A Dinâmica do Emprego na Região do Norte

cef.up working paper 2013-12 SERIAL ENTREPRENEURSHIP, LEARNING BY DOING AND SELF-SELECTION Vera Rocha Anabela Carneiro Celeste Amorim Varum Serial entrepreneurship, learning by doing and self-selection Vera Rocha Universidade do Porto, CEF.UPyand CIPES Anabela Carneiroz Universidade do Porto and CEF.UP Celeste Amorim Varum Universidade de Aveiro, DEGEI and GOVCOPP July 11, 2013 Abstract It remains a question whether serial entrepreneurs typically perform better than their novice counterparts owing to learning by doing e¤ects or mostly because they are a selected sample of higher-than-average ability entrepreneurs. This paper tries to unravel these two e¤ects by exploring a novel empirical strategy based on continuous time duration models with selection. We use a large longitudinal matched employer-employee dataset that allows us to track almost 220,000 individuals who have left their …rst entrepreneurial experience. Over 35,000 serial entrepreneurs are identi…ed and followed in their second business, in order to evaluate how entrepreneurial experience acquired in the previous business improves persistence by reducing their exit rates. Our results show that serial entrepreneurs are not a random selection of ex-business-owners. The positive association found between prior experience and serial entrepreneurs’survival is mainly due to selection on ability, rather than the result of learning by doing. Keywords: Serial Entrepreneurship, Entrepreneurial Experience, Learning, Selection JEL Codes: D83, J24, L26 We acknowledge GEP-MSSS (Gabinete de Estratégia e Planeamento – Ministério da Solidariedade e Segurança Social) for allowing the use of Quadros de Pessoal dataset. The …rst author also acknowledges Fundação para a Ciência e Tecnologia (FCT) for …nancial support through the doctoral grant SFRH/BD/71556/2010. yCEF.UP –Centre for Economics and Finance at the University of Porto is funded by Programa Compete – POFC (Fundação para a Ciência e Tecnologia and European Regional Development Fund); project reference: PEst-C/EGE/UI4105/2011. zCorresponding author: Anabela Carneiro, Faculdade de Economia, Universidade do Porto, Rua Dr. Roberto Frias, 4200-464 Porto, Portugal. E-mail: [email protected] 1 Introduction Most of the seminal theories of industry evolution that incorporate entrepreneurship as a means of market entry assume that exit is a …nal event: once it has occurred, reentry (into the same or a di¤erent industry) is not an option (e.g., Jovanovic, 1982; Hopenhayn, 1992; Jovanovic and MacDonald, 1994). Nevertheless, not only there is a growing awareness that entrepreneurship is not solely con…ned to the creation of a new business as a singleaction event (Ucbasaran et al., 2006; Plehn-Dujowich, 2010; Sarasvathy et al., 2013), as also empirical evidence con…rms that a signi…cant part of entrepreneurial activity around the world is conducted by serial (or renascent) entrepreneurs.1;2 As a result, serial entrepreneurs have been gaining an increasing attention of scholars and, especially, policymakers, who have enlarged entrepreneurial incentives so as to target both experienced and novice entrepreneurs (see Westhead et al., 2003, 2005a, 2005b). Moved by the widespread beliefs that serial entrepreneurs will perform better owing to the learning e¤ects from past entrepreneurial experiences (Cope and Watts, 2000; Minniti and Bygrave, 2001; Cope, 2005, 2011), many European countries in particular have launched new programs to promote a fresh restart of entrepreneurs who performed poorly in the past and who would, otherwise, feel prevented to try again due to the so-called “stigma of failure” (European Commission, 2002, 2011). However, and despite the early recognized value of studying serial entrepreneurship (e.g., MacMillan, 1986), empirical research on this topic is still at the beginning stage due to the lack of suitable data (Zhang, 2011; Parker, 2012; Sarasvathy et al., 2013), as most data collection e¤orts cover only one of a series of businesses, or track …rms rather than entrepreneurs. Consequently, most of the existing research on serial entrepreneurship is by and large descriptive, based on case studies (e.g., Ucbasaran et al., 2003) or using small samples of individuals to establish comparisons between serial and nascent entrepreneurs (e.g., Westhead et al., 2005a, 2005b; Li et al., 2009; Ucbasaran et al., 2010; Robson et al., 2012; Kirschenhofer and Lechner, 2012). While a more recent stream of work has been particularly more concerned with the determinants of reentering into entrepreneurship (see Hyytinen and Ilmakunnas, 2007; Stam et al., 2008; Amaral et al., 2011; Hessels et al., 2011), substantial knowledge about how – or whether –entrepreneurs e¤ectively learn with past experience is still lacking. On the one 1Serial entrepreneurs have been broadly de…ned as individuals who have sold or closed a business in which they had a minority or majority ownership stake in the past, and who currently own (alone or with others) a di¤erent independent business that is either new, purchased or inherited (Westhead et al., 2005a). 