. European Economic Review 40, 1305-1323. Mata, J., Portugal, P., 1994. Life duration of new firms. Journal of Industrial Economics 42, 227245. 22

Mata, J., Portugal, P., Guimaraes, P., 1995. The survival of new plants: start-up conditions and postentry evolution. International Journal of Industrial Organization 13, 459-482. Meyer, B.D., 1990. Why are there so few black entrepreneurs?. NBER Working paper n. 3537, Cambridge, Mass. Muthen, B., Jöreskog, K., 1983. Selectivity problems in quasi-experimental studies. Evaluation Review 7, 139-174. Philips, B.D., Kirchhoff, B.A., 1989. Formation, growth and survival: Small firm dynamics in the U.S. economy. Small Business Economics 1, 65-74. Pindyck, R.S., 1988. Irreversible investment, capacity choice, and the value of the firm. American Economic Review 78, 969-985. Pindyck, R.S., 1991. Irreversibility, uncertainty, and investment. Journal of Economic Literature 29, 1110-1148. Pindyck, R.S., 1993. A note on competitive investment under uncertainty. American Economic Review 83, 273-277. Preisendörfer, P., Voss, T., 1990. Organizational mortality of small firms: The effects of entrepreneurial age and human capital. Organizational Studies 11, 107-129. Santarelli, E., Vivarelli, M., 2002. Is subsidizing entry an optimal policy?. Industrial and Corporate Change 11, 39-52. Siegel, R., Siegel, E., Macmillan, I.C., 1993. Characteristics distinguishing high growth ventures. Journal of Business Venturing 8, 169-180. Storey, D.J., 1994. Understanding the small business sector. London: Thomson Learning. Stuart, R.W., Abetti, P.A., 1990. Impact of entrepreneurial and management experience on early performance. Journal of Business Venturing 5, 151-162. Sutton, J., 1997. Gibrat’s legacy. Journal of Economic Literature 35, 40-59. Wagner, J., 1994. The post-entry performance of new small firms in German manufacturing industries. Journal of Industrial Economics 42, 141-154. Westhead, P., Cowling M., 1995. Employment change in independent owner-managed high technology firms in Great Britain. Small Business Economics 7, 111-140. Westhead, P., Storey, D.J., 1997. Financial constraints on the growth of high technology small firms in the United Kingdom. Applied Financial Economics 7, 197-201. Xu, B., 1998. A reestimation of the Evans-Jovanovic entrepreneurial choice model. Economics Letters 58, 91-95.

23

Table 1 - The explanatory variables of firm start-up size Variable

Description

Education Ecoeducation

Average number of years of founders’ education Average number of years of founders’ economic, law and/or managerial education at graduate and post-graduate level Average number of years of founders’ scientific and/or technical education at graduate and post-graduate level Average number of years of founders’ working experience before firm’s foundation Average number of years of working experience gained by founders in the same sector of the start-up before firm’s foundation Average number of years of working experience gained by founders in other sectors than the one of the start-up before firm’s foundation One for firms with one ore more founders with a prior management position in a large or medium company (i.e. number of employees greater than 100) One for firms with one or more founders with a previous self-employment experience One for firms that at start-up time, received some aid by a “mother” company One for firms located in a technology incubator Real interest rate in the year of firm’s foundation Value of the index measuring regional infrastructures in 1992 (mean value among Italian regions=100) Minimum efficient scale in the sector of the start-up in the year in which the firm was created (or in the nearest year for which data were available) measured by the log of the average number of employeesa Proportion of employment in the sector of the start-up absorbed by firms that operate at sub-optimal scale in the year in which the firm was created (or in the nearest year for which data were available) a Industry average of the normalised standard deviation of the market price of newly listed firms in the 50 days following the IPO

Techeducation Workexp Specworkexp Genworkexp DManager DEntrepreneur DMother company DIncubated Rreal Infrastructure Mes

Suboptimal

Uncertainty DBiopha

One for firms in the biotechnology or pharmaceutical industry

DSemic

One for firms in the semiconductor industry

DTLCequip

One for firms in the telecommunication equipment industry

DInstrument

One for firms in medical, optical and electrical instruments industry

DAerospace

One for firms in the aerospace industry

DSoftware

One for firms in the software industry

DElpub

One for firms in the electronic publishing industry

DInternet

One for firms in Internet and telecommunication services industry

Legend. aAvailable industry data refer to years 1981, 1991 and 1996. Data are from the ISTAT Census of firms.

