Historical roots of innovative entrepreneurial culture: the case of Italian regions

ABSTRACT This paper investigates persistence in entrepreneurship across Italian provinces over the period 1927–2017. We distinguish between a ‘general’ dimension (sociopsychological) and a ‘specific’ dimension (learning) of the role model effect. Based on data obtained from a comprehensive survey conducted in 1927 by the Italian National Statistical Institute, we find different drivers of persistence across sectors: the only driver of persistence in high-technology sectors is the share of entrepreneurs using motive power (‘specific’ role model effect), while the share of industrial entrepreneurs (‘general’ role model effect) is an important driver of persistence in low-technology manufacturing and service sectors.


INTRODUCTION
Recent research has documented marked persistence patterns of regional levels of entrepreneurial activity over time in countries such as Germany (Fritsch & Mueller, 2007;Fritsch & Wyrwich, 2014), Sweden (Andersson & Koster, 2011), the Netherlands (van Stel & Suddle, 2008), the UK (Fotopoulos, 2014;Fotopoulos & Storey, 2017) and the United States (Glaeser et al., 2015). We investigate here the historical roots of contemporaneous provincial differences in new business formation in Italy. In particular, we study whether the persistence of entrepreneurship over time depends on the intensity of the historical technology of the various industries and if there are differences in the determinants of persistence between sectors with different technological intensity.
Italy represents an interesting case study for several reasons. The country experienced the 1929 world economic crisis, the Second World War and the end of Fascism, the introduction of a new constitutional base, and the reconstruction of the economy. The post-war Italian governments adopted a very 'dirigiste' industrial policy aimed at shaping the sectoral and geographical distributions of industrial production. They promoted an intense subsidization programme in favour of southern regions, a governmental infrastructural plan with a particular emphasis on roads, with the aim of supporting the national motor vehicle industry, and a selective credit policy based on a strict regulation that imposed administrative portfolio constraints on the banking system's activity. All these measures, which were aimed at affecting the geographical distribution of industrial production and its sectoral composition, make it highly unlikely that persistence in Italian provincial entrepreneurship is the result of persistence in its determinants. Therefore, Italy may be considered a suitable context in which to study whether entrepreneurship is the result of a self-reinforcing process (Minniti, 2005), that is, if the present start-up rates are higher where entrepreneurs in the past began to transmit their 'models' and their knowledge to other potential entrepreneurs. A region with high levels of entrepreneurship may in fact encourage new entrepreneurial initiatives because it is easy to find an appropriate example or obtain information from other entrepreneurs (Davidsson & Wiklund, 1997;Mueller, 2006), thus the current differences in entrepreneurship among regions may depend on the historical availability of 'role models' (Fornahl, 2003;Lafuente et al., 2007;Sternberg, 2009).
We extend earlier works in two ways. First, we introduce, for the first time, to the best of our knowledge, a measure of the historical technological intensity of regional entrepreneurship in Italy and investigate its effect on the current start-up rates. Thanks to the Industrial Census (Censimento Generale dell'Industria e dei Servizi) published in 1932 by the Italian National Institute of Statistics (ISTAT), provincial data are available on the percentage of industrial firms that used motive power in 1927. The use of motive power reflected the diffusion of advanced production technologies in Italian industry, as also pointed out by Italian statisticians who have commented on the Italian Industrial Census, and it may therefore be a good proxy of the propensity of entrepreneurs to adopt new technologies at the beginning of the 20th century.
Second, referring to the literature that focuses on the influence of entrepreneurial role models on the decision to start a firm, we distinguish between a 'general' and a 'specific' dimension of the role model effect. The 'general' dimension is related to the rise of individual self-confidence and of social acceptance of entrepreneurship caused by the observation of the success of other entrepreneurs in a region. This is the sociopsychological dimension of role modelling that Obschonka et al. (2015) also emphasized. The larger the number of successful entrepreneurs in a region, the larger the number of individuals who perceive entrepreneurship as a favourable career option because they think, 'if they can do it, I can do it too' (Sørenson & Audia, 2000). The 'specific' dimension of the role model effect is that linked to learning and knowledge (Bosma et al., 2012;Gibson, 2004). The knowledge transmitted through social contact with an entrepreneur is supposed to build a local culture that predisposes entrepreneurs to act upon new knowledge. Innovating entrepreneurs may create a 'specific' role model that fosters the development of important capabilities, such as the recognition of entrepreneurial opportunities and the absorptive capacity for new knowledge (Qian & Acs, 2013).
