@prefix vivo: . @prefix edm: . @prefix ns0: . @prefix dcterms: . @prefix skos: . vivo:departmentOrSchool "Business, Sauder School of"@en ; edm:dataProvider "DSpace"@en ; ns0:degreeCampus "UBCV"@en ; dcterms:creator "Pe’er, Aviad A."@en ; dcterms:issued "2010-01-16T17:54:27Z"@en, "2005"@en ; vivo:relatedDegree "Doctor of Philosophy - PhD"@en ; ns0:degreeGrantor "University of British Columbia"@en ; dcterms:description "This thesis includes three distinct manuscripts which help to elucidate the impacts of location externalities on entry strategies and performance of heterogeneous new ventures, as well as on the overall regional entrepreneurial activity. The first manuscript studies location choices of de novo entrants as a function of their initial resources and capabilities. It reveals that weak and strong entrants have distinct preferences for various location attributes reflecting both the differences in externalities they face and the value of such externalities in facilitating entry, maturation, and future prospects of profitability. The second manuscript reveals that geographical industry clustering matters to survival of new entrants. Moreover, firm specific factors and strategies which enhance survival vary significantly between different levels of industrial clustering. It shows that the initial endowments of resources and capabilities provide longer adolescence period for firms in clusters. The third manuscript argues that the causal links between entry and failure rates also flow from failure to entry. It shows that exit of local older firms stimulate entry and renewal. For the empirical analyses I used a longitudinal data set developed by Statistics Canada, which provides detailed firm level data for all firms operating in Canada from 1984 to 1998 as well as their employment, financial characteristics, industry affiliation, and location."@en ; edm:aggregatedCHO "https://circle.library.ubc.ca/rest/handle/2429/18286?expand=metadata"@en ; skos:note "ENTRY STRATEGIES AND PERFORMANE OF NEW VENTURES IN CLUSTERS AND ISOLATION by A V I A D A . PE 'ER B.Sc. Technion Israel Institute of Technology, 1991 B.Sc. Technion Israel Institute of Technology, 1992 M.Sc. Technion Israel Institute of Technology, 1999 A THESIS SUBMITTED IN PARTIAL F U L F I L L M E N T OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR IN PHILOSOPHY in THE F A C U L T Y OF G R A D U A T E STUDIES (Business Administration) THE UNIVERSITY OF BRITISH C O L U M B I A June, 2005 © Aviad Pe'er, 2005 ABSTRACT This thesis includes three distinct manuscripts which help to elucidate the impacts of location externalities on entry strategies and performance of heterogeneous new ventures, as well as on the overall regional entrepreneurial activity. The first manuscript studies location choices of de novo entrants as a function of their initial resources and capabilities. It reveals that weak and strong entrants have distinct preferences for various location attributes reflecting both the differences in externalities they face and the value of such externalities in facilitating entry, maturation, and future prospects of profitability. The second manuscript reveals that geographical industry clustering matters to survival of new entrants. Moreover, firm specific factors and strategies which enhance survival vary significantly between different levels of industrial clustering. It shows that the initial endowments of resources and capabilities provide longer adolescence period for firms in clusters. The third manuscript argues that the causal links between entry and failure rates also flow from failure to entry. It shows that exit of local older firms stimulate entry and renewal. For the empirical analyses I used a longitudinal data set developed by Statistics Canada, which provides detailed firm level data for all firms operating in Canada from 1984 to 1998 as well as their employment, financial characteristics, industry affiliation, and location. TABLE OF CONTENTS ABSTRACT ii TABLE OF CONTENTS iii LIST OF TABLES v LIST OF FIGURES v ACKNOWLEDGEMENTS vi INTRODUCTION vii CHAPTER 1: WHO ENTERS, WHERE, AND WHY? THE INFLUENCE OF CAPABILITIES AND INITIAL RESOURCES ON THE LOCATION CHOICES OF NEW ENTERPRISES 1 Introduction 2 Theory and Hypotheses 4 Firm Initial Characteristics 4 Location Choices and Firms Heterogeneity 7 Location Attributes and Initial Resources and Capability Profiles 8 Cost of Entry 11 Research Methods 14 Econometric Framework 14 Data Sources 19 Variable Definitions 20 Firm-Specific Attributes 20 Location Characteristics 23 Empirical Results 26 Limitations and Robustness Checks 36 Conclusion 37 References 39 CHAPTER 2: THE SURVIVAL VALUE OF CLUSTERS 45 Introduction '. 46 Related Literature 47 Hypotheses Development 52 Size 52 Age 54 Human Resources 56 Sunk Costs 56 Competition 57 Controls Variables 59 Firm Level 59 Industry Level 60 Methodology 61 Empirical Model 61 Data 64 Cluster Definition 66 Covariates 68 Firm Specific Attributes 68 Industry Level 71 Empirical Results 72 in Exit of New Firms 75 Testing for Structural Differences between Levels of Local Industry Clustering 78 The Honeymoon Period 85 Robustness 87 Conclusion 88 References 90 APPENDIX 2.A: Variables Definition 98 APPENDIX 2.B: Correlation Table 100 CHAPTER 3: FIRM FAILURES AS A DETERMINANT OF NEW ENTRY: IS THERE EVIDENCE OF L O C A L CREATIVE DESTRUCTION? 101 Introduction 102 Theory Development 103 Failures 103 Agglomeration Economies and Competition 107 Method 109 Model 109 Measures 113 Data Sources 116 The Geographic Units 118 Industries 1' 9 Results 119 Robustness of Results • 124 Limitations of the Study 126 Conclusions 127 References 130 APPENDIX 3.A: Descriptive Statistics and Definitions 135 APPENDDC 3.B: Labor Tracking Procedure 136 APPENDIX 3.C: Industries 137 APPENDDC 3.D: OLS Fixed Effects Estimates of New Entry 138 LIST OF TABLES T A B L E 1.1: Variables Definition and Descriptive Statistics 27 T A B L E 1.2: Correlation Matrix 28 T A B L E 1.3: Nested Logit Model of Census Division/Economic Region Choice 30 T A B L E 2.1: Life-Table Estimates of Survival Rates 73 T A B L E 2.2: Cox Hazard Model - unconditional impact of clustering 77 T A B L E 2.3: The Determinants of Firm Exit: Semi-proportional Cox Hazard Model with Interaction Terms between Levels of Local Industry Clustering and other Covariates 80 T A B L E 3.1: Tobit Estimates of New Entry: Investigating the Impact of Failures 122 LIST OF FIGURES FIGURE 2.1: Survival of New Entrants Operating in Locations with Different Levels of Industry Clustering 74 FIGURE 2.2: Smoothed Hazard Function of new Entrants Operating in Locations with Different Level of Industry Clustering 86 FIGURE 3.1: Location Choice Set 112 A C K N O W L E D G E M E N T S I am deeply indebted to Han Vertinsky for the guidance, direction, friendship, and intellectual support that he has provided during my program of study at U B C . I am also very grateful to J im Brander, Bob Helsley, and Daniel Muzyka for insightful comments and nurturing my development as a researcher. Many thanks to the members of the Strategy and Business Economics Division, particularly to Werner Antweiler, Keith Head, John Ries, and W i l l Strange (University of Toronto), their comments and suggestions were most appreciated. The access to the excellent data resources of Statistics Canada has been invaluable. Financial support of the Entrepreneurship Research Alliance and the Social Sciences and Humanities Research Council of Canada (grant # 752-2000-2117), and M C R I (grant #412 93 005) is kindly acknowledged. I have benefited immensely from the friendship and encouragement of Ofer Arazy, Sabrina Deutsch Salamon, Yuval Deutsch, Cynthia Holmes, Anton Kaplan, Benny Mantin, and Hakan Ozcelik. Finally, I am deeply grateful to Talia, Arie and Efrat Pe'er, and Ronit Barzilai for their unconditional love, patient support, and wonderful personal and professional example. This thesis is dedicated to them. INTRODUCTION This thesis stands at the intersection of the fields of strategic management, entrepreneurship and economic geography. It is comprised of three distinct manuscripts which help to elucidate the impacts of agglomeration economies on entry decisions, strategies and performance of heterogeneous new enterprises. The first manuscript, \"Who enters, where, and why? The influence of capabilities and initial resource endowments on the location choices of new enterprises,\" stems from the observation that new enterprises rarely possess an optimal endowment of resources and capabilities upon entry. Often they face severe constraints of financial and physical resources, human capital and knowledge. Different locations present new entrants with different competitive environments, opportunities and threats. In trying to understand the nature of entry strategy, I answer the following questions: a) what locational characteristics are associated with low and high entry barriers?, b) what types of externalities new firms seek to exploit in order to overcome liabilities of smallness and newness?, and c) how the pattern of trade-offs between negative and positive externalities alters as a function of heterogeneity in the initial resources and capabilities of the firm? I model location choices as constrained decision processes where new firms select locations so as to obtain the most profitable, feasible match of firm and location characteristics. Entry decisions are informed by a process of information gathering and evaluation of profitability. Entry occurs i f an entrepreneur expects the discounted value of entering to be greater than the cost of entry. The evaluation of post-entry profits is conditional on the characteristics of the entering firm, expected benefits from agglomeration economies, and the observed competitive environment. In a cluster, entrepreneurs may benefit from agglomeration economies but face competition from incumbents. In isolation, new entrants may shield V l l themselves from intense competition but may need to develop some crucial specialized resources (e.g., provide training programs). The econometric model that I develop to examine the strategic tradeoffs new entrants make in choosing a location incorporates specific interaction patterns between location characteristics, and strengths and weaknesses, defined along different resource and capability dimensions. I estimate nested logit models using a unique data set made available by Statistics Canada that covers all new entrants into the Canadian manufacturing sectors during 1984-1998. The results reveal that differences in resource and capability profiles play an important role in strategic entry decisions. I find a favorable selection process in cluster entry where the stronger entrants expect higher agglomerative benefits than the weaker ones. However, incumbents' propensity to engage in entry deterring activities has a higher detractive influence on larger firms, those with higher-quality human capital, and the more productive ones. Less concentrated markets and lower sunk costs are more attractive factors to smaller entrants and those with lower quality human capital. I show the existence of both favorable and adverse entry selection processes which are dominant at different phases of the evolution of a cluster. The results deepen our understanding about the strategic matching process of resources and capabilities and location traits. They confirm the importance that new entrants place upon the strategic value of different location characteristics, in building new capabilities, and in avoiding externalities that erode competitive advantage. The second manuscript, \"The survival value of clusters\" investigates the impacts of location externalities on survival. I examine whether there are differences between the determinants of failure among firms operating in locations with different levels of industrial clustering. The modeling approach uses longitudinal firm-level data at fine-grained, empirically determined classification of the level of industrial agglomeration. I control for endogeneity in the location choice, and the impact of initial endowments of resources and capabilities. The empirical strategy is semi-proportional Cox hazard model. The results show that survival rates of de novo entrants operating in locations with high levels of industry clustering are higher than those of entrants in locations with lower levels of same industry concentrations. Extending prior research, the findings reveal differences in firm-specific survival enhancing strategies among firms located at different levels of clustering. Larger initial size, higher growth, better initial and current quality of human capital, lower leverage, and higher initial and current productivity have survival advantages in all locations irrespective of clustering levels. However, survival prospects of firms operating in clusters are more positively affected by initial relative size, and initial quality of human capital than those of firms in lower levels of clustering. Larger entrants most likely enjoy economies to scale in identifying, accessing, and exploiting localization externalities. The ability of new firms to screen, assess, absorb, and internalize knowledge spillovers is largely determined by the initial relative quality of their employees. In isolation, current resource endowments, productivity, and recent growth have higher impacts on the survival of firms. Survival of firms in isolation depends to a larger extent on the firm's internal resources and capabilities while firms operating in clusters also benefit from their ability to utilize externalities which reside in the cluster. B y outsourcing activities to specialized firms operating in the cluster, forging local strategic alliances, and limiting their internal growth to specific activities which are more likely to generate competitive advantage, firms in clusters optimize and gradually adjust their internal growth thus minimizing damages and adjustment costs associated with excessively rapid growth. Finally, the analysis demonstrates that clusters extend the adolescence period of new entrants. The third manuscript, \"F i rm failures as a determinant of new entry: Is there evidence for creative destruction?\" investigates the spatial relationships between aggregated enterprise failures and new firm entry. I argue that the causal links between entry and failure may flow not only from new entry to failure, reflecting the liability of newness, but also from failure to entry. Indeed such link is consistent with Schumpeter's idea of creative destruction where failures of old firms create opportunities for new entrants. I estimate tobit models of entry events of five manufacturing sectors in Canada between 1984 and 1998 at the census subdivision level. While controlling for agglomeration economies, competition, census subdivision area, time, industry, and economic-region fixed effects I find that failures of older firms in a location may intensify search for entrepreneurial opportunities and/or generate more opportunities in that location thus, attracting new entry. The study shows that the impact of failure levels attenuates with distance. The average rate of failure is an indicator of risks associated with a particular location and therefore deters entrepreneurs from entering that location i f it is judged riskier than other locations. In contrast, a higher risk of failure associated with neighboring locations within a region wi l l increase the relative attractiveness of the location. The study contributes to the understanding of the role that enterprise failure plays in the industrial regional renewal process. Policy implications are drawn. Taken together, these three manuscripts provide distinct but complementary angles which aggregate into a complex set of interlacing impacts of location externalities on firm-level and regional entrepreneurial activity. CHAPTER ONE WHO ENTERS, WHERE, AND WHY? THE INFLUENCE OF CAPABILITIES AND INITIAL RESOURCES ON THE LOCATION CHOICES OF NEW ENTERPRISES A B S T R A C T Using data about all de novo entrants into Canadian manufacturing sectors during 1984-1998, we studied location choices as a function of firms' initial resources and capabilities. Employing nested logit estimation, we examined the impact of various location traits such as: agglomeration, competition, deterrence, and sunk costs, on location choices. Findings reveal that stronger entrants value more locations with positive cluster externalities, but are more detracted by local competition and incumbents' deterrence strategies. Weaker firms are attracted to places with lower entry barriers and sunk costs. The findings imply the existence of both favorable and adverse entry selection processes which are dominant at different phases of the evolution of a cluster. Introduction New enterprises rarely possess an optimal endowment of resources and capabilities upon entry. Often they face severe constraints of financial and physical resources, human capital and knowledge. Thus, their entry strategies must accommodate three imperatives; they must be able to (1) overcome entry barriers; (2) find a protected environment where they are relatively shielded from the 'liability of newness' (Stinchcombe 1965); and (3) find an environment which facilitates their process of maturation (Helfat and Peteraf 2003). The strategic choice of a new enterprise requires a careful fit of firm's initial resources and capabilities and environmental characteristics. Different locations present new entrants with different competitive environments, opportunities and threats. Thus, location choice is often a critical factor in their entry strategies and future performance. If we are to understand the nature of entry strategy, we must answer the following questions: (1) what location characteristics are associated with low entry barriers?; (2) what types of externalities new firms seek to exploit in order to overcome the 'liability of newness'?; (3) how the pattern of trade-offs firms are wil l ing to make between negative (e.g., competition) and positive externalities (e.g., knowledge spillovers) alters as a function of heterogeneity in the initial resources and capabilities of the firm?; and (4) how the fit between new venture characteristics and the environment is affected by the prevailing macro-economic and industry conditions at the time of entry? This paper studies the impacts enterprise initial resources and capabilities have on location choices of new firms. Previous studies have modeled location choice as a function of demand, the intensity of competition (e.g., Baum and Haveman 1997), and location externalities (e.g., Carlton 1983, Rosenthal and Strange 2003). Shaver and Flyer (2000) using a sample of multinationals entry decisions to the United States pointed out that firms may evaluate location externalities on the basis of the strength of their resources and capabilities. They suggested that \"firm heterogeneity leads to asymmetric contributions to agglomeration economies and, in turn, adverse selection of which firms agglomerate\" (p. 1190). They argued that 'strong' firms have an incentive to locate in isolation to sustain their competitive advantages, while 'weak' firms are more likely to choose a cluster location to benefit from agglomeration. In this paper we focus on entry location choices of domestic de novo enterprises. We argue that location choices of new enterprises may be distinctly different from those of mature firms (especially those of foreign multinationals). Our paper explores the dimensionality of strengths and weaknesses examining whether strengths and weaknesses in terms of different resources and capabilities have specific patterns of interaction with location attributes. The paper also extends the analysis beyond agglomeration economies to other location attributes. We model location choices as constrained decision processes where new firms select locations so as to obtain the most profitable, feasible match of firm and location characteristics. Entry decisions are informed by a process of information gathering and evaluation of profitability. Entry occurs i f an entrepreneur expects the discounted value of entering to be greater than the cost of entry. The evaluation of post-entry profits is conditional on the characteristics of the entering firm, expected