24

Table 2 - Descriptive statistics of variables of the econometric models Variables

Mean

S.D.

Min

Max

Start-up size: n. of salaried employees

4.4245

15.4808

0

230

Start-up size: sum of the n. of founders and salaried employees

6.7800

15.4966

1

233

Education

14.8661

2.5604

8.0000

21.5000

Ecoeducation

0.3695

0.9585

0

5.0000

Techeducation

1.4936

1.9333

0

8.0000

Workexp

13.3468

8.6397

0

55.0000

Specworkexp

3.7624

7.2085

0

55.0000

Genworkexp

9.5844

8.9516

0

46.0000

DManager

0.0895

0.2859

0

1

DEntrepreneur

0.4038

0.4916

0

1

DMother company

0.1125

0.3164

0

1

DIncubated

0.1253

0.3315

0

1

Rreal

5.0604

1.9412

-4.8400

7.9300

116.2590

27.6970

43.7000

174.7000

Mes

0.9823

0.3789

0.7363

2.0230

Suboptimal

0.2262

0.0232

0.1525

0.2863

Uncertainty

0.0353

0.0032

0.0300

0.0391

DBiopha

0.0485

0.2152

0

1

DSemic

0.0255

0.1580

0

1

DTLCequip

0.0434

0.2041

0

1

DInstrument

0.0920

0.2894

0

1

DAerospace

0.0127

0.1125

0

1

DSoftware

0.2864

0.4526

0

1

DElpub

0.0588

0.2355

0

1

DInternet

0.3989

0.4903

0

1

Infrastructure

25

Table 3. Correlation matrix of the explanatory variables Educationn

Education

1.0000

Ecoeducation

0.2514

1.0000

Techeducation

0.8085

-0.0798

1.0000

Workexp

-0.1407

-0.0754

-0.0049

1.0000

Specworkexp

-0.1811

-0.0958

-0.0440

0.4294

1.0000

Genworkexp

0.0238

0.0112

0.0336

0.5927

-0.4729

1.0000

0.0374

0.0557

0.0262

0.0573

-0.0459

0.0969

1.0000

-0.0768

0.0031

-0.0951

0.1522

0.0013

0.1474

0.0896

1.0000

DMother company

0.1592

-0.0008

0.1614

0.1042

0.0413

0.0675

0.0301

0.1439

DIncubated

0.1380

-0.0278

0.1209

0.0267

-0.0385

0.0561

-0.0646

-0.0033

Rreal

-0.0017

-0.0294

0.0178

-0.0419

0.0218

-0.0575

-0.0512

0.0064

Infrastructure

0.0447

0.1344

-0.0093

0.0924

0.0051

0.0849

0.0042

-0.0062

Mes

0.0549

-0.1080

0.1209

0.1293

0.1999

-0.0361

-0.0564

-0.0831

Suboptimal

0.0254

-0.0473

0.0015

-0.1171

-0.0695

-0.0572

-0.0191

0.0116

Uncertainty

-0.0725

0.1346

-0.1999

-0.0959

-0.2802

0.1322

0.1294

0.1029

DBiopha

0.1851

-0.0375

0.2119

0.1070

0.1129

0.0118

-0.0709

-0.0922

DSemic

-0.1668

0.0051

-0.0974

-0.0030

-0.0397

0.0296

-0.0508

-0.0221

DTLCequip

-0.1277

-0.0299

-0.0404

0.1097

0.2100

-0.0621

-0.0669

-0.0316

DInstrument

0.0843

-0.0600

0.1185

0.1088

0.0271

0.0830

-0.0379

0.0083

DAerospace

0.0267

-0.0122

0.0259

0.0198

0.0032

0.0168

0.0441

0.0173

DSoftware

0.0582

-0.0905

0.1420

-0.0740

0.0600

-0.1188

-0.0798

-0.0455

DElpub

0.0918

-0.0132

0.0359

0.0027

-0.0932

0.0771

-0.0022

0.0701

DInternet

-0.0965

0.1816

-0.2368

-0.0917

-0.1756

0.0522

0.1653

0.0574

DManager DEntrepreneur

b

Variables

Ecoeducation

DMother company

Techeducation

Workexp

DIncubated

RReal

Specworkexp

Genworkexp

Infrastructure

Mes

DManager

DEntrepreneurb

Variables

Suboptimal

Uncertainty

DMother company

1.0000

DIncubated

0.0363

1.0000