2The increasing relative importance of serial entrepreneurs around the world has been documented by several studies: see, for instance, Westhead and Wright (1998) and Westhead et al. (2005a) for UK, Wagner (2003) for Germany, Hyytinen and Ilmakunnas (2007) for Finland and Headd (2003) for USA. 1 hand, latest empirical results have o¤ered little con…dence on signi…cant learning by doing e¤ects (e.g., Gompers et al., 2010; Nielsen and Sarasvathy, 2011; Parker, 2012; Frankish et al., 2012). On the other hand, an emerging concern about self-selection among serial entrepreneurs is blurring the broad expectations about true entrepreneurial learning (Chen, 2013). Accordingly, it remains a question whether serial entrepreneurs typically perform better than their novice counterparts because they have learned about running a business and have improved their entrepreneurial skills with their past experience or, instead, mostly because they are a selected sample of high-ability entrepreneurs. Accordingly, this study contributes to the current debate and tries to unravel these two e¤ects by exploring a novel empirical strategy. Using the methodology developed by Boehmke et al. (2006), we estimate continuous time duration models that account for selection bias to study how previous entrepreneurial experience in‡uences the survival of serial entrepreneurs in their second business. The analysis is based on a large longitudinal matched employer-employee dataset for Portugal, where serial entrepreneurs constitute an important component of the entrepreneurial activity in the country, by accounting for more than 20% of all new businesses created over the most recent years. About 220,000 exbusiness-owners are identi…ed, out of which 35,202 have tried again, by becoming serial entrepreneurs. Our results con…rm that serial entrepreneurs are not a random sample, which signi…cantly moderates any learning by doing e¤ects that might be transferred between subsequent entrepreneurial experiences. To the best of our knowledge, this is the …rst comprehensive study employing duration models with sample selection and using such a unique and rich dataset that attempts to evaluate entrepreneurial learning and self-selection issues among serial entrepreneurs. The remaining sections of the paper are organized as follows. Section 2 brie‡y presents prior literature on serial entrepreneurship and entrepreneurial learning and exposes the research design of the paper. Section 3 describes the data and the methodological procedures. The empirical results are presented and discussed in section 4. Section 5 concludes. 2 Previous Literature and Research Design The belief that entrepreneurs learn from experience is not recent (see Von Hayek, 1937), and this idea has been expressed in several highly in‡uential models on dynamic industrial organization over the last thirty years (e.g., Jovanovic, 1982; Frank, 1988; Hopenhayn, 1992; Ericsson and Pakes, 1995). As Minniti and Bygrave (2001: 7) explicitly state, “entrepreneurship is a process of learning, and a theory of entrepreneurship requires a theory of learning”. Consequently, several theoretical and conceptual attempts were developed dur2 ing the last decade, trying to identify the mechanisms through which entrepreneurs learn (e.g., Cope, 2005; Parker, 2006; Petkova, 2009; Plehn-Dujowich, 2010; Ucbasaran et al., 2012). In contrast, learning and its implications were largely ignored for long time (Fraser and Greene, 2006; Parker, 2007), mainly in empirical work due to data limitations (Zhang, 2011; Sarasvathy et al., 2013). Early studies, largely relying on small samples of entrepreneurs, self-reported data and simple statistics, tried to show that experienced entrepreneurs have all the conditions to perform better than their novice counterparts, by being more able to access …nancial resources, more alert to business opportunities and better endowed with a larger set of skills –particularly with those acquired through entrepreneurial experience –which make them o¤er more attractive growth prospects than …rst time entrepreneurs (Wright et al., 1997; Westhead and Wright, 1998; Westhead et al., 2003, 2005a, 2005b). Additional comparisons between experienced and nascent entrepreneurs performed by more recent studies have also shown that the former are better at developing networks (Li et al., 2009), more likely to quickly gain access to venture capital (Zhang, 2011), more prone to take risks and pursue innovative activities (Robson et al., 2012), and also more capable of building more e¤ective and diversi…ed teams (Kirschenhofer and Lechner, 2012). While these studies have signi…cantly contributed to our knowledge about how di¤erent are serial and novice entrepreneurs, they have o¤ered no systematic empirical test of learning e¤ects resulting from entrepreneurs’experience. Some of the most recent studies have been trying to address the issue of entrepreneurial learning, either by comparing the outcomes of entrepreneurs with and without experience (Frankish et al., 2012), or by evaluating how serial entrepreneurs’performance in one venture is related to their performance in a subsequent venture (Gompers et al., 2010; Nielsen and Sarasvathy, 2011; Parker, 2012; Chen, 2013). Nevertheless, none of their results has undoubtedly supported that serial entrepreneurs perform better owing to their accumulated experience. Both Frankish et al. (2012) and Nielsen and Sarasvathy (2011) have analyzed the 3-year survival rate of new businesses for UK and Denmark, respectively. While the former found no signi…cant survival di¤erences between those who have previously owned a business and those without such experience, the latter concluded that some form of absorptive capacity (in terms of education and prior industry background) is necessary for entrepreneurs to bene…t from any learning opportunities. For US, Gompers et al. (2010) showed that entrepreneurs with a track record of success are much more likely to succeed –by being more able to obtain venture capital funding at an earlier stage of their company’s development –than nascent entrepreneurs and those who have previously failed. However, they also suggest that such result was, in part, driven by some type of skills of serial entrepreneurs –in particular, their “market timing ability”(p. 3 19) –which allow them to start a company at the right time in the right industry. Parker’s (2012) results, in turn, showed that despite good performance (in terms of …rm pro…ts) in one venture appears to be associated with good performance in subsequent ventures, the e¤ects are temporary and depreciate over time. Finally, also for US, Chen (2013) took a step forward in this issue and tried to disentangle learning