Export activity, innovation and institutions in Southern European nascent entrepreneurship

This paper studies the role of personal characteristics, perceptual variables and countrylevel conditioning (financial environment, government quality and support, education quality and entrepreneurship know-how, innovation environment and support, business infrastructure, entrepreneurial culture and society, and gender roles) in explaining the export propensity and intensity of nascent entrepreneurs in four Southern European countries (Portugal, Spain, Italy and Greece), using Total Early-stage Entrepreneurial Activity (TEA) data from the Global Entrepreneurship Monitor (GEM) dataset in 2003– 2010. Due to the nascent nature of the business, export activity is starting or about to start at the time of the survey and, for that reason, it cannot be studied using theoretical frameworks based on productivity heterogeneity, which has not yet been measured. In this sample of nascent businesses, there is no evidence of a selection effect into exporting and the individual-level factors influencing export propensity and intensity are identical. The most relevant individual-level variables facilitating export activity are new products, new technology, graduate education, and entrepreneurship networks. The most relevant country-level factors facilitating export activity are the availability of funding, the national government’s macroeconomic support, and the support for new technology. JEL F14 J24 L26 O43


Introduction
Moreover, the international economics literature has studied the direction of causality between innovation and exports, finding a clear and strong association between the two variables (Clerides et al., 1998;Helpman, 2006;Greenaway and Kneller, 2007;Bernard et al., 2012;Melitz and Redding, 2012;Silva et al., 2013). Van Beveren and Vandenbussche (2010) find that innovation makes firms more likely to start exporting. In this way, there is support for seeing innovation as exogenous with respect to exporting. The studies cited above that link exporting and innovation all deal with established firms and thus are able to link innovation to productivity and from there to exporting. However, none of them deals with nascent entrepreneurs, which is what this paper does, using the GEM database. The definition of nascent entrepreneur in the GEM database is that of an individual carrying out Total Early-Stage Entrepreneurial Activity (TEA). This definition pools the startups (SU), involved in setting up a business in the 12 months preceding the survey, and the owners-managers (OM), who owned and managed a business started less than 3.5 years prior to the survey. The SU group answers the survey question: "What proportion of your customers will normally live outside the country?" The OM group answers the survey question: "What proportion of your customers normally live outside the country?". Due to the nascent nature of the business, export activity is starting or about to start at the time of the survey. For that reason, it cannot be studied using theoretical frameworks based on productivity heterogeneity, which has not yet been measured. Thus, in the context of nascent businesses, the study of export activity means asking which factors hinder or facilitate the willingness to export and the development of the export activity. In businesses that strive to be successful in a globalized world, the decision to export and by how much may be confounded with the decision to start the business itself.
In this context, the research benchmark is the GEM model, which has been established to form the basis of the research on the relationship between entrepreneurship and growth (Reynolds et al., 2005;Levie and Autio, 2008;Alvarez et al., 2014), as well as between entrepreneurship and innovation (Bosma et al., 2012). The transmission channels of that relationship include: (i) objective factors, such as personal socio-economic characteristics of the entrepreneur (gender, age, prior work status, education and income levels); (ii) subjective factors related to the entrepreneur's attitudes, motives and perceptions; (iii) and national institutions and social values having a strong direct effect on entrepreneurship by providing a normative and institutional context.
The development of the GEM database originated a great deal of research (see reviews in Arenius and Minitti (2005), van der Sluis et al. (2008), Alvarez et al. (2014), Bergmann et al. (2014)). Van Stel et al. (2005) and Hessels and van Stel (2011) found that the contribution of TEA to economic growth, moreover if it is export-oriented, is stronger in high-income countries, possibly because these have higher levels of human capital. Bogenhold et al. (2014) 2 focus on the so-called professions and, using the results of a survey distributed in Finland in 2006, present evidence that those professionals have very specific personal characteristics (89% worked alone, 71% were women and 72% held a university degree). 2 Bogenhold et al. (2014) discuss self-employment, which implies a deliberate decision to start an activity but applies to those working alone. Entrepreneurship may include self-employment, but it also refers to businesses that employ workers in addition to the entrepreneur. Moreover, individuals with similar socio-economic status are differently influenced by perceptual variables such as knowing other entrepreneurs, alertness to opportunities, fear of failure, confidence about one's own skills (Arenius and Minitti, 2005;Koellinger et al., 2013) and motives labelled as necessity (Bogenhold et al., 2014) and opportunity-seeking (Block et al., 2015). Whereas opportunity-seeking entrepreneurship is innovative and carried out by individuals who are employees or students, necessity-based entrepreneurship is defined as more incremental or imitative and it is typically carried out by individuals who are unemployed (Acs et al., 2008). Bogenhold et al. (2014) present evidence that Finnish professionals choosing to be self-employed have mostly opportunity-seeking motivations. Block et al. (2015) focus on necessity entrepreneurship and conclude from a survey given to German nascent entrepreneurs in 2008 that this motive is linked to pursuing a low-cost price strategy, having a lower education level, and being present in more labour-intensive sectors.