Rreal

-0.0163

0.0947

1.0000

Infrastructure

0.0318

0.1089

-0.0484

1.0000

Mes

0.1010

0.1577

-0.0233

0.0978

1.0000

Suboptimal

-0.0750

-0.0099

0.1499

0.0490

-0.1190

1.0000

Uncertainty

-0.0811

-0.1082

-0.0363

-0.0631

-0.4486

0.0016

1.0000

DBiopha

0.1077

0.1659

0.0410

0.0998

0.5897

0.2371

-0.1831

DSemic

-0.0064

-0.0124

0.0584

-0.0006

0.1086

-0.1183

-0.1163

DTLCequip

-0.0362

-0.0049

0.0560

0.0566

0.3711

-0.3367

-0.1763

DInstrument

0.0825

0.1199

-0.0613

0.0565

0.2187

-0.2576

-0.2286

DAerospace

0.0315

0.0944

-0.1185

-0.0697

0.2099

-0.2068

-0.0817

DSoftware

-0.0108

-0.0006

0.1138

0.0012

-0.2577

0.2402

-0.4711

DElpub

-0.0202

0.0367

-0.0013

-0.0303

-0.0951

-0.0196

0.1411

-0.1664

-0.0447

-0.0841

-0.4580

0.0537

0.7561

DInternet -0.0753 Legend. bData available for a subset of 260 firms.

26

Table 4 - The determinants of start-up size: the effect of founders’ education and working experience N. of salaried employees (log)

Sum of the n. of founders and salaried employees (log)

I

II

III

IV

a0

Constant

-6.9071 (5.4403)

-10.7244 (3.5071)c

1.2355 (0.9990)

-0.0800 (0.4787)

a1

Education

0.0987 (0.1018)

0.2014 (0.0956)b

0.0288 (0.0209)

0.0487 (0.0209)b

a2

Workexp

0.1285 (0.0402)c

0.1192 (0.0317)c

0.0182 (0.0061)c

0.0201 (0.0059)c

c

c

c

a3

DMother company

2.968 (0.9402)

2.7639 (0.7517)

0.6495 (0.1562)

0.6645 (0.1490)c

a4

DIncubated

-2.4624 (1.1488)b

-1.7252 (0.8613)b

-0.2262 (0.1647)

-0.1800 (0.1584)

a5

Rreal

-0.1232 (0.1351)

-0.1485 (0.1154)

-0.0315 (0.0264)

-0.0422 (0.0258)

a6

Infrastructure

0.0460 (0.0170)c

0.0385 (0.0127)c

0.0042 (0.0019)b

0.0042 (0.0018)b

a7

Mes

1.3139 (0.7712)a

-

0.2212 (0.1528)

-

a8

Subotpimal

-8.2059 (10.1735)

-

-1.6796 (2.2217)

a9

Uncertainty

-120.9366 (105.0171)

-

-24.3427 (18.3051)

-

a10

DBiopha

-

0.0953 (1.3688)

-

0.3232 (0.3545)

a11

DSemic

-

4.2149 (1.6102)c

-

1.3267 (0.3913)c

a12

DTLCequip

-

1.6246 (1.4094)

-

0.4234 (0.3564)

a13

DInstrument

-

-1.4942 (1.3676)

-

-0.2744 (0.3239)

a14

DAerospace

-

1.4664 (1.8878)

-

0.5844 (0.4904)

a15

DSoftware

-

-0.7506 (1.1957)

-

-0.0170 (0.2921)

a16

DElpub

-

-1.1578 (1.5352)

-

-0.0800 (0.3472)

a17

DInternet

-

-0.5363 (1.1599)

-

0.0061 (0.2856)

Sigma

2.1238 (0.3519)c

1.8713 (0.2636)c

0.8879 (0.0462)c

0.8515 (0.0431)c

Number of observations

391

391

391

391

Log-likelihood

-335.7741

-325.5592

-426.8874

-415.7872

H0: industry slopes = 0 29.6270c (8) 29.8268c (8) H0: all slopes (except LR test 122.5564c (9) 144.4298c (14) 53.4690c (9) 75.6694c (14) costant term) = 0 Legend. Results of the truncated regression model. a Significance level greater than 90%; b Significance level greater than 95%; c Significance level greater than 99%. Standard errors and number of restrictions in parentheses. In Model I and II, the number of salaried employees has been augmented by one in order to permit the use of a logarithmic form. LR test

