DOCUMENT DE TREBALL XREAP2009-7 The Black Box of Business Dynamics María Callejón (PPRE-IREA) Vicente Ortún THE BLACK-BOX OF BUSINESS DYNAMICS María Callejón* (UB) and Vicente Ortún** 1 September 2009 Adreça correspondència: Maria Callejón Facultat d’Economia i Empresa Avda. Diagonal, 690 Despatx 6314 08034 Barcelona maria.callejon@ub.edu 1 This paper has benefited from the financial support of ECO2009-08735 (Spanish Ministry of Science and Innovation). It has benefited also from both authors’ period (2009) as visiting scholars at Harvard University and MIT respectively. Professor of Applied Economics, University of Barcelona. ** Professor in the Department of Economics and Business, Universidad Pompeu Fabra, Barcelona. * Abstract Research in business dynamics has been advancing rapidly in the last years but the translation of the new knowledge to industrial policy design is slow. One striking aspect in the policy area is that although research and analysis do not identify the existence of an specific optimal rate of business creation and business exit, governments everywhere have adopted business start-up support programs with the implicit principle that the more the better. The purpose of this article is to contribute to understand the implications of the available research for policy design. Economic analysis has identified firm heterogeneity as being the most salient characteristic of industrial dynamics, and so a better knowledge of the different types of entrepreneur, their behavior and their specific contribution to innovation and growth would enable us to see into the ‘black box’ of business dynamics and improve the design of appropriate public policies. The empirical analysis performed here shows that not all new business have the same impact on relevant economic variables, and that self-employment is of quite a different economic nature to that of firms with employees. It is argued that public programs should not promote indiscriminate entry but rather give priority to able entrants with survival capacities. Survival of entrants is positively related to their size at birth. Innovation and investment improve the likelihood of survival of new manufacturing start-ups. Investment in R&D increases the risk of failure in new firms, although it improves the competitiveness of incumbents. Resumen: La investigación en dinámica industrial empresarial ha avanzado con rapidez en los últimos años, pero la traslación de los nuevos conocimientos hacia el diseño de la política económica es todavía escasa. Un aspecto llamativo relativo a la política hacia el entorno productivo es que aunque la investigación y el análisis no identifican la existencia de alguna tasa óptima específica de creación de empresas ni de salida de empresas, los gobiernos de muchos países han adoptado programas para el apoyo de la creación de empresas bajo el principio implícito de que más es mejor. La heterogeneidad empresarial ha sido identificada por el análisis económico como la característica más sobresaliente de la dinámica industrial y, por tanto, un mejor conocimiento de las diferentes tipologías de emprendedor, de su comportamiento y contribución específica a la innovación y el crecimiento, nos permitiría abrir “la caja negra” de la dinámica empresarial, y nos ayudaría a mejorar el diseño de las políticas ii públicas. El análisis empírico realizado en el presente trabajo señala que todas las empresas nuevas no tienen el mismo impacto sobre variables económicas relevantes, y que el autoempleo presenta una naturaleza económica muy diferente de las empresas con empleados. Se defiende que los programas públicos no deberían promover la entrada indiscriminada sino priorizar los entrantes con mayor capacidad empresarial y mayor probabilidad de sobrevivir. La supervivencia de las nuevas empresas se encuentra relacionada positivamente con su tamaño al nacer. El esfuerzo innovador e inversor mejoran la probabilidad de sobrevivir de las nuevas empresas manufactureras. La intensidad en I+D aumenta el riesgo de cierre de las nuevas empresas, pero muy probablemente mejora la competitividad de las empresas una vez establecidas. JEL Classification: D21, L16, L52, M13, O25. Key words: Industrial dynamics, industrial policy, creative destruction, business demography. iii Introduction Many governments refer to new venture creation as one of the goals of their policy for creating jobs and improving competitiveness. International organizations issuing economic policy recommendations, including the European Commission, the OECD and the World Bank, report favorably on the strengthening of entrepreneurial initiatives in public programs. Yet the analytical bases for such programs are not always entirely clear. More often than not the programs fail to take into consideration, or are quite unaware of, the nature and contents of existing research that has been undertaken in business dynamics and its impact on efficiency. This article seeks to add to thinking on the contribution research in business dynamics can make to industrial policy. This study comprises two parts. The first