by doing from selection e¤ects by combining …xed-e¤ects panel data models and IV estimation for a sample of about 3,200 serial entrepreneurs. She found that self-selection on ability –rather than learning by doing –is the key determinant of the early performance of new businesses (measured by their …rst-year income), except when such businesses are established in an industry closely related to the individual’s past entrepreneurial experience. Accordingly, not only learning by doing became a questionable hypothesis, as also only more contemporary studies have tried to empirically address the possible coexistence of learning by doing and learning about own ability (e.g., Chen, 2013). Even so, selection on ability as a source of serial entrepreneurship had been early proposed in theory (e.g., Holmes and Schmitz, 1990). Actually, theoretical grounds suggest that unobserved talent shapes entrepreneurial entry (Inci, 2013) and that, moreover, individuals learn and update beliefs about their entrepreneurial ability as they accumulate experience in entrepreneurship (Jovanovic, 1982; Parker, 2006). This is expected to generate a dynamic process where high-skill entrepreneurs tend to shut down businesses of low quality to become serial entrepreneurs, launching and subsequently closing …rms until a high quality business is found, while low-skill entrepreneurs are expected to shut down their businesses of low quality to enter the labor market (Plehn-Dujowich, 2010). In view of that, this paper contributes to this emerging literature by evaluating to what extent serial entrepreneurs learn with past experience, after controlling for potential biases driven by self-selection into serial entrepreneurship. We characterize entrepreneurial experience using three variables: i) the cumulative years the individual has survived as BOs in the …rst entrepreneurial experience; ii) previous experience in creating a start-up venture; and iii) industry-speci…c experience. Learning by doing e¤ects are evaluated by studying whether these three particular measures improve the persistence of serial entrepreneurs in their second business, by reducing their exit rates. According to the literature, we would anticipate positive (negative) and signi…cant e¤ects from each of these variables on serial entrepreneurs’survival (hazards). First, by running a business, entrepreneurs acquire unique speci…c resources, knowledge, skills and capabilities that can be used to start and/or acquire subsequent businesses. Hence, the longer an individual has been in a business in the past –successful or not –the more s/he is likely to have learned about being an entrepreneur, and the larger the stock of knowledge that may be accumulated about customers and suppliers, networks of contacts, as well as market4 speci…c information (Cope and Watts, 2000; Ucbasaran et al., 2006; Frankish et al., 2012; Parker, 2012), which may constitute important resources when they decide to try again as entrepreneurs. Second, the particular experience of founding a …rm – given that not all individuals become entrepreneurs by creating a start-up venture (see, Parker and Van Praag, 2011; Bastié et al., 2013; Rocha et al., 2013) – may also deliver greater opportunities to learn about the overall entrepreneurial process. Starting a …rm from scratch requires a wide range of skills, and prior …rm-founding experience is believed to help an entrepreneur to acquire and enhance such skills (Zhang, 2011). In addition, learning experiences are expected to be mostly relevant during business’infancy, i.e., during the …rst few years of the …rm (Van Gelderen et al., 2005). Finally, learning by doing may also arise from industry-speci…c experience (Frankish et al., 2012; Chen, 2013). Individuals becoming serial entrepreneurs by establishing a business in the same industry where they operated in the past may bene…t from informational advantages and su¤er from lower uncertainty, which may also contribute to their resilience in their second entrepreneurial attempt. 3 Data and Methodological Issues 3.1 Data This study uses data from Quadros de Pessoal (QP), a longitudinal matched employeremployee administrative dataset from the Portuguese Ministry of Employment. QP is an annual mandatory employment survey that all …rms in the private sector employing at least one wage earner are legally obliged to …ll in.3Requested data cover …rms/establishments (e.g., location, employment, industry, sales, ownership, among others) and each of its workers (for instance, gender, age, education, occupational category and skill levels), thus providing very rich information on individual’s backgrounds, career paths and transitions between …rms and industries. Firms/establishments and individuals (both workers and businessowners) are identi…ed by a unique identi…cation number, so that they can be tracked and matched over time. All these characteristics of the dataset make QP a suitable database for a study on serial entrepreneurship. Raw QP …les are available for the period 1986-2009, though there is a gap for the particular years of 1990 and 2001, for which there is no available information at the individual-level. 3For this reason, self-employed individuals without employees are not covered by QP. 5 For this reason, we restrict our study to serial entrepreneurs reentering into entrepreneurship between 1993 and 2007.4Data for the period 1986-1992 was only used to characterize individuals’previous experiences as entrepreneurs. The entrepreneur de…nition used in this study corresponds to Business-Owners (BOs) of …rms with at least one wage earner (i.e., employers). We identify serial entrepreneurs in particular as those ex-BOs who become BOs again, in a di¤erent …rm, after leaving their …rst entrepreneurial experience. 