27

Table 5a - The determinants of start-up size: the effect of founders’ generic and specific human capital N. of salaried employees (log)

Sum of the n. of founders and salaried employees (log)

I

II

III

IV

a0 a1

Constant Ecoeducation

-4.1389 (3.8472) 0.7155 (0.2002)c

0.2136 (5.3366) 1.1399 (0.3682)c

1.7495 (0.8891)b 0.2187 (0.0485)c

1.9764 (1.0716)a 0.2599 (0.0680)c

a2

Techeducation

-0.0661 (0.1060)

-0.1633 (0.1491)

-0.0150 (0.0261)

-0.0161 (0.0302)

c

c

c

0.0199 (0.0082)b

a3

Specworkexp

0.1174 (0.0310)

b

0.1155 (0.0402)

b

0.0267 (0.0074)

a4

Genworkexp

0.0578 (0.0241)

0.0734 (0.0346)

0.0061 (0.0060)

0.0036 (0.0073)

a5

DManager

1.7204 (0.6446)c

1.8658 (0.8853)b

0.6150 (0.1597)c

0.4218 (0.1810)b

a6

DEntrepreneur

-

1.4655 (0.6579)b

-

0.2529 (0.1152)b

a7

DMother company

2.6157 (0.6590)c

2.4486 (0.8701)c

0.6767 (0.1438)c

0.5983 (0.1725)c

a

a8

DIncubated

-1.4435 (0.7567)

-1.4604 (0.9332)

-0.0941 (0.1501)

-0.0738 (0.1573)

a9

Rreal

-0.0697 (0.1006)

-0.1283 (0.1372)

-0.0300 (0.0244)

-0.0232 (0.0296)

a10

Infrastructure

0.0291 (0.0101)c

0.0306 (0.0136)b

0.0030 (0.0018)a

0.0024 (0.0020)

b

a

a

0.2934 (0.1584)a

a11

Mes

1.5280 (0.5941)

1.5060 (0.7932)

0.2544 (0.1397)

a12

Subotpimal

-1.9776 (7.5057)

-13.8004 (10.4418)

-0.8079 (2.0459)

-3.2971 (2.2994)

a13

Uncertainty

-105.4176 (76.8836)

-186.7860 (120.4778)

-29.9990 (17.5107)a

-21.2714 (21.0864)

Sigma

1.7652 (0.2310)c

1.8002 (0.3046)c

0.8266 (0.0411)c

0.7819 (0.0463)c

Number of observations

391

260

391

260

Log-likelihood H0: all slopes (except costant term) = 0 H0: a3-a4=0

-320.8993

-190.6995

-407.1531

-261.7236

LR test Wald test

c

153.7496 (12)

c

95.5376 (13)

92.9376 (12)

63.7704c (13)

1.76 (1)

c

4.27b (1)

b

5.18 (1)

c

8.34 (1)

Legend. Results of the truncated regression model. a Significance level greater than 90%; b Significance level greater than 95%; c Significance level greater than 99%. Standard errors and number of restrictions in parentheses. In Model I and II, the number of salaried employees has been augmented by one in order to permit the use of a logarithmic form.

28

Table 5b - The determinants of start-up size: the effect of founders’ generic and specific human capital N. of salaried employees (log)

Sum of the n. of founders and salaried employees (log)

I

II

III

IV

a0 a1

Constant Ecoeducation

-5.4739 (1.8574)c 0.6175 (0.1640)c

-7.9564 (3.1329)b 0.9185 (0.2907)c

0.7220 (0.3472)b 0.2071 (0.0465)c

0.4992 (0.4316) 0.2320 (0.0655)c

a2

Techeducation

0.0067 (0.0956)

-0.0302 (0.1349)

-0.0029 (0.0259)

0.0019 (0.0299)

c

c

c

0.0206 (0.0081)b

a3

Specworkexp

0.1169 (0.0271)

c

0.1099 (0.0362)

b

0.0295 (0.0071)

a4

Genworkexp

0.0547 (0.0210)

0.0684 (0.0320)

0.0056 (0.0058)