benefits from agglomeration economies, and the observed competitive environment. In a cluster, entrepreneurs may benefit from agglomeration economies but face competition from incumbents. In isolation, new entrants may shield themselves from intense competition but may need to develop some crucial specialized resources (e.g., provide training programs). The econometric model that we test to examine the strategic tradeoffs new entrants make in choosing a location incorporates specific interaction patterns between location characteristics, and strengths and weaknesses, defined along different resource and capability dimensions. We estimate nested logit models using a unique data set made available to us by Statistics Canada that covers all new entrants into the Canadian manufacturing sectors during 1984-1998. We show that differences in resource and capability profiles play an important role in strategic entry decisions. We find a favorable selection process in cluster entry where the stronger entrants expect higher agglomerative benefits than the weaker ones. However, incumbents' propensity to engage in entry deterring activities has a higher detractive influence on larger firms, those with higher-quality human capital, and the more productive ones. Less concentrated markets and lower sunk costs are more attractive factors to smaller entrants and those with lower quality human capital. Findings also reveal some interesting implications with respect to cluster dynamics and sustainability. We show the existence of both favorable and adverse entry selection processes which are dominant at different phases of the evolution of a cluster. Our results are robust to the use of alternative specifications and econometric techniques, different location choice sets, and alternative measures of the covariates. The results deepen our understanding about the strategic matching process of resources and capabilities and location traits. They confirm the importance that new entrants place upon the strategic value of different location characteristics, in building new capabilities, and in avoiding externalities that erode competitive advantage. Theory and Hypotheses Firm Initial Characteristics Entrants are heterogeneous with respect to their initial characteristics and availability of resources and capabilities (Barney 1991; Helfat and Peteraf, 2003). The initial state of resources has a lasting impact on performance (Dunne, et al. 2004). Geroski, et al. (2002), found that initial resources are more important than current values in explaining survival of new firms, and that their effects do not decay rapidly over the first five to ten years of a new firm's life. Several scholars argue that a lack of financial resources early on in the life of an enterprise is a key component of the 'liability of newness' (e.g., Stinchcombe 1965). A slack in financial resources provides high flexibility in the choice of strategic options (Amit and Schoemaker 1993), allowing firms to take calculated risks with long-term projects and to experiment with new products or new markets (March and Shapira 1987). Such experimentation in entrepreneurial firms has been linked to improved performance (Christensen, et al. 1998). The amount of financial resources accessible to the firm at founding is a source of competitive advantage1. The lack of internal financing and imperfections in capital markets may compel new enterprises to enter at a smaller scale than they would have chosen i f they had the funds. Myers (1977) and Gertler (1994) suggested that the liquidity requirements created by external debt may limit a firm's ability to finance new projects or to further invest in existing projects, even those with a positive net present value. B y limiting the firm's ability to finance new growth opportunities, the liquidity constraint imposed by existing leverage may reduce its chance of survival. Firms with high levels of leverage face substantial cash flow demands to service their debt and these costs infringe on their ability to take advantage of opportunities and cope with threats. Zingales (1998) demonstrated that, even when controlling for firm's efficiency level, leverage is still relevant to the probability of survival. Given the importance of access to financial resources, firms with weak financial resources may seek to enter locations where access to capital is easy (Sorenson and Stuart 2001). Repeated interactions, and dense social and professional networks of financial institutions and nascent entrepreneurs, that exist in clusters, 1 Other studies claim that resource constraints force entrepreneurs to improve allocation efficiency which positively influences performance (e.g., Baker, et al. 2003). may decrease the perceived risk associated with this class of borrowers. In addition, since the ability of lenders to monitor the behavior of borrowers is higher in clusters, they may be wil l ing to lend more readily. Thus, the access to capital for entrepreneurs in clusters may be higher than in isolation. A large initial size of a new establishment decreases the cost disadvantage confronting firms operating at sub-optimal scale levels of output (Audretsch and Mahmood 1995, Audretsch and Mata 1995) 2. Organizational theorists assert that larger organizations have superior access to capital and trained workers, higher legitimacy with external stakeholders, and a better ability to chart a clear course and stick to it (Baum and Oliver 1991, Hannan and Freeman 1977). Small new firms are more likely to employ less able or inexperienced managers who are more likely to make the greatest mistakes in estimating their true ability level (Mata and Portugal 1994). Entry size may also signal the prior expectations of entrepreneurs - large scale entry indicates a greater a priory expectation of ability to compete (Frank 1988). While a large size, at the point of entry, provides many benefits, firms may choose to enter on a small scale to avoid the aggressive behaviors of incumbents. This entry strategy is referred to as a \"Judo strategy\" (Scherer and Ross 1990). B y entering with a small capacity, the firm signals that it w i l l not play out an aggressive price strategy (i.e., projecting a friendly image) which, in turn, may soften the response of incumbents to the entry, as the cost of a response may be viewed to be higher than the expected benefit (Gelman and Salop 1983). New entrants may also choose to be small because they are uncertain about their efficiency and profitability. Being small allows entrants to learn from experience (Jovanovic 1982) and to develop learning abilities (Ericson & Pakes 1995, Pakes & Ericson 1998), which may reduce the risk of failure and 2 Sub-optimal scaled entrants, however, may overcome the scale-related disadvantages by developing cost-reducing strategies, or deploying and remunerating their productive and managerial structures differently so that they are more flexible and less hierarchical. If entrants can implement such compensatory mechanisms, size disadvantages would be less likely to determine the entry mode choice or expected profitability (Audretsch and Yamawaki 1992). increase the possibility of growth. Initial investment (commitment) in physical resources may be viewed as sunk costs (Cabral 1995). Confident entrepreneurs (holding more positive expectations about their capabilities and opportunities) make a larger initial commitment. Less confident entrepreneurs, on the other hand, limit their sunk costs (forgoing the option to invest heavily) until their untested capabilities can be revealed 3. The Resource-Based View of the firm emphasizes that the ability of firms to compete successfully is largely determined by the ability of the firm to develop and sustain human capital assets that cannot be imitated by competitors and which contribute to post-entry performance (Tirole 1988). Initial human resources have permanent and profound effects on the performance of firms (Cooper 1984, Eisenhardt & Schoonhoven 1990). In their research of the U S biotechnology industry, Zucker, et al. (1998) found a tight connection between the intellectual human capital created by frontier research and the founding of firms in that industry. Moreover, the growth and location of human capital were principal determinants of the growth and location of the industry itself through the diffusion of intellectual human capital. Location Choices and Firms Heterogeneity In choosing a location new entrants face three major concerns. The immediate short-term concern is overcoming barriers to entry. The perceived severity of entry barriers depends in part on the initial endowment of resources and capabilities of the enterprise. For example, abundance of financial resources allows firms to locate in sites that may not be affordable to those with severe financial constraints. The second concern relates to the intermediate phase of firm entry, This view is consistent with the principle of \"asset parsimony,\" where investments in fixed assets are kept to a minimum, thus increasing flexibility and minimizing potential loses (Hambrick and MacMillan 1984). the period when the firm grows until it reaches the minimum efficient size, and develops unique competitive portfolio of resource and capabilities. Locations are evaluated on the basis of those attributes that hasten the maturation process and those that provide shelter from risks during the initial and intermediate phases. The reliance of firms on location characteristics to develop dynamic capabilities may be higher for firms which lack those capabilities (Helfat and Peteraf 2003). Differences in initial capabilities may lead, however, to differences in exposure to certain risks and also may explain differences in the benefits firms can derive from various externalities or the costs they impose. Vulnerable firms may also be concerned about the expected costs associated with failure, costs that can be mitigated i f a market for used assets is well developed in a location. The third concern of a location choice is one that is common to both mature and new enterprises - identifying those attributes which contribute to long-term profit maximization. Thus, the choice of a location is a process of evaluation of the different costs and benefits that accrue from alternative location characteristics assuring that barriers to entry can be overcome, that maturation can be achieved rapidly and with safety, and that long-term survival and productivity are enhanced. Below we explore in more detail the interactions between specific location attributes and initial profiles of resources and capabilities. Location Attributes and Initial Resources and Capability Profiles Localization Economies Firms in the same industry tend to agglomerate or cluster in particular regions. The agglomerating (centripetal) forces are 'localization externalities' that can be categorized into production enhancements and heightened demand. Production enhancements originate from: (1) labor market pooling (Marshall 1920); (2) advantages of backward and forward linkages associated with large local markets (i.e., improving the quality of matching crucial inputs and intermediate goods between suppliers and demanders) (Fujita 2000); (3) shared infrastructure (broadly defined) that is available to firms that locate close to each other (Helsley and Strange 2002); (4) technological and knowledge spillovers (Jaffee et al. 1993, Almeida and Kogut 1999); (5) information externalities about demand or the feasibility of production at a particular location that are available to prospective entrants who observe incumbents operating there profitably; and (6) lower exit barriers, which, in turn, may facilitate entry with lower costs. Agglomeration wi l l also cause heightened demand in industries where consumers need to personally inspect goods or compare prices. Clustering may reduce consumers' search costs, thus increasing the likelihood of visitation and purchase, compared to firms in a separate location (Baum and Haveman 1997, Chung and Kalnins 2001, Kalnins and Chung 2004) 4. Clusters foster entry of entrepreneurial enterprises by reducing the costs of inputs associated with introducing products or services to the market (Helsley and Strange 2002). Entrepreneurs contemplating entry may benefit from employing existing (sometimes shared) infrastructure, logistics, and inputs (such as specialized input providers, and input supply networks) at a lower cost than developing or employing new inputs. New enterprises in clusters may increase their efficiency by focusing on their internal competencies while outsourcing, leasing, or renting other activities. Firms in clusters are more exposed to collaboration opportunities with other firms within the same industry. While some diseconomies are often associated with clustering, such as congestion costs and competition, empirical economic studies show that incentives for clustering tend to outweigh 4 Those studies focused on the hotel and motel industries and thus on agglomerative benefits due to heightened demand. Moreover, the majority of entrants to these sectors are not independent operators. Our focus is on manufacturing sectors and thus location externalities which stem from performance enhancement and on independent entrants. the incentives for isolation (e.g., Carlton 1983, Dumais, et al. 2002, Head, et al. 1995, Krugman 1991, Porter, 1998, Rosenthal and Strange 2003). The existence of agglomeration economies may lead to adverse selection by reducing entry barriers. Marginal investment opportunities that nascent entrepreneurs may reject in isolation may pass thresholds of expected profit in clusters once localization externalities are factored in. Nascent entrepreneurs with weak resource positions may benefit more than the stronger ones from knowledge spillovers, access to specialized inputs (e.g. skilled labor), and heightened demand (Baum and Haveman 1997, Chung and Kalnins 2001, Kalnins and Chung 2004) that localization economies facilitate. The link between information flows and spatial proximity was vastly supported in the literature. Jaffe, et al. (1993) showed that firms which are physically located closer to each other are more likely to cite each others' patents. These types of information flows wi l l enhance organizational performance5. The extent to which establishments are involved in information spillovers is inversely related to their size and age (Scott and Kwok 1989, Almeida and Kogut 1997, Acs , Audretsch and Feldman 1994, Aharonson et al. 2004). Companies with low-quality human capital must rely more heavily on markets of skilled and specialized labor in developing their competencies. Locating in a cluster, allows them to benefit from access to skilled workers that were trained by their competitors (Almazan, et al., 2003). In contrast, entrants possessing initial human capital with higher than average quality may strategically refrain from locating in a cluster since labor mobility can erode their competitive advantage (Shaver and Flyer 2000) 6. Consistent with this idea, strong foreign entrants may choose to locate distantly to slow information spillovers that might erode their competitive advantage (Shaver and Flyer 2000). 6 Isolation may also be preferred by firms that expect to have less favorable growth prospects than their competitors in order to avoid the competition with their more productive counterparts for labor (Almazan et al. 2003). Hypothesis 1 - The weaker the resources and capabilities position of prospective entrants the higher the likelihood they enter a location with strong localization economies. Cost of Entry New establishments, limited by their internal capabilities and by their access to available resources (Shane and Stuart 2002, Shane and Venkatataman 2000) are likely to prefer locations where entry barriers are low. Entry costs and other barriers vary between cluster and isolation locations. Entrepreneurs may choose not to bear, or are unable to afford the costs associated with entering a cluster or isolation. Such costs are related to land rents and wage rates (higher in clusters), and sunk costs (higher in isolation). Entry barriers are also determined by the competitiveness of markets for specialized inputs (e.g. skilled labor), and the deterrence strategies of incumbents. Land Rents Land rents represent a major cost facing entrepreneurs. Local agglomeration suggests an increase in relative land rents in clusters for both industrial and residential uses (Wheaton and Lewis 2002). Firms w i l l tend to enter a place with high land rents only i f the region yields higher productivity benefits in comparison to regions with lower land rents (Roback 1982). While one would expect that in equilibrium land rents would incorporate the true value of the location externalities, new entrants with limited financial resources may face an imperfect world where concerns for affordability dominate. Thus, a smaller initial endowment of financial resources may increase the tendency of a new entrant to locate in low rent areas. Hypothesis 2 - The likelihood of new entry by firms with weaker initial financial resources is lower in locations with high land rents. Sunk Costs The Theory of Contestable Markets (Baumol and W i l l i g 1981) asserts that the extent to which new firms enter a market in response to observed abnormal profits depends on the level of sunk costs in that particular market (Dixit 1989, Kessides 1990, Sutton 1991). The magnitude of sunk costs of entry is of primary importance in determining the steady state rate of firm births and deaths within the industry. L o w costs of entry and exit encourage potential entrants, who typically do not know their full relative efficiency before entering a market, to enter and experiment ('learning by doing') so that the true ex-post efficiency can be revealed (Jovanovic 1982). The extent to which initial investments can be recovered upon exit (i.e., exit barriers) depends on the mobility of tangible resources employed by the firm and on entry rates to that location. When resources are immobile (e.g., infrastructure, heavy machinery), high entry rates to this location wi l l increase the demand for second hand assets and thus increase the salvage value. Firms with weak resource endowments are exposed to higher risks of failure, thus their estimates of sunk costs is l ikely to have a higher influence on their location decisions. Hypothesis 3 - The impact of sunk cost on location choice decisions of companies with weaker initial resource endowments will be higher than on the decisions of firms with strong resource endowments. Incumbents' propensity to deter entry Incumbents have incentives to behave aggressively towards entry when they have a profitable market position that is threatened by potential entrants and can be protected (Tirole 1988). A n aggressive response towards entry needs to be funded. Thus, more profitable incumbents, with surplus funds at their disposal, may be more inclined to protect their positions in the market (Kessides 1990). A n important element of market structure that affect the motive and ability of incumbents to deter entry is market concentration. Incumbents may find it easier and more effective to engage in, and coordinate, entry deterring activities where they are spatially concentrated, since firms' incentives are more aligned and the free rider problem may be smaller (Bunch and Smiley 1992). Coordinating and monitoring entry deterrence activities is more effective in clusters than in isolation. Thus, we expect that the propensity of incumbents to engage in costly strategic deterrence is stronger in clusters than in isolation. Entry deterrence, however, w i l l be more effective when directed at the less-productive, high-cost entrants. Hypothesis 4 - The negative effect of incumbents' deterrence behavior on entry to a location is higher for new entrants that have weaker initial resources and capabilities. Competition Entrants with fewer resources and lower productivity are less able to compete. In competitive markets only the most efficient firms can survive. Less efficient new entrants may choose to enter a highly competitive market only i f they have superior human capital or deeper pockets to ensure that they can climb the learning curve fast enough and cut their costs. Entrants may use