Skip to main content

Determinants of novice, portfolio, and serial entrepreneurship: an occupational choice approach


In this paper, we first develop an original theory in which, based on their individual skills and the quality of their business, entrepreneurs can keep their original business (and thus remain novice entrepreneurs), start and keep a new business in the same or another sector along their current business (therefore becoming portfolio entrepreneurs), transfer or shut their original business down to either start a new one (turning themselves into serial entrepreneurs), or enter the labor market as wage workers. We then use the insights from our theory to develop three main hypotheses that are finally tested for a 10-year panel dataset (2001 to 2010) of more than 4000 Vietnamese manufacturing firms. We estimate an occupational choice model and a survival model and find that (i) a greater endowment of human capital is associated with a higher likelihood of a business owner to become a serial or a portfolio entrepreneur; (ii) a higher quality of the new business is associated to a higher likelihood that it is run by any type of habitual entrepreneur. Particularly, high entrepreneurial skills together with a high-quality business positively influence the likelihood of an individual to be serial or portfolio entrepreneur; (iii) ceteris paribus, firms run by serial or portfolio entrepreneurs tend to stay in business longer, although high-quality ones run by novice entrepreneurs endowed with high entrepreneurial skills are those with the lowest probability to leave the market.

This is a preview of subscription content, access via your institution.

Fig. 1
Fig. 2
Fig. 3


  1. For these three types of entrepreneurship, we follow the definitions earlier adopted, among others, by Ucbasaran et al. (2006, 2008).

  2. Scholars have mostly explored either the choice between entering entrepreneurship for the first time and engaging in paid employment (Evans and Leighton 1989; Parker 2009; Santarelli et al. 2009) or habitual entrepreneurship in general without distinguishing between serial and portfolio (Lafontaine and Shaw 2016).

  3. Among those finding a superior performance of habitual entrepreneurship, there are Blanchflower and Oswald (1998), Wright et al. (1998), Åstebro and Bernhardt (2003), Chen (2013), and Rocha et al. (2015). Among those providing opposite evidence, we find Rerup (2005) and Gottschalk et al. (2017).

  4. With the exception of Plehn-Dujowich (2010), who, however, does not consider portfolio entrepreneurship.

  5. In particular, Parker (2014) analyzes how an entrepreneur’s recognition and exploitation abilities impact on the variance and the mean of the payoffs generated by their entrepreneurial activity. He finds that (i) higher opportunity-exploitation ability is associated with portfolio entrepreneurship rather than with serial and novice entrepreneurship (see also Parker and van Praag 2010); (ii) a sequence of opportunities whose returns covary negatively (diversifying opportunities) promotes portfolio entrepreneurship at the expense of both novice and serial entrepreneurship; (iii) synergies between successive opportunities promote portfolio entrepreneurship over novice entrepreneurship, unless the initial opportunity is sufficiently valuable; and (iv) individuals with low (moderate) (high) risk aversion are more likely to be serial (portfolio) (novice) entrepreneurs, respectively (for experimental lines of evidence, see Koudstaal et al. 2016).

  6. These are: “Accessing, multiplying, redeploying, incubating, decoupling, aligning, complementing, and coupling” (Baert et al. 2016, p. 354).

  7. Cases with s > 1 have been left out because they imply increasing returns to capital. While it may be interesting to study economies in which tensions toward market concentration exist, entrepreneurship in Vietnam is mostly characterized by SMEs. Hence, limiting our theoretical analysis to s ≤ 1 seems more consistent with the data we use to validate our theory.

  8. We assume that, once made, the entrepreneur’s occupational choice is not reversible. This assumption is common in occupational choice models (see Plehn-Dujowich 2010) and is also suitable for our dataset, which registers whether, at a given date, an entrepreneur is novice (still owns her/his original business), serial, or portfolio but does not give account for her/his future decisions.

  9. The parameter τ can be interpreted as the “start-up cost” (see Plehn-Dujowich 2010; Holmes and Schmitz 1990).

  10. In fact, \( {V}^S\left(q,s,\widehat{q}\right)={qK}^s+\beta \underset{0}{\overset{\infty }{\int }}\frac{q{\left[\left(1-\tau \right)K\right]}^s}{1-\beta}\mathrm{d}Q(q) \), where the last integral can be rewritten as \( \frac{{\left[\left(1-\tau \right)K\right]}^s}{1-\beta}\underset{0}{\overset{\infty }{\int }}q\mathrm{d}Q(q)=\frac{{\left[\left(1-\tau \right)K\right]}^s}{1-\beta } Eq \).