identifies the main lines of research in industrial dynamics developed to date and their contribution to knowledge in the field of economic policy design. It does not seek to undertake a review of the literature and its findings, rather its aim is to identify and differentiate analytical approaches and their implications. It is argued that aggregate analyzes of the impact of business creation on growth fail to provide any substantive insights, since the heterogeneity of entrants means that the effects of each new firm on aggregate efficiency is very different. What is needed is for the “black box” of business dynamics to be opened up so as to know how and when the creation and destruction of businesses improves economic efficiency and well-being. The key seems to be that the more innovative a new firm is, the greater its contribution will be to society within a competitive environment. The second part comprises an empirical study undertaken using databases available for Spain. The aim here is twofold. First, we seek to show the 1 difference between self-employment (those who work for themselves without hiring any employees) and businesses (where the hiring of employees is indicative of the existence of a project with expectations of permanence and growth). It should be clear to governments that programs fostering self-employment are not the same as those that foster the creation and growth of firms. The other aim of the empirical analysis considers survival as an observable characteristic that enables us to appraise the efficiency of a new firm. By applying Cox’s proportional-hazards regression we determine the effects that certain individual or sectoral variables might have on a firm’s survival in the years following start up. Government programs would be more efficient if they strove to maximize the survival and growth rates of new start-ups, rather than indiscriminately seeking to maximize rates of entry. 1. Contributions from the field of business dynamics to the design of economic policy. 1.1 How much business turbulence? A healthy economy is dynamic. One of the implications is that new innovative firms replace those that have become inefficient or obsolete. This dynamism entails a lot of turbulence when firms are simultaneously entering and exiting the market. If the number of new innovative firms is too small, medium and long-term growth will be too low and the country will lag behind. It may also happen that, if the economy experiences high entry of traditional non-innovative firms, the consequences may be no better if market competition is exacerbated, survival made too difficult, and a great number of firms are forced out of the market. Churning can be higher than optimal. 2 Given that the process of structural change involves benefits (new efficient firms) and costs (closures of firms), it is the experts’ task to analyze and to seek to determine the limits and specific conditions under which business dynamics has a positive impact on economic growth and social prosperity. Over the last decade, research into business dynamics has grown notably in quantity and quality. However, we are still some way from understanding the complex, noisy and confused phenomenon of business dynamics and so economists interested in these deep underlying forces face a stimulating challenge. A number of firms equivalent to 20 per cent of all existing firms are created and destroyed, enter and exit the markets each year. Between 20 and 40 per cent of new firms close within two years. In terms of employment, the figures are also notable. Calculations made for the period 1977 to 2005 in the United States (Haltiwanger et al. 2008) report that the mean annual rate of new job creation stands at almost 18 per cent, while the equivalent rate for job destruction stands at around 16 per cent. This means there is a need for the reassignment of 34 per cent of total employment to achieve a net increase in employment of less than 2 per cent each year. Of the new jobs created each year, 7 per cent approximately correspond to jobs generated by new start-ups and the rest is employment originated by firms that are growing. But the destruction of employment is equally high. Of the 16 per cent of jobs destroyed on average each year, around 6 per cent of these are lost following the closure of firms while the rest is attributable to the reduction in employment in existing firms. The above data are drawn from the new database, Business Demography Statistics (BDS), which is prepared by the US Census Bureau and which gives us a very clear picture of business demography and its growth in the 3 US. Other countries, including those in Europe (A1 Annex), present similar levels of turbulence but with the fundamental difference that in the US new firms grow more rapidly and that their survival rate is higher (Bartelsman et al. 2005). Given the intensity of the turbulence in business and employment it may seem odd to learn that, in the past, the dominant lines in economic research have paid relatively scant regard to it, with the obvious exception of Schumpeter and a part of the Austrian School. Fortunately, in recent years an abundant and sophisticated body of economic research has grown up in the field of business dynamics. A phenomenon of such complexity might be tackled from various perspectives. In this article we adopt two broad points of view. First we look at the great regularities observed in business demography and at the individual patterns of firm behavior which result in the regularities observed. Second, we look at specific analyses of the effect of business dynamics and certain types of new start-ups on efficiency and growth. 