In this framework, the present paper investigates the effects of the provincial entrepreneurship rate ('general' role model effect) and of the share of entrepreneurs who used motive power in 1927 ('specific' role model effect) on the current provincial start-up rates  in Italy. We examine the impact of the effects of the two role models separately for the high-and medium-hightechnology manufacturing (HTM) sectors, for the low-and medium-low-technology manufacturing (LTM) sectors and for the service sector. Interestingly, we find evidence of a significant 'specific' role model effect, but not of the 'general' role model effect, on the current level of new business formation in the HTM sectors. On the other hand, we find evidence of both effects in the LTM sectors. We only find evidence of the 'general' role model effect for the service sector. This suggests that the persistence of entrepreneurship over time has different drivers in different sectors. Where technology is very complex, as in the HTM sector, the effect of the general role model may be too weak to increase self-confidence and to sufficiently reduce the fear of failure, while regional start-up rates are likely to depend to a great extent on the presence of a 'specific' kind of entrepreneurial culture, that is, on the knowledge accumulation process activated by entrepreneurs who, in the past, adopted more advanced production technologies.
The remainder of the paper is structured as follows. The next section presents the conceptual framework and formulates the testable hypotheses. The third section presents the data and the methodology. The fourth section discusses the results. The last section concludes.

CONCEPTUAL FRAMEWORK AND TESTABLE HYPOTHESES
Many studies have investigated individual entrepreneurial behaviour in its regional context and, in particular, the mutually reinforcing nature of entrepreneurship (for a survey, see Fritsch & Storey, 2014). On the one hand, persistence in the entrepreneurship rates may be explained by the fact that regional determinants of new business formation tend to remain constant over time; on the other hand, it may be the result of the presence of an 'entrepreneurial culture' that only changes over a long period of time (Fritsch & Falck, 2007;North, 1994;Williamson, 2000). In particular, regions may be more likely to have higher start-up rates today simply because past entrepreneurship fostered the creation of a regional environment in which one person's decision to start a successful business encouraged others to follow (Chlosta et al., 2012;Fornahl, 2003;Lafuente et al., 2007;Minniti, 2005;Nanda & Sørenson, 2010;Stuetzer et al., 2016;Wagner & Sternberg, 2004). Entrepreneurial culture may be defined as an informal institution that is 'in the air' in some regions more than in other ones. It is higher where the population has a positive attitude toward entrepreneurial activity, shares entrepreneurial values, such as individualism and autonomy, contains a high number of people who are able to bear risks and many entrepreneurial role models that generate demonstration and peer effects (Sørenson, 2017). The presence of entrepreneurial role models that weaken the social stigma of entrepreneurial failure can also reduce the fear of failure and therefore increase regional entrepreneurship rates (Vaillant & Lafuente, 2007;. Gibson (2004) argued that the term 'role model' draws on two prominent theoretical constructs, since individuals are attracted to role models who are perceived to be similar in terms of their characteristics, behaviour or goals (the 'role' aspect), and from whom they are able to learn certain abilities or skills (the 'model' aspect). Bosma et al. (2012) argued that entrepreneurial role models may perform four interrelated functions: (1) inspiration and motivation (i.e., the role model creates awareness and motivates people to get started); (2) increasing self-efficacy (i.e., the role model makes people confident that they too can achieve a certain goal); (3) learning by example (i.e., the role model provides guidelines for action); and (4) learning by support (i.e., the role model provides hands-on support or advice). The first two functions result (indirectly) from the role identification theory, whereas the third and fourth are implied by the social learning theory and are related to knowledge transmission.