_________________________
At the country-level, macroeconomic characteristics such as GDP per capita and the business cycle have been considered an influence on the entrepreneurship decision, together with institutions and policies (see the summary provided by Levie et al., 2014). The importance of institutions in shaping different entrepreneurship choices has been examined by Acs et al. (2008) and Herrmann (2010), whilst the importance of inherited values and institutions in shaping individual decisions is discussed at length in Guiso et al. (2006), Liñán and Fernandez-Serrano (2014), and Urbano and Alvarez (2014). Some of the more recent empirical entrepreneurship research has combined individual-level GEM data with country-level data from various other sources and focused on the role of country-level variables such as intellectual property protection (Autio and Acs, 2010), regulatory framework (Ardagna and Lusardi, 2008) and national institutions (van Stel et al., 2005;Levie and Autio, 2008;Liñán and Fernandez-Serrano, 2014;Urbano and Alvarez, 2014).
The proposed model of export activity in entrepreneurship is summarized in Figure 2. There are three groups of covariates that influence the choice between exporting or not (export propensity), and by how much (export intensity): personal characteristics of the entrepreneur, perceptual variables, and country-level culture (institutions and social norms). These variables are listed in Table A1 (personal characteristics and perceptual variables) and in Table A4 (country-level factors). As shown in Figure 2, country-level culture has both a direct and an indirect effect because it also influences personal characteristics and perceptions. In turn, export propensity may mediate the effects of these covariates on export intensity if the characteristics of entrepreneurs who export are significantly different from those of the population. However, if nascent businesses are set-up with a view to exporting (born to export), the entrepreneurs who want to export may be the actual population of entrepreneurs, and then there may be no mediation through the propensity to export. As a consequence, both export propensity and intensity are determined by the direct effects of the covariates and the two models become the same.
Indeed, the export intensity and export propensity models are identical for the SE4 sample used in this paper. In this way, as will be seen, job growth, the existence of a new product or a new technology, the number of owners of the business, a graduate education level, and knowing other nascent entrepreneurs, are individual-level factors that increase the probability of exporting as well as its intensity. The existence of a new product or a new technology makes a business different from its competitors and thus it is more competitive in foreign markets. A graduate education level provides the entrepreneur with more knowledge, know-how and skill, which may be required in foreign markets. The number of owners of the business provides it with a pool of knowledge, know-how and skill, as well as more financial resources, allowing for venturing into foreign markets. Knowing other nascent entrepreneurs may have similar advantages, in addition to providing a role-model to newcomers. The psychological process of the entrepreneur is described in detail in Marques (2017b). Export propensity Export intensity strategy implies that entrepreneurs in SE4 countries share common individual characteristics and thus they can be pooled together for analysis, whilst controlling for characteristics that are specific to each country by means of country fixed effects.

Data and methodology
The definition of nascent entrepreneur in the GEM database is that of an individual carrying out Total Early-Stage Entrepreneurial Activity (TEA). This definition pools the start-ups (SU), involved in setting up a business in the 12 months preceding the survey, and the ownersmanagers (OM), who owned and managed a business started less than 3.5 years prior to the survey. Therefore, the entrepreneurs surveyed are individuals who carried out their activity, including the setting-up process, in the period 2000-2010. There is a maximum of 7,463 individual observations, each corresponding to a nascent entrepreneur, although due to missing values the number of observations diminishes as covariates are included in the regressions. All individual-level variables are described in Table A1, with descriptive statistics provided in TableA2 and the association between individual-level covariates shown in Table A3. Information relative to country-level factors is provided on Tables A4-A5.
The dependent variables are export intensity and export propensity. In the APS database, export intensity is a categorical variable for the percentage of customers abroad with categories given by intervals (None; 1-10%; 11-25%; 26-50%; 51-75%; 76-90%; over 90%). From export intensity it is possible to construct a binary variable for export propensity taking None as the reference group and making all other categories equal to one. The nature of the dependent variables conditions the methodology that can be used. The regression models for export propensity and for export intensity may be related by means of a second stage ordered probit for positive export intensity with a Heckman selection correction estimated as a first stage binary probit for export propensity (see De Luca and Perotti, 2011). 3 However, in this sample, the correlation between the errors of the two equations is not significant, so the selection model is not required. 4 Therefore, the export intensity and propensity models are estimated separately, with the former being estimated as ordered logit, and the latter being estimated as a binary logit. In interpreting the results from the export intensity estimation, it is necessary to take into account that the dependent variable, as given in the survey data, is an ordered ranking and no cardinal meaning can be attributed to it in terms of regression. By conducting the export intensity and propensity regressions separately, the same set of covariates can be used in both cases, thus minimizing the conditioning in the choice of covariates and the problems of perfect prediction in the export propensity model due to the existence of missing values.