1. 2. Stochastic dynamic models Stochastic dynamic models analyze firm size and age distributions and seek to reveal individual behaviors and sectoral conditioning factors compatible with the regularities of observed distributions. A persistent characteristic of business demography in practically all studies is the simultaneous occurrence of strong spatial and temporal stability in the distribution of firm sizes, and yet great turbulence among the individual firms. The first proposal that sought to account for this dual phenomenon of turbulence and stability was provided by Robert Gibrat (1931). Gibrat’s Law, or the rule of proportionate growth, claims that the highly asymmetric 4 distribution observed in firm sizes – a multitude of small firms and just a few large ones – can be explained if the growth of each firm in each time period is proportional to its size. The result would be identical if the growth of each firm followed a random pattern. Gibrat’s Law has conventionally been represented by a normal distribution, presenting an inverted-U shape relationship of the firms’ growth rate. However, recent studies claim that a Laplace distribution, with an invertedV shape, provides a better fit of the data observed (Teitelbaum & Axtell, 2005). The difference between the two distributions is significant. The Laplace distribution implies that the tails of the growth rate distribution are somewhat denser than those in a normal distribution, and that there is a greater number of firms growing at intermediate rates than is the case with a normal distribution, where the bulk of firms grow at rates close to the mean and a small number grow either very slowly or very quickly. The consequences of this dynamic, which undoubtedly are of great interest, have yet to be analyzed in depth. Segarra et al. (2008) provide an excellent presentation and discussion of alternative density functions representative of the distribution of firms by size and by age. Teruel (2009), moreover, analyzes the growth of Spanish firms in relation to Gibrat’s Law. The idea that firms follow random growth rates is disturbing and barely compatible with usual economic models. It is assumed that agents, and in particular business managers, far from operating in a random fashion, take rational and maximizing decisions. And, yet, the profession should be grateful to Gibrat for his provocative and stimulating ideas, even more so as it has not been possible to rule out random growth in those firms that have survived their first few years of activity (Loti, et al. 2009) 5 The most recent research has been based on two observable, and systematic phenomena. First, that the market structure varies from one industry to another and, second, that the differences in market structure between industries are similar from one country to another. The intersectoral variation suggests that the specific patterns of entry, exit and behavior (survival, growth, innovation) are also related to the characteristics of each industry such as their size, the degree of scale economies, the prevalence of product differentiation, and the level of investment in I+D. These nonrandom approaches suggest that agents behave so as to optimize their opportunities, albeit that the great regularities highlighted by Gibrat continue to be found. The convergence of both theoretical approaches has been resolved, according to Sutton (1997), by introducing stochastic elements in maximizing models. The stochastic dynamic models that have had greatest repercussions in the subsequent literature are those proposed by Jovanovic (1982), Hopenhayn (1992), and Erikson & Pakes (1995). These models incorporate idiosyncratic firm characteristics that give rise to the individual variability observed. According to Jovanovic, firms might follow a pattern of passive learning in the sense that they learn to evaluate their own efficiency and so decide whether to grow, not to grow, to reduce their size or to close, but their levels of effectiveness do not change over time. A further, more complex vision is provided by the model of active learning proposed by Erikson & Pakes, where firms seek to improve their relative efficiency by investing in innovation. Here each firm can acquire either more, less or the same level of efficiency as its competitors, and the result is a Markov equilibrium. The theoretical development has continued with the proposals of the likes of Melitz (2003), and Asplund & Nocke (2006) who add specific differentiating elements to their stochastic models. 