Among the studies on the relationship between knowledge and entrepreneurship, Qian and Acs (2013) introduced the concept of entrepreneurial absorptive capacity, defined as a critical factor that affects the process of transmitting knowledge spillovers by entrepreneurs. It allows entrepreneurs to understand new knowledge, recognize its value and commercialize it by creating a firm. Since knowledge tends to be 'sticky' in space, the regional knowledge base may represent a determinant of regional innovative start-up rates (Acs et al., 2009). This knowledge may have different sources, including the educational attainment of the workforce, the presence of private and public research, and the work experience of the population (Helfat & Lieberman, 2002). Del Monte and Pennacchio (2020) studied the historical factors that shape the current levels and regional differences in new firm formation in innovative industries in Italy and found that regional knowledge and creativity affect entrepreneurial activity in the long term. Furthermore, Obschonka et al. (2015) outlined that regional differences in entrepreneurial culture may be a hidden key factor in moderating the relationship between knowledge and economic activity; specifically, people in some regions may live in a local culture that enables them to act upon new knowledge by starting a new business. Consistently, Fritsch and Wyrwich (2018) found that the interaction of the historical knowledge base with an entrepreneurial tradition has an enduring effect on the formation of innovative new businesses in Germany.
We argue that the historical presence in a region of innovative entrepreneurs may represent a particular kind of role model that fosters the absorptive capacity for new knowledge, and that new knowledge may therefore need to be interacted with the historical innovative entrepreneurial tradition to stimulate the birth of innovative new firms. Following the simple sender-receiver model of , an entrepreneur (sender) may transmit two signals to an observing non-entrepreneurial peer (receiver): (1) a 'general' signal, that is, that entrepreneurship is attractive and feasible (sociopsychological dimension); and (2) a 'specific' signal, that is, how to become an entrepreneur, and how to organize the resources and the production technology that has to be adopted (learning dimension). The receiver chooses to either engage in entrepreneurship or opt for paid employment. Since sending and receiving information may be considered as a local working process, entrepreneurship may become self-perpetuating over time. As stated in the introduction, we use the historical entrepreneurship rate (number of entrepreneurs over the population) to represent the 'general' signal ('general' role model effect) and the historical share of entrepreneurs who used motive power (number of entrepreneurs who used motive power over the total number of entrepreneurs) to represent the 'specific' signal ('specific' role model effect). On the basis of these arguments, we test the following hypotheses: Hypothesis 1a. Nowadays provincial start-up rate depends on the historical entrepreneurship rate ('general' role model effect).
Hypothesis 1b. Nowadays provincial start-up rate depends on the historical share of entrepreneurs who used motive power ('specific' role model effect).
We test these hypotheses separately for the HTM, LTM and service sectors in order to investigate whether persistence in entrepreneurship over time depends on the technology intensity of the various industries, and if there are differences between the manufacturing and the service sectors. In the case of German regions, Fritsch and Wyrwich (2016) found that not all types of self-employment are equally important for the long-term persistence of regional entrepreneurship: while the overall historical level of selfemployment is only weakly related to the current level of new business formation, the share of self-employed males in knowledge-intensive industries has long-lasting effects. We ask whether the differences in the determinants of persistence across industries are related to the complexity of the knowledge base. In particular, in sectors with complex technologies, regional start-up rates are likely to depend to a great extent on the presence of a 'specific' kind of entrepreneurial culture, that is, on the knowledge-accumulation process activated by entrepreneurs that in the past adopted more advanced production technologies, more than on a 'general' entrepreneurial tradition. Where technology is particularly complex, as in the HTM sector, entrepreneurship requires knowledge transfer, and the effect of the general signal may be too weak to increase self-confidence and to reduce the fear of failure to a sufficient extent to induce the receiver to decide to engage in entrepreneurship. Entrepreneurial activity in less technologically advanced sectors requires less specific knowledge and results in a lower degree of risk. Therefore, the creation of new LTM and service firms may be fostered simply by the presence of role models that make people confident that they too can achieve a certain goal, that is, the effect of the general signal may be strong enough to foster entrepreneurship. 1