3.2 Identifying the entry and exit of serial entrepreneurs We started by identifying in QP …les all BOs who left their …rst business ownership experience.5From these, portfolio BOs (i.e., those BOs who simultaneously owned two or more …rms) were excluded from the current analysis.6A total of 219,462 ex-BOs, aged between 16 and 50 years old, were identi…ed and tracked over time, in order to …nd out who has reentered and became BO in a second …rm, and who did not.7About 16% of them (precisely 35,202 ex-BOs) have tried a second chance by transiting into serial entrepreneurship during the period 1993-2007. For each of those 219,462 ex-BOs, we have retained a set of information related to the previous business-ownership experience, namely regarding the time (in years) the individual has survived as BO in the business, the entry mode in the …rst business (start-up versus acquisition) and the respective industry. Additional information regarding the size of the …rm at the moment of exit, the location of the …rm, the ownership structure of the previous business and the exit mode8adopted by the BO was also gathered, as those variables may 4To avoid measurement errors on the time spent until reentry into entrepreneurship –one of the variables to be included in our estimations – we have also excluded those reentries occurring in 2002, as we are not able to ensure whether the reentry occurs in 2001 (for which no data are available at the worker-level) or in 2002. Besides, we exclude reentries occurring after 2007 because, given the criteria adopted to identify business-owner’s exit (section 3.2), we need at least two years of available information after his/her reentry to clearly identify individual’s exit. 5A BO was considered to have left the previous business if s/he has de…nitely exited the BO status in the previous …rm. To consider that a de…nite exit has taken place, we have imposed an absence of the BO from the …rm larger or equal to two consecutive years. Accordingly, the identi…cation of ex-BOs’exits had to stop in 2007, as data for 2008 and 2009 were only used to check the presence/absence of each BO in the respective business. Even so, as this study covers reentries occurring between 1993 and 2007, we must restrict the analysis to ex-BOs who have left their prior business until 2006. 6Ex-BOs identi…ed as portfolio BOs in the past were excluded from our analysis because we are interested in particular characteristics of the …rst business ownership experience –which must be unique –when taking into account the potential non-randomness of the sample of serial entrepreneurs. Even so, portfolio BOs corresponded to less than 1% of all ex-BOs identi…ed, so their exclusion from the analysis has no signi…cant impact on the results. 7By imposing the upper limit of 50 years old by the time of the exit from the …rst business we are minimizing the reentries of serial entrepreneurs after attaining the retirement age. Even so, when performing the analysis with all individuals from all age cohorts, the results were not found to be signi…cantly di¤erent. 8Regarding the exit mode followed by the BO in the …rst entrepreneurial experience, QP dataset allows 6 play a role when explaining the decision of reentering into entrepreneurship (i.e., the selection process of serial entrepreneurs). We have then followed each of those 35,202 serial entrepreneurs over time, since the moment of their reentry until their last record in QP …les, which may either correspond to the moment of their exit from the second business, or to the last year of available information in the dataset –right-censored cases (Lancaster, 1990; Jenkins, 2005). Following the same procedures adopted to identify the exit of the BOs from the …rst business, we have required an absence of each BO from the …rm, or from the BO position, larger or equal to two consecutive years in order to identify serial BOs’exit year. 3.3 Empirical Strategy As the primary variable of interest is the time spent by serial BOs in their second business, hazard models were considered. The …nal dataset was thus constructed in a continuous survival time format so as to employ continuous time duration models that correct for selection bias. A spell starts when an ex-BO becomes a serial entrepreneur, by entering a new entrepreneurial experience. The duration of that spell corresponds to the time elapsed until the exit of the entrepreneur. Single-spell duration data were thus obtained by ‡ow sampling, so left censoring is not an issue in our analysis. We started by testing the suitability of semi-parametric and several parametric survival models (see, for instance, Lee and Wang, 2003; Jenkins, 2005; Cleves et al., 2010) accounting as well for individual-level unobserved heterogeneity, which may produce biased results when ignored (Hougaard, 1995; Jenkins, 2005). All the estimated models were evaluated and compared in terms of their Log-Likelihood, Akaike Information Criteria (AIC) and CoxSnell residuals. According to these preliminary tests, Weibull proportional hazard model was found to provide a very satisfactory …t to the data.9 us to distinguish those who have left by closing down the …rm from those who have exited by transferring the business to others. However, despite some studies associate an exit by dissolution to failure and an exit by ownership transfer to a more successful entrepreneurial experience (e.g., Stam et al., 2008; Amaral et al., 2011; Nielsen and Sarasvathy, 2011), we cannot ensure that this was actually the case, as QP does not provide …nancial data at the …rm-level. Actually, as Amaral et al. (2011: 7) also recognize, business failure may be understood as the failure to attain or exceed a performance threshold required by the entrepreneur to keep the business running (Gimeno et al., 1997; McCann and Folta, 2012), which does not necessarily indicate that the business is economically unviable. Consequently, some businesses may be transferred to other entrepreneurs with a lower performance threshold. Accordingly, we do not associate more (un)successful experiences to any of these cases in particular and we use the information on BO’s exit mode from the …rst business just as a control variable when studying individuals’reentry decisions. 