6 In Marc Melitz’s model the progressive opening up to trade and international competition forces the exporting firms to improve their productivity, resulting in the exit of the most inefficient, non-international firms; while Asplund & Nocke find that, in a context of monopolistic competition, the bigger the market grows the greater the rate of entry and the level of competition and the lower the survival rate. A common element in the above models is that the sum of agents’ decisions with markedly heterogeneous degrees of efficiency gives rise to stable distributions of sizes and growth rates. 1. 3. Business dynamics and efficiency Discovering the laws of behavior and the patterns underlying business demography constitutes an essential insight, but from the perspective of economic policy what is most urgent is understanding how the entry of new start-ups and the closure of incumbents can influence the economy, and what those responsible for economic policy can do to influence business dynamics and its effects. Existing studies seek to analyze the relationship between business demography and dynamics (entry, exit, growth, survival) and other economic variables such as: business cycle, sectoral growth, innovation, employment, productivity, competitiveness and structural change. Again two types of approach can be distinguished. The first approach is general, and includes studies that are concerned with the effect derived from new venture creation in general. The analysis of entrepreneurship belongs to this school. Other approaches are more selective in character, that is, they are concerned with the contribution that specific types of new 7 start-ups, in particular innovative or high-growth firms, make to growth and innovation. The economic models of new start-ups are in general in a phase of revision on finding that the enormous heterogeneity of entrants makes difficult the interpretation of the economic effects. Many existing empirical studies show that the relationship between the firms’ rate of creation and economic efficiency is not direct. The analysis of specific aspects of firm creation and entrepreneurship seems to have greater scientific potential. This shift can also be detected in the public policies of a number of countries where new programs are being oriented towards new firms that can grow (named as high growth firms), and towards new technology based firms from which it is expected an impact on the speed of innovation, on the improvement of the productive structure and on internationalization capacity. 1.3.1. The “black box” of business demography It is reasonable to assume that the efficient rate of business creation belongs to a range with both an upper and lower limit. And it is also reasonable and important to consider that the efficient rate of entry should have some economic relationship with the efficient rate (range) of exit. This is because firm failures and closures imply social costs of more or less importance, so there must be a rate of exit beyond which social losses dominate social gains. Given that overall rates of entry and exit are closely correlated, efficient rates of entry and exit have to be related to each other. Now, given the heterogeneity of agents, it is necessary to determine what type of entrants contribute most to growth and what type of entrants contribute little, nothing or negatively. An entrant impacts negatively if it displaces a 8 more efficient firm, and this can happen in very turbulent environments. In other words, if we borrow the terminology coined by Rosenberg (1982) to a different context, it is essential to “open the black box” of business entry and exit. Several studies show that it makes little sense undertaking analyses that fail to take into account the type of entrants (Santarelli & Vivarelli, 2007; Headd & Saade, 2008). For example, using Global Entrepreneurship Monitor (GEM) data, a number of studies find that while self-employment out of necessity acts as a negative economic indicator, the creation of new ventures with an ambition to grow, prepared to invest their resources in physical and human capital is desirable (Stemberg & Wennekers, 2005). However, useful knowledge remains scarce, fragmented and of limited dissemination. Economic prosperity has been related with entrepreneurship in many studies and even among public opinion. The existing literature mentions various reasons why the rate of new business creation is related positively with economic growth. It is often claimed that a dynamic society is less averse to risk and presents a greater proportion of individuals that prefer self-employment to dependent work (European Commission, 2003). The formalization of entrepreneurial initiative as a growth motor was made by Audretsch & Keilbach (2004) who propose a model of economic growth in which entrepreneurship capital appears as a factor of production within the neoclassical model of the production function. In this model entrepreneurship is measured by the rate of new business start-ups. A different perspective involves assuming that business churning, that