Data sources
The empirical analysis was based on a dataset that combined different sources of data. The analysis units were Italian provinces, according to the NUTS-3 classification. By considering the changes that have taken place over time in the administrative partition of Italy, a total of 95 provinces were included in the sample. The data sources were the Italian National Institute of Statistics (ISTAT) and the Business Register of the chambers of commerce, the latter of which referred to the number of start-ups in the 2000s. The industrial census cited above (Censimento Generale dell'Industria e dei Servizi), published in 1932 when the famous statistician Corrado Gini (1884Gini ( -1965 was head of ISTAT, is an important source of historical data. Early in the survey, Gini defined the 1927 Census Historical roots of innovative entrepreneurial culture: the case of Italian regions 1685 as 'a grandiose experiment in the field of economic statistics', since, for the first time ever, a complete survey had been carried out of the industrial and commercial life of the nation. The census contains detailed data about the number of establishments, entrepreneurs and employees in Italy, and the use of motive power. We used the related classification adopted by Eurostat 2 for the analyses on the technology intensity of the manufacturing sectors (HTM: high and medium-high; LTM: low and medium-low 3 ). A detailed description of all the variables (definitions and sources) and of the summary statistics is reported in Table 1.

Dependent variable
We estimated the start-up rate at the provincial level (for the period 2000-17) for different groups: (1) the high-and medium-high-technology manufacturing sector (HTM); (2) the low-and medium-low-technology manufacturing sector (LTM); (3) the total manufacturing sector (Manuf); and (4) the total service sector (Services). The start-up rate is given by the number of start-ups divided by the active population (which represents the group of potential entrepreneurs in the population) in each province (NUTS-3). Start-ups are measured by newly founded businesses irrespective of the size, operating in all manufacturing and services sectors, according to the entire Business Register of the chambers of commerce. 4 Figure A1(c) in the supplemental data online shows 'three Italies' with respect to the start-up rate in the HTM sector in the period 2000-17: most of north-west Italy, and the northern Adriatic and central provinces are characterized by high levels, while the remaining central provinces are characterized by intermediate levels and most southern provinces by low levels. However, the map also shows important variations between provinces within the macro-areas.

Independent variables
In order to test Hypotheses H1a and H1b, we used our proxy of the 'general' role model effect, that is, the ratio between the number of entrepreneurs in 1927 and the population in 1931 5 in industry (Histentr_ind) and in services (Histentr_ser), and our proxy of the 'specific' role model effect, that is, the ratio between the number of entrepreneurs who used motive power and the total number of entrepreneurs in 1927, in industry (Histmpower_ind) and in services (Histmpower_ser) as independent variables.
Italy was still an agrarian country at the time of its unification in 1861. The Italian industrial system in 1871 was mainly made up of home-based manufacturing and artisan work, which were carried out in the cities to satisfy the demand of the local population (Cafagna, 1989). During the 1880s, the Italian economy entered a phase of intensive growth, fostered by both the unitary state and the introduction of new energy sources, especially imported coal and hydroelectricity (Malanima, 2013(Malanima, , 2016. Thanks to the changes in the energy exploitation paradigm, the capital per worker began to increase rapidly and, with it, productivity and per capita GDP. The first wave of Italian industrialization was witnessed between 1871 and 1911: industrial production rose by 50% in 40 years and then remained substantially stable until 1936 (Daniele & Malanima, 2011;Baffigi, 2013Baffigi, , 2015. 6 As stated in the industrial census of 1927, the percentage of Italian industrial firms that used motive power reflected the degree of technological advancement of the adopted production methods, which differed to a great extent among sectors. The percentage of industrial firms that used motive power in Italy was on average 17.6%, but there were significant differences between the more technologically advanced northern area (24.7%), the centre (15.6%) and the more backward areas of the south (8.5%) and islands (7.2%). The data reported in Table A2 in the supplemental data online show that, in the south, manual work prevailed in industries in which the use of motive power was more prevalent in the north (such as in the textile, paper and mechanical industries). Furthermore, only about 10% of the mechanical industries used motive power in the south, against about 35% in the north.