isolation to shield themselves from rivalry since distant competitors may drive down payoffs less than proximate competitors (Baum and Mezias, 1992). Even when firms compete in national product markets, spatial concentration may imply competition in factor markets (e.g., intermediate goods, skilled labor) and thus, may drive weaker firms to seek isolation. Hypothesis 5 - The negative effect of local competition on entry to a location is higher for new entrants with weaker initial resources and capabilities. Unemployment To the extent that the unemployment rate is an indicator of labor availability, as well as has a dampening influence on wages, it w i l l l ikely be positively related to new entry into a location . In addition, the public sector in regions with high unemployment may also encourage entrepreneurial entry by providing monetary incentives (e.g., tax relief, improved access to risk capital), or more flexible regulatory environments. The inducements for entry offered by local authorities and the lower wages associated with high levels of unemployment are more important to those entrants with fewer resources. Hypothesis 6 - The likelihood of new entry by firms with weaker initial resources is higher in locations with high unemployment rates. Research Methods Econometric Framework Nascent entrepreneurs contemplating entry in a given industry decide whether to enter or not, and choose locations. Evaluating and comparing the set of feasible (affordable) locations, entrepreneurs wi l l choose the location that w i l l yield the highest expected profits. The evaluation of entry to a location considers entry costs and post-entry profits. Entry costs include land rents, cost of resource acquisition, search for markets, and coping with incumbents' entry deterring activities. Expected post-entry profits reflect factors such as the expected benefits from local externalities (e.g., knowledge and technology spillovers, outsourcing and collaboration opportunities), exit barriers, the competitive environment, and industry-specific characteristics in terms of costs and demand conditions. Incumbent firms are assumed to anticipate entry and its effects on the location competitive environment and idiosyncratic profitability, and choose an entry-deterring strategy before an entry occurs. The timing of the decisions made by incumbents and entrepreneurs allows us to assume that any entry-deterring strategy and information regarding the characteristics of a location are correctly observed, interpreted and incorporated by 7 Long or frequent spells of unemployment, however, may lead to deterioration in quality of labor and input and service providers. entrepreneurs at the time of entry. Therefore, entrepreneurs' entry strategy can be inferred from the location choice at the moment of entry8. The empirical approach that we use to model location choice was developed by Carlton (1983) based on McFadden's (1974) multinomial logit model. The econometric specification of the discrete choice methodology explicitly ties it to the theoretical foundations of the Random Util i ty (profit) Maximization framework. Given a set of potential mutually exclusive locations 7=7,...,/, entrepreneur i selects the one that w i l l maximize profits. The chosen site reveals the preferences of the entrepreneur for location attributes (i.e. their weights) since the relative level of profits that can be derived in this location is higher compared with all other feasible locations (Figueiredo, et al., 2002) 9 . Consider an economy with indifferent industrial sectors (k=l,...,K), and TV new entrants (i=l,...,N), who independently select a location j from a location choice set (j=l,...,J). Upon entry to sector k, entrepreneur i's profit in location j is given by the following reduced form: (1) xijk=a'Zijk+£ijk where a is a vector of unknown parameters, Zijk is a vector of observable factors that have impact on profit, and eijk is a random term that absorbs all unobserved heterogeneity. Accordingly, the profit for entrepreneur i of locating in j is composed of deterministic and stochastic components. When the entrepreneur decides not to enter, equation (1) becomes (2) K..k =7C0+£m 8 The likelihood of future entry to a location and its effects on profits are also critical to entry decision. We assume, however, that nascent entrepreneurs cannot know or predict if, and how many, other potential entrants will choose to enter a location in the future. where mean profits from not entering (i.e. no location was chosen j=0), x0, are normalized to zero across firms and locations so that entrepreneurs deciding not to enter the market earn only their idiosyncratic profitability draw eiQk, corresponding to the option value of engaging in different pursuits. The entrepreneur would choose to locate in site r such that: (3) 7tirk >7Tijk Vj,j*r Likely , the random term is correlated among two alternative Census Divisions (CDs): j and r, in the same Economic Region (ER) - i.e. alternatives in the same nest. Thus, we use a nested logit model (a generalization of the multinomial logit) that permits proportional substitution within a nest (such that the independence of Irrelevant Alternatives assumption is valid within a nest 1 0 ) (Train 2003, Head and Mayer 2004). If the random term has a multivariate Extreme Value distribution, then the probability pijk that location (CD) j in E R / yields entrepreneur i, belonging to sector k, the highest profits among all locations within the E R and may be represented as the product of two standard logit probabilities (McFadden 1978)\". (4) Pijk = Pijk ER, *PER, The conditional logit approach assumes the Independence of Irrelevant Alternatives - i.e. the error terms are independent across entrepreneurs and choices. Specifically, the relative probabilities of various alternatives depend only on the observed characteristics of these alternatives regardless of the choice set (i.e., substitution patterns are imposed by the functional form). If this assumption is violated, due to unobserved location characteristics that may include correlation across choices, it can lead to biased coefficient estimates (Figueiredo et al. 2002). Consequently, some authors have introduced dummy variables for larger regions (e.g., Bartik 1985, Carlton 1983, Head, et al. 1995). 1 1 The construction of a comprehensive sample consisting of all feasible locations poses practical problems as it is extremely taxing. To illustrate this point, consider our data: it consists of 46,620 events of entry. The number of feasible CDs in Canada is 289, thus the number of all feasible locations is 13,473,180. There are two additional alternatives to the construction of a practical choice set. The first is to include all locations in the province (see robustness section). The second is a choice-based sampling, where the sample includes all the realized locations and a randomly chosen sub-sample of the unrealized locations. Choice-based sampling approach can yield biased estimates. By construction, the ratio of chosen to non-chosen locations is different than its proportion in the population. where pijk ERi is the conditional probability of choosing C D j, given that ERl was chosen (je ERt), and pER is the marginal probability of choosing economic region / (an appendix describing the development of equation 4 is available from the authors upon request). Maximum likelihood techniques are used to estimate the parameters a (Equation 1). Previous research has typically specified vector Zijk as a linear combination of location characteristics such as factor costs and agglomeration economies that affect the profit function of entrants (e.g., Bartik 1985, Carlton 1983, Head, et al. 1995). The assumption was that entrants are homogeneous, free to choose among locations and thus wi l l have similar preferences for location attributes. Entrants, however, are heterogeneous with respect to the initial profiles of resources they can command, and to their capabilities. These profiles are determined by the firm's initial financial, physical and human capital resources and capabilities. Entrants may: (1) face different constraints that could limit their location choices; and (2) have different production functions and thus different preferences for location attributes. To address the problem of heterogeneity of entrepreneurial entrants, we specify a profit function that recognizes the interactions between firm, location, and industry characteristics. For example, all firms may prefer locations that provide agglomerative benefits. Resource-poor entrants, however, may have a stronger incentive to enter such a location, as their benefits from agglomeration economies may be higher (e.g., they may be able to outsource some activities or share an existing infrastructure). This effect is captured by an interaction term involving a location characteristic (e.g., concentration) and a firm-specific attribute (e.g., initial financial resources). The profit function can be presented as (5) K..k = /3xLjk + fcFy + J33lk + XlLjkFy +ZihLjk + dER< +dk+d,+ £ijk where the first vector Ljk includes location covariates. It represents all observed systematic location-sector specific characteristics that have an impact on profits for all potential new entrants. Specifically, this vector is a linear combination of factors such as localization and urbanization economies, land rents, sunk costs, competition, incumbents' expected entry deterrence strategy, and unemployment rate. The second set of vectors includes interaction terms. The vector Fijk specifies measures of firm-specific attributes including relative scale efficiency, leverage, productivity, and quality of human capital. Following Geroski, et al. (2002) all our firm-specific variables are measured at the end of the first year of operation 1 2. The vector Ik is specified as the linear combination of industry traits that affect the profitability of all entrants to that industry. Our location and industry variables are measured at period t-1, before entry is observed. The third set of variables control for many permanent and time-varying factors such as location, industry, and time-fixed effects. E R effects dER/, captures all other permanent factors common to the location (ER1) that are not observed by the econometrician 1 3, dk absorbs permanent industry characteristics, and dt is a time effect that captures the macroeconomic conditions as well as general trends in technology 1 4. Last, the idiosyncratic component of entrant /'s profits from operating in location j belonging to sector k, as well as the intrinsic randomness associated with the entrant and the choice is denoted eiik. 1 2 Note, for the median firm, these measures will represent its resource endowments and capabilities after six month of operation (see data sources section), which is a good proxy for the initial conditions. Nevertheless, the qualitative pattern of the results is similar when we re-estimate our models using firms' first full year as corporations to measure initial resource endowments and capabilities. 1 3 We use location fixed-effects to avoid identifying and measuring all other factors common to the location that may influence the profitability of a location such as public infrastructure, efficiency of local authorities, and weather conditions. 1 4 See robustness section for additional interaction dummy effects. Data Sources The data sets used to estimate our model are T 2 - L E A P and the Census of Population (for years 1981, 1986, 1991, and 1996). T 2 - L E A P is a merger of two different Canadian databases. The first database, the Longitudinal Employment Analysis Program ( L E A P ) , is used to identify new entries, 3-digit SIC codes, number of employees, and their location choice. The second database is The Corporate Tax Statistical Universe File (T2SUF). This database is used to assess initial firm-specific financial variables such as equity, assets, sales, and closing inventories, which are converted to constant Canadian dollars using a 1986 price index. T 2 - L E A P is a unique, firm-level database that includes all incorporated employers in Canada. The database tracks employment and payroll characteristics of individual firms from their year of entry to their year of exit and allows to determine the time of entry and exit with precision 1 5 . Births (entrants) in any given year are firms that have current payroll data, but did not have payroll data in the previous year. In our empirical estimation, we include only new entry (also referred as 'de novo', independent, or entrepreneurial entry 1 6); we do not include births of diversifying entrants or parent-company ventures (Helfat and Lieberman 2002) . New 1 5 Every employer in Canada is required to register a payroll deduction account with Revenue Canada for the purpose of unemployment insurance), and issue a T4 slip to each employee that summarizes earnings received in a given fiscal year. The LEAP database includes every business that issues a T4 taxation slip. For each year, total payroll and employment are calculated. The latter is the average annual count of employees within the firm, or Average Labor Units (ALUs). ALUs are calculated by taking the total payroll of the enterprise for the year, then dividing by the average annual income for workers in the relevant province, size class, and industry (at the 3-digit SIC level). 1 6 Self-employed accounted for 17% of total employment in Canada in 1997. However, a substantial proportion of the self-employed are not creating production entities of any substance - either in terms of sales, employment or capital formation (Statistics Canada 1997). Because of difficulties in measuring self-employment and the conceptual problems in equating it to the creation of new enterprises, we follow previous research and capture a new firm when it first hires employees. 1 7 We exclude from our sample spin-offs that started as subsidiaries or divisions of incumbents and later transformed to independent establishments (also known as parent spin-offs). We cannot, however, differentiate between new entry and firms which started by executives of an incumbent leaving to start their own firms (also known as entrepreneurial spin-offs). Such spin-offs were found by Klepper and Sleeper (2000), and Klepper (2002, 2003) to have an important determinant of geographic concentration in the U.S. automobile and tire industries. entry accounts for 85% to 94% of all newly created establishments, depending on the sector 1 8. Similarly, deaths (exits) in any given year are identified by the absence of current payroll data, where such data had existed in the previous year. A special labor tracking mechanism allows us to exclude mergers, changes in control, changes of name or location as (false) exits and subsequent (false) entries (Appendix describing this mechanism is available upon request). Our database covers the years 1984 to 1998. As with all studies of location choices, it is important to include only feasible locations in our sample. Since regulations and zoning restrictions prevent some CDs from having industrial activity, we excluded from our choice set those CDs that did not exhibit any existing manufacturing activity or new entry over fifteen years. A l l of these characteristics make this data set an excellent source for studying the characteristics of new entrants and their location choices. Aside from allowing us to identify new entries and their location choice, our database permits us to compute a number of covariates that we wi l l use to test the hypotheses formulated above. The Census of Population is used to derive a measurement of local unemployment, site area, average value of dwellings, and population density. The Annual Survey of Manufacturing is used to derive a measurement of capital share. Variable Definitions Firm-Specific Attributes Size. We represent the initial endowment of physical resources through R E L A T I V E S C A L E , which indicates the size of the firm relative to the industry's average. The industry average 1 8 Establishment is not necessarily equivalent to a firm as some firms have more than one establishment. However, 89% - 96% of new entries are independent firms. reflects the importance of scale economies and provides a measure of the Min imum Efficient Scale (MES) in the industry. The indicator is defined as the ratio of the size of the firm at the end of the first year to the average size of incumbents (both measured in A L U s ) . A ratio of 1 or higher suggests that the firm operates at an efficient scale. Lower ratios indicate \"weakness\" in the initial endowment of physical resources. Since, during the initial period (entry), firms may not be able to realize the full scale of their planned entry, we use the R E L A T I V E I N I T I A L SIZE as an alternative test, which is measured by a firm's A L U s , relative to all other entering firms within the sector at the same year. Financial Resources. Several accounting ratios can be used to measure the initial financial position of a firm, including: leverage and liquidity ratios. Leverage indicates a limit on the ability of an entrant to obtain further financing for growth 1 9 . Liquidity ratios test the degree of firm solvency and may also indicate the existence of slack resources. We use in our model L E V E R A G E to measure financial resources while the current ratio is used to check the robustness of our results to the choice of this proxy. Quality of Human Resources. Higher wages tend to reflect a greater investment in certain labor-related sunk costs, such as training and firm-specific human capital. The literature on wage efficiency shows that firms tend to pay a wage rate above the market clearing wage to attract and retain high-quality labor and to provide incentives for workers to exert more effort. As an indirect measure of the initial quality of human capital resources available to a firm, we employ R E L A T I V E A V E R A G E W A G E , which is defined as the average wages paid by a firm divided by the industry's average wages in the census division. A ratio of 1 or higher suggests that the quality of human capital that the firm employs is at least comparable to its competitors. Lower 1 9 Firms can have debts that exceed the book value of assets. For example, firms may borrow money based on their intellectual property or business plan that is not reported as an asset. In order to avoid outliers in our regressions we eliminate all observations for which debt to asset ratio exceeds 3 which are likely to reflect a measurement error. ratios indicate \"weakness\" in the initial quality of human resource endowments. Since new firms may have to pay higher wages than incumbents, to compensate for risks, an alternative measure of R E L A T I V E I N I T I A L W A G E was defined where the comparison group consists of all new entrants to the sector 2 0. Productivity. The effective deployment of resources is measured by the P R O D U C T I V I T Y of the firm. Our database does not contain sufficient data for classical measures of Total Factor Productivity (TFP); however, it is possible to calculate Approximate Total Factor Productivity (ATFP) . A s originally suggested by Griliches (1990), and more recently by Ha l l (1999), this measure of productivity is derived from a simple Cobb-Douglas production function. Suppose that firm i has a certain productivity level A , and produces output F, using capital Kt and labour Li. The firm's production function is: (6) Yt = AiKiaL}~a. If we solve for productivity, A,-, and take the natural log of both sides, the equation can be rewritten as: - G r i n U J U J (7) ln (A) = ln Equation (7) describes the efficiency of the firm at turning inputs into outputs. This is comprised of the firm's labor productivity and the amount of capital each worker has at their disposal. Labor productivity is measured as total sales divided by the number of employees ( A L U ) . N o measure of capital per worker is present in T 2 L E A P , however, a measure of total assets is available. We use total assets minus closing inventory and divide the result by the number of 2 0 It should be noted that compensation may include equity and stock options. However, because of competitive pressures, compensation practices seem to converge within a location and a sector. Thus, our