is the combined process of the entry and exit of firms, strengthens growth by substituting the less efficient firms with more efficient business. Callejón & 9 Segarra (1999) and Segarra & Callejón (2002) have estimated a production function with data for Spain where both the entry and exit rates appear positively related with productivity. The authors adopt the hypothesis that the entrants operate with latest capital equipment available which incorporates the newest technologies and, therefore, they are more efficient than the firms that have been displaced. The most widely disseminated study on the relationship between the rate of entrepreneurship and the level of income is based on various GEM reports. The 2007 report is based on a survey of 42 rich and poor countries. From the transversal data a U-shaped curve of best fit is obtained between the rate of new entrepreneurs and the level of income. In other words, in countries with low income levels the rate of entrepreneurship (young firms and start-ups compared to the number of adult firms) is very high. The rate then falls as a country’s income level increases until a minimum is reached, and rises again in richer countries (Bosma et al. 2008). However, the interpretation of this pattern, were it to be confirmed, is not at all clear. The reason is that the positively sloped right part of the curve is associated in some studies with an increase in the proportion of selfemployment – or persons that work for themselves without receiving any salary– and could indicate a reduction in adequately paid job opportunities, and a greater dependence on subcontracting on the part of businesses. According to the aforementioned GEM report, 40 per cent of new entrepreneurs in OECD countries do not choose self-employment for reasons of opportunity or personal preference, but rather out of necessity. Carree et al. (2007) and Wennekers et al. (2008) undertake an alternative test of the connection between the rate of entrepreneurship and income levels over time in a group of developed countries. They use OECD 10 databases and do not rely on a survey as was the case of the GEM report. The rate of entrepreneurship is captured by the number of business owners as a proportion of the active population between 1972 and 2004 in a set of eight rich OECD countries. What is interesting in this empirical study is that the fit of data to a U-shaped curve is somewhat inferior to that of an Lshaped curve (repositioned with the long horizontal line to the right). The implications in both cases are very different: with a long-term tendency represented by a L shape, the proportion of business owners would not grow with rising income levels but rather would tend to become stable towards an asymptote at around 7 per cent; and, with a growing Ushaped trend, in 25 years the proportion of business owners would reach 12 per cent. The specific data for the US indicate a point of stabilization would be reached at around 10 per cent. Further research carried out by Acs and Szerb (2009) finds that the relationship between entrepreneurship and economic growth might be neither U-shaped nor L-shaped but slightly S-shaped. Acs and Szerb sensibly advise about the convenience of moving away from simple measures of entrepreneurship across countries to more complex measures. It should be borne in mind, however, that the series analyzed in studies carried out until now do not present such severe economic disruptions as those that have appeared since 2008, which means that the equilibrium values for entrepreneurship indexes could change drastically in the future due to the likely increase in self-employment if business employment opportunities decrease in many OCDE countries for the next years. One clear fact is that the analyses undertaken to date are still largely inconclusive and the reason for this is the lack of longitudinal business data that can be compared between countries. The motivation and skills of 11 entrepreneurs are certainly highly relevant and we know that they can differ notably between new start-ups. A study published by Stam et al. (2007) shows that new ambitious entrepreneurs contribute more to macroeconomic results than the creation of firms in general does. The analysis of business skills is the aim of a study undertaken by Salas & Sánchez-Asín (2009) who draw on data for the sectors and regions of Spain. Salas & Sanchez-Asín build an interesting model in which the decision of an individual to become either an employee, a selfemployed or an employer depends on what his business skills are with respect to the environment and how he or she perceives and evaluates them. Their model predicts that relatively high rates of employers to employees are associated with higher levels of productivity in a society, and that relatively high rates of self-employment are associated with lower levels of social productivity. These results provide evidence that what is important is not the number of entrepreneurs but rather their quality. 