0.0028 (0.0072)

a5

DManager

1.7176 (0.5709)c

1.9353 (0.8397)b

0.6367 (0.1540)c

0.4343 (0.1792)b

a6

DEntrepreneur

-

1.4257 (0.5948)b

-

0.2690 (0.1115)b

a7

DMother company

2.4201 (0.5447)c

2.2163 (0.7600)c

0.6844 (0.1377)c

0.5805 (0.1697)c

a

a8

DIncubated

-1.1133 (0.6293)

-1.3733 (0.8502)

-0.0801 (0.1453)

-0.0827 (0.1546)

a9

Rreal

-0.0805 (0.0888)

-0.1151 (0.1243)

-0.0403 (0.0238)a

-0.0318 (0.0288)

a10

Infrastructure

0.0272 (0.0086)c

0.0348 (0.0136)b

0.0031 (0.0017)a

0.0034 (0.0020)a

a11

DBiopha

0.6896 (1.0644)

1.5358 (1.4608)

0.4884 (0.3261)

0.7382 (0.3684)

c

3.1313 (1.1510)

b

3.6451 (1.6525)

c

1.2800 (0.3602)

1.1474 (0.4306)c

a12

DSemic

a13

DTLCequip

0.8655 (1.0798)

2.2842 (1.5119)

0.3382 (0.3278)

0.5766 (0.3733)

a14

DInstrument

-0.9909 (1.0433)

0.0990 (1.3418)

-0.1330 (0.2963)

0.0629 (0.3410)

a15

DAerospace

1.5926 (1.4706)

2.7937 (2.4535)

0.5607 (0.4478)

1.0183 (0.6413)

a16

DSoftware

-0.6724 (0.9146)

-0.3264 (1.2128)

0.0328 (0.2668)

0.1247 (0.3048)

a27

DElpub

-0.5681 (1.1618)

-0.8709 (1.5667)

0.0668 (0.3175)

0.0961 (0.3549)

a28

DInternet

-0.9788 (0.9185)

-0.8581 (1.2502)

-0.0546 (0.2630)

0.1100 (0.3054)

Sigma

1.5997 (0.1856)c

1.8658 (0.2625)c

0.7940 (0.0386)c

0.7589 (0.0442)c

LR test LR test Wald test

Number of observations

391

260

391

260

Log-likelihood

-311.4081

-186.6319

-396.1276

-256.2134

H0: industry slopes = 0 H0: all slopes (except costant term) = 0 H0: a3-a4=0

34.2092c (8)

23.5634c (8)

32.4680c (8)

20.5436c (8)

172.732c (17)

103.6728c (18)

114.9886c (17)

74.7896c (18)

7.68c (1)

2.00 (1)

12.27c (1)

5.27b (1)

Legend. Results of the truncated regression model. a Significance level greater than 90%; b Significance level greater than 95%; c Significance level greater than 99%. Standard errors and number of restrictions in parentheses. In Model I and II, the number of salaried employees has been augmented by one in order to permit the use of a logarithmic form.

29

APPENDIX Table A1 - The determinants of start-up size: the effect of founders’ generic and specific human capital (OLS and sample selection models)

N. of salaried employees (log) Ia

IIa

OLS

Sum of the n. of founders and salaried employees (log) IIIa

IIIb

OLS

IVa

IVb Sample selection model

OLS

Sample selection model

OLS

a0

Constant

0.6653 (0.8127)

1.5205 (1.0232)

1.7773 (0.7013)b

-2.8889 (5.9235)

1.9620 (0.8704)b

1.1132 (1.1274)

a1

Ecoeducation

0.2277 (0.0468)c

0.2922 (0.0690)c

0.1857 (0.0404)c

1.1145 (1.0051)

0.2238 (0.0587)c

0.4268 (0.1569)c

a2

Techeducation

-0.0066 (0.0236)

-0.0117 (0.0287)

-0.0105 (0.0203)

0.4324 (0.5313)

-0.0103 (0.0244)

-0.0235 (0.0735)

a3

Specworkexp

0.0353 (0.0070)c

0.0298 (0.0081)c

0.0223 (0.0061)c

0.1300 (0.1301)

0.0168 (0.0069)b

0.0419 (0.0333)

a4

Genworkexp

0.0170 (0.0055)c

0.0168 (0.0070)b

0.0051 (0.0048)

-0.0136 (0.0610)

0.0030 (0.0060)