measure which is based on the relative position of firms within a sector in the location is likely to reflect this heterogeneity. Differences in firm specific practices are idiosyncratic and are included in the error term. employees. Removing closing inventories leaves us with a good measure of the efficiency with which workers turn inputs into outputs, using their available resources. The optimal capital share a, varies significantly from industry to industry in the manufacturing sector. The Annual Survey of Manufacturing is used to derive this share. The natural log of A T F P for a given firm is defined as: (8) ln(ATFPi) = \\n r sales, ^ f -aln l K alui j V assets (. — inventories i alu-where alu-, is the average labor units (i.e., total employees) of the firm, sales-, is its total sales, assets, is its total assets, and inventories. is the closing inventories of the firm (all measured by the end of the first year of operation). Location Characteristics Localization Economies. We follow the tradition in the literature of regional economics and use the share of manufacturing employment in the C D belonging to the 3-digit SIC to measure L O C A L I Z A T I O N E C O N O M I E S . This measure of localization economies is widely used, for example see Feser and Bergman (2000), Glassman and Voelzkow (2001), Porter (2001), and Shaver and Flyer (2000). As a robustness check, we used location quotients. Urbanization Economies. Urbanization externalities reflect the attractiveness of a location to entrants due to industrial cross-fertilization through information spillovers, social networks, and other sources of diversity of the economic activity (Henderson 2000, Jacobs 1969). Glaeser et al. (1992) found that spillovers across industries are more important for startups than are spillovers within industries. Following Henderson (1986), and Rosenthal and Strange (2003), we measure U R B A N I Z A T I O N E C O N O M I E S as employment outside of the industry 2 1. Investment Recoverability. We derive a proxy for the magnitude of I N V E S T M E N T R E C O V E R A B I L I T Y by computing the product of the entry and exit rates of the industry in the location in the year prior to entry (Mata, 1996). Nascent entrepreneurs may infer from high entry and exit rates that they can sell their tangible assets in case of failure and recover a significant proportion of their initial investment. High turnover (reflected in high past entry and exit rates and thus their product) suggest the presence of a volatile ('churning') marketplace characterized by a continual reorganization of firms. Low local turnover indicates high sunk costs, since sunk costs create a zone of inaction. The advantage of using both entry and exit rates to indicate sunk costs is that both exogenous and endogenous sunk costs are accounted for (Sutton, 1991) 2 2. Land Rents: We use establishment density to proxy industrial L A N D R E N T S . A s alternative proxies, we use in robustness tests, L A N D A R E A of the C D , P O P U L A T I O N D E N S I T Y , and A V E R A G E V A L U E of dwelling in a C D 2 3 . Unemployment Rate: We measure the unemployment rate at the C D where entry occurred at time t-124. Market Competitiveness. As an indicator of M A R K E T C O M P E T I T I V E N E S S , we use the number of establishments per worker at the location in the focal sector (this measure was used in Alternative measures for urbanization economies include total employment per square kilometre (Bartik 1985, Coughlin, et al. 1991, Figueiredo, et al. 2002) and a Herfindahl-Hirschman index of employment diversity at the MSA level (Henderson, 1995). 2 2 An industry with exogenous sunk costs has costs and demands that are given. An industry with endogenous sunk costs is one in where the strategic decision of whether or not to sink certain costs is a key in determining the firm's competitive position (Sutton, 1991). Outlays such as establishing trademark, goodwill, or buyers' perceptions of superior product quality, enlarge the firm's market share and put pressure on rivals to either imitate the strategy or to exit. 2 3 Due to availability of data, AVERAGE VALUE of dwelling is used for the sub-sample of the population. 2 4 Unemployed data were available census years 1981, 1986, 1991, 1996 and interpolated between census years. Glaeser et al. 1992, Rosenthal and Strange 2003). As this ratio increases, the local market in a given industry in the location is thought to become more competitive. Incumbents' Propensity to Deter Entry. We use the product of industries' average profit margins and an index of concentration in the C D as a proxy for the propensity of incumbents to engage in strategic behavior towards entry . The costs of coordination are lower when there is smaller number of large incumbents in a location. Concentration may also indicate the presence of economies to scale which facilitate preemption. High levels of profits suggest a strong motive to deter entry as well as deep pockets to fund deterrence. Industry profit margin is measured by the ratio of profits over sales of all incumbents in the sector in the C D . The location concentration ratio is defined as the employment of the four largest firms in the C D over the total local industry employment 2 6. Industry Growth. Our measure for I N D U S T R Y G R O W T H R A T E is defined as the percentage change in employment in the location during the year prior to entry (this measure was used by Audretsch and Mahmood 1995a). Fixed Effects. We use time dummies to control for macroeconomic conditions, and 2-digit SIC industry-fixed effects in all of our estimations. Industry-specific characteristics that may affect 27 entrepreneurial enterprises' profitability are similar to those affecting incumbents' profits . Industry characteristics were found to account for about 75 percent of the variation in industry average profitability (Schamalensee, 1985). Thus, we expect that the likelihood of entry is higher in more profitable, innovative, and growing industries. While other measures may be used to predict the intensity of incumbents' deterrence activity such as patenting and branding expenditures, or investments in capacity expansion in the pre-entry period, they are likely to be correlated with price-cost margin (Barbosa 2002). 2 6 We used Herfindahl index as an alternative measure of local concentration as a robustness check. 2 7 Since some of the covariates vary by 3-digit SIC level we use 2-digit SIC level fixed effects. Moreover, industry fixed effects are expected to operate at higher aggregation level than 3-digit. Empirical Results The population of new entrants in Canada consists of predominate small enterprises (59% of all new entrants have less than 10 employees while only 3% have more than 100 employees). The average size of a census division in our sample is 5,704 square kilometers. Descriptive and summary statistics of the explanatory variables are provided in Table 1.1. The location and industry characteristics are lagged by one year. Table 1.2 presents the correlation among the variables. TABLE 1.1: Variables Definition and Descriptive Statistics Variable Definition Mean S.D. Location attributes Localization Economies Share of manufacturing employment in the Census Division (CD) in the same 3-digit SIC as the entrant. Source: T2LEAP 0.027 0.088 Urbanization Economies * Number of manufacturing employees outside the sector in scope in the CD. Source: T2LEAP 9.610 2.729 Market Competitiveness Establishments per worker in the sector at the CD. Source: T2LEAP 0.055 0.096 Incumbents' propensity to deter entry Share of manufacturing employment of the four largest establishments in the focal sector in the CD multiplied by the ratio of profits over sales of all incumbents in the sector in the CD. Source: T2LEAP 3.628 66.73 7 Investment recoverability Product of the entry rate and the exit rate in the focal sector (SIC3) in the CD. Source: T2LEAP 0.029 0.234 Land rents Establishments density: number of establishments in a CD divided by the area of the CD [square kilometers]. Source: T2LEAP, Census of Population 1981, 1986, 1991, 1996 0.685 1.649 Unemployment rate Unemployment rate at the CD. Sources: Census of population 1981, 1986, 1991, 1996. 9.810 3.511 Industry Industry growth rate Percentage change in employment in the CD in the focal sector. Source: T2LEAP 0.266 5.633 Initial Firm Attributes Relative scale (physical resources)* Firm size measured by Average Labor Units (ALUs) divided by incumbents' average size for the sector. Source: T2LEAP -2.613 1.885 Leverage (financial structure)* Leverage. Source: T2LEAP -0.891 0.677 Relative average wage (human resources)* Average wages paid by a firm divided by the sector's average wages in the CD -0.024 0.155 Source: T2LEAP * Natural logarithm is used in all regressions. Descriptive statistics for initial firm attributes are reported for the sample of 46,620 entrants. Productivity ln(ATFPi) = ln salesi ^ -aln assets( - inventoriesi 2.469 1.197 alut where alu-, is the average labor units or total employees of the firm, sales, is the total sales of the firm, assets, its total assets of that firm and inventories. is its closing inventories, all measured by the end of the first year of operation. Source: T2LEAP; Annual Survey of Manufacturing Fixed-effects Economic Region Economic region fixed-effects Industry 2-digit Industry fixed-effects Time Year dummies TABLE 1.2: Correlation Matrix (1) (2) (3) (4) (5) (6) (7) (1) Localization economies 1.0000 (2) Urbanization Economies -0.1453 1.0000 (3) Market Competitiveness -0.0851 0.1277 1.0000 (4) Incumbents' propensity -0.0123 -0.0014 0.1549 1.0000 (5) Recoverable Costs 0.0015 -0.0063 0.0397 0.0041 1.0000 (6) Land Rents 0.0340 0.3552 -0.1274 -0.0161 -0.0248 1.0000 (7) Unemployment Rate 0.1182 -0.3213 0.0199 -0.0179 -0.0255 0.0707 1.0000 (8) Industry Growth Rate 0.0011 -0.0069 -0.0141 -0.0011 0.0043 -0.0124 0.0037 (1) Relative Scale (2) Leverage (3) Relative Average Wages (4) Productivity (1) (2) (3) 1.0000 -0.0045 1.0000 0.3019 -0.0072 1.0000 -0.2059 -0.0043 -0.1218 Table 1.3 presents the coefficients generated by the maximum likelihood estimation of the nested logit model (equation 4). In this specification, the choice sets are defined as all alternative Census Divisions (CDs) in the Economic Region (ER) where entry is observed 2 8 . Eleven ERs in the dataset included only one C D and therefore were removed. This reduces the set of choosers in our sample from 46,620 to 43,324. The choice sets vary in size from 2 to 15. On average, 6.32 location choices were available per new entrant. Additional location attributes may also affect the expected profitability of locating in a C D . These include political stability, labor market policies, quality of workforce, infrastructure systems, wage rates, and natural attributes of the E R such as climate. Given the wide range of attributes, identification and measurement of all of them are infeasible, thus, the estimation would result in an omitted variable bias. Since these attributes are common factors to the E R , we introduce ER-specific fixed effects in our estimation. We also include sector-specific and time-fixed effects. The first column of Table 1.3 provides the benchmark model estimates where we assess the impact of location characteristics on choices of new entrants. The estimated coefficients reveal that new entrants locate in CDs where strong localization and urbanization economies exist and where the sunk costs associated with the investment are low. The probability elasticity that entrant i chooses C D j in sector k given a percentage change in location characteristics Ljk is: On average, Pijk is the inverse of the number of choices available to each entrant. Evidence in the literature suggest that entrepreneurs choose among locations close to their existing area of economic activity. This preference can be explained by costs associated with acquiring information about the characteristics of feasible locations (Pascal and McCall 1980, Figueiredo et al 2002), social capital and professional networks (Sorenson and Audia 2000), and private information about local opportunities. E (Train 2002, p. 63) TABLE 1.3: Nested Logit Model of Census Division / Economic Region Choice Dependent variable: Location choice Specification 1 2 3 Localization Economies 3.079*\" (0.755) 5.838*** (0.196) 6.219\"* (0.211) Urbanization Economies 0.498\"\" (0.005) 0.506\"* (0.005) 0.505*'* (0.005) Market Competitiveness -2.542\"' (0.112) -2.558*\" (0.112) -2.535\" (0.112) Land Rents 0.145\"' (0.005) 0.143\"* (0.005) 0.143\"\"\" (0.005) Incumbents' propensity to deter entry -0.005\"\" (0.001) -0.005**' (0.001) -0.006\" (0.001) Investment Recoverability 0.535*\" (0.023) 0:533*\" (0.022) 0.534*\" (0.022) Unemployment -0.047*** (0.004) -0.047*** (0.003) -0.047\" (0.003) Relative scale & Localization 0.557*\" (0.043) 1.197*\" (0.010) (Relative scale) A 2 & Localization -0.109\" (0.015) Leverage & Localization -0.074 (0.123) -0.035 (0.143) Relative average wage & Localization 6.311\"' (0.501) 3.365*\" (0.651) (Relative average wage) A 2 & Localization -11.416* (1.617) Productivity & Localization 0.135 (0.088) 0.105 (0.092) Leverage & Land rents Relative scale & Investment recoverability 4 5 6 7 8 9 6.217'\" (0.212) 6.233*** (0.212) 7.220*\" (0.217) 7.274*** (0.220) 6.770*\" (0.216) 6.651*\" (0.215) 0.504\"* (0.005) 0.504**\" (0.005) 0.510\"' (0.005) 0.511**' (0.005) 0.512*\" (0.005) 0.515*** (0.005) -2.534'\" (0.113) -2.559\"* (0.113) -3.865'\" (0.125) -3.870\"' (0.126) -9.649*** (0.423) -10.385\" (0.439) 0.140*\" (0.005) 0.140*\" (0.005) 0.149\"* (0.005) 0.149\"* (0.005) 0.147*'* (0.005) 0.145*** (0.005) -0.006*'* (0.001) -0.007\"* (0.001) -0.009\"' (0.001) -0.006\"*\" (0.001) -0.010*** (0.001) -0.010*' (0.002) 0.534\"' (0.023) 0.170\"' (0.058) 0220*\" (0.058) 0.238\"* (0.058) 0.285*** (0.057) 0297\"* (0.059) -0.046\"* (0.003) -0.046\"* (0.004) -0.054\"* (0.008) -0.055\"*\" (0.008) -0.060*** (0.008) -0.058** (0.008) 1.198— (0.096) 1.213\"* (0.096) 1.633\"* (0.096) 1.666\"** (0.102) 1.515*** (0.100) 1.509\"* (0.100) -0.109\"** (0.015) -0.111\"\" (0.015) -0.141*** (0.015) -0.146*'* (0.015) -0.137'*' (0.015) -0.137\" (0.015) -0.129 (0.125) -0.117 (0.126) -0.178 (0.125) -0.165 (0.126) -0.121 (0.126) -0.096 (0.127) 3.303*\"\" (0.652) 3.228\"*\" (0.652) 3.041\"* (0.658) 3.185*** (0.661) 3.923*** (0.674) 4.005'*' (0.675) 11.509*\" (1.619) -11.521*** (1.619) -12.449*** (1.555) -12.481'** (1.552) -12.549*\" (1.579) -12.744\" (1.589) 0.102 (0.091) 0.100 (0.091) 0.090 (0.096) 0.084 (0.097) 0.108 (0.095) 0.110 (0.093) 0.071'\" (0.013) 0.071*\" (0.013) 0.073*\" (0.013) 0.072*** (0.013) 0.066**' (0.013) 0.067— (0.013) -0.142\"* (0.020) -0.116\"* (0.020) -0.108*\" (0.020) -0.092'** (0.020) -0.092\" (0.020) Relative scale & Investment recoverability -0.142\"* (0.020) -0.116*\" (0.020) -0.108*\" (0.020) -0.092\"* (0.020) -0.092*** (0.020) Relative average wage & Investment recoverability 0.561 (0.296) 0.009 (0.319) -0.006 (0.320) -0.034 (0.316) -0.059 (0.322) Relative scale & Unemployment -0.014\"* (0.0020) -0.014\"* (0.0020) -0.016\"* (0.0017) -0.016— (0.0018) Relative average wage & Unemployment 0.381\"* (0.008) 0.382\"* (0.009) 0.429*\" (0.009) 0.423*\" (0.009) Productivity & Unemployment -0.0071\" (0.003) -0.0074\" (0.003) -0.0073\" (0.003) -0.0079\" (0.002) Productivity & Incumbents' propensity -0.0017\"* (0.0005) -0.0018*** (0.0005) -0.0017\" (0.0005) Relative scale & Incumbents' propensity -0.0009*\" (0.0003) -0.0014*\" (0.0002) -0.0014\" (0.0002) Relative average wage & Incumbents' propensity -0.012\" (0.005) -0.011\" (0.005) -0.013\" (0.005) Productivity & Market competitiveness 0.833*\" (0.117) 0.872\"* (0.122) Relative scale & Market competitiveness -1.132\"* (0.076) -1.143— (0.079) Relative average wage & Market competitiveness -3.172\"* (0.162) -3.400— (0.165) Localization & Industry growth rate 0 .507— (0.078) Market Competitiveness & Industry growth rate 3.310— (0.160) Investment recoverability & Industry growth rate -0.086— (0.027) Log-likelihood -47,371 -46,959 -46,425 -46,383 -46,131 -45,576 -45,543 -45,185 -44,646 Likelihood Ratio Index 0.227 0.234 0.242 0.243 0.247 0.256 0.257 0.263 0.271 number of choosers 43,324 43,324 43,324 43,324 43,324 43,324 43,324 43,324 43,324 number of observations 215,668 215,668 215,668 215,668 215,668 215,668 215,668 215,668 215,668 Standard Errors in parentheses \"* significant at 1% level; \" significant at 5% level ER, SIC and Year fixed effects are included in all models The Likelihood ratio test is computed as 1-L1/L0, where L0 is the constant only log-likelihood and L1 is the full model log-likelihood. Firm-specific attributes are included in specifications 2-9 Examining the probability elasticity reveals that an increase of 1 percent in the share of manufacturing employment in the focal sector in a C D would increase the likelihood of an average C D being chosen by entrant from the same sector by approximately 2.59 percent. The results also indicate that entrants avoid competition, prefer locations where incumbents are less likely to engage in entry deterrence, and are detracted from locations with higher rates of unemployment. The results indicate that increasing market competitiveness by 1 percent for the average C D (measured by establishments per worker in the sector at the C D ) would decrease the probability of it being chosen by entrants to that sector by 2.14 percent. One surprising result is the positive and significant coefficient of land rents . This result suggests that new entrants interpret land rents as signals of the quality of a location in their location search. Alternatively it is possible that land rents are correlated with other unobserved location attributes that are not controlled in the model 3 0 . Columns 2-9 (Table 1.3) present estimated results of models that include interactions of location and firm-specific attributes. Column 2 adds the interactions of localization economies with the initial resources and capabilities of entrants. The interaction effects of productivity and leverage with localization economies are not significant. The positive and significant coefficients of the interactions of relative scale 3 1 and relative average wage with localization economies suggest that larger firms, and those with a higher quality of human capital, expect to benefit more from agglomeration externalities than those which are smaller and/or have lower-quality of human capital. The result confirms that a favorable, rather than the hypothesized adverse selection process is present in entry to location with high agglomeration economies . This suggests that: (1) economies to scale are present in exploiting agglomeration externalities; and (2) higher-quality human capital facilitates absorption of knowledge spillovers. Possibly, as firms increase in size, the marginal benefits of agglomeration externalities may decline. For example, as the size of a firm increases, it is more difficult for it to exploit the We have examined the robustness of the relationship by using two alternative proxies for land rents: land area of the CD (intended to represent differences in land supply and therefore land prices), and population density (since residential and industrial users compete for land, population density should reflect the price of this factor). We also estimated the model using data about entrants during 1997 for which we had the average value of dwelling in a CD, derived from the Census of Population of 1996. The estimation of these three alternative models yielded coefficients of the same sign and similar magnitude as those reported here. 3 0 We attempted to refute the second explanation by including an economic region X time interaction effects in order to absorb time-varying ER wide shocks such as fluctuation in commercial real estate prices. The results do not materially change. 31 Since, during entry, firms may not be able to realize their full planned scale, we experimented with an alternative measure for initial physical resource endowments - RELATIVE INITIAL SIZE. This covariate compares the firm size to all other entering firms within the sector at the same year. The mean of this covariate is 0.952 (compared with 0.795 for relative size). Nevertheless, the coefficient of this interaction term retains the same level of statistical significance and approximate magnitude, to suggest the robustness of our result. 