Consistently with the data reported above, Figure A1 (b) in the supplemental data online shows a dualistic Italian economy, where the share of entrepreneurs who used motive power in the western part of northern Italy is much higher than the average, while most Adriatic and central provinces show intermediate rates. Figure A1 (a) online, which shows the total industrial entrepreneurship rate for 1927, describes a far less dualistic country.
In 1927, the northern province of Varese was the area with the highest share of entrepreneurs who used motive power in Italy. The evolution of its industries represents an interesting example of the link between historic and contemporary variables. Industrialization in this area started at the end of the 18th century with silk spinning and continued in the 19th century with the impressive development of cotton-textile and mechanical industries. It continued in the 20th century with an exemplary flowering of manufacturing in several production sectors, such as the rubber-plastic, chemistry, food, wood and building material sectors. Entrepreneurs did not hesitate to commit to pioneering initiatives, such as the introduction of the mechanical spinning and weaving of cotton, and later on to cutting-edge productions in the mechanical-engine sector and, subsequently, to futuristic projects in the aviation field. Today, we also witness the entry into the aerospace system supply chain. The province of Varese is now one of the main industrial districts in Italy, but this is in part due to its ancient passion for anything new, which is also confirmed by the large share of firms that used motive power in 1927.

Controls
We included in the estimates control variables pertaining to both the current period and to the past. As far as the current period is concerned, we controlled for the local economic development, as measured by per capita value added (Va), and for the 15-64-year-old population (Pop), which may be used to represent the pool of potential entrepreneurs (Fritsch & Falck, 2007). Moreover, we Historical roots of innovative entrepreneurial culture: the case of Italian regions controlled for agglomeration economies by including the population density (Popdens) for regional knowledge, as measured by means of research and development (R&D) expenditure as a percentage of GDP (R&D), and for the quality of human capital, by including the share of employees with a tertiary degree (Tertiary).
As for the past, we controlled for historical differences among provinces in the socio-economic development level and in the industrial structure. Among historical socioeconomic variables, we included in our estimates per capita value added in 1931 (Histva), population in 1931 (Histpop) and the share of illiterate people out of the total people aged six years and over in 1931 (Histilliteracy). Furthermore, we controlled for historical provincial differences in the agglomeration by including the population density in 1921 (Histpopdens), and in the development of the financial system by including the share of firms in the financial sector (Histfirms_cred). Finally, we included a dummy in our estimates for provinces with a long tradition of academic institutions (i.e., that had a university in 1895), which may have fostered the accumulation of large stocks of knowledge over time (Fritsch, 2011). We High motive power industries are those in sectors with a Histmpower_ind higher than the mean level (25%). b High motive power services are those in the public performances sector (characterized by a Histmpower_ser of 41.9%, which is much higher than the average across services at 3%).
also included two variables to measure the ratio between the number of female employees and the total number of employees in industry (Histfemale_ind) and in services (Histfemale_ser) in 1927, since individuals whose parents were both employed may have been able to obtain more financial and psychological support to start a new business. In order to control for differences in the provincial industrial structure, we included a variable in the estimates to measure the historical presence of large firms, that is, the number of firms with more than 250 employees over the total number of firms in 1927 (Histfirms_large). On the one hand, the high division of labour in largescale industries may have reduced the availability of a skilled labour force and entrepreneurial spirit; on the other hand, institutions may have been tuned to the needs of firms in large-scale industries and not the needs of smaller and younger firms (Glaeser et al., 2015;Stuetzer et al., 2016). In addition, we included the historical share of firms in the manufacturing and service sectors on the total economy (Hist_manuf and Hist_ser, respectively), the share of HTM (Hist_HTM) and LTM (Hist_LTM) on the total economy, and the share of firms belonging to sectors in which the use of motive power was higher than a certain threshold out of both the total industry (Histmpowersett_ind) and the total services (Histmpower-sett_ser). 7 Finally, we controlled for differences in the geographical location: north-west (nw); north-east (ne); south (so) (centre is the reference category) and over time (by introducing year dummies).