9Conventional wisdom (e.g., Cleves et al., 2010) suggests that the best-…tting model is the one with the largest Log-Likelihood and the smallest AIC value. In our data, the model ful…lling these two requirements would be an Accelerated Failure Time (AFT) model with a log-logistic distribution for the hazard rate. However, as we are mainly interested in incorporating selection bias in the estimation of duration models, 7 Table 3. Estimation results from the Weibull proportional hazard model (cont.) (Portugal 1993-2007) Model 1 Model 2 Firm-level characteristics (cont.) Primary Sectorb-0.1681** -0.1675** (0.0845) (0.0845) Energy & Constructionb0.0782** 0.0824** (0.0385) (0.0385) Servicesb0.0145 0.0181 (0.0317) (0.0317) Reenter in a year of crisis 0.0339 0.0382 (0.0366) (0.0366) Constant 0.1298 0.2433 (0.2465) (0.2462) Number of observations 35202 35202 p (duration dependence) 1.8281*** 1.8262*** Log Likelihood -42267.06 -42259.52 Theta 5.4656*** 5.4363*** Notes: Serial entrepreneurs entering in 2001 or 2002 are excluded. All the speci…cations include an individual-level inverse Gaussian distributed unobserved heterogeneity term. Theta corresponds to the variance of this term. Reference categories: aLess than 9 years of schooling; bManufacturing. *, ** and *** denote signi…cant at 10%, 5% and 1% levels, respectively. Values in parentheses correspond to Huber-White standard errors. Regarding the several individual-level characteristics taken into account in our estimations, results con…rm that men survive longer as serial BOs than women, and that serial BOs’age exerts an U-shaped e¤ect on hazard rates.10 Higher levels of education are found to be associated with greater exit rates, which may be related to the higher opportunity costs that highly educated individuals have by remaining in the business, given that they may also be more likely to …nd more satisfactory alternatives (in the form of less risk-taking and better remunerated options) in the labor market (Gimeno et al., 1997; Georgellis et al., 2007). Those reentering into entrepreneurship at a smaller (larger) scale tend to face somewhat higher (lower) hazard rates –in line with the liability of smallness argument - though the 10 Our estimations suggest that exit rates are decreasing in BOs’age until they reach about 38 years old, starting to increase thereafter. 14 e¤ects are not statistically signi…cant. Being located in an urban center also increases exit rates, probably due to the greater competition characterizing large urban regions (Stearns et al., 1995). As expected, sharing the ownership of the second business with other BO(s), by reducing the risk and potentially increasing the sources of capital and knowledge, is found to reduce entrepreneurs’ exit rates. Those who reentered into entrepreneurship in times of crisis seem to have somewhat higher hazards, but the di¤erences are not statistically signi…cant. Finally, our results show that serial entrepreneurs’exits present positive duration dependence (the estimated value for pis higher than 1), which means that the exit of the entrepreneur becomes more likely as time goes by. However, this result is mainly capturing the relatively higher and increasing hazard rates su¤ered during the initial years in business, when the liabilities of newness and smallness play a particularly signi…cant, and thus dominant, role. If, instead, the baseline hazard rate was parameterized according to a non-linear distribution, serial BOs’exit rates would rather show an inverted U-shaped dependence.11 4.3 Self-selection and serial entrepreneurs’persistence We now take into account that serial entrepreneurs may be a nonrandom sample of ex-BOs. As there may be unobserved factors in‡uencing the decision of reentering into entrepreneurship, ignoring possible self-selection of serial entrepreneurs may actually bias the results and overestimate learning by doing e¤ects (Chen, 2013). A brief characterization of ex-BOs according to their reentry decision is provided in Table A.II in the Appendix. The data show that those who became serial entrepreneurs correspond to i) those who survived for longer periods in the …rst business; ii) those with more experience as start-up founders; iii) those with higher levels of education on average; and iv) those who owned larger …rms at the time of exit from the …rst business (see Table A.II). In this sense, it becomes crucial to understand whether observed and unobserved factors that have in‡uenced the decision of reentering into entrepreneurship were also correlated with the performance shown by serial entrepreneurs after their reentry. Table 4 presents the results obtained from the estimation of Models 1 and 2 previously discussed, now using the two-staged Full Information Maximum Likelihood Weibull duration model with selection developed by Boehmke et al. (2006). The results for the estimated selection equation (reported in Table A.III in the Appendix) con…rm that those who established a start-up venture before and those who have survived for a longer period in the 11 Alternative estimations of a loglogistic AFT model showed that the estimated hazard rates would be increasing during the …rst three to four years of the BO in the …rm, starting to decrease thereafter. The remaining results were not signi…cantly di¤erent from those obtained with the Weibull model. 