3 2 In order to evaluate the effects of location attributes in models with interaction terms we used the deviation from the mean of the relevant choice set (or the mean of the whole sample as a robustness check). existing infrastructure without investing to extend it. Similarly, firms with high-quality human capital may experience a decrease in their marginal benefits from knowledge spillovers as their proprietary knowledge is leaked to rivals. As the quality of their human capital increases, they also may become more attractive targets for recruitment efforts by rivals and see high-quality employees being lured by their local competitors. Clustering diseconomies, indeed, can dominate cluster benefits, resulting in preferences of larger and higher quality firms for isolation. To investigate the balance between localization economies and diseconomies as a function of size and quality of human capital, we have introduced quadratic terms to the model. The results show that adding the quadratic terms significantly improves the explanatory power of the model (see column 3, Table 1.3). The likelihood ratio test (two degrees of freedom) rejects the hypothesis that the added terms have no explanatory power at the 1 percent level. Negative and highly significant coefficients of the quadratic interaction terms imply that the benefits from localization have a diminishing effect, as relative scale, and quality of human capital increase. Taking the derivative of perceived benefits of localization (reflected in an increased likelihood of choosing clusters) with respect to relative average wage, we confirm that the perceived benefits increase at a decreasing rate 3 3. On average, the maximum benefits from localization are achieved by entrants with wage rates 28.1 percent higher than their sectoral average wage in the C D . Beyond this maximum, the perceived advantage of localization declines. Analyzing the nature of the relationship between relative size and localization reveals that the marginal benefits of localization increase at a decreasing rate for all relative sizes within our population 3 4 . These nonlinear relationships help reconcile the apparent inconsistencies between the findings of Shaver and Flyer (2000) and the results of this paper. Shaver and Flyer have used in their study a 3 3 holding other firm characteristics constant Xrelate_wageMocanw,ion = (5Pr/(^localization) I drelative _ wage < 0 3 4 Extrapolation suggests that the function reaches a maximum outside the range of relative size observed in our sample. sample of multinational companies which are larger and have a higher initial quality of human capital than domestic new independent entrants used in this study (e.g., Mitchel l et al. 1994, Mata and Portugal 2000, 2002). Indeed the models we estimated suggest that at certain thresholds of size and quality of human capital the favorable selection of entrants in clusters changes to adverse selection. Column 4 adds an interaction term between leverage and land rents. The coefficient suggests that entrants who can borrow more (i.e. credit-worthy companies) are more attracted to higher-quality locations than entrants with low leverage 3 5. Typically, the assumption is made that higher leverage represents financial weakness as it may constrain further borrowing (Myers 1977, Gertler 1994, Zingales 1998). This may not be the case for new entrants. Lack of a performance history and absence of public information about initial stage companies prevent them from accessing affordable debt making equity such as: personal savings, family and friends, angels, and venture capital the only feasible sources of funding. The ability of entrants to raise debt in their early stages may suggest relative financial strength. Thus, the general notion suggesting that high-leverage firms are more financially constrained, may not apply to new ventures. Column 5 adds the interaction effects of relative scale and human capital with the degree of investment recoverability. As we hypothesized, the coefficient of the interaction is negative and significant, suggesting that investment recoverability receives more attention in the evaluation of location by weaker entrants. Less confident entrepreneurs make smaller initial investments and prefer locations where the perceived salvage value of their investments is higher. The interaction between relative average wage and recoverability is not significant. The magnitude and level of significance of this interaction term does not change when we use an acid-test ratio instead of leverage as a proxy for financial structure. The interactions of relative scale, relative average wage, and productivity with unemployment are introduced in column 6. The results indicate that the likelihood of choosing a location with high unemployment is higher for smaller and less productive entrants, as predicted in hypothesis 6. The interaction between relative average wage and unemployment suggests that firms who pursue a strategy of building their competencies through investment in human capital are attracted to C D s with a higher unemployment rate where they may obtain qualified employees at lower wage rates than in other locations. Columns 7 and 8 present estimation results for models that include interaction effects of incumbents' propensity to deter entry and the competitiveness of the local market with entrant's attributes. The results suggest that entrants with higher productivity, larger scale, and higher-quality of human capital are more concerned about potential incumbents' entry deterrence strategies3 6. Thus, hypothesis 4 is rejected. The findings are consistent with the observations of Gelman and Salop (1983) and Scherer and Ross (1990) who suggested that larger and stronger firms are usually the targets of incumbents' reaction while smaller or weaker companies present less of a threat to them and therefore encounter less of an effort to deter their entry in concentrated markets. Consistent with our hypothesis 5, firms with high productivity are less detracted from competitive markers. However, smaller entrants and those with lower-quality human capital are less detracted by competitiveness of the market, in contradiction to the prediction of the hypothesis. This may reflect the fact that the existence of many small firms, rather than a few large ones, may reduce entry barriers making clusters somewhat more welcoming to new enterprises (Saxanian 1994, Porter 1998). The stronger firms see less value in lower barriers of entry offered by the competitive markets. 3 6 We also experimented with an interaction between leverage and incumbents' propensity. The coefficient is not significant and therefore omitted from the table. In the last column of Table 1.3, we examine the effect of industry growth on location attributes. The results indicate that during periods of sectoral expansion, the perceived benefits from locating in CDs with higher localization economies increase. Furthermore, the detractive effect of local competition decreases, and the impact of investment recoverability is reduced. Limitations and Robustness Checks We assumed that entrants choose locations within their economic regions. We tested whether expanding the choice sets may lead to different results by relaxing this assumption and re-estimating the model with an extended choice set. The extended choice set consists of all alternative CDs in a province - on average, 28.2 location choices are available per new entrant. Our results did not materially change, to indicate that they are robust for changes in the choice set. The assumption underlying an estimation of conditional logit is Independence from Irrelevant Alternatives (IIA). This property implies that the ratio of the probabilities of choosing 37 one alternative over the other does not depend on the availability (or lack of) other alternatives . The Hausman and McFadden (1984) test revealed that there is no serious violation of the 38 assumption. We also tested the model on various sub samples of the choice set . The restricted models for all specifications generated coefficients which were not substantially different from those obtained for the unrestricted models. Violation of the IIA property due to unobserved attributes of the choosers, or locations, and thus correlated error terms, will lead to biased estimates (Train, 2003: chapters 3,4). Note that using ER and SIC fixed-effects may reduce the bias. 3 8 Specifically, for the choice set of alternative CDs within the ER, we ran tests for all models by excluding several CDs from the choice set. For the choice set that contained all alternative CDs in the province, we excluded ERs that may share some common unobserved characteristics (e.g., Toronto ER in Ontario, or Montreal in Quebec have more supporting services for new ventures than do other ERs in the provinces). Additionally, we eliminated several sectors while maintaining the completed location choice set since entrepreneurs in those sectors may be attracted (detracted) to certain locations due to availability of natural resources, clusters of related industries or research institutions. Specifically, we eliminated the wood pulp and sawmill sectors, transportation sectors, and pharmaceutical and medicine sector In addition to the fixed effects described in the econometric models above we also experimented with the following interactions: E R x year effects to absorb any E R specific time varying shocks that are shared by all firms operating in the E R such as the construction of rail l ink or the opening of an airport; sector x year effects to capture sector specific time-varying shocks, such as the introduction of an industry-specific new technology; and E R x year x industry effects. The qualitative pattern of the results is similar. We also tested the model using the alternative proxies discussed previously to represent key variables and obtained similar results. Conclusion Entry location choice is a multifaceted decision process. Heterogeneous entrants with distinct production functions, entry strategies, resources, and capabilities may value location attributes differently. Our analysis confirms that firms' strengths and weaknesses in different dimensions have distinctive patterns of interaction with location attributes. The paper makes a number of contributions. On the management practice front, our evidence sheds light on the strategic importance of location choices of new entrants. The study unveils the impacts of various positive and negative location attributes that heterogeneous new entrants incorporate in their strategies. The paper provides the patterns of relationships between the profile of initial resources and capabilities and the benefits and costs that accrue from different externalities offered in a location. They also indicate which firms perceive themselves to be more vulnerable to competition and which place more importance on entry barriers. The paper finds that location of new ventures is endogenously determined. Larger entrants and those with a higher-quality human capital, expect to benefit more from localization externalities. The larger companies most likely enjoy economies to scale in identifying, accessing, and exploiting localization externalities, while those with high-quality human capital 37 probably are more capable to absorb knowledge spillovers. As the initial \"strength\" (relative size and quality of human capital) of the firm increases, the positive marginal value of agglomeration economies decreases and may become negative. We found three other factors which explain significant differences in location preferences between weak and strong entrants. Weaker entrants prefer locations with better opportunities for salvaging initial investments. The thresholds to enter and experiment are lower when the potential to recover one's investment in case of failure is higher. Stronger entrants, who do not contemplate failure as being probable, are less concerned with the salvage value of their investment. They are more concerned with incumbents' propensity to engage in entry-deterring activities than weaker firms as they are more likely to draw attention and be targeted. Less concentrated markets are found to be more attractive to smaller firms and those with lower-quality human capital as they have lower entry barriers and more accessible business networks. Our results have interesting implications to cluster dynamics. Initially, as a cluster starts to grow and its density increases, its attractiveness to potential entrants is heightened (Krugman 1991, 1993). A t this stage, a favorable entry selection process increases the share of strong entrants, reflecting the higher value they derive from localization economies. A s density further increases, the formation of a concentration of strong incumbents, or the emergence of a more competitive market, trigger an adverse selection process that may counterbalance the favorable one. A t this stage, as the cluster grows, location preferences of stronger entrants may shift in favor of isolation, to avoid hostile reaction of incumbents and/or the erosion of their competitive advantage. As the cluster approaches its carrying capacity, increases in competition lead to higher mortality rates, and a reduction of its attractiveness to all entrants, but especially the stronger ones. The resulting equilibrium is likely to be characterized by high turnover of weak enterprises. References Aharonson, B.S . , Baum, J . A . C . and M . P . Feldman 2004. Borrowing from Neighbors: The location choice of entrepreneurs. Rotman School of Management, University of Toronto Working Paper. Acs, Z .J . , D . B . Audretsch and M . P . Feldman 1994. R & D spillovers and recipient firm size. Review of Economics and Statistics, 76(2): 336-340. Almazan, A . , A . de Motta and S. Titman 2003. A theory of location choice and the utilization of human capital. Working Paper, Almeida, P. and B . Kogut 1997. The exploration of technological diversity and the geographic localization of innovation. Small Business Economics, 9(1): 21-31. Amit , R. and P. J. H . S. Schoemaker 1993. Strategic assets and organizational rent. Strategic Management Journal, 14:33-46. Audretsch, D . B . and M . P. Feldman 1996. R & D spillovers and the geography of innovation and production. American Economic Review, 630-640. Audretsch, D . B . and M . Fritsch 2002. Growth regimes over time and space. Regional Studies, 36(2): 113-124. Audretsch, D . B . and J. Mata 1995. The Post-entry Performance of Firms: Introduction. International Journal of Industrial Organization, 13(4): 413-19. Barney, J .V. 1991. Fi rm resources and sustained competitive advantage. Journal of Management, 17:99-120. Bartik, T. 1985. Business location decisions in the United States: Estimates of the effects of unionization, taxes, and other characteristics of states. Journal of Business and Economic Statistics, 3:14-22. Baum, J . A . C . and H . A . Haveman 1997. Love thy neighbor? Differentiation and agglomeration in the Manhattan hotel industry, 1898-1990. Administrative Science Quarterly, 42(June): 304-338. Baum, J . A . C . and C . Oliver 1991. Institutional linkages and organizational mortality. Administrative Science Quarterly, 36:187-218. Baumol, W . and R. W i l l i g 1981. Fixed costs, sunk costs, entry barriers and sustainability of monopoly. Quarterly Journal of Economics, 96(3): 95-106. Boeri, T. and L . Bellmann 1995. Post-entry behavior and the cycle: Evidence from Germany. International Journal of Industrial Organization, 13(4): 483-500. Bunch, D . and R. Smiley 1992. Who deters entry? Evidence on the use of strategic entry deterrents. Review of Economics and Statistics, 74:509-521. Cable, J. and J. Schwalbach 1991. International comparison of entry and exit, In P. A . Geroski and J. Schwalbach (Ed.), Entry and Market Contestability: An International Comparison, Blackwell Press: Cambridge, 257-281. Cabral, L. 1995. Sunk Costs, F i rm Size and F i rm Growth. Journal of Industrial Economics, 43(2): 161-72. Carlton, D . W . 1983. The location and employment choices of new firms: A n econometric model with discrete and continuous endogenous variables. Review of Economics and Statistics, 65(3): 440-49. Christensen, C , F. Suarez and J. Utterback 1998. Strategies for survival in fast-changing industries. Management Science, 42(12): 207-220. Chung, W . and A . Kalnins 2001. Agglomeration effects and performance: A test of the Texas lodging industry. Strategic Management Journal, 22:969-988. Cooper, A . C . 1984. Contrasts in the role of incubator organizations in the founding of growth-oriented companies, Babson College, 159-174. Coughlin, C , V . Terza and V . Arromdee 1991. State characteristics and the location of foreign direct investment within the United States. The Review of Economics and Statistics, 73:675-683. Dixi t , A . 1980. The role of investment in entry deterrence. The Economic Journal, 90:95-106. Dudey, M . 1990. Competition by choice: the effect of consumer search on firm location decisions. American Economic Review, 80:1092-1104. Dumais, G . , G . Ell ison and E . L . Glaeser 2002. Geographic concentration as a dynamic process. The Review of Economics and Statistics, 84(2): 193-204. Dunne, T., S.D. Kl imek and M . J Roberts 2004. Exit from regional manufacturing markets: The role of entrant experience. Working Paper. Duranton, G . and D . Puga 2001. Nursery Cities: Urban diversity, process innovation, and the life-cycle of products. American Economic Review, 91(5): 1454-1477. Eaton, B . and R. Lipsey 1980. Entry barriers are exit barriers: The durability of capital as a barrier to entry. Bell Journal of Economics, 11:721-729. Eisenhardt, K . M . and C . B . Schoonhoven 1990. Organizational growth: linking founding team, strategy, environment, and growth among U.S . semiconductor ventures, 1978-1988. Administrative Science Quarterly, 35:504-529. Ericson, R. and A . Pakes 1995. Markov-perfect industry dynamics: A framework for empirical work. Review of Economic Studies, 62(1): 53-82. Frank, M . 1988. A n intertemporal model of industrial exit. Quarterly Journal of Economics, 103(2): 333-344. Gelman, J.R. and S.C. Salop 1983. Judo economics: Capacity limitation and coupon competition. Bell Journal of Economics, 14:315-325. Geroski, P .A . , J. Mata and P. Portugal 2002. Founding conditions and the survival of new firms. Working Paper, Gilbert, R. and D . Newberry 1982. Preemptive patenting and the persistence of monopoly. American Economic Review, 72:514-526. Glaeser, E . L . , H . Kal la l , J. Scheinkman and A . Schleifer 1992. Growth in cities. Journal of Political Economy, 100(6): 1126-1152. Gort, M . and S. Klepper 1982. Time paths in the diffusion of product innovations. Economic Journal, 92(367): 630-653. Griliches, Z . , Mairesse, J. 1990. R&D and Productivity Growth: Comparing Japanese and U.S. Manufacturing Firms. The University of Chicago Press: Chicago. Hannan, M . T . and J. Freeman 1977. The population ecology of organizations. American Journal of Sociology, 82(5): 929-964. Hannan, M . T . , J. Ranger-Moore and J. Banaszak-Holl 1990. Competition and the evolution of organizational size distributions, In J. V . Singh (Ed.), Organizational Evolution: New Directions, Sage: C A , 246-268. Hausman, J. and D . McFadden 1984. Specification tests for the multinomial logit model. Econometrica, 52(5): 1219-1240. Head, K . , J. Ries and D . Swenson 1995. Agglomeration benefits and location choice: Evidence from Japanese manufacturing investments in the United States. Journal of International Economics, 38:223-247. Head, K . , and T. Mayer 2004. Market Potential and the Location of Japanese Investment in Europe. Forthcoming at Review of Economics and Statistics. Helfat C E . and M . B . Lieberman 2002. The birth of capabilities: Market entry and the importance of pre-history. Industrial and Corporate Change, 11(4): 725-760. Helfat, C E , and M . Peteraf 2003. The dynamic resource-based view: Capability lifecycles. Strategic Management Journal, 24: 997-1010. Helsley, R . W and W . C . Strange 2002. Innovation and input sharing. Journal of Urban Economics, 51(1): 22-45. Henderson, R . M . and I . M . Cockburn 1996. Scale, Scope, and Spillovers: The Determinants of Research Productivity in Drug Discovery. Rand Journal of Economics, 27(1): 32-59. Jacobs, J. 1969. The Economics of Cities. Vintage: New York. Jaffe, A . , M . Trajtenberg and R. Henderson 1993. Geographic localization of knowledge spillovers as evidenced by patent citations. Quarterly Journal of Economics, 108(3): 577-598. Jovanovic, B . 