_________________________
3 The Heckman selection correction proposed by DeLuca and Perotti (2011) has been estimated in STATA using the command heckoprobit.
4 Nevertheless, some interesting additional results are that in the second equation for positive export intensity, the significant covariates are the number of owners of the business and being motivated by increasing income. Notably, there are no gender differences in either export propensity or positive export intensity when the two-step selection model is used. Due to collinearity, country, industry and year fixed effects have to be included in either the main equation or in the selection equation. When including them in the main equation, having new technology, the level of competition and the entrepreneur's education level were relevant selection variables (as well as knowing other entrepreneurs and being motivated by maintaining income, although these two variables do not produce a valid selection model, that is, the F-test is not significant). If instead fixed effects are included in the selection equation, being motivated by maintaining income is the only relevant selection variable, producing a valid model (significant F-test). These results do not, therefore, contradict the main results presented in the paper.
Both models are estimated with survey weights that match the gender-age distribution of the adult population in the 18-64 age range in each sample country-year pair. Hence, the group of survey-weighted individual observations pertaining to each country-year cell is representative of the working-age population in that cell (Levie and Autio, 2008;Levie et al., 2014). This allows the pooling of observations across countries and years, taking into account unobserved heterogeneity by means of country, industry and year fixed effects. Country-year fixed effects are included whenever possible in order to capture the diverse macroeconomic evolution of the SE4 countries shown in Figure 1.
With respect to nascent entrepreneurship, Portugal is the most export-oriented (54.8% of the sample businesses export) and Spain is the least export-oriented (31.2% of the sample businesses export). 5 The distribution of the export intensity categories by country is shown in Figure 3 and it significantly differs across the SE4 countries (Pearson Chi2 = 219, p-value = 0.000). It can be observed that the distribution of export intensity categories has the fattest tails in the case of Portugal, given that this country not only has the largest share of nascent businesses exporting not more than 10% of their total sales, but it also has the largest share of nascent businesses exporting over 75% of their total sales. For the whole SE4 group, 33.9% of the businesses export, of which 17.7% do not export more than 10% of their total sales and 5.3% export over 75% of their total sales. _________________________ 5 Note that the GEM data for nascent entrepreneurship does not necessarily coincide with aggregate trade data.
According to Eurostat data, the aggregate share of exports of goods and services in GDP in the 2007-2015 period was 23.1-33.6% for Spain, 19.0-32.4% for Greece,.6% for Portugal. Therefore, the data seems to imply that nascent businesses are more export-oriented than the average business in some countries (for example, Portugal), but are in line with existing businesses in other countries (for example, Spain). These data hint at the existence of two groups of exporters in the SE4 countries: a larger group that has lower export intensity, and a smaller group that has higher export intensity. This characterization is most pronounced in the case of Portugal and it is compatible with the coexistence of low and stagnating aggregate productivity with a few very internationalized large firms (Gopinath et al., 2017;Nogueira and Inácio, 2017;Pinheiro Alves, 2017). The nonlinearity existent in the data for export intensity can be well captured by an ordered logit model that takes into account all the intervals of export intensity provided by the GEM data, with the binary logit for export propensity revealing whether the extensive and intensive margin decisions are significantly different.
The individual-level covariates obtained from the APS comprise personal characteristics of the entrepreneur (gender, age, education level, and income level), motives for entrepreneurship (greater independence, increase personal income, and just maintain income) 6 and a perception variable (knowing other entrepreneurs). Other individual-level variables related to the characteristics of the business are its number of owners, the contemporaneous number of jobs created and its expected five-year job growth rate, as well as variables related to technology, innovation and competition (use of technological innovations, selling a new product, and the number of competitors). The contemporaneous number of jobs, as well as the age and income level of the entrepreneur were not significant covariates. 7 Neither were other perceptual variables such as the alertness to opportunities, fear of failure, and confidence about one's own skills (see Arenius and Minitti (2005) for an assessment of perceptual variables). In this paper, the covariates that were not relevant were not included in the final regressions.
As shown in Table A2, an average of 55.6% of the sampled entrepreneurs claims to know other entrepreneurs. It can also be seen that on average 38.4% are women, 31.1% have some secondary studies and 51.2% are motivated by greater independence. The average business in the sample has 1.8 owners, employs 3.4 people and expects to employ 351 people within five years. In the sample, 58% of businesses do not offer a new product, 73.8% do not use any new technology and 59.46% face many competitors.
The country-level variables obtained from the NES are country mean scores measured in 5point Likert scales and grouped into topics A to R, covering issues such as the country's financial environment (topic A), government quality and support (topics B, C and Q), education quality and entrepreneurship know-how (topics D and L), innovation environment and support (topics E, N and R), business infrastructure (topics F, G and H), cultural and social views on entrepreneurship (topics I, K and M), and views on gender roles (topic P). In order to facilitate the analysis and restrict multicollinearity problems, country-level normalized factor scores are obtained for each of the seven variable groups described. The country-level variables were reduced to 22 factors that were obtained applying principal components with varimax rotation and Kaiser normalization within each of the seven NES aggregates to facilitate interpretation. The details on those factors are provided in Table A4.