1.3.2. Firms and self-employment Public programs should be clear as to whether they seek to foster selfemployment or strengthen the creation of new firms with high-growth capacity or, also, whether they propose promoting the growth of existing firms. Frequently public programs that try to stimulate the creation of new start ups, in fact foster self-employment. It is not unusual for many of these programs to be ambiguous or confused in their objectives. As discussed earlier, the GEM report points to the existence of a high proportion of new entrepreneurs that emerge out of necessity, owing to limitations in the number of adequately paid jobs. The empirical analysis in section 2.1 12 confirms that the net rate of entry of the self-employed grows as the net entry of firms with employed workers falls. The clear implication is that each group – employers and self-employed - react to the effects of the business cycle in a different and opposite way. This also demonstrates that employers and self-employed have different economic incentives. When demand grows and more new firms with employees enter the market, the rate of new self-employed decreases, and vice-versa. The question, nevertheless, should still be raised as to whether the many firms that prosper, grow and eventually generate employment in fact started from self-employment opportunities. The answer though is no. A study published by Davis et al. (2007), made possible thanks to the new longitudinal database of the US Census Bureau, finds that in 2000 only 3.15 per cent of the existing self-employed migrated to the group of businesses with employees. The authors deduce from their study that it is not correct for firm statistics and data bases to mix together both type of agents. We have presently access to a set of recent studies (Davis et al. 2007, Wennekers et al. 2008 and Salas & Sánchez Asín, 2009) that present persuasive evidence regarding the need to separate analytically the selfemployed from firms. Likewise, the new joint OECD-Eurostat database (2008) only includes establishments with at least one employee. 1.3.3. New firms and innovation It is in their contribution to innovation and in their promotion of productive change that new start-ups acquire all their meaning. Joseph Schumpeter contributed the seminal model that associates the creation of new firms with innovation and growth. However, Schumpeter 13 proposed two different growth models each driven by innovation. His first model (1911), known as Schumpeter Mark I, introduces the notion of the innovative entrepreneur, the creator of his own firm, who uses existing knowledge to launch an innovation on to the market. The innovator brings about the obsolescence and decline of some incumbent firms producing the older good, and then enjoys a period of monopoly that is terminated by the entry of imitators, in what is known as a process of creative destruction. The innovative entrepreneurs are a minority since most of the entrants are imitators. In Schumpeter’s second technological model (1942), known as Mark II, innovation becomes a systematic activity of big companies, that have the capacity to invest in R+D. Schumpeter observes and describes the intense process of market concentration occurring between the First and Second World Wars as a consequence of the growing economies of scale of the new production technologies and organization. The new model was described as creative accumulation. In the last years, however, it is Schumpeter’s concept of creative destruction that has inspired many of the analyses and models related to business turbulence and its innovative effect. Some authors argue that the phase of market concentration in the first part of the 20th C has been followed by a period were innovation is dominated by small firms. According to Audretsch and Thurik (2001) the period of managed economy has been followed in present times by the entrepreneurial economy where the role of new firms is crucial in a context of systematic change. In Agarwal, Audretsch and Sarka (2008) it is argued that the industrial dynamics of the present time cannot be identified neither with creative destruction nor creative accumulation, but with the new entrepreneurial model of creative creation. 14 Although the emphasis of many studies typically falls on the positive effect of innovative entry, some important contributions take into consideration the social costs of the closures of firms that are forced out of the market and those incurred by firms that have to reduce their activities. Aghion & Howit (1992) have analyzed how a firm’s incentives to invest in innovation can be affected negatively or positively according to the characteristics of the market in which they operate. Firms react differently to the threat of seeing themselves being pushed out by new competitors. While competition can act as a disincentive to innovative investment in traditional industries, it can stimulate it in cutting-edge industries. Traditional activities usually include many firms with just minor product differentiation. In advanced activities, a smaller number of firms compete with significant product differentiation and they have to