1982. Selection and the evolution of industry. Econometrica, 50(3): 649-70. Kalnins A . and W . Chung 2004. Resource-seeking agglomeration: A study of market entry in the lodging industry. Strategic Management Journal, 25:689-699. Kessides, I .N. 1990. Towards a testable model of entry: a study of the U S manufacturing industries. Economica, 57:219-238. Klepper, S. 2002. The capabilities of new firms and the evolution of the U S automobile industry. Industrial and Corporate Change, 11: 645-666. Krugman, P. 1991. Geography and Trade. M I T Press: Cambridge. Krugman, P. 1993. Increasing returns and economic geography. Journal of Political Economy, 99(3): 483-99. March, J. G . and Z . Shapira 1987. Managerial perspectives on risk and risk taking. Management Science, 1404-1418. Marshall, A . 1920. Principles of Economics. MacMi l l an : London. Mata, J. 1996. Business conditions and business starts. International Journal of the Economics of Business, 3(3): 295-305. Mata, J. and P. Portugal 1994. Life duration of new firms. Journal of Industrial Economics, 42(3): 227-45. McFadden, D . 1974. Conditional Logit Analysis of Qualitative Choice Behavior, In C . F. M . a. D . McFadden (Ed.), Structural Analysis of Discrete Data with Econometric Applications, M I T Press: Cambridge, Massachusetts, 105-142. McPherson, J . M 1983. The ecology of affiliation. American Sociological Review, 48:519-532. Mitchel l , W , Shaver J, Yeung B . 1994. Foreign entrant survival and foreign market share: Canadian companies' experience in United States medical sector markets. Strategic Management Journal, 15(7): 555-567. Moretti, E . 2004. Workers' education, spillovers and productivity: Evidence from plant-level production functions. UCLA andNBER Working Paper. Pakes, A . and R. Ericson 1998. Empirical implications of alternative models of firm dynamics. Journal of Economic Theory, 79(1): 1-45. Porter, M . E . 1998. Clusters and the new economics of competition. Harvard Business Review, Nov-Dec:77-90. Porter, M . E . 2000. Location, competition, and economic development: Local clusters in a global economy. Economic Development Quarterly, 14(1): 15-34. Rosenthal, S. and W . C. Strange 2001. The determinants of agglomeration. Journal of Urban Economics, 20(2): 191-229. Rosenthal, S. and W . C. Strange 2003. Geography, industrial organization, and agglomeration. Review of Economics and Statistics, 85(2): 377-393. Scherer, F . M . and D . Ross 1990. Industrial Market Structure and Economic Performance. Houghton Mif f l in : Boston Mass. Schumpeter, J. A 1934. The Theory of Economic Development: An Inquiry into Profits, Capital, Credit, Interest, and the Business Cycle. Harvard University Press: Cambridge. Scott, A . and E . C . Kwok 1989. Inter-Firm Subcontracting and Locational Agglomeration: A Case Study of the Printed Circuits Industry in Southern California. Regional Studies, 25:405-416. Shane, S. and T. Stuart 2002. Organizational endowments and the performance of university start-ups. Management Science, 48(1): 154-170. Shane, S. and S. Venkatataman 2000. The promise of entrepreneurship as a field of research. Academy of Management Review, 25(1): 217-226. Shaver, J . M . and F. Flyer 2000. Agglomeration economies, firm heterogeneity, and foreign direct investment in the United States. Strategic Management Journal, 21(12): 1175-1193. Sorenson, O. and T .E . Stuart 2001. Syndication networks and the spatial distribution of venture capital investments. American Journal of Sociology, 106(6): 1546-1588. Starr, J. , MacMi l l an , I. 1990. Resource cooptation via social contracting: Resource acquisition strategies for new ventures. Strategic Management Journal, 21: 277-292. Stinchcombe, A . L . 1965. Social structure and organizations, In J. G . March (Ed.), Handbook of Organizations, Rand McNal ly : Chicago IL, 153-193. Sutton, J. 1991. Sunk Costs and Market Structure: Price Competition, Advertising, and the Evolution of Concentration. M I T Press: Tirole, J. 1988. The Theory of Industrial Organization. The M I T Press: Cambridge, M A . Train, K . 1986. Qualitative Choice Analysis: Theory Econometrics, and an Application to Automobile Demand. M I T Press: Cambridge, M A . Train, K . 2003. Discrete Choice Methods with Simulation. Cambridge University Press. Webber, M . J . 1972. Impact of Uncertainty on Location. MPT Press: Cambridge, M A . Zucker, L . G . , M . R . Darby and M . B . Brewer 1998. Intellectual human capital and the birth of U . S . biotechnology enterprises. The American Economic Review, 88(1): 290-306. CHAPTER TWO THE SURVIVAL VALUE OF CLUSTERS ABSTRACT We analyze the survival of all de novo entrants into Canadian manufacturing sectors between 1984-1998. We classify firms according to the level of industry clustering in their location. The central questions are whether new firms in cluster and in non-cluster locations experience different chances of survival? If they do, what are the locational and firm specific characteristics associated with survival of new firms that differentiate cluster and non-cluster firms? Controlling for the possibility of selection in cluster entry we find that initial endowment of resources of new entrants provides a safety net for a longer period of time, and a higher survival rates in clusters. Findings also suggest that different strategies enhance enterprise survival in clusters and isolation. Introduction A large percentage of new enterprises fail within a short time after entry (Amit and Thornhill 1999; Audretsch and Mahmood 1993, Caves 1998; Dunne Roberts and Samuelson 1988, 1989; Mata and Portugal 1994, 2004; Romanelli 1989; Shane and Foo 1999; Zingales 1998). Previous research has shown that location choice is an important early strategic decision (Baum and Mezias 1992; Shaver and Flyer 2000). Firms choose locations to leverage their initial endowments of resources and capabilities and overcome entry barriers. They seek locations which provide them with a shelter during their infancy and helping them mature faster. They also consider the longer-term prospects of profitability and survival. The literature provides contradicting findings concerning the impacts of locational attributes on survival. Some studies emphasized survival benefits of co-location derived from agglomeration externalities such as: localized knowledge spillovers, specialization among industry-specific suppliers, and shared pool of skilled labor (e.g., Henderson 1994, 2000; Dumais, Ell ison and Glaeser 2002; Duranton and Puga 2001). Yet, other studies found survival disadvantages which stem from intensified competition (Baum and Mezias 1992) and amplified erosion of competitive advantages (Shaver and Flyer 2000). This paper examines whether there are differences in survival values and whether the path to enhancing survival prospects is similar in clusters and isolation. New extensive longitudinal database which covers all de novo entrants into the Canadian manufacturing sector during the period 1984-1998 allowed us to follow their paths and model for the first time (to the best of our knowledge) the differences between the determinants of failure among firms operating in locations with different levels of industrial clustering. Following findings suggesting that the effects of agglomeration attenuate rapidly with distance (e.g., Zucker, Darby and Brewer 1998, Rosenthal and Strange 2003) this study investigated the effects of co-location in fine-grained geographic units rather than the state or metropolitan statistical area used in previous research. The analysis reveals that clustering matters to survival of new entrants. Moreover, firm specific factors and strategies which enhance survival vary significantly between clusters and isolation. The article is structured as follows. We start by reviewing the relevant literature, and deriving the hypotheses to be tested from various literatures including: entrepreneurship, strategy, organizational ecology, economic geography, and industrial organization. We proceed by presenting the empirical models of the determinants of failure of firms operating in various levels of industry clustering. Next we introduce our data and discuss alternative definitions for the covariates. Results of two different specifications, discussion and description of robustness checks follow. Finally, we draw our conclusions. Related Literature The determinants of firm survival have been researched extensively in the empirical and theoretical literatures. Traditionally, the literature has examined the causes of failure by focusing on the aggregate (industry level) constructs. More recently, as data became available, the focus of the analysis moved to the firm and plant levels. A large empirical literature documents the impact on survival of firm's specific attributes and industry environmental characteristics: The attributes found to be positively correlated with firms survival included: size, age, productivity, growth, employment of high skilled workers, exports, capital intensity and diversity of product mix (e.g., Dunne Roberts and Samuelson 1988, 1989; Audrestch 1995; Mata, Portugal, and Guimaraes 1995; Freeman, Carrol and Hannan 1983; Geroski 1995). Several industry characteristics were found to have negative impact on survival, including: concentration, high entry rates, low sunk costs, and stagnated demand (e.g., Geroski, Mata and Portugal 2002, Mata and Portugal 2002; Carroll and Hannan 2000). These empirical studies stimulated the development of theoretical models of heterogeneous firms operating under different industry structures, exposed to different sources of uncertainty 3 9. Only recently the literature has started to examine the impacts of locational externalities, in particular the clustering of economic activity, on survival. This literature presents opposing theoretical arguments and empirical findings. On the one hand, new firms located in clusters (spatial concentration of firms operating in the same or closely related industries) may benefit from agglomeration externalities. Those externalities can reduce the costs associated with search for knowledge and the creation of innovations (Duranton and Puga 2001, Helsley and Strange 2002), increase their productivity (e.g., Henderson 1994, Dumais, Ell ison and Glaeser 2002), and consequently decrease the hazard facing them. Agglomeration externalities can be categorized into production enhancements and heightened demand. Production enhancements originate from: (1) labor market pooling (Marshall 1920); (2) advantages of backward and forward linkages associated with large local markets (i.e., improving the quality of matching crucial inputs and intermediate goods between suppliers and demanders) (Fujita 2000); (3) shared infrastructure (broadly defined) that is available to firms that locate close to each other (Helsley and Strange \" Underlying these theoretical models is the evidence that firms make their entry investments unsure of their quality and use noisy cost and profit signals to learn about their true productivity or efficiency. Jovanovic (1982) offered a model where firms learn how well they can compete in the market. Firms discovering they are efficient grow and survive; the inefficient decline and fail. The model is termed \"passive learning\" model since it assumes that efficiency remains constant over the lifetime of the firm. Ericson and Pakes (1995) argued that firms (actively) learn how to become more efficient through R&D and exploration (i.e., investments with uncertain outcomes). The model is termed \"active learning\" model since it assumes that a firm can change its efficiency through potentially quality enhancing investments. Pakes and Ericson (1998) checked the implications of the two models against an eight-year panel of Wisconsin firms. The results indicate that the active learning model is consistent with the data on manufacturing firms while the passive learning model consistent with the data on retail trade. Baldwin and Rafizussaman (1995) examine the maturation process of firms that enter an industry by constructing a new plant and investigated the extent to which improvements in the performance of any entry cohort are the result of a selection process that culls out the most inefficient entrants or of a learning process that allows survivors to improve their performance relative to incumbent firms. Both selection and evolutionary learning are found to affect post-entry performance but selection per se is a more important contributor to the overall growth of a cohort. In Hopenhayn (1992) model firms are faced with idiosyncratic productivity shocks based on which they optimize their decisions when to exit the industry. As firms exit, new ones enter. In equilibrium, entry and exit rates are equal, and the distributions of firm size, profits and values are stationary. 2002); (4) technological and knowledge spillovers (Jaffee Trajtenberg and Henderson 1993, Zucker, Darby and Brewer 1998, Almeida and Kogut 1999) 4 0; and (5) information externalities about local demand or the feasibility of production at a particular location that are available to prospective entrants who observe incumbents operating there profitably. Agglomeration wi l l also cause heightened demand in industries where consumers need to personally inspect goods or compare prices. Clustering may reduce consumers' search costs, thus increasing the likelihood of visitation and purchase, compared to firms in a separate location (Baum and Haveman 1997, Chung and Kalnins 2001, Kalnins and Chung 2004). Examining the dynamics of U .S . manufacturing concentration between the years 1972 and 1992 using the Census Bureau's Longitudinal Research Database (LRD) , Dumais et al. found that plants are less likely to close in states that have high concentration of the industry's employment. Bernard and Jensen (2004) used two panels from the U.S . census of manufacturing (with five years interval). Controlling for firm and plant specific characteristics they found that an increase in regional relative specialization in an industry (measured at the labor market area level) decreased the probability of plant exit. Examining the survival of 209 firms operating in Greece, Fotopoulos and Louri (2000) found that firms located in greater Athens (where spatial concentration of industrial activity is very high) faced lower hazard when compared to firms located in the rest of Greece. On the other hand agglomeration also entails diseconomies. Firms that co-locate in physical space experience more intense competition (Carroll and Wade, 1991; Baum and Singh, 1992, 1993), reduced growth rates (Baum and Mezias 1992), and suffer from negative externalities associated with density and congestion such as potential bidding-up prices over Jaffe et al. (1993) found that patent citations are more likely to come from the same state or Metropolitan Statistical Area (MSA) as the originating patent. Zucker el al. (1998) found that the stock of human capital of outstanding scientists in life-sciences impact the entry decisions of new biotechnology firms in a city. specialized and limited (or fixed supply) resources (e.g., wages, land). Moreover, new firms may lose their proprietary knowledge to neighboring incumbents as incumbent firms are better positioned to exploit knowledge (Aharonson, Baum and Feldman 2004). These diseconomies suggest higher failure rates. Shaver and Flyer (2000) examined the survival patterns of 101 foreign direct investment entrants in the manufacturing sector into the United States in 1987. Controlling for industry factors and firm specific dummies at the time of entry, the proportion of industry establishments in the state from which the firm operates (their agglomeration measure) was found to be negatively associated with survival. Their findings could not be generalized to domestic de novo entrants: since foreign investors tend to have established business operations in their home country they differ from domestic entrants in various aspects. Their entry size tends to be bigger, the quality of their employees and productivity higher, they also have better access to financial resources (Mata and Portugal 2002; Bernard and Jensen 2004). Moreover, almost every parent firm in their sample controlled more than one subsidiary in the U S . Multinational multi-establishment firms may shutdown plants to shift production across locations within the firm in order to adjust to changing (national, global) market conditions and are more likely to close a plant (Rodrik 1997, Braconier and Ekholm 2000, Bernard and Jensen 2004). Sorenson and Audia (2000) examined mortality rates of shoe manufacturers operating between 1940 to 1989 in the U.S . Controlling for age, size, number of plants operated by the same parent firm, carrying capacity (import, export and domestic production) they found that local density is negatively associated with survival. Specifically, young shoe plants located in or near large concentrated regions of shoe manufacturing failed at a higher rate than isolated plants. Since the footwear manufacturing industry is characterized by lack of strong scale economies and barriers to entry, limited importance of human capital, and low rate of innovation it is not a typical industry where agglomeration economies may have significant impact on survival of new entrants. Baum and Mezias (1992) explored the survival patterns of 614 hotels operating in Manhattan between 1898 and 1990. They examined the effects of localized competition in terms of size, geographic location, and price on survival. They demonstrated that establishments that are more similar in location, price, and physical size reduce each other's probability of survival. Proximity to tourist attractions or business activity (resulting in heightened demand) impose constrains on the location choice set available to new entrants in Manhattan's hotel industry and practically eliminate the benefits associated with isolation. Moreover, the majority of entrants to this sector are diversifying entrants (Helfat and Liberman 2002, Chung and Kalnins 2001). Thus their findings may not be generalizable to new entrants in the manufacturing sectors. In this paper, we take a more direct approach to the estimation of the impacts of location externalities on survival. Our investigation differs from previous research in several dimensions. First, rather than including a covariate which represents the level of clustering (e.g., Shaver and Flyer 2000, Sorenson and Audia 2000, Bernard and Jensen 2004), or a dummy variable representing locations which exhibit high clustering levels (e.g., Fotopoulos and Louri 2000, Shaver and Flyer 2000), we employ empirical methodology that allows us to test for the differences between the determinants of failure among firms (i.e., structural differences) operating in locations with different levels of industrial clustering. Second, since location of new ventures is endogenously determined (Baum and Haveman 1997, Shaver and Flyer 2000), and differences in resources and capabilities that existed in founding have persistent impacts on survival chances (Dunne, Roberts and Samuelson 1988, Audretsch 1995, Swaminathan 1996, Geroski et al. 2002, Klepper 2002), we control for the effects of entry selection processes. Third, in addition to agglomeration we control for other location-industry factors such as market structure and entry rate. Fourth, we use fine-grained agglomeration measures that are not segmented by political boundaries or statistical units that may bisect clusters. Data constraints forced previous research to use state (e.g., Shaver and Flyer 2000), or Metropolitan Statistical Area ( M S A ) (e.g., Bernard and Jensen 2004) level measures of the agglomerative effect. However, the 'new economic geography' literature suggests that agglomerative forces operate at a much smaller geographic units and may not be captured by crude measures (e.g., Jaffe, Trajtenberg and Henderson 1993; Porter 2000; Rosenthal and Strange 2003; Dumais et al. 2002; Chung and