_________________________
6 To achieve greater independence and to increase personal income may be seen as opportunity motives, whereas to just maintain income is more akin to a necessity motive.
7 As characterization of the entrepreneurs in the sample it is relevant to mention that 28.86% and 31.37% are in the 25-34 and 35-44 age ranges, respectively, and that 40.38% belong to the upper income tercile. Table A5 shows the relative strengths and weaknesses of SE4 countries for the 22 factors obtained. Due to the normalization of the factors, the means are always very close to zero, although the 5% confidence intervals are quite wide, revealing implicit country differences. Factors that have confidence intervals that span over positive values longer (shorter) than over negative values can be regarded as strengths (weaknesses). Thus, it is possible to identify as relative strengths the availability of private funding, the role of the education system and knowhow in starting a business, the level of support to innovation and new technology, market structure and functioning, and the existence of social support to entrepreneurship.
Some relative weaknesses are the weak governmental support to entrepreneurship, in particular through the tax system and bureaucratic system (length of time required for new firms to obtain permits and licenses), the weak know-how in managing a business, the weak infrastructure affordability and efficiency (utilities such as electricity, gas and water, and communications networks, as well as business support services), the perceived lack of entrepreneurial opportunities, the low social status of entrepreneurs in the context of national culture, and the perceived lack of gender equality of opportunities through the level of knowledge and skills required to start a new business.

4
Export activity of nascent businesses in SE4 countries

Results for individual-level variables
Tables 1 and 2 show the regression results for, respectively, export intensity and export propensity, considering intercept differences in the SE4 countries by means of country fixed effects and controlling also for industry and year fixed effects. In Table 1, the export intensity model is built-up progressively, with the benchmark model (5) being compared to two robustness checks: model (6) adding the knowledge of other entrepreneurs, which eliminates roughly half of the observations, and model (7) adding country-year fixed effects to account for the different evolution of the SE4 countries during the sample period. In Table2, only these three models are shown for export propensity. The results on Table 1 reveal that, on average, job growth expectations are positively associated to high export intensity, but this link is weakened when the use of new technology and especially selling a new product are considered. Therefore, it can be said that these two variables, with a positive coefficient always significant at 1%, fully mediate any link between the entrepreneur's expectations for business growth and export intensity. Using new technology and selling a new product have associated odd-ratios of 0.62-0.82 and 0.44-0.68, respectively, making it more than twice as likely that a business exports a high share of total sales. The effect of competition is weakly significant at 10% and possibly non-linear as related to the degree of innovation. Women entrepreneurs are less likely to export a high share of total sales, although this is a complex effect, itself mediated by other variables, as shown by Marques (2017aMarques ( , 2017b, and it loses significance if a two-stage model is used. To have graduate studies is positively associated with export intensity but the association is significant only at the 10% The regression method is ordered logit with survey weights that match the gender-age distribution of the adult population between 18-64 years in each country-year pair. Standard errors in parentheses with *** p<0.01, ** p<0.05, * p<0.1. The definitions of all variables are provided in Appendix Table A1, with expint being the dependent variable. All regressions include country, industry and year fixed effects. Cut points are omitted but are all significant. level, as education may be partly mediated by innovation and partly a signal of higher entrepreneurial ability and business know-how. To maintain the income level is a motivation positively associated to export intensity, hinting at the role of export markets in compensating for weak domestic demand in the SE4 countries during the period under analysis. Finally, the number of business owners is positively associated with export intensity, but this effect is fully mediated by knowing other entrepreneurs, which indicates that the incorporation of several owners to the business is equivalent to having an entrepreneurial network. The results in Table 2 reveal that the determinants of binary export propensity behave similarly to those of ordered export intensity. Using new technology and selling a new product have associated odd-ratios of 1.32-1.52 and 1.42-2.03, respectively, making it up to twice as likely that a business exports. The education level of the entrepreneur is positively associated with export propensity. This effect is found at the level of graduate studies with 5% significance, although it drops to 10% if knowing other entrepreneurs is taken into account, hinting that having access to an entrepreneurial network may partly compensate for ability and know-how. The education effect on export propensity is also found at the level of postsecondary studies with 10% significance, although this education level did not relate to export intensity. Whereas to maintain the income level is positively associated to export propensity, as it was to export intensity, to increase income is negatively associated to export propensity (and this motive had no effect on export intensity). Since its significance level drops from 5% to 10% when considering access to entrepreneurial networks, it could be associated to the perceived costs of exporting, which entrepreneurs who are motivated by increasing income are not willing to incur.