be radically innovative so as not to be pushed out of the market. From a different perspective, aligned with evolutionary economics, David Audretsch (1995) highlights the way in which new start-ups promote aggregate innovation. The notion of a technological trajectory is at the base of Audretsch’s model in which it is the new firms that experiment with and take to the market innovations that the established incumbents consider too far removed from their nucleus of experience and know how. In this model, the big established firms find themselves subject to a technological dependence (Nelson & Winter, 1982) and, therefore, it is the new start-ups that introduce the innovations that are associated with greatest risk. Also differing from the above focuses, the concept of “cost discovery”, proposed by Hausmann, Hwang & Rodrik (2006), offers an excellent explanation as to how the market can fail because of a lack of innovative business projects. If, as it seems, countries are what they produce, we should concern ourselves with the mechanisms that determine their 15 productive specialization and their position in the international division of labor. When an entrepreneur seeks to produce something for the first time, he or she experiences uncertainty as to the costs that will be incurred. Even when a standard technology is adopted, it needs to be adapted to the local conditions governing costs and access to specialist inputs. The entrepreneur explores the cost structure – in its broadest sense – of the economy in which he or she is operating this new activity, and the search presents considerable external economies for those that subsequently follow his or her lead. If the entrepreneur is successful, others will emulate him and thereby avoid part of the experimentation costs. If he fails, nobody will imitate him and the cost of failure will remain private. The presence of external economies of knowledge means that the levels of private investment in “cost discovery” are sub-optimal at the aggregate scale, if the government does not manage to stimulate them. Therefore, the number of entrepreneurs that become involved in new activities which involve “cost discovery” becomes an important variable. Policies aimed at stimulating entrepreneurship, at creating innovative firms, at the undertaking of innovative projects by incumbent firms, therefore, take on special importance. 1.4. Productivity and survival Measuring the impact of new firms and business rotation on aggregate productivity, the main determinant of sustained economic growth, faces considerable obstacles. Mairesse & Jaumandreu (2005) have highlighted the problems involved in measuring the productivity of individual firms on the basis of their sales income alone, without knowing their price variations. 16 One of the most frequently reported findings is the enormous heterogeneity in the productivity levels of new start-ups (Haltiwanger et al. 2000). Studies undertaken with data from Spain (Fariñas & Ruano, 2004, Segarra et al. 2008) find that new firms are less productive than incumbents, but that they achieve higher productivity growth rates in their first years (Huergo & Jaumandreu, 2004), which reduces the variance over time and supports the assumption of learning. However, these results can vary if the productivity of each firm is measured in terms of its income or the quantities it produces. Foster et al. (2008) have been able to calculate both types of individual productivity and, as expected, they find that firms that close are less productive than incumbents whatever their size, but that the distance is greater in revenue based productivity. By contrast, new firms present greater physical productivity than incumbents but their advantage in revenue productivity is lower or non-existent. What is most interesting is that the authors believe the reason for this discrepancy lies in the fact that new firms charge lower prices than incumbents. Therefore, revenue productivity studies that do not include individual prices tend to understate the contribution of new start-ups to aggregate productivity growth. The second interesting finding reported by Foster et al. (2008) is that although the firms that close are less productive when applying both measures, it is the differential in revenue productivity that contributes most to their failure. Firms that suffer falls in demand face greater risks of closure when adjusting for technical productivity. The study estimates that variations in demand between firms are the principal determinant of survival. 