  11. See Ljungqvist and Sargent (2004, chapter 6).

  12. Proofs are in Appendix 1.

  13. As clarified above, we do not consider cases in which s > 1. Results would be qualitatively similar to those in Proposition 2. In fact, with s > 1, the value \( {V}^P\left(q,s,\widehat{q},\gamma \right) \) in expression (4) would be convex in γ. \( {V}^P\left(q,s,\widehat{q},\gamma \right) \) would then be maximized at either γ = 0 or γ = 1, according to the relative values of \( q,\widehat{q} \), and τ, exactly as in Proposition 2.

  14. This is in line with the data we use in our empirical analysis, according to which entrepreneurial exit encompasses both entrepreneurs closing their business and those transferring it. Transfer and closing down can be justified by either leaving entrepreneurship or starting a new firm (serial entrepreneurship).

  15. The function \( {V}^P\left(q,s,\widehat{q},\gamma \right) \) is increasing in \( \widehat{q} \) and linear and is therefore the upward sloping line in Fig. 1. In fact, applying the envelope theorem, \( \frac{\partial {V}^P}{\partial \widehat{q}}=\frac{\beta }{1-\beta }{\left[\left(1-\gamma \right)\left(1-\tau \right)K\right]}^s>0 \). The function Vw(q, s, w) is invariant with respect to \( \widehat{q} \). In Fig. 1, we have neglected the case in which Vw(q, s, w) (hence w) is so low that there is no \( {\widehat{q}}^P>0 \) such that the entrepreneur is willing to exit if \( \widehat{q}<{\widehat{q}}^P \) (this case would occur if Vw(q, s, w) lies below the vertical intercept of \( {V}^P\left(q,s,\widehat{q},\gamma \right) \) at \( \widehat{q}=0 \)).

  16. Private ownership was experimentally permitted to operate in 1987–1988 in small-scale industries. By the promulgation of the Law on Foreign Investment in Vietnam 1987, the Company Law in 1990, and the Law on Private Enterprises in 1991, there has been a sharp increase in the number of private enterprises.

  17. The surveys were designed in the way that all firms are surely traced over time. Firms exit the surveys for a definite reported reason. This keeps the number of enterprises being lost during the sampling to the minimal. Indeed, given an average annual exit rate of around 10% across the five surveys, only about 20% of these exits are justified with the sentence “no specific reason,” which means that the enterprise could not be found, or the owner declined to answer the questionnaire.

  18. Appendix 3 contains a formal description of the three models employed and of their differences.

  19. Given that our duration data are right censored, we cannot analyze them by means of conventional methods such as a linear regression. Survival times tend to have a positively skewed distribution, which violates the normal distribution assumption of the conventional linear regression.

  20. These multiple enterprises could be either in the same or different industries. Thus, portfolio entrepreneurs are those running at least two different businesses at the same time, whatever their sector.

  21. Given the structure of our dataset, we can observe only the quality of the current business which is the new business for serial entrepreneurs, the main business for portfolio entrepreneurs, and the old business for novice entrepreneurs in case they decide to exit entrepreneurship during the observation years.

  22. While Opler and Titman (1994) find that highly leveraged firms lose a substantial market value and impose greater risks to owners and creditors than their more conservatively financed competitors. Teece (1982) finds that debts reduce the chances of bankruptcy through flexible asset deployment.

  23. Since skill affects productivity and productivity is an aspect of the firm (technological) quality, there might be a positive correlation between skill and firm quality.

  24. Duration, in terms of number of years, of the period that an individual stays in the current business or in entrepreneurship.

  25. Ronnas and Ramamurthy (2001: 328) describe a typical urban entrepreneur in Vietnam as “a middle-aged male with at least 10 years of education and prior experience in similar fields in a position of responsibility”.

  26. The two-tailed t-test for the comparison of mean ‘education’ between novice and habitual entrepreneurs significantly rejects the equality and supports the superiority of habitual entrepreneurs. Results are reported in Appendix 2.