Results for country-level variables
The regression results for each of the 22 country-level factors obtained are provided in Table 3. Those factors that were not significantly associated to the dependent variables before controlling for other variables were not included in the regressions. All regressions include as control variables those individual-level variables that were relevant in the regressions of Tables 1 and 2: using a new technology, offering a new product, and the entrepreneur's gender and education level. The inclusion of these variables does not affect factor significance. Out of the 22 factors obtained, 17 and 13 are significantly associated with export intensity and propensity, respectively. The "knowing other entrepreneurs" variable is included in model (3) as it halves the number of observations. It only affects the significance of the social status of entrepreneurs, as presumably knowing other entrepreneurs who can be taken as role-model removes relevance from the country's views on the social status of entrepreneurs as a group.
In model (2), country, industry and year fixed effects are included. Some factors are fully explained by those fixed effects and lose significance, notably in education and entrepreneurship know-how, and in business infrastructure. Still, there are 8 factors that remain significantly associated with both export intensity and propensity, 5 of them positively (Availability of equity funding, debt funding, and government subsidies; National government's macroeconomic support to entrepreneurship; New technology's interest and affordability; Business interest in new technology and respect for IPRs; Social status of entrepreneurs) and 3   Model F 14.38*** 11.31*** 14.13*** 11.90*** 13.40*** 11.01*** 15.98*** 11.66*** 15.85*** 11.07*** 14.37*** 11.98*** 9.142*** 15.11*** 11.17***    The regression method is binary logit, with survey weights that match the gender-age distribution of the adult population between 18-64 years in each country-year pair. Standard errors in parentheses with *** p<0.01, ** p<0.05, * p<0.1. The definitions of all variables are provided in Appendix Table A1, with exporter obtained from expint making positive export intensity equal to 1. All regressions include significant individual-level variables (newtech, newprod, owoman, educ and knowent).
of them negatively (Local government's microeconomic support to entrepreneurship; National culture; Equality of opportunities for men and women). Whilst the positive association is quite intuitive and logical, the negative association can be explained by the fact that the dependent variables represent export activity decisions, not the entrepreneurship decision in itself, such that those factors may contribute to higher entrepreneurship rates whilst those new businesses do not have an export orientation. 8

Robustness checks on the individual-level export intensity model
The export intensity model of Table 1 is estimated again for a variety of robustness checks with respect to other samples, various subsamples and methods. Although the specific results of each check are discussed below, in general the main results are very robust and confirm the role of innovation and technology, as well as the existence of entrepreneurship networks, in fostering export intensity.

Micro (0-1 jobs) versus small businesses (10-49 jobs)
In the full dataset, 47.25% of the businesses are micro businesses (0-1 jobs), whereas only 6.39% are small businesses (10-49 jobs). It could be that larger businesses are driven by different factors compared to smaller businesses. As a consequence, Table 1 regressions have been repeated for the two subsamples separately and the results summarized on Tables A6 and  A7. It can be seen that the results for micro businesses replicate by and large those presented in Table 1, but for small businesses only the number of owners remains relevant due to the small size of the subsample.

Services versus manufacturing
In the full dataset, 87.76% of the businesses belong to the services sector, whereas only 12.24% can be found in the manufacturing sector. It could be that the services sector is moved by different determinants compared to the manufacturing sector. As a consequence, Table 1 regressions have been repeated for the two subsamples and the results are shown on Tables A8 and A9. It can be seen that the results for the services subsample replicate by and large those presented in Table 1, but for the manufacturing subsample only selling a new product and having many competitors remain relevant due to the small size of the subsample.