17 2 Empirical analysis applied to Spain With data drawn from the Directorio Central de Empresas (DIRCE), a directory including data for all Spanish firms, we undertake three types of empirical analysis. First, we compare the behavior of the net entry rate (entries minus exits) for self-employment (no employees) with respect to that for firms that are created with employees, and we find that these rates are diametrically opposed. This can be interpreted as evidence of the distinct nature – incentives and behavior – of self-employment compared to that of firms that hire workers. Second we build life tables – of duration and hazards– for all start-ups and instances of self-employment created in 1994. The resulting data show that self-employed entrants face a considerably greater risk of failure than entrants with employees. Third, we estimate Cox’s proportional-hazards regression to determine whether a firm’s endowment of physical and immaterial resources and the particular environment provided by their business sector contribute to their chances of survival. In this case we only include entrants to the manufacturing sectors from the 1994 cohort. 2.1. Differential behavior of self-employed and firms Self-employment accounts for the majority of the business units in the DIRCE data base. The DIRCE includes business establishments with and without employees. What we call firms and self-employed. Around 50 per cent of all establishments and around 70 per cent of the new establishments correspond to self-employed. 18 The correlation tests performed for a panel of 40 industries (manufacturers and services) for the series of years 1994 to 2001 confirm that the decision to enter the market is made for different reasons among the self-employed and businesses with employees. The technique used is a panel regression with fixed effects that controls for the specific effects of each industry in the net entry. The net rate of entry is defined as the difference between the gross rate of entry (number of establishments that are created each year divided by the stock of establishments) and the gross rate of exit (number of establishments that close divided by the stock of firms). The aim is to compare the signs and the significance (not the value) of the regression coefficients. Table 1 shows that: - The variation in net rates of entry between self-employed (entrants with 0 employees) and entrants with 1-2 employees presents opposite signs. That is, if in a given year the net entry rate of firms grows, the net entry rate of self-employed shrinks. - The same opposite variation is observed for self-employment and entrants with 3 or more employees - By contrast, the net rates of entry vary in the same direction between firms with 1-2 employees and entrants with 3 or more employees. So, entrants with employees react in the same way to the business cycle. When demand grows the rate of gross entry (exit) is higher (lower) and when demand falls the rate of gross entry (exit) shrinks (grows). A plausible interpretation for these three correlations is that the decision to contract employees implies a great difference in firm type. It is reasonable to treat the group of firms without employees as a distinct group, which reacts differently to the economic stimuli that govern the entry and exit of businesses with employees. 19 When the net entry is greater in businesses with employees because there are better opportunities for the survival of such firms, self-employment presents lower net rates of entry. This finding is compatible with the theory that the majority of self-employed are not in fact entrepreneurs, but rather they seek a means of living. Table 1 Correlation test (fixed effects regression) Net entry of Net entry of self-employed firms with 1-2 employees Net entry of firms with 1-0.2767*** 2 employees Net entry of firms with 3 -0.8720*** or more employees 1.3707*** Number of observations: 320 F test (p-value)< 0.0002 or less for the three coefficients This result might be interpreted as an indicator that public policies which seek to strengthen business capacity should include specific programs that go above and beyond facilitating self-employment. In the academic field this result supports the recommendation that analyses of the dynamics of self-employment should be conducted separately from the dynamics of businesses. Headd & Saade (2008) reach the same conclusion, and most convincingly, following their analysis of business dynamics data in the US. 2.2. Survival analysis Survival analyses using individual, censored panel data enjoy a long tradition in economic analyses, both in labor economics and industrial 20 economics. Below we present the most usual concepts included in tables of business survival. The survival rate of a cohort of firms in financial year “t” shows the number of firms that remain active in “t” in relation to the initial number of firms. In other words, S ( t )= Firms active at " t " Firms born at t = 0 The probability of a firm ceasing to trade in “t” is the hazard rate to which the firm is exposed. If time adopts a discrete dimension we can express the hazard rate using the following expression, ⎡ S (t ) ⎤ h(t ) = 1 − ⎢ ⎥ ⎣ S (t − 1) ⎦ The hazard rate h(t) indicates the probability of a firm that has survived financial year “t” exiting the market in time period “t+Δt”. When time adopts a discrete dimension we have, h ( t ) = lim P (t ≤ T ≤ t + Δ t T ≥ t ) Δt = Δt→ 0 f (t ) S (t ) where t = 1,2,...T, is the discrete time; f(t)=dF(t)/dt is the density function corresponding to the distribution of exits as regards the initial number of firms in the group; F(t)=Pr(T