15 …rst business are more likely to try again and become serial entrepreneurs. In addition, higher levels of education and a larger size of the previous business, among other factors, are also associated with a greater likelihood of reentering into entrepreneurship. Those who (re)entered into paid employment after leaving their …rst entrepreneurial experience, in turn, are found to be signi…cantly less likely to reenter into entrepreneurship, as they may have higher opportunity costs of becoming entrepreneurs again (see, for instance, Baptista et al., 2012). These second results attest that selection should not be overlooked, as a negative and signi…cant correlation is found between the error terms (see the estimated values for rho at the bottom of Table 4). In other words, there are unobserved factors that positively a¤ect reentry into entrepreneurship and simultaneously decrease subsequent hazard rates. This …nding is in line with the theories predicting that those involved in serial entrepreneurship correspond to individuals with higher than average innate ability and skills (Holmes and Schmitz, 1990; Plehn-Dujowich, 2010). Additionally, accounting for serial entrepreneurs’self-selection has important implications on the conclusions derived from potential learning by doing e¤ects. First, results now show that the cumulative experience acquired in the …rst business does not exert any signi…cant e¤ect on serial BOs’hazards. The signi…cant negative e¤ects previously found are now shown to be irrelevant (Model 1) or vanishing in a very short period of time (two years after leaving the previous business, according to Model 2). The e¤ects of industry-speci…c experience are also found to be overestimated when self-selection is ignored –those who tried their luck in the same sector have actually 15% lower hazard rates than those who moved to a di¤erent sector, instead of 22% lower hazard rates as suggested by the “naïve”Weibull model (Model 1 from Table 3). Even so, this comparative advantage seems to vanish after eight to nine years elapsed since the exit from the …rst business. The presence of self-selection also changes the magnitude of almost all coe¢ cients, which were considerably overestimated in the “naïve”Weibull model. The same is applicable to the duration dependence – serial BOs’hazards are found to increase over time, but at a much lower rate when accounting for self-selection. The constant term also decreases considerably when correcting selection bias, con…rming that exit rates of serial entrepreneurs were arti…cially increased in previous "naïve" Weibull models. In sum, once we account for the decision of reentering into entrepreneurship, the estimated exit rates of serial entrepreneurs decrease, since the negative error correlation biases the baseline hazard rates upwards, when ignored. 16 Table 4. Estimation results from the Weibull proportional hazard model with selection (Portugal 1993-2007) Model 1 Model 2 Speci…cities of the …rst entrepreneurial experience Cumulative years as BO 0.0015 -0.0149** (0.0039) (0.0050) Start-up Experience 0.0796*** 0.0792*** (0.0158) (0.0158) Same Sector Experience -0.1610*** -0.2644*** (0.0165) (0.0231) Years elapsed between 1st and 2nd experiences 0.0314*** (0.0032) Cumulative years as BO*Years elapsed 0.0062*** (0.0016) Same Sector Experience*Years elapsed 0.0304*** (0.0051) Individual-level characteristics Male -0.0776*** -0.0788*** (0.0182) (0.0182) Age -0.0703*** -0.0697*** (0.0089) (0.0089) Age squared/100 0.0900*** 0.0896*** (0.0111) (0.0112) 9 years of schoolinga-0.0096 -0.0086 (0.0219) (0.0219) 12 years of schoolinga0.1075*** 0.1078*** (0.0212) (0.0212) College educationa0.0843*** 0.0860*** (0.0254) (0.0255) Firm-level characteristics Firm size at reentry -0.0128 -0.0137 (0.0100) (0.0100) Urban location 0.0717*** 0.0713*** (0.0160) (0.0160) Shared ownership -0.0914*** -0.0928*** (0.0165) (0.0165) It continues in the next page... 17 Table 4. Estimation results from the Weibull proportional hazard model with selection (Portugal 1993-2007) (cont.) Model 1 Model 2 Firm-level characteristics (cont.) Primary Sectorb-0.1303** -0.1292** (0.0615) (0.0614) Energy & Constructionb0.0582** 0.0616** (0.0269) (0.0269) Servicesb0.0056 0.0082 (0.0223) (0.0224) Reenter in a year of crisis 0.0261 0.0303 (0.0255) (0.0255) Constant -1.0611*** -0.9700*** (0.1751) (0.1749) No. of observations 219462 219462 Uncensored Obervations 35202 35202 p (duration dependence) 1.1777*** 1.1766*** Log Likelihood -151079.18 -151075.82 Rho (error correlation) -0.1463*** -0.1464*** Notes: Serial entrepreneurs entering in 2001-2002 are excluded. Both speci…cations have an associated selection equation. The results for the selection equation of the global speci…cation (Model 2) are provided in the Appendix (Table A.III). Reference categories: aLess than 9 years of schooling; bManufacturing. *, ** and *** denote signi…cant at 10%, 5% and 1% levels, respectively. Values in parentheses correspond to Huber-White standard errors. All the estimations were performed with the program DURSEL for Stata, for right-censored survival time data, written by F. Boehmke, D. Morey and M. Shannon and available at: http://myweb.uiowa.edu/fboehmke/methods.html (see also Boehmke et al., 2006). Overall, our results show that neglecting self-selection of serial entrepreneurs may produce biased conclusions about learning by doing e¤ects that might be transferable from past entrepreneurial experiences to the current ones. The positive association between prior experience and the performance of serial BOs in subsequent entrepreneurial attempts is mainly the result of selection on ability, more than the result of learning by doing. Some learning by doing is found only through industry-speci…c experience (see also Frankish et al., 2012; Chen, 2013). Otherwise, learning e¤ects are really modest.12 12 As robustness checks, additional estimations were performed for particular samples of serial BOs according to their entry mode (start-up versus acquisition) and exit mode from the second business (…rm 18 5 Concluding Remarks The topic of entrepreneurial learning has been nurturing a growing debate in the midst of both scholars and policymakers over the most recent years. Entrepreneurs are believed to accumulate unique knowledge and skills by creating and running new ventures, and