Nascent versus young businesses
As stated before, the dataset used is cross-sectional and it is not always possible to address any problems of reverse causality. However, the full sample is composed of 46.22% nascent entrepreneurs ("respondents (18-64) involved in nascent business, defined as active, expect to be a full or part owner, and no salaries or wages paid for over three months") and 53.78% baby business owner-managers ("respondents (18-64) involved as owner and manager in new firms for which salaries or wages have been paid between 3 and 42 months"). The survey question for the former group is "What proportion of your customers will normally live outside the country?", whereas for the latter group is "What proportion of your customers normally live outside the country?" Although the variable is not forward as such, in fact it has a prospective nature. Therefore, whilst it is recognized that in cross-sectional data only correlations can be discussed, the individual-level regressions have been repeated for the two subsamples of nascent and young businesses, one prospective and another contemporaneous. The regressions results are presented on Tables A10 and A11.
Although having a new product and a new technology increase export intensity for the two groups, there are differences in several variables. Start-ups (nascent) are negatively affected by the number of competitors, whilst owner-manager young businesses are positively affected. Also, women are negatively affected in start-ups but not in young businesses, whereas the number of owners has a positive effect only in the latter group. Tertiary education increases export intensity only in the start-up group. In this group, greater independence goes against export intensity, but just maintaining income fosters it, whilst in the owner-manager group increasing income makes export intensity lower. These differences can be interpreted in terms of prospective versus contemporaneous behaviour.

Spain
In the dataset used initially for the period 2003-2010, Spain takes up 82.97% of the observations (Table A15) and, if there are no outliers driving the results, it would be expected that the results obtained for Spain replicates the overall results. Indeed, repeating the estimations for Spain does not change the main overall results, as can be seen in Table A12. An exception is the graduate education level, which appeared with a coefficient significant at the 10% level in the full sample and loses significance for Spain. The full sample significance is due to Greece, especially with respect to export propensity, where the coefficient of the graduate education level reaches the 5% significance level.

Interactions
Many of the covariates are correlated, which may cause simultaneity and multicollinearity. One way of finding out whether there are pairs of covariates that have only a joint effect is to add to the individual-level regressions one interaction between each pair of covariates at a time. In this way, it can be seen whether an interaction removes the individual effects of the two interacted covariates. The estimation results are displayed in Table A13. When the export intensity models of Table 1 are considered, a few interactions survive the inclusion of the remaining variables: with a positive coefficient, newtech#compete, new-tech#motive2, newprod#nowners, newprod#motive1; with a negative coefficient, new-tech#nowners, newprod#gemeduc, compete#nowners. The predominance of having a new technology or a new product in the significant interactions confirms the results of the paper regarding the importance of those characteristics for exporting. The inclusion of the interaction removes significance from the coefficient of at least one of the covariates present in that interaction, so that it is possible that synergic effects are present as second order.

Zero-inflated probit
The results of Tables 1 and 2 are based on, respectively, the ordered logit and binary logit models. As stated in Section 3, the regression models for export propensity and for export intensity may be related by means of a second stage ordered probit for positive export intensity with a Heckman selection correction estimated as a first stage binary probit for export propensity. However, in this sample, the correlation between the errors of the two equations is not significant, so the selection model is not required. Therefore, the export intensity and propensity models are estimated separately, with the former being estimated as ordered logit, and the latter being estimated as a binary logit. It could be the case that the Heckman selection model is not the most suitable because export intensity is measured on an ordinal scale that includes the value zero, which is observable, that is, it is known that the business does not export. The Heckman model is most suitable when the zeros correspond to non-observable values because of truncation, that is, the value for a business's export intensity is not known. It could be zero or not. Due to this issue, the most appropriate selection model may be the zero-inflated ordered probit model (Harris and Zhao, 2007). Table 1 regressions have been repeated for this alternative method and the estimation results are displayed in Table A14. It can be seen that having a new product or a new technology, having a tertiary education level, or knowing other entrepreneurs, positively influences the first stage of exporting. On the contrary, being a woman has a negative influence on the first stage of exporting. However, this behavior of the covariates corresponds to what is found in the model of Table 2 for export propensity. Therefore, the zero-inflation probit model returns a very similar result to Tables 1 and 2: for these data, export propensity and export intensity follow similar models.