  27. chi2(2) = 165.14; Pr > chi2 = 0.0000

  28. We have also performed two robustness checks: one for the occupational choice equation and one for the survival equation. In the first one, we have tried to separate the effects of different skills: education and managerial experience. We have distinguished serial entrepreneurs between those who launched a new business after closing their previous one and those who acquired an existing one. Literature (Parker and Van Praag 2010) claimed that the first ones require education, whereas the second require managerial experience. We have then excluded those who acquired an existing enterprise and rerun the multinomial logit regression. In the second robustness check, we sort entrepreneurs according to their exit modality: some went bankrupt and some others sold their enterprise. While bankruptcy is typically attributed to skill and financial resources (or lack thereof), ownership transfer may not. We exclude those who sold their firm from the sample and rerun the survival analysis. The results of the two robustness checks (available upon request) are consistent with our main results.


Download references


Previous versions of this paper have been presented at the 8th Annual Conference for the Academy of Innovation and Entrepreneurship (Toronto, Ryerson University) and at seminars held at University of Saarland, University of Luxembourg, University of Trento, and Bucerius Law School (Hamburg). We thank seminar participants and, in particular, Bettina Müller for helpful suggestions.

Author information

Authors and Affiliations


Corresponding author

Correspondence to Enrico Santarelli.

Additional information

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.


Appendix 1. Proofs

Proof of proposition 1

Differentiating expression (4) with respect to γ, the first-order condition for the maximization of \( {V}^P\left(q,s,\widehat{q},\gamma \right) \) is

$$ q{K}^s{\gamma}^{\ast \left(s-1\right)}-\widehat{q}{\left[\left(1-\tau \right)K\right]}^s{\left(1-{\gamma}^{\ast}\right)}^{s-1}=0 $$

which can be rewritten as

$$ \frac{\gamma^{\left(s-1\right)}}{{\left(1-\gamma \right)}^{s-1}}=\frac{\widehat{q}}{q}{\left(1-\tau \right)}^s $$

Solving Eq. (11) for γ always yields an interior solution γ ∈ (0, 1). In fact, when s < 1, \( {R}_1\left(\gamma \right)=\frac{\gamma^{\left(s-1\right)}}{{\left(1-\gamma \right)}^{s-1}} \) is always decreasing in γ, with \( \underset{\gamma \to 0}{\lim }{R}_1\left(\gamma \right)=+\infty \), and R1(1) = 0. Since the right-hand side of (11) is constant with respect to γ and positive, there will always exist a value 0 < γ < 1 such that \( \frac{\gamma^{\left(s-1\right)}}{{\left(1-\gamma \right)}^{s-1}}=\frac{\widehat{q}}{q}{\left(1-\tau \right)}^s \).

The second-order condition for a maximum is

$$ \left(s-1\right)q{K}^s{\gamma}^{s-2}+\left(s-1\right)\widehat{q}\left(1-\tau \right){\left(1-\gamma \right)}^{s-2}<0 $$

Being (s − 1) < 0, the second-order condition is always satisfied and γ is a maximum.

Finally, totally differentiating the first-order condition in (7) with respect to γ, \( \widehat{q}, \) and τ yields

$$ \frac{{\mathrm{d}\gamma}^{\ast }}{\mathrm{d}\widehat{q}}=\frac{{\left(1-\tau \right)}^s{\left(1-{\gamma}^{\ast}\right)}^{s-1}}{\left[{q\gamma}^{\ast \left(s-2\right)}+\widehat{q}{\left(1-\tau \right)}^s{\left(1-{\gamma}^{\ast}\right)}^{s-2}\right]\left(s-1\right)}<0 $$

since (s − 1) < 0. Similarly

$$ \frac{{\mathrm{d}\gamma}^{\ast }}{\mathrm{d}q}=-\frac{{\gamma^{\ast}}^{s-1}}{\left[{q\gamma}^{\ast \left(s-2\right)}+\widehat{q}{\left(1-\tau \right)}^s{\left(1-{\gamma}^{\ast}\right)}^{s-2}\right]\left(s-1\right)}>0 $$


$$ \frac{{\mathrm{d}\gamma}^{\ast }}{\mathrm{d}\tau }=\frac{\widehat{q}{\left(1-\tau \right)}^{s-1}{{\left(1-\gamma \right)}^{\ast}}^{s-1}}{\left[{q\gamma}^{\ast \left(s-2\right)}+\widehat{q}{\left(1-\tau \right)}^s{\left(1-{\gamma}^{\ast}\right)}^{s-2}\right]\left(s-1\right)}<0 $$

Thus, γ is increasing in q and decreasing in \( \widehat{q} \) and τ when s < 1.