Estimated equation
Equation (1) was estimated for the total manufacturing and service sectors and for different groups of sectors (total manufacturing, HTM and LTM sectors, total services): It was estimated using the proxy to capture the existence of the 'general' role model effect (Histentr) and the proxy for the 'specific' role model effect (Histmpower). The dependent variable and a part of the explanatory variables vary across both the provinces (n ¼ 95) and over time , while the historical variables are mostly measured with reference to 1927 (Table 1).
Since the main variables of interest are the historical data of 1927, equation (1) cannot be estimated using fixed effects. We therefore estimated a random effect (RE) model (Wooldridge, 2011) with clustered standard errors, which exploits the variability of the time varying dependent and independent variables. However, in order to take into account the issue of unobserved heterogeneity, we checked for robustness of the results using a correlated random effect model (CRE) in which the group means of the variables that varied within groups were added to regressors (the so-called Mundlak correction). We introduced this correction in order to combine the advantages of both random and fixed effects. This technique was proposed by Mundlak (1978) as a way of relaxing the assumption in the RE estimator that the observed variables are uncorrelated with the unobserved variables (Bell & Jones, 2015;Mavromaras et al., 2013). The introduction of means should capture the correlation between the unobserved heterogeneity and the covariates that could render the RE model inconsistent. Finally, to account for a possible spatial correlation of the dependent variables across provinces, we introduced robustness checks by estimating a spatial lag model. 8 Table 2 reports the results of the estimation of equation (1) using RE and CRE models for the HTM sectors (columns 1 and 2), LTM sectors (columns 3 and 4), total manufacturing (columns 5 and 6) and total services (columns 7 and 8), respectively. We report estimations for each group of sectors in which we included our proxies as regressors to capture the existence of the 'general' role model effect (Histentr) and of the 'specific' role model effect (Histmpower).

RESULTS
The results clearly show that new business formation depends significantly on the historical entrepreneurship rate (Histentr_ind and Histentr_ser) in the LTM and service sectors (Table 2, columns 3, 4, 7 and 8), but not in the HTM sectors (columns 1 and 2). Therefore, the evidence does not support Hypothesis 1a for the sectors with complex technologies. This suggests that the 'general' signal, that is, that entrepreneurship is attractive and feasible, might not be sufficient to increase self-confidence and reduce the fear of failure in sectors where knowledge is more complex and the perceived risk is therefore higher. We have found evidence in support of Hypothesis 1b for both the HTM and LTM sectors: new business formation depends significantly in both sectors on the share of entrepreneurs who used motive power (Histmpower_ind). On the other hand, we have not found evidence of a 'specific' role model effect for the service sector.
These results suggest that persistence in entrepreneurship varies according to the complexity of knowledge that is transmitted over time. Today, there is more HTM entrepreneurship in those Italian provinces where in the past there was a higher share of innovating entrepreneurs, and not simply where there were 'more' entrepreneurs. A possible interpretation of this result is that the historical presence of innovative entrepreneurs, which fostered the absorptive capacity for new knowledge, played a crucial role in stimulating the birth of innovative new firms. We have also found evidence of the 'specific' role model effect in the LTM sectors. The importance of the specific role model effect for medium and LTM sectors can be explained by considering that some of these sectors, such as the food, textile and paper industries, had relatively high percentages of entrepreneurs who used motive power in 1927, especially in the northern and central Italian provinces (see Table A2 in the supplemental data online). Finally, as stated above, we have not found any     Historical roots of innovative entrepreneurial culture: the case of Italian regions evidence of a 'specific' role model effect in the service sector. Considering that the service activities in 1927 consisted mainly of wholesale, retail and trade services, it is clear that such activities required a limited degree of knowledge complexity and capital intensity, and therefore resulted in a limited degree of risk. On the other hand, manufacturing activities (including the more traditional ones, such as textiles and food industries) required a higher degree of knowledge complexity and the use of specialized employees. We must outline that motive power might be a good proxy for the learning dimension in manufacturing but not in services. The use of motive power in services was naturally very limited.