Kalnins 2001). Moreover, our measures allow us to compare the resource endowments of a firm to other neighboring firms operating in the relevant sector rather than using absolute levels or industry comparisons at the national level. Lastly, datasets used in previous research may suffer from survival bias since time intervals between two observations (census years) are five years. One consequence is that large firms are more likely to be in the sample 4 1. Since our dataset includes annual observations we do not have such bias. Hypotheses Development The goal of this section is to discuss the firm specific and local-industry characteristics which are likely to affect the survival of entrepreneurial firms in clusters and non-clusters locations and to develop a set of hypotheses about the expected effects. Size The assumption that large new establishments have better survival prospects than small new establishments is referred by organizational ecologists as the \"liability of smallness\". There are several explanations to this observation. First, larger entrants are more likely to be closer to the minimum efficient scale (MES) of their sector and thus suffer less from cost disadvantage 4 1 Moretti (2004) used the U.S. Census of Manufactures in 1982 and 1992. He found only 42% of the plants exist in both years. resulting from operating further up the cost curve. Firms with larger initial size relative to the M E S w i l l need to grow less (or not at all) to exploit potential scale economies and to become competitive and thus have better survival prospects 4 2. Using the Dun and Bradstreet market identifier file that covers U.S . establishments between 1976 and 1986, Audretsch and Mahmood (1995) found that as the gap between M E S and initial size decreases, the hazard rate also decreases4 3. Second, larger firms tend to be more diversified than smaller firms and thus exposed to lower risks associated with shocks in specific markets. Third, size may reflect the beliefs held by entrepreneurs about their ability to compete. Larger investments are typically made by those entrepreneurs who are more confident about their abilities and/or those with better access to resources. Moreover, in the case of large scale entry, more periods with bad results w i l l be needed to eliminate the ex ante positive profit expectations and lead to exit (Frank 1988). Larger start-ups are better endowed to weather environmental fluctuations than firms which enter relatively small. Forth, small new firms are more likely to employ the less able or experienced managers who are more likely to commit the greatest mistakes (Mata and Portugal 1994). Last, selection processes favor large organizations' access to capital, valuable knowledge and trained workers. Larger organizations tend to be perceived as more legitimate with external stakeholders and thus engender trust. Legitimate firms can obtain resources of higher quality and at more favorable terms, have less difficulty in recruiting alliance partners, and generating consumer goodwill. They can more easily chart a clear course and stick to it (e.g., Baum and Oliver 1991, Hannan and Freeman 1977). Clusters allow nascent entrepreneurs to accumulate the knowledge, social ties, and confidence necessary to mobilize vital resources for a new enterprise (Sorenson and Audia 4 2 However, distinct barriers to entry such as a large MES or high capital intensity could also induce a self-selection process that results in few but high-quality start-ups with above-average survival rates (Dunne and Roberts 1991). 4 3 Baldwin (1995: 210) found that entrepreneurial entrant average initial size was about 17 per cent of that of existing firms, suggesting a large gap to the MES. However, in markets where the majority of entrepreneurial entrants are substantially smaller than MES, we expect relative size to contribute less to the hazard rates. 2000). Since in clusters entrepreneurial entry is more common, the legitimacy of new entrants is likely to be higher. Thus, the disadvantages of small size in the recruitment of partners, securing resources from others (e.g., intermediate goods, venture capital (Stuart and Sorenson 2003)), or gaining good wi l l with customers are particularly mitigated. Entrants in clusters have ample opportunities for outsourcing many activities while focusing on their core competencies. Moreover, firms in clusters have incentives to specialize rather than acquiring broad generalized capabilities. While high cost of capital, and other cost disadvantages have negative effects on the success of market entry, prospering growth in the cluster level may be conductive to survival (Audretsch 1995, p.70-73; Boeri and Bellmann 1995). Thus, we expect that, everything else being equal, the negative effect of size on exit is reduced in clusters. Hypothesis 1. The liability of smallness has lower impact on survival of entrants in clusters relative to isolation locations. Age The \"liability of newness\" (Stinchcombe 1965) argument suggests that new organizations are more prone to fail than older more established ones. Looking across industries, Dunne, et al. (1988, 1989) found that exit rates decline as age increases. Age was found to be more important than size in explaining small firms' exit rates (Wedervang 1965) 4 4 . The \"liability of newness\" is attributed to the time required for new entrants to accumulate resources, knowledge and build capabilities. New firms lack stable ties with customers, and must build trust relationships. In addition they must often cope with inefficiencies and conflicts during the formation process (Stinchcombe 1965, Freeman, et al. 1983). Lack of knowledge and networks with suppliers may result in challenges in maintaining flows of resources from the environment (Hannan and Carroll 1992: 37), and inability to cope with extreme environmental challenges. 4 4 See also surveys of the literature by Caves 1998, Geroski 1995, Sutton 1997. The relationships between age and hazards facing new entrants may not be monotonic. Fichman and Levinthal (1991) and Levinthal (1991) argued that there exist a \"honeymoon\" period during which the initial resource endowment (e.g., pool of financial resources) buffers new entrants and improves their ability to cope with random shocks from the environment. This early stage life-cycle is termed \"adolescence \". Firms with larger initial endowments have longer adolescence, or time to develop their own capital, establish structure, and gain the confidence of external resource holders in the venture's success, and adapt to the environment. A s firms spend these initial endowments mortality rises. Thus, the relationship between age and hazards has an inverted U-shape mortality pattern 4 5. This proposition is supported by empirical evidence (e.g., Mahmood 2000, Henderson 1999, Ranger-Moore 1997). If clusters accelerate learning processes and compensate for some of the disadvantages of youth, than locating in a cluster may moderate the relationship between age and survival. New entrants in a cluster may benefit from knowledge spillovers (e.g., Henderson 2000, Krugman 1991b, Rosenthal and Strange 2003). They may imitate practices of incumbents, for example adopting similar production methods and techniques, acquiring intermediate goods from the same suppliers, and use similar marketing channels (Helsley and Strange 2002). Skilled labor mobility within a cluster may facilitate transfer of know how and encourage learning (Glaeser 1999). In isolation the speed of learning is likely to be lower while the knowledge gap higher in the absence of spillover from incumbents. Thus in a cluster a firm may be able to conserve its initial resources for a longer period and expand its adolescence period. Hypothesis 2. The initial endowments of resources of new entrants provide a safety net for a longer period of time in clusters than in isolation 4 5 This pattern is also consistent with learning models since ill-fated firms may need some experience to be sure of their unfitness (Pakes and Ericson 1998). Human Resources The Resource-Based View of the firm stresses that the competitiveness of the firm is largely determined by its ability to develop and sustain human capital assets that cannot be imitated by competitors and thus contribute to its survival (Barney 1991, Tirole 1988). This assertion was confirmed by recent empirical studies that found that human capital is a good predictor of survival (e.g., Mata and Portugal 2002, Chang 1996, Bogner, Thomas and McGee 1996). Mobi l i ty of high-skilled labor is higher in clusters since labor market search and matching are more efficient and there are many opportunities. Skilled employees may enter smaller or younger companies seeking promotion opportunities confident that in case of failure they can find new jobs with ease. Thus while recruiting high-skilled employees is easier in clusters it may be difficult to retain them. In isolation firms must invest in training of new employees but can more easily retain them (and pay them lower salaries) since the search for alternative employment opportunities is much more expensive (Almazan, de Motta and Titman 2003). Thus, the advantage of possessing high quality human capital is more secure in isolation, while in cluster the chance of appropriation of the firm's competitive advantage by rivals is higher. The survival value of higher quality human capital is therefore higher in isolation. Hypothesis 3. The positive impact of the quality of human capital on survival of new entrants is higher for firms in isolation relative to firms in cluster locations. Sunk Costs High sectoral entry rates suggest that resources can migrate to new firms, increasing the level of investment recoverability and decreasing exit barriers 4 6. The presence of high sunk costs has High entry rate may expose a new firm to more intense competition from those of its own kind. It also may imply that each generation of entrants has to face a continuously renewed challenge posed by the new waves of entrants each year. Thus high entry rates imply high exit rates. Empirical evidence support this proposition. Dunne, et al. negative effects on entry and exit and therefore prolongs survival. This would result in a lower probability of failure for incumbents (Hopenhayn 1992). When resources are immobile (e.g., infrastructure, heavy machinery) or specialized (i.e., tailored to specific industry), high entry rates w i l l increase the demand for second hand assets and thus decrease sunk costs by increasing the salvage value. Vibrant markets for used, specialized and immobile assets in clusters reduce sunk costs (Dixit and Pindyck 1994, Helsley and Strange 1991). We expect that for two otherwise identical new entrants, the one located in a cluster is likely to exit voluntarily sooner than the one located in isolation i f the expected level of profits has not been reached since the value of the option of waiting is lower. Hypothesis 4a) Low sunk costs result in higher exit rates, b) This relationship is stronger in clusters. Competition Competition is a force that increases hazards facing a firm. In competitive markets entry barriers are low and any abnormal profits dissipate as new entrants are attracted to the market. L o w entry barriers also allow weaker companies to enter 4 7. Both low selectivity of entrants and low profit margins explain higher exit rates in competitive markets. Moreover, competitive markets exert a strong disciplinary effect and drive inefficient firms out of the market. Competition intensifies as the degree of overlap in resource requirements between organizations increases (Hannan and Freeman 1977) . Since firms in clusters are likely to draw their resources from the same local (1988) and Mata and Portugal (1994), for example, found that there is a strong positive correlation between the flows of entry and exit across markets. 4 7 When entry barriers are high and markets become more concentrated, the barriers shield incumbents from competition, allowing them to sustain above normal profits, and thus result in higher survival rates. New entrants who can jump the entry barriers are also shielded from competition. 4 8 Since organizations of different sizes in a population may use different mixes of resources, the intensity of competition between local organizations depends on their sizes (Hannan and Freeman 1977, Hannan, et al. 1990). pool (e.g., labor, intermediate goods) they compete more intensely with local rivals (e.g., Baum and Mezias 1992, Deephouse 1999). Competition on scarce resources limits the performance of the firm and increase failure rates. Empirically, Carroll and Wade (1991) and Baum and Singh (1994) found that the addition of a competitor to a population had a greater impact on the failure rates at the local level. Investigating the competition among Manhattan hotels Baum and Mezias (1992) found that it intensifies with geographic location, size and price similarities. Moreover, the \"density delay\" argument (e.g., Carroll and Hannan 2000; Ch . 11, Ranger-Moore 1997) suggests that organizations founded in periods when markets are very crowded are weaker (i.e. have higher age specific rates of mortality) than organizations founded in periods when the market is less densely populated. Geroski, Mata and Portugal (2002) suggested two explanations for this phenomenon. The first explanation, called the \"liability of scarcity\", asserts that organizations created in unfavorable circumstances are in relatively bad shape and less robust: \"...are unlikely to be anywhere near their optimal structure configuration and, in addition, may not be able to find the right kinds of resources, make the correct organization specific investments, or design the right kinds of routines\" (p.5). The second explanation suggests that organizations which have been set up under crowded market conditions may be pushed into a \"tight niche packing\" - unpromising niches which may be transitory or may lead them to develop knowledge and routines which are so specialized that they wi l l never be able to reposition themselves into more favorable parts of the market later on. Entering a niche is more common in clusters than in isolation since industry concentration generates demand for specialized products and services. Hypothesis 5 The negative impact of competition on survival of new entrants is higher in clusters relative to isolation locations. Controls Variables Other factors may influence the survival of new entrants, which i f uncontrolled, may lead to spurious findings. Accordingly, we control for various additional firm and industry level characteristics. Firm level Growth. Adjustment costs prevent firms from changing their sizes instantaneously in order to achieve optimal size reflecting their past and predicted outcomes (Penrose 1959). Therefore, firm converge gradually to their desired size (Bogner et al. 1996). Past growth may signal positive past performance, positive expectations, the intention to grow further in the future, and thus lower hazard than the current size indicates (Mata et al. 1995). Productivity. Less productive firms exit the market and are being replaced by more productive ones. Using Canadian manufacturing data for the period between 1970-1979, Baldwin (1995) found that at birth, the labor productivity of new entrants averaged about 73 per cent of incumbents and after a decade, the productivity of survivors and incumbents was about equal. The probability of exit by less productive new firms (on average 79% as productive as continuing firms) was found to be significantly higher than the probability of exit by more productive new firms. Financing Structure. The financing structure of a firm has been hypothesized to affect survival. Several scholars argued that a lack of financial resources early on in an enterprise life is a key component of the liability of newness (e.g., Stinchcombe 1965). Slack in financial resources provides high flexibility to managers and their strategic options (Amit and Schoemaker 1993), and allows firms to take calculated risks with long-term projects and experiment with new products or new markets (March and Shapira 1987). Such experimentation in entrepreneurial firms has been linked to survival (Christensen, et al. 1998). Researchers have argued that slack funds can be directed toward projects with uncertain outcomes, and provide flexibility and speed in reacting to exogenous or endogenous triggers for change, thus, supporting survival (Levinthal and March 1981). Myers (1977) and Gertler (1994) suggested that the liquidity requirements created by external debt may limit a firm's ability to finance new projects or further invest in existing projects, even those with positive net present value. B y limiting the firm's ability to finance new growth opportunities, the liquidity constraint imposed by existing leverage may reduce chance of survival. Firms with high levels of leverage face substantial cash flow demands to service their debt and these costs infringe on their ability to take advantage of opportunities and cope with threats. Highly leveraged firms are more likely to exit since debt holders w i l l exert pressure to liquidate or sell assets when the firm fails to meet its interest payment. Moreover, debt holders, being the first on the creditors list, w i l l fight against continued operation i f the net value they receive from selling assets exceeds their expected proceeds from continued operations (Schary 1991). Zingales (1998) demonstrates that even when controlling for firm's efficiency level, leverage is still relevant to the probability of survival. Thus, everything else equal, we expect firms with high leverage to have lower survival rates. Industry Level Market Growth. Increasing demand in a specific industry is expected to decrease competitive pressure and permit growth and therefore should decrease the hazard rate confronting them. In fast growing industries firms may grow without inflicting market share losses on their rivals and, therefore, the likelihood of aggressive reactions by incumbents is lower. Macroeconomic Environment. Macro-economic conditions reflecting the impacts of business cycles, movement in exchange rates, policy changes and free trade agreements have long been indicated as an important force affecting the survival of firms (Geroski, et al. 2002). Adverse conditions such as increases in interest rates, a drop in income may affect costs, demand and other critical determinants of firm survival. New firms are more vulnerable than incumbents. Indeed studies suggest that entry rates decline during recessions reducing the pressures on incumbents (Caballero and Hammour 1994). Conversely, all firms, including entrants, may reap the benefits of a robust economy. Empirically, Keasy (1991) showed a positive relationship between the survival of new small firms and national economic growth. Methodology Empirical Model To test our hypotheses we estimate the hazard facing de novo entrants overtime in locations with various levels of industrial clustering. For those firms that have not exited at the end of our period of analysis, we do not have information on how long they are going to last. This is known as 'right censoring', as for those firms we know only that they survive longer than the age they had at the end of the sample. In addition, firms are added to our sample in different points in time. We observe whether or not a firm is active at the survey year, and therefore, our measured durations are grouped into annual intervals. Consequently, we need to employ a statistical model that is capable of accommodating data with incomplete discrete durations as well as entries in different times (i.e., cohorts of entrants). There are several empirical approaches that can be applied (e.g., surveys by Kiefer 1988, Lancaster 1990). We choose to use the Cox proportional hazards model since it does not assume particular distributional form for the probability of exit (such as required by probit and logit) or firm age (required by tobit). The basic (static) version of the Cox model can be extended to allow for time-varying covariates along with time-invariant ones. Specifically, the Cox proportional model estimates the hazard h(t) — the probability that a firm with covariates vector Xt_x exits in t, conditional on the fact that it survived to period t-1. (1) h(t) = Pfexit at t | survive to t-1) = P(T = t\\T >t ) for t=l,...K Consequently, the probability of reaching the fth interval (year) is provided by the survival function (2) S(t) = f[[l-h(j)] In order to account for the effects of the covariates we define the hazard rate (3) log h(t |• *,_,)= Xt + fix t_x where the parameters A, identify the baseline hazard function providing the annual exit probability for a firm whose current covariates are denoted by the vector Xt assume a zero value 4 9 . The fi vector represents the regression coefficients measuring the impact of the set of explanatory variables included in the vector Xt. We model the probability of an event (exit) between time t-1 and t as a function of state variables observed at time t-150. In this paper we are interested in exploring how the factors influencing firm survival vary with the degree of local industry clustering. Since previous empirical research suggested that location of new ventures is endogenously determined (Shaver and Flyer 2000), and that differences in resources and capabilities that existed in founding have persistent impacts on 4 9 Thus, /^ j gives the probability of exit within the first year of activity, X2 gives the probability of exit in the second year provided that the firm did not exit during the first year. Note, A varies with time but not across firms. 