by establishing networks with suppliers, customers and other business-owners. All this knowhow accumulated through experience is believed to make serial entrepreneurs more able to run successful ventures than novice (i.e., inexperienced) entrepreneurs. Nevertheless, if on the one hand, the lack of suitable data has prevented in-depth empirical analyses about entrepreneurial learning, on the other hand more recent empirical studies addressing these issues have been …nding limited support for entrepreneurial learning hypotheses (e.g., Frankish et al., 2012; Parker, 2012). While the signi…cance of learning by doing remains a question, a new debate has been emerging regarding the potential selection bias associated with the reentry of individuals into entrepreneurship. In fact, do entrepreneurs really learn with their past experience or are those who try again a selected sample of higher-thanaverage ability entrepreneurs? Whether their outperformance comes from learning by doing or self-selection according to their own innate ability thus remains a pertinent query. This paper thus contributes to this debate by using a large longitudinal matched employeremployee dataset that allows us to track individuals and their entrepreneurial experiences over time. We evaluated how previous entrepreneurial experience impacts on serial entrepreneurs’persistence in the second business, exploring a novel empirical strategy based on continuous time duration models that take into account selection bias issues. Our results seem to con…rm that serial entrepreneurs are not a random sample of individuals. Instead, they possess some unobserved characteristics that not only make them more likely to try again as entrepreneurs, as also reduce their exit rates in their second entrepreneurial experience. After correcting this bias in their selection process, the cumulative experience as business-owners exerts no signi…cant e¤ect on their survival in the second business. Besides, the comparative advantages associated with industry-speci…c experience are found to be overestimated when ignoring self-selection problems. In short, our study does not o¤er support for the widespread expectations related to signi…cant entrepreneurial learning. While part of the performance shown by serial entrepreneurs may result from the entrepreneurial knowledge acquired in the previous business – especially when the second entrepreneurial try occurs in the same industry –, learning by doing e¤ects seem to be much less important than self-selection e¤ects. Innate entrepreneurial ability seems to play an essential, and possibly dominant, e¤ect. dissolution versus ownership transfer), and also for a subsample of younger serial BOs. The overall results and conclusions remained consistent across the several sub-samples, being available upon request. 19 References [1] Amaral, A. M., Baptista, R., Lima, F. (2011), “Serial entrepreneurship: impact of human capital on time to re-entry”, Small Business Economics, 37(1), 1-21. [2] Baptista, R., Lima, F., Preto, M. T. (2012), “How former business owners fare in the labor market? Job assignment and earnings”, European Economic Review, 56(2), 263-275. [3] Bastié, F., Cieply, S., Cussy, P. (2013), “The entrepreneur’s mode of entry: the e¤ect of social and …nancial capital”, Small Business Economics, 40(4), 865-877. [4] Boehmke, F., Morey, D. S., Shannon, M. (2006), “Selection bias and continuous-time duration models: Consequences and a proposed solution”, American Journal of Political Science, 50(1), 192-207. [5] Brüderl., J., Preisendörfer, P., Ziegler, R. (1992), “Survival chances of newly founded business organizations”, American Sociological Review, 57(2), 227-242. [6] Chen, J. (2013), “Selection and serial entrepreneurs”, Journal of Economics & Management Strategy, 22(2), 281-311. [7] Cleves, M., Gould, W. W., Gutierrez, R. G., Marchenko, Y. (2010), An Introduction to Survival Analysis using Stata, 3rd edition. Texas: Stata Press. [8] Cope, J. (2005), “Toward a dynamic learning perspective of entrepreneurship”, Entrepreneurship Theory & Practice, 29(4), 373-397. [9] Cope, J. (2011), “Entrepreneurial learning from failure: An interpretative phenomenological analysis”, Journal of Business Venturing, 26(6), 604-623. [10] Cope, J., Watts, G. (2000), “Learning by doing: An exploration of experience, critical incidents and re‡ection in entrepreneurial learning”, International Journal of Entrepreneurial Behaviour and Research, 6(3), 104–124. [11] Ericsson, R., Pakes, A. (1995), “Markov perfect industry dynamics: a framework for empirical work”, The Review of Economic Studies, 62(1), 53–82. [12] European Commission (2002). Best project on restructuring, bankruptcy and a fresh start. Enterprise and Industry Directorate-General. Brussels. [13] European Commission (2011). A second chance for entrepreneurs. Prevention of bankruptcy, simpli…cation of bankruptcy procedures and support for a fresh start. Enterprise and Industry Directorate-General, Brussels. [14] Frank, M. Z. (1988), “An intertemporal model of industrial exit”, Quarterly Journal of Economics, 103(2), 333-344. 