We have found an interesting result (which is robust across the estimation methods in the HTM industries) when considering the historical control variables: the larger the share of female employees was in 1927 (Histfema-le_ind and Histfemale_ser), the higher the new business formation is nowadays. A possible interpretation of this result is that working mothers, encouraged by the success of their employers, may in turn have encouraged their sons to also become entrepreneurs. However, the existence of such an indirect role model effect requires further investigation in future studies. A complementary explanation of the positive effect of the share of female employees in 1927 on entrepreneurship may be that individuals whose parents were both employed had more financial and psychological support to start a new business. Finally, the historic female employment rate could be a proxy of a less patriarchic society. The persistence of this type of culture may lead to increases in the supply of entrepreneurs to high tech sectors. 9 We can notice that the variables Histfemale_ind and Histfemale_ser might represent also a proxy of the historical strength of those sectors employing a larger share of women after the First World War. Further research should investigate these possible sectoral effects.
Regarding the industrial sector, we have surprisingly found that the current HTM entrepreneurship rate is negatively influenced by the past per capita income (Histva). However, this result should be considered with caution because we have also found a positive significant effect of current per capita income (Va) on the dependent variable. This means that, for a given current per capita income, those Italian provinces that were poorer in the past now register a higher current innovative entrepreneurship rate and that, therefore, in the manufacturing sector, the most innovating provinces are probably those that historically were more dynamic and more able to converge. The current industrial entrepreneurship rates are positively affected by the past and current population densities. Finally, we have found that, in manufacturing, entrepreneurship rates are higher where employees were not concentrated in large firms in the past (as predicted by the literature) and where education and R&D expenditure is now greater. We have found, together with an expected positive effect of the current tertiary education, an unexpected positive effect of the historical illiteracy rate on the entrepreneurship rate for the LTM (in the CRE model). A possible explanation of this result may be that the growth of traditional LTM sectors was driven in many Italian provinces, during the post war period, by new entrepreneurs skilled in exploiting the benefits of backwardness, that is, of a large availability of manpower at a low cost. 10 Another possible explanation is that historical illiteracy rates could lead to higher levels of necessity driven entrepreneurship contributing to foster entrepreneurial role models.
The current provincial start-up rates in the service sector depend positively and significantly on both the current and past per capita income (Table 2, columns 7 and 8). We have found, as expected, a significant positive effect of the current population density, but also a significant negative effect of the past population density on the current entrepreneurship rates. The decision to start a commercial activity in a specific place depends to a great extent on the market opportunities, and the literature on the localization of knowledge-intensive services (KIS) has emphasized the role played by agglomeration economies and urban areas (Ciarli et al., 2012;Polese & Shearmur, 2004;Rodriguez-Pose & Crescenzi, 2008;Shearmur & Doloreux, 2009). The two results could suggest that those provinces that underwent a more relevant urbanization process in the past decades are now those characterized by higher entrepreneurship rates. Moreover, we have found, for the service sector, that the past illiteracy rates positively affect the current entrepreneurship rates.
The results of the spatial model reported in Table 3 confirm the main findings reported above, and show that although there is a significant spatial dependence in the service sector, this is not the case in the HTM and LTM sectors. It seems that entrepreneurship tends to remain a local phenomenon in the manufacturing sector, where knowledge is more complex, while entrepreneurship tends to spread to neighbouring provinces in the service sector, where we have not found any evidence of a 'specific' role model effect.