Data for 2011-2015
Although the paper used data for 2003-2010, more recent data is available for the four countries up to 2015. 9 In the new data for 2011-2015, the country structure is not very different from that of the first period, with Spain having 81.48% of the observations, Greece 7.69%, Portugal 6.17%, and Italy 4.66% in 2011-2015 (Table A15). The descriptive statistics for the post-recession period 2011-2015 are shown in Table A16. It can be said that 65.42% of the businesses does not export, whilst 5.15% exports more than 75% of sales. There are 57.44% of businesses with two owners and 25.14% with three owners. Only 4.47% of the surveyed entrepreneurs are women. A very recent (recent) technology is used by 13.10% (20.07%) of the businesses, whilst 36.13% of the businesses sell a new product. The same product is offered by few (none) of other businesses for 35.78% (10.66%) of the sample. Regarding the education level of the entrepreneurs, 33.05%, 60.74% and 2.90% has, respectively, primary-level, secondary-level and tertiary-level studies. To increase income, gain independence or just out of necessity are the motives for setting up a new business for, respectively, 21.94%, 16.57% and 34.23% of the entrepreneurs surveyed. To have known other entrepreneurs was considered as important in setting up a new business for 30.41% of the sample.
The individual-level regressions have been repeated for 2011-2015 and the estimation results are shown in Table A17. Some adjustments had to be made as some variables that were reported as continuous are now reported in intervals and the dummy for woman entrepreneur had to be removed for the reason explained above. Besides, the export intensity intervals differ 10 and the industry classification is not the same (now Industry ISIC version 4, 1-digit is used). As such, it does not make sense to simply extend the sample and the regressions for 2011-2015 have been added separately. The results show that job growth, the number of owners, being motivated by increasing income and gaining independence, as well as having a new product and, to a lesser extent, a new technology, all influence export intensity positively.

6
Discussion and concluding remarks This paper finds that, in the SE4 countries, the ability of a nascent business to offer a new product, or a new technology, is very significantly associated to both export propensity and intensity. There are several transmission channels from innovation to exporting. On the one hand, there is a supply argument linked to a higher productivity derived from innovation. On the other hand, there is a demand-side argument through which innovation leads to higher product competitiveness in foreign markets, thus fostering the exporting activity. This transmission channel may be observed through product upgrading or through a combination of product, process and marketing innovations. Exporting is also positively influenced by having a graduate education, the number of owners and knowing other nascent entrepreneurs. A graduate education level provides the entrepreneur with more knowledge, know-how and skill, which may be required in foreign markets. The number of owners of the business provides it with a pool of knowledge, knowhow and skill, as well as more financial resources, allowing for venturing into foreign markets. Knowing other nascent entrepreneurs may have similar advantages, in addition to providing a role-model to newcomers. Additionally, Grossman et al. (2017) argue that productivity depends on the ability of manager and workers, proxied by their education level, and Dinopoulos and Unel (2017) explain that high-ability entrepreneurs invest in managerial capital, increasing firm productivity that leads to exporting.
In the 2003-2010 data used, 42.02% of the surveyed entrepreneurs in the SE4 countries claim to be offering a new product. If new businesses follow the tendency of established businesses of exporting low value-added products, that can be harmful for the country's longrun growth (Dimelis et al., 2017). Given that change may be easier for nascent businesses than for long-established ones, it would be important to actively promote innovation among the former and target the profile of entrepreneur most prone to innovation, at the same time facilitating change for those profiles less likely to innovate by understanding the mechanisms that hinder them.
As an example, Marques and Moreira (2013) found that a typical innovative entrepreneur is a man who graduated in engineering, was motivated by opportunity, greater independence and increasing income, had high social capital, and was likely to start a business in knowledgeintensive technological sectors; whilst a typical non-innovative entrepreneur is a woman who graduated in social science, law or education, was motivated by necessity, had low social capital, and was likely to start a business in health care and education sectors. Moreover, another study found that 30.5% of the participants in entrepreneurship support programs in 2007-2013 were engineering, computer-science and technology graduates, and 19% were economics and business studies graduates . As a consequence, those programs do not provide equal support to different target groups.
There is evidence that the education system can play a crucial role in the promotion of entrepreneurship, namely by promoting adequate attention to entrepreneurship in primary and secondary education, providing good and adequate knowledge in business and management education, good preparation for self-employment in vocational, professional and continuing education, and teaching how to start and manage a small business (Marques, 2015;Marques et al., 2015). The education system can also work towards improving the views that starting a new business is a socially acceptable career option for women, as well as ensuring gender equality in the provision of knowledge and skills to start a new business. In conjunction to the role of the education system in providing that knowledge and skills, it is fundamental to implement realistic and operative forms of conciliation, such as the provision of sufficient social support services when there are dependent people in the family, including children, disabled or older people. The presence of aged dependents in the household is a more and more common situation to which European countries need to provide growing attention and, in turn, the need for work-family balance is one of the most relevant factors for women's entrepreneurship (Santos et al., 2017).
Other policy areas in which greater effort needs to be made is in reducing the length of time to obtain required permits and licenses, and in protecting copyrights and trademarks. Since these issues depend essentially on adequate legislation and monitoring, it seems that there is good scope for policy intervention in these areas.
The type of survey data used in this paper has the potential to relate internationalization, innovation, competition and competitiveness, and allow a policy-relevant analysis with international comparability, given that it is harmonized across countries and it contains policy variables in the same topics as the individual-level variables. The GEM data used in this paper samples the 18-64 age range from the whole population, but it presents some caveats. On the one hand, the GEM APS and NES indicators have the advantage of picking-up newly formed businesses, which have a different character to older ones, in particular are more prone to innovation, but to pick up cohorts of newly formed businesses it forms cross-sectional waves. On the other hand, the number of nascent entrepreneurs participating in the GEM surveys is very unbalanced across countries. As an example, in Portugal the sample size is 153 in 2005-2010 and it represents only 0.22% of the full worldwide sample. In comparison, the number of surveyed businesses in Spain was 7,268 in the same period, which represented 10.57% of the worldwide total. Taking into account the different economic size of the two countries as measured by GDP, a participation of around 1,198 businesses would be expected in Portugal. Instead, participation was only 13% of the expected number. Therefore, in order to achieve more robust policy conclusions, it would be desirable to have representative samples for each country. Otherwise, the small sample size would mean that there is a gap in the entrepreneurship rate, implying in that case that there is potential for entrepreneurship promotion relative to the country's economic size.
Given the links between the educational level of the entrepreneur, using a new technology, offering a new product, and export propensity and intensity, it seems that, in SE4 countries, entrepreneurship may be a way of increasing and improving young people's employment, renewing those countries' business structure, fostering product and process innovation, thereby allowing future increases in productivity, greater international competitiveness and, all in all, promoting long-run growth. Dummy variable taking value 1 if the entrepreneur knows someone who started a business in the past 2 years. NOTE: All survey questions can be found at http://www.gemconsortium.org.  1.4e+03*** NOTE: The Pearson Chi-squared tests for association between each pair of variables. Rejection of the null hypothesis of no association at the 1%, 5% or 10% significance levels is represented by ***, **, *, respectively. The jobnow and jobgrow variables were discretized at, respectively, less than 5 jobs, between 5 and 10, and more than 10 jobs, and for negative, null and positive growth.