$$ \frac{{\mathrm{d}\gamma}^{\ast }}{\mathrm{d}s}=-\frac{{q\gamma}^{\ast \left(s-1\right)}\ln \left(\gamma \right)-\widehat{q}{\left(1-\tau \right)}^s{\left(1-{\gamma}^{\ast}\right)}^{s-1}\Big[\ln \left(1-\gamma \right)+\ln \left[1-\tau \Big)\right]}{\left[{q\gamma}^{\ast \left(s-2\right)}+\widehat{q}{\left(1-\tau \right)}^s{\left(1-{\gamma}^{\ast}\right)}^{s-2}\right]\left(s-1\right)} $$

being γ and τ ∈(0, 1), then all the natural logarithms in expression (16) are negative numbers. Therefore, the numerator is positive if \( \widehat{q} \) is large relative to q and τ is small. Given the minus sign in front of the r.h.s., a positive numerator implies that γ is increasing in s (since s < 1 and the denominator is negative). Vice versa, the numerator is negative if \( \widehat{q} \) is small relative to q and τ is large. This implies that, in this case, γ is decreasing in s.■

Proof of Proposition 2

Given s = 1, the value \( {V}^P\left(q,s,\widehat{q},\gamma \right) \) in expression (4) is linear in γ and can be written as

$$ {V}^P\left(q,1,\widehat{q},\gamma \right)=\frac{qK}{1-\beta }-\frac{\beta \left(1-\gamma \right)K}{1-\beta}\left[q-\left(1-\tau \right)\widehat{q}\right] $$

which is increasing in γ (and thus maximized at γ = 1) if \( q\ge \widehat{q}\left(1-\tau \right) \), whereas it is decreasing in γ (and thus maximized at γ = 0) if \( q<\widehat{q}\left(1-\tau \right) \).■

Appendix 2. t test on the equality of means of age, firm age, and education among novice, serial, and portfolio; and tabulation of ownership types adopted by novice, serial, and portfolio

Analysis of the statistical differences in age, firm age, and education among novice, serial, and portfolio entrepreneurs

Table 4 Age—novice and habitual entrepreneurs
Table 5 Age—serial and portfolio entrepreneurs
Table 6 Firm age—novice and habitual entrepreneurs
Table 7 Firm age—serial and portfolio entrepreneurs
Table 8 Education—novice and habitual entrepreneurs
Table 9 Education—serial and portfolio entrepreneurs
Table 10 Tabulation of legal ownership

Appendix 3. The survival equation. A formal analysis

In this paper, we use three different estimation models: the nonparametric Kaplan–Meier estimator, the semiparametric Cox proportional hazards regression, and the parametric Weibull model.

The Kaplan–Meier estimator is a nonparametric estimator of the survival function S(t). If all the failure times are ordered and labeled t(j) such that t(1) ≤ t(2) …  ≤ t(n), the estimator is given by \( \widehat{S}(t)=\prod \limits_{j\mid {t}_{(j)}\le t}\left(1-\frac{d_j}{n_j}\right) \), where dj is the number of entrepreneurs who exit at time t(j), and njis the number of entrepreneurs who are still in the business at the time and are therefore still “at risk” of experiencing exit.

The Cox hazard function for entrepreneur i is hi(t) = h0(t) exp(si qi Ki, θ), where h0(t) is the baseline hazard function when all covariates are zero. The parameters θ are estimated by maximizing the partial log likelihood given by \( \sum \limits_f\log \left(\frac{\exp \left({s}_i\ {q}_i\ {K}_i,\theta \right)}{\sum_{i\in r(f)}\exp \left({s}_i\ {q}_i\ {K}_i,\theta \right)}\right) \), where the first summation is over all failures exit f, and the second summation is over all entrepreneurs r(f) who are still at risk at the time of failure.

The Weibull model assumes the Weibull distribution for T with parameters λ and p, denoted T~W(λ, p), if Tp~E(λ). The cumulative hazard is H(t) = (λt)p, the survivor function S(t) = exp(−(λt)p), and the hazard is λ(t) = λpptp − 1. Both semiparametric and parametric survival models are estimated by maximum likelihood estimation technique.

Appendix 4

Table 11 Summary statistics and matrix of correlation

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Carbonara, E., Tran, H.T. & Santarelli, E. Determinants of novice, portfolio, and serial entrepreneurship: an occupational choice approach. Small Bus Econ 55, 123–151 (2020).

Download citation

  • Accepted:

  • Published:

  • Issue Date:

  • DOI:


  • Portfolio entrepreneurship
  • Serial entrepreneurship
  • Occupational choice
  • Industrial policy

JEL codes

  • F02
  • L26
  • L53