CONCLUSIONS
In this paper we have investigated persistence in entrepreneurship across Italian provinces over the period 1927-2017 by focusing on the role of past innovative entrepreneurship in fostering the present start-up rates. We have found evidence of the crucial role of this particular kind of entrepreneurship in building a ground that is able to foster the start-up rate of innovating manufacturing firms. We have also found that the current start-up rate in the HTM sector depends significantly on the historical propensity of provincial entrepreneurs to innovate, that is, on the share of entrepreneurs who used motive power (the 'specific' role model effect), and not on the historical manufacturing entrepreneurship rate (the 'general' role model effect). On the other hand, we have also found that the 'general' role model effect is an important driver of entrepreneurship in both the LTM and service sectors. According to our interpretation of these results, such differences show that persistence in the HTM sector Historical roots of innovative entrepreneurial culture: the case of Italian regions cannot be explained as just simple imitation or identification effects or as a general 'culture' of entrepreneurship.
Our results support the hypothesis that learning plays an important role in increasing self-confidence and reducing the fear of failure in sectors where knowledge is more complex and, therefore, the perceived risk is higher. This is in line with the results of Fritsch and Wyrwich (2016), who found that not all types of self-employment are equally important for the long-term persistence of regional entrepreneurship. Since the transmission of competencies requires physical proximity, collective learning at the territorial level may be the basis of a sustained regional competitive advantage. According with our theoretical framework, the higher the riskiness of productive sectors, the more role models' effects are important. Our results suggest that although 'specific' role model effects are crucial in riskier productive sectors (HTM) to induce individuals to become entrepreneurs, 'general' role model effects may be strong enough to decide to become entrepreneurs in those sectors that make use of traditional technology (LTM and services) in which the perceived risk is lower. Overall, what emerges from this study is that the more complex knowledge is, the more entrepreneurship requires role models based on learning rooted in history. The distribution of entrepreneurs who used motive power in 1927 for advanced production technologies may be considered as a proxy of the historical distribution of innovative entrepreneurship in the country. Even after about 90 years, this distribution still affects new business formation, even though the Italian post-war governments adopted a very 'dirigiste' industrial policy aimed at shaping the sectoral and geographical distribution of industrial production.
Our analysis suffers from some limitations and some aspects deserve in-depth exploration in future research. For example, it is unclear to what extent the share of firms that used motive power is a proxy of a specific role model effect linked to the knowledge transfer within specific sectors or rather a proxy of the innovation propensity of entrepreneurs, which may have an effect across different sectors. Further research should test such an effect using historical sectorial data. In addition, innovative start-up rates could be used as a dependent variable for different sectors. In particular, further research might identify other possible proxies supporting the presence of a specific role model effect in services, not linked to technological knowledge. The 'specific signal' sent to potential entrepreneurs in the service sector might be about how to organize resources or about general knowledge that may not be proxied by the use of motive power. This analysis could also help understanding whether the specific effects are more about risk (which is higher in capital intensive activities) or about complexity (which is higher in knowledge intensive activities). Another factor that needs further analysis is what we have named the 'working mother effect'. Did women employed in 1927 foster total entrepreneurship and female entrepreneurship in particular? It would be interesting to investigate this effect separately for the different sectors. Finally, a limitation of our study is that it has not found an answer to what can be done to improve regional innovating entrepreneurship. Our paper supports the evidence that past attitudes toward using advanced technologies vary considerably by region and are associated with current innovating entrepreneurial activity. Further analyses should be conducted to study whether attitudes toward innovating entrepreneurship are a cause or an outcome. As argued by Fritsch and Storey (2014), if they are a cause, then they should be addressed directly, perhaps through awareness-raising of the benefits of innovating entrepreneurship. If, instead, they are an outcome, then the valid policy response should be to address economic disadvantage issues, thereby directly Standard errors clustered at the provincial level are shown in parentheses ***p < 0.01, **p < 0.05, *p < 0.1.
improving economic conditions. Our paper points out the need to consider the 'historical entrepreneurial culture endowment' of a region in industrial plans. A policymaker should be like a good gardener who studies what plants grew in a garden in the past, as he/she is aware that those plants are the most likely to find an environment that is able to foster their growth. The next step should be to find out what can be done to improve that environment, which seems now so much predetermined by history.