-0.0253 (0.0263)

a5

DManager

0.4292 (0.1548)c

0.3247 (0.1835)a

0.5146 (0.1336)c

1.9339 (1.8557)

0.3486 (0.1561)b

1.2125 (0.6605)a

a6

DEntrepreneur

-

0.2381 (0.1106)b

-

-

0.2010 (0.0941)b

0.4105 (0.2698)

c

a7

DMother company

c

0.7713 (0.1402)

b

c

0.6681 (0.1768)

b

c

c

0.5774 (0.1209)

0.5349 (0.1200)

0.5279 (0.1504)

0.6173 (0.1663)c

a8

DIncubated

-0.3017 (0.1350)

-0.3192 (0.1499)

-0.0803 (0.1165)

-0.0153 (0.1208)

-0.0711 (0.1275)

-0.0487 (0.1457)

a9

Rreal

-0.0141 (0.0227)

-0.0184 (0.0288)

-0.0236 (0.0196)

-0.1908 (0.2817)

-0.0182 (0.0245)

-0.0830 (0.0791)

a10

Infrastructure

0.0047 (0.0016)c

0.0043 (0.0019)b

0.0022 (0.0013)

0.0015 (0.0014)

0.0019 (0.0016)

0.0014 (0.0017)

b

b

a

a

a

0.2865 (0.1246)b

a11

Mes

0.3335 (0.1314)

0.3594 (0.1554)

a

0.2014 (0.1134)

0.1852 (0.1032)

0.2468 (0.1322)

a12

Subotpimal

-1.0458 (1.9260)

-4.2285 (2.2551)

-0.7456 (1.6619)

1.1305 (1.6223)

-2.9006 (1.9183)

-1.4150 (2.0340)

a13

Uncertainty Number of observations

-20.3929 (15.6718)

-26.1024 (19.7710)

-22.9234 (13.5230)a

-22.2889 (13.5141)a

-16.2767 (16.8189)

-15.5003 (16.9790)

391

260

391

391

F test LR test

260

260

b

-

0.8775 (0.1352)c

c

0.9106 (0.0811)

-

0.8008 (0.0951)c

σ YY σ SY

-

-

-

R2

0.2809

0.3053

0.2229

-

0.2276

-

Log-likelihood H0: all slopes (except costant term) = 0 H0: all slopes (except costant term) = 0 H0: a3-a4=0

-

-

-

-410.2328

-

-263.9068

12.31c (12)

8.32c (13)

9.04c (12)

-

5.58c (13)

-

-

42.988c (13)

-

-

-

-

1.3746 (0.5989)

c

-

67.842 (12)

F test 7.23c (1) 2.75a (1) 8.63c (1) 4.31b (1) Wald H0: a3-a4=0 1.09 (1) 3.66a (1) test Legend. a Significance level greater than 90%; b Significance level greater than 95%; c Significance level greater than 99%. Standard errors and number of restrictions in parentheses. Sample selection models Ib and IIb are not identified (i.e both σYY and σSY are not statistically different from zero at 95%). Parameters of the selectivity portion of the sample selection model are not reported (see footnote 9). In Model Ia and IIa, the number of salaried employees has been augmented by one in order to permit the use of a logarithmic form.

30

Table A2 - The determinants of start-up size: the effect of founders’ generic and specific human capital with industry dummies (OLS and sample selection models) N. of salaried employees (log)

a0

Constant

Sum of the n. of founders and salaried employees (log)

Ia

Ib

IIa

IIIa

OLS

Sample selection model

OLS

OLS

Sample selection model

OLS

-1.9339 (3.7548)

0.7459 (0.3550)b

0.2756 (0.0684)

c

0.1791 (0.0396)

0.8174 (0.6271)

0.2038 (0.0579)c

0.0612 (0.4907)

-0.0763 (0.4194)

c

c

c

0.2852 (0.0579)

IVa

c

0.1165 (0.3198) 0.2204 (0.0460)

IIIb

0.9484 (0.2757)

a1

Ecoeducation

a2

Techeducation

0.0084 (0.0240)

0.0311 (0.0322)

0.0114 (0.0291)

-0.0011 (0.0207)

0.2680 (0.3067)

0.0044 (0.0247)

a3

Specworkexp

0.0376 (0.0069)c

0.0624 (0.0095)c

0.0307 (0.0082)c

0.0247 (0.0059)c

0.1089 (0.0846)