20 [15] Frankish, J. S., Roberts, R. G., Coad, A., Spears, T. C., Storey, D. J. (2012), “Do entrepreneurs really learn? Or do they just tell us that they do?”, Industrial and Corporate Change, 22(1),73-106. [16] Fraser, S., Greene, F. J. (2006), “The e¤ects of experience on entrepreneurial optimism and uncertainty”, Economica, 73(290), 169–192. [17] Georgellis, Y., Sessions, J., Tsitsianis, N. (2007), “Pecuniary and non-pecuniary aspects of self-employment survival”, The Quarterly Review of Economics and Finance, 47(1), 94–112. [18] Gimeno, J., Folta, T. B., Cooper, A. C., Woo, C. Y. (1997), “Survival of the …ttest? Entrepreneurial human capital and the persistence of underperforming …rms”, Administrative Science Quarterly, 42(4), 750–783. [19] Gompers, P., Kovner, A., Lerner, J., Scharfstein, D. (2010), “Performance persistence in entrepreneurship”, Journal of Financial Economics, 96(1), 18-32. [20] Headd, B. (2003), “Rede…ning business success: Distinguishing between closure and failure”, Small Business Economics, 21(1): 51–61. [21] Hessels, J., Grilo, I., Thurik, R., van der Zwan, P. (2011), “Entrepreneurial exit and entrepreneurial engagement”, Journal of Evolutionary Economics, 21(3), 447-471. [22] Holmes, T.J., Schmitz, J. A. (1990), “A theory of entrepreneurship and its application to the study of business transfers”, Journal of Political Economy, 98(2), 265–294. [23] Hopenhayn, H. (1992), “Entry, exit, and …rm dynamics in long run equilibrium”, Econometrica, 60(5), 1127–1150. [24] Hougaard, P. (1995), “Frailty models for survival data”, Lifetime Data Analysis, 1(3), 255-273. [25] Hyytinen, A., Ilmakunnas, P. (2007), “What distinguishes a serial entrepreneur?”, Industrial and Corporate Change, 16(5), 793–821. [26] Inci, E. (2013), “Occupational choice and the quality of entrepreneurs”, Journal of Economic Behavior & Organization, 92, 1-21, doi: http://dx.doi.org/10.1016/j.jebo.2013.04.015. [27] Jenkins, S. P. (2005), Survival Analysis, unpublished Lecture Notes manuscript, Institute for Social and Economic Research, University of Essex. (Available at https://www.iser.essex.ac.uk/…les/teaching/stephenj/ec968/pdfs/ec968lnotesv6.pdf) [28] Jovanovic, B. (1982), “Selection and the evolution of industry”, Econometrica, 50(3), 649–670. 21 [29] Jovanovic, B., MacDonald, G. (1994), “The life cycle of a competitive industry”, Journal of Political Economy, 102(2), 322–347. [30] Kalb‡eisch, J., Prentice, R. (1980), The statistical analysis of failure data. New York: Wiley. [31] Kirschenhofer, F., Lechner, C. (2012), “Performance drivers of serial entrepreneurs: Entrepreneurial and team experience”, International Journal of Entrepreneurial Behavior & Research, 18(3), 305-329. [32] Lancaster, T. (1990), The Econometric Analysis of Transition Data. Econometric Society Monographs, 17, Cambridge University Press. [33] Lee, E. T., Wang, J. W. (2003), Statistical Methods for Survival Data Analysis, 3rd edition. New York: Wiley. [34] Li, S., Schulze, W., Li, Z. (2009), “Plunging into the sea, again? A study of serial entrepreneurship in China”, Asia Paci…c Journal of Management, 26(4), 667-680. [35] MacMillan, I. (1986), “To really learn about entrepreneurship, let’s study serial entrepreneurs”, Journal of Business Venturing, 1(3), 241 –243. [36] McCann, B. T., Folta, T. B. (2012), “Entrepreneurial entry thresholds”, Journal of Economic Behavior & Organization, 84(3), 782-800. [37] Minniti, M., Bygrave, W. (2001), “A dynamic model of entrepreneurial learning”, Entrepreneurship Theory and Practice, 25(3), 5-16. [38] Nielsen, K., Sarasvathy, S. D., (2011), “Who reenters entrepreneurship? And who ought to? An empirical study of success after failure”, Paper presented at the DIME-DRUID ACADEMY Winter Conference, Denmark, January, 2011. [39] Parker, S. C. (2006), “Learning about the unknown: How fast do entrepreneurs adjust their beliefs?”, Journal of Business Venturing, 21(1), 1-26. [40] Parker, S. C. (2007), “Entrepreneurial learning and the existence of credit markets”, Journal of Economic Behavior & Organization, 62(1), 37-46. [41] Parker, S. C. (2012), “Do serial entrepreneurs run successively better-performing businesses?”, Journal of Business Venturing, DOI: 10.1016/j.jbusvent.2012.08.001. [42] Parker, S. C., Van Praag, C. M. (2011), “The entrepreneur’s mode of entry: Business takeover or new venture start?”, Journal of Business Venturing¸ DOI:10.1016/j.jbusvent.2010.08.002. [43] Petkova, A. P. (2009), “A theory of entrepreneurial learning from performance errors”, International Entrepreneurship Management Journal, 5(4), 345-367. 22 [44] Plehn-Dujowich, J. (2010), “A theory of serial entrepreneurship”, Small Business Economics, 35(4), 377-398. [45] Robson, P. J. A., Akuetteh, C. K., Westhead, P., Wright, M. (2012), Exporting intensity, human capital and business ownership experience”, International Small Business Journal, 30(4), 367–387. [46] Rocha, V., Carneiro, A., Varum, C. A. (2013), “Entrepreneurship dynamics: Entry routes, business-owner’s persistence and exit modes”, CEF.UP Working Paper No. 2013-10. [47] Sarasvathy, S. D., Menon, A. R., Kuechle, G. (2013), “Failing …rms and successful entrepreneurs: serial entrepreneurship as a temporal portfolio”, Small Business Economics, 40(2), 417-434. [48] Stam, E., Audretsch, D., Meijaard, J. (2008), “Renascent entrepreneurship”, Journal of Evolutionary Economics, 18(3-4), 493-507. [49] Stearns, T. M., Carter, N. M., Reynolds, P. D., Williams, M. L. (1995), “New …rm survival: industry, strategy and location”, Journal of Business Venturing, 10(1), 23-42. [50] Ucbasaran, D. Wright, M., Westhead, P. (2003), “A longitudinal study of habitual entrepreneurs: starters and acquirers”, Entrepreneurship & Regional Development, 15(3), 207-228. [51] Ucbasaran, D., Shepherd, D. A., Lockett, A., Lyon, J. (2012), “Life after business failure: The processes and consequences of business failure for entrepreneurs”, Journal of Management, DOI: 10.1177/0149206312457823. [52] Ucbasaran, D., Westhead, P., Wright, M., Binks, M. (2006), Habitual Entrepreneurs. Cheltenham: Edward Elgar. [53] Ucbasaran, D., Westhead, P., Wright, M., Flores, M. (2010), “The nature of entrepreneurial experience, business failure and comparative optimism”, Journal of Business Venturing, 25(6), 541-555. [54] Van Gelderen, M., van der Sluis, L., Jansen, P. (2005), “Learning opportunities and learning behaviours of small business starters: Relations with goal achievement, skill development and satisfaction”, Small Business Economics, 25(1), 97-108. [55] Von Hayek, F. A. (1937), “Economics and knowledge”, Economica, 4(13), 33–54. [56] Wagner, J. (2003). “Taking a second chance: Entrepreneurial restarters in Germany”, Applied Economics Quarterly, 49(3), 255–272. 23