Appendix Tables
Economics: The Open-Access, Open-Assessment E-Journal 13 (2019-53) www.economics-ejournal.org 30 Qp01 there are sufficient social services available so that women can continue to work even after they start a family 0.31985 Qp02 starting a new business is a socially acceptable career option for women 0.66538 Qp03 women are encouraged to become self-employed or start a new business F72: Equality of opportunities for men and women 0.57884 Qp04 men and women get equally exposed to good opportunities to start a new business 0.54982 Qp05 men and women have the same level of knowledge and skills to start a new business NOTE: All factors were obtained applying principal components with varimax rotation and Kaiser normalization within each NES aggregate described in the paper to facilitate interpretation. Those factors that were not significantly associated to the dependent variables before controlling for other variables were not included in the regressions. In the table, each question is assigned to the factor where it enters with the highest loading, which is also indicated on the table. All survey questions can be found at http://www.gemconsortium.org.  jobgrow -3.10e-05 -5.91e-05 -8.74e-05 -5.81e-05 -5.84e-05 -0.000103 -6.77e-05 (6.32e-05) (6.55e-05) (7.13e-05) (6.58e-05) (6.49e-05) (6.93e-05) (6.78e-05) newtech ( The regression method is ordered logit with survey weights that match the gender-age distribution of the adult population between 18-64 years in each country-year pair. Standard errors in parentheses with *** p<0.01, ** p<0.05, * p<0.1. The definitions of all variables are provided in Appendix Table A1, with expint being the dependent variable. All regressions include country, industry and year fixed effects. Cut points are omitted but are all significant.     Table A1, with expint being the dependent variable. All regressions include country, industry and year fixed effects. Cut points are omitted but are all significant. Nascent entrepreneurs are "respondents (18-64) involved in nascent business, defined as active, expect to be a full or part owner, and no salaries or wages paid for over three months".   The regression method is ordered logit with survey weights that match the gender-age distribution of the adult population between 18-64 years in each country-year pair. Standard errors in parentheses with *** p<0.01, ** p<0.05, * p<0.1. The definitions of all variables are provided in Appendix Table A1, with expint being the dependent variable. All regressions include country, industry and year fixed effects. Cut points are omitted but are all significant.
Economics: The Open-Access, Open-Assessment E-Journal 13 (2019-53) www.economics-ejournal.org 45 The regression method is zero-inflated probit with survey weights that match the gender-age distribution of the adult population between 18-64 years in each country-year pair. Standard errors in parentheses with *** p<0.01, ** p<0.05, * p<0.1. The definitions of all variables are provided in Appendix Table A1, with expint being the dependent variable. All regressions include country, industry and year fixed effects. Cut points are omitted but are all significant.   Table A1, with expint being the dependent variable. All regressions include country, industry and year fixed effects. Cut points are omitted but are all significant.