0.0174 (0.0069)b

a4

Genworkexp

0.0168 (0.0054)c

0.0119 (0.0076)

0.0172 (0.0071)b

c

c

0.0046 (0.0047)

-0.0107 (0.0486)

0.0023 (0.0060)

a5

DManager

0.4659 (0.1528)

0.7130 (0.1862)

0.3653 (0.1859)a

0.5409 (0.1317)c

1.8765 (1.4729)

0.3642 (0.1574)b

a6

DEntrepreneur

-

-

0.2439 (0.1102)b

-

-

0.2145 (0.0932)b

a7

DMother company

c

0.7925 (0.1371)

b

0.1790 (0.1230)

c

0.5907 (0.1182)

0.5612 (0.1135)

0.5185 (0.1508)c

b

0.6804 (0.1782)

c

c

a8

DIncubated

-0.2734 (0.1340)

-0.1311 (0.1783)

-0.3155 (0.1513)

-0.0712 (0.1155)

-0.0094 (0.1154)

-0.0802 (0.1281)

a9

Rreal

-0.0246 (0.0226)

-0.0094 (0.0380)

-0.0249 (0.0288)

-0.0329 (0.0195)a

-0.1715 (0.2084)

-0.0264 (0.0244)

a10

Infrastructure

0.0047 (0.0015)c

0.0011 (0.0020)

0.0048 (0.0019)b

0.0024 (0.0013)a

0.0018 (0.0014)

0.0027 (0.0016)

a11

DBiopha

0.3294 (0.3143)

0.1507 (0.4242)

0.4358 (0.3725)

0.4266 (0.2710)

0.3262 (0.2432)

0.6419 (0.3153)b

a12

DSemic

1.2274 (0.3568)c

1.0274 (0.4913)b

1.1986 (0.4475)c

1.1504 (0.3076)c

0.9106 (0.2717)c

1.0314 (0.3788)c

a

a13

DTLCequip

0.3420 (0.3143)

0.0512 (0.4427)

0.6380 (0.3766)

0.2886 (0.2709)

0.2091 (0.2799)

0.4962 (0.3188)

a14

DInstrument

-0.2369 (0.2765)

-0.1340 (0.4263)

0.0010 (0.3356)

-0.0996 (0.2384)

-0.1351 (0.2237)

0.0528 (0.2841)

a15

DAerospace

0.3230 (0.4452)

0.5991 (0.5427)

0.5071 (0.6808)

0.4787 (0.3838)

0.5552 (0.4410)

0.9074 (0.5763)

a16

DSoftware

-0.1352 (0.2515)

-0.0738 (0.3932)

-0.0588 (0.3011)

0.0339 (0.2168)

0.0203 (0.2097)

0.1066 (0.2549)

a27

DElpub

-0.1304 (0.2965)

-0.0264 (0.4546)

-0.1669 (0.3482)

0.0563 (0.2556)

0.0277 (0.2431)

0.0839 (0.2947)

a28

DInternet Number of observations

-0.1722 (0.2472)

-0.1469 (0.3947)

-0.1178 (0.3012)

-0.0325 (0.2131)

-0.1053 (0.2090)

0.0971 (0.2550)

391

391

260

391

391

260

σ YY σ SY

-

0.9805 (0.0577)

-

-

1.2096 (0.4735)

-

0.9989 (0.0001)c

-

-

0.8866 (0.0970)c

-

R2

0.3232

-

0.3305

0.2702

-

0.2629

c

c

-

Log-likelihood

-

-429.018

-

-

-400.7237

-

F test

H0: industry slopes = 0

4.66c (8)

-

3.08c (8)

4.36c (1)

-

2.68c (8)

LR test

H0: industry slopes = 0

-

20.10c (8)

-

-

36.92c (8)

-

c

a

c

F test H0: a3-a4=0 9.89 (1) 2.96 (1) 12.38 (1) 5.15b (1) Wald H0: a3-a4=0 34.76c (1) 1.44 (1) test a b c Legend: Significance level greater than 90%; Significance level greater than 95%; Significance level greater than 99%. Standard errors and number of restrictions in parentheses. Sample selection models IIb and IVb are not identified (i.e both σYY and σSY are not statistically different from zero at 95%). Parameters of the selectivity portion of the sample selection model are not reported (see footnote 9). In Model Ia, Ib and IIa, the number of salaried employees has been augmented by one in order to permit the use of a logarithmic form.

31