5 0 Note, with time varying covariates the model is no longer proportional. survival chances (Dunne, Roberts and Samuelson 1989, Audretsch 1995, Swaminathan 1996, Geroski et al. 2002, Klepper 2002) 5 1 , we include founding conditions X0 in our models. Our empirical strategy of choice is a semz'-proportional Cox model, where the baseline hazard rate is stratified by the level of local industry clustering, and interaction terms between the stratification variable and the covariates are included. Specifically the model is formulated as: (4) log h(t | Xt_l;X0;s];s2)= Xt + P0Xt_x + XoX0 + s ^ X + Z^0)+ s2{fi2Xt_x + %2X0) where the hazard function is calculated for a firm with lagged covariate vector and initial covariates X0 that belongs to a given level of local industry clustering (i.e., stratum; s[ =1 for firms located in moderate clustering and zero elsewhere, and s2 =1 for firms located in strong clustering and zero elsewhere). The model preserves proportionality of hazards functions within each stratum while letting the hazard ratio vary between strata as well as introducing non-proportionality between hazards functions of firms from different strata. Testing the interaction terms corresponds to testing structural differences between firms from different strata in a very general framework. The model is estimated by maximum likelihood methods. 5 1 Passive leaning models predict a significant effect on initial conditions even at higher ages. The active leaning models predict that the effects of initial conditions diminish faster with age. Data The data set used to estimate our model is T 2 - L E A P which is a merger of two different Canadian databases. The first database, the Longitudinal Employment Analysis File ( L E A P ) , is used to identify new entrepreneurial entry and exit, 3-digit SIC code 5 2 , number of employees, and their location coordinates. The second database is The Corporate Tax Statistical Universe File (T2SUF). This database is used to assess firm specific annual financial variables such as equity, assets, sales and closing inventories, converted to constant Canadian dollars using a 1985 price index. T 2 - L E A P is a unique, firm-level database that includes all incorporated employers in Canada. The database tracks the employment and payroll characteristics of individual firms from their year of entry to their year of exit. The employment record of each business is derived from administrative taxation records that each Canadian employer must file . The payroll data are associated with a Revenue Canada employer identification number. Accordingly, firms enter the L E A P database in the year they first hire employees, and record their last entry in the database in the last year they have employees. For each year, total payroll and employment are calculated. The latter is the average annual count of employees within the firm, or Average Labor Units ( A L U ) . This payroll and employment information is then organized longitudinally; each observation in the database corresponds to a particular firm whose employment, payroll and industry characteristics are recorded annually. The longitudinal nature of L E A P allows entry and exit times to be measured with precision. Births (entrants) in any given year are firms that have current payroll data, but did not have payroll data in the previous year. In our empirical estimation, we include only de novo 5 2 Firms may produce products belonging to different 3-digit SIC codes, however, our data include only the primary sector in which a firm operates. 5 3 Every employer in Canada is required to register a payroll deduction account (for the purpose of unemployment insurance), and issue a T4 slip to each employee that summarizes earnings received in a given fiscal year. The LEAP database includes every business that issues a T4 taxation slip. entry (also referred as new entry, greenfield, or independent) ; we do not include births of establishments that are owned by a firm that had establishments in previous years (also referred as dependent, subsidiary), or firms that were classified as belonging to another industry at time t-1 (Helfat and Lieberman 2002). De novo entry accounts for 85% to 94% of all newly created establishments, depending on the sector 5 5. Similarly, exits in any given year are identified by the absence of current payroll data, where such data had existed in the previous year. L E A P distinguishes 'real' birth or death from 'false' ones. Real births and deaths reflect the creation of new establishments and the failure of existing ones. False births and deaths can reflect organizational restructuring within a firm, a change in the name of the firm, change in location 5 6 spin-offs or merger and acquisitions. While almost every study undertaken analyzing entry and exit suffers from false qualification (comprehensive literature reviews by Caves, 1998; Sutton 1996; and Geroski, 1995), L E A P identifies false births and deaths using a method of 'labor tracking' . A firm is considered to have survived year t i f year t is not the year in which the firm is unincorporated, i f it has assets greater than zero in year t+1, and i f the firm has one or more employees in year t+1. For a firm to survive it must meet all of these criteria, otherwise it is We exclude from our sample spin-offs that started as subsidiaries or divisions of incumbents and later transformed to independent establishments (also known as parent spin-offs). We cannot, however, differentiate between new entry and firms which started by executives of an incumbent leaving to start their own firms (also known as entrepreneurial spin-offs). 5 5 Examining the experience of new entrants across food industries and construction-related product sectors in the US, Dunne, Kilmek and Roberts (2004) found that de novo entry characterizes the majority of entrants however they account for 42-88% of all newly created establishments, 5 6 We excluded firms for which location change was observed. However, our results do not change when excluding only those firms that moved to other locations 10 miles or more from their founding locations. 5 7 This means that if, for example, firm A merged with firm B in year t, then a new firm, C, is created and given a synthetic history aggregated from the histories of firms A and B. The individual histories of A and B disappear from the data base and firm C represents their joint operations both before and after year t. recorded as exiting during year t . A firm is added to our sample in year t i f it is incorporated in that year and hires its first employees at that year. T 2 L E A P contains business operating in all sectors of the economy. Our database covers the years 1984 to 1998. A l l of these characteristics make this data set an excellent source for studying the factors affecting the survival of new entrants in cluster and isolation locations. Aside from allowing us to identify exits and their location, our database permits us to compute a number of firm level covariates which we wi l l use to test the hypotheses formulated above. Cluster Definition We used two alternative techniques to identify clusters: relative distances and Location Quotients. The first cluster identification technique is based on the relative distance between firms in a sector in each observed year. This method follows the logic that in order to benefit from agglomeration externalities firms seek to locate nearby similar entities based on spatial proximity 5 9 . We created concentric rings with radius of 20 miles around each firm for a given year 6 0 . Radius of twenty miles was chosen (hereinafter critical chosen radius) based on findings of pervious studies showing that cluster's externalities attenuate rapidly beyond that distance 5 8 Firms that appear to cease operations for one or more years and then start up again in subsequent years are dropped from the data set. Adding employment greater than one and assets greater than zero to the existing measures of exit in T2LEAP, corrects for cases in which firms did not legally unincorporated in the same year their operations ceased, and for cases when production had ended but a low level of employment was retained while the firm was shutting down. 5 9 Segmenting the data into political jurisdictions produces arbitrary spatial boundaries that may bisect clusters, and thus may misrepresent industrial concentrations. The distance between any two firms A and B, is adjusted to the earth's curvature using spherical geometry and is computed as: d{A,B) = 3437x < arccos sin(latitudeA) x sin(latitudeB) + co^latitudeA)x cos{latitudeB)x cos longitudeA - longitude^ where latitude and longitude are measured in radians, the constant is the earth's radius and the linear distance units are in miles. 6 0 See robustness section for discussion about the sensitivity of our analysis to concentric rings of different radius. (Abrahamson, Baum and Feldman 2004a,b; Keller 2002; Rosenthal and Strange 2003) 6 1 . We calculated the number of firms operating in the 3-SIC sector for a given year in a ring (density) and ranked the level of local industry clustering according to three categories: isolation, moderate clustering, and strong clustering. Covariates were calculated with respect to the relevant population of firms operating in 20 miles ring and outer rings. The second technique employs a geographic classification which divides Canada to 289 census divisions. Following the tradition in regional economics we calculate Location Quotients LQs) using employees or number of establishments to identify and measure clusters . For a census division (CD) c, the share of employment (number of establishments) in a 3 SIC sector j is defined as Fcj. The sector's share of total national employment (number of establishments) is Fj. Wi th these quotients we compute the L Q defined as Fcj/ Fj. When a C D has a share of employment (establishments) in a specific sector that corresponds to the national average the L Q equal to one. Thus, i f L Q is larger than one it implies that the C D has a disproportional large 63 share of employment in the sector (the sector is a \"basic\" or export sector) . The table provides a division to different clustering categories. 6 1 Identifying the boundaries of a cluster is outside the scope of this paper. 6 2 There are other methods for identifying and measuring clusters. The Gini coefficient is defined as the degree to which a particular industry is unequally dispersed over localities. It measures the degree to which the regional distribution of employment in a particular industry deviates from the regional distribution of the same variable as a whole. The index may take on values between zero (i.e. totally even distribution) and 1.0 (total concentration in one industry). This method was used in Audretsch and Feldman 1996, Krugman 1991a among others. However, the method has the same weaknesses as LQ. The results only indicates the degree to which an industry deviates from a situation where its employment is distributed over regions precisely in the same way as the total employment in the country. Results are sensitive to the size of the particular industry and the number and size of the geographical areas within the country (Malmberg and Maskell 1997). Moreover the method does not distinguish random concentration arising from industrial structure from concentration arising from agglomerative externalities or natural advantage (Rosenthal and Strange 2001). 6 3 This definition of clusters is widely used in the literature, for example, Feser and Bergman (2000), Glassman and Voelzkow (2001), Porter (2001), Shaver and Flyer (2000). LQ LQ<1 Isolation \\ mean(LQ) Strong clustering If the L Q lies between 0 and 1 there is no extraordinary concentration of specific sector in the C D and thus it is classified as isolation. If the L Q exceeds 1 but is lower than the mean level of L Q for that sector, the share of activity in the sector in the C D is larger than the national distribution of that activity and this it is classified as a moderate clustering. Strong clustering is defined where L Q of a C D exceeds the mean level for than sector. Covariates Firm Specific Attributes Size. The I N I T I A L R E L A T I V E SIZE and current size of a firm are calculated as the size of the firm relative to the industry's average at the critical chosen radius around the firm for a given year. The industry average reflects the importance of scale economies and provides a measure of the Min imum Efficient Scale (MES) in the industry. A ratio of 1 or higher suggests that the firm operates at an efficient scale. Lower ratios indicate that the firm is operating at a suboptimal size. Growth. Growth of a firm in terms of employees (Z t ) in year t relative to year t-2 is measured by arc growth rate. (5) Z , = Z , ~ Z \" \\ n Since this measure is defined with respect to the average size of the firm rather than size in the initial year, it has the advantage of being bounded in the interval [-2,2], thus moderating the effect of mismeasurment of initial size and consequently a large error in growth rate (Brander et al. 1999). Quality of Human Resources. Higher wages tend to reflect a higher quality of human capital. The efficiency wage literature shows that firms tend to pay a wage rate above the market clearing wage in order to attract and retain high quality labor and to provide incentives for workers to exert more effort. A s an indirect measure of human capital resources available to a firm we employ current and initial R E L A T I V E Q U A L I T Y O F H U M A N C A P I T A L . It is defined as the average wages paid by a firm divided by the industry average wages within the critical chosen radius around the firm for a given year. A ratio of 1 or higher suggests that the quality of human capital that the firm employs is at least comparable to its competitors 6 4. Leverage. L E V E R A G E is the ratio of debt to assets. Debt is defined as assets (book value) less equity. Equity consists of common and preferred shares and accumulated retained earnings (or losses). Productivity. The current and initial effective deployment of resources is measured by the P R O D U C T r V I T Y of the firm. Our database does not contain sufficient data for classical measures of Total Factor Productivity (TFP); however, it is possible to calculate Approximate Total Factor Productivity (ATFP) . As originally suggested by Griliches (1990), and more recently by Hal l (1999), this measure of productivity is derived from a simple Cobb-Douglas production function. Suppose that firm i has a certain productivity level A, and produces output Yi using capital Ki and labour L, . The firm's production function is: 6 4 Since it is possible that new firms must pay higher wages than incumbents to compensate for risks, an alternative measure of relative average wage during the initial stages may define the comparison group as all entrants to the sector in the same cohort within the R radius circle around the firm. It should be noted that compensation may include equity and stock options. However, because of competitive pressures, compensation practices seem to converge within a location and a sector. Thus, our measure which is based on the relative position of firms within a sector in the location is likely to reflect this heterogeneity. Differences in firm specific practices are idiosyncratic and are included in the error term. (6) Yi=AiK\"L>~a. I -aln U J U J If we solve for productivity, A,-, and take the natural log of both sides, the equation can be rewritten as: (7) ln(A,.) = ln Equation (7) describes the efficiency of the firm at turning inputs into outputs. This is comprised of the firm's labor productivity and the amount of capital each worker has at their disposal. Labor productivity is measured as total sales divided by the number of employees ( A L U ) . N o measure of capital per worker is present in T 2 L E A P , however, a measure of total assets is available. We use total assets minus closing inventory and divide the result by the number of employees. Removing closing inventories leaves us with a good measure of the efficiency with which workers turn inputs into outputs, using their available resources 6 5. The optimal capital share a, varies significantly from industry to industry in the manufacturing sector. The Annual Survey of Manufacturing is used to derive this share. The natural log of A T F P for a given firm is defined as: (8) l n ( A r F ^ ) = ln r sales. ( - a l n v. a^ut V assetsi - inventories. alu. where alui is the average labor units (i.e., total employees) of the firm, salesi is its total sales, assetsi is its total assets, and inventoriest is the closing inventories of the firm. 6 5 As a measure of capital we use total assets minus closing inventory. The value of assets is drawn from the firm's balance sheet and consists of capital assets like land, building and machinery; short-term financial assets like cash; long-term financial assets like bonds; and intangible assets like goodwill and reputation. Industry Level Local Market Competitiveness (Structure). A s an indicator of L O C A L M A R K E T C O M P E T I T I V E N E S S we use the number of establishments per worker in the focal sector within the critical chosen radius around the firm (this measure was used in Glaeser et al. 1992 and Rosenthal & Strange, 2003). A s this ratio increases the environment in a given industry and in a given location is thought to become more competitive (i.e., the market is populated by large number of firms). A s an alternative measure of size of firms in relationship to the local industry and as an indicator to the amount of local competition we use the inverse Herfindahl index. Increases in this index generally indicate a loss of pricing power and an increase in competition. Local Entry Rate. We compute the two year average L O C A L E N T R Y R A T E to the focal sector within the critical chosen radius. Industry Growth. We use two measures for I N D U S T R Y G R O W T H : average percentage change in employment over 2 years in the 3 digit SIC sector, or the 2 year average of sectoral percentage sales growth. Macroeconomic Conditions. We use time dummies to control for macroeconomic conditions. Industry-Specific Characteristics. We employ industry fixed effects in all of our estimations. Industry specific characteristics that may affect entrepreneurial enterprises' profitability are similar to those affecting incumbents' profits. Sample correlations between the independent variables are reported in Appendix 2.A. Empirical Results Table 2.1 and Figure 2.1 present survival rates for different levels of local industry clustering. The data suggest that clustering is a discriminating factor for survival. For each year the survival rate is always lower for isolated firms than for moderate or strong clustered firms. Over time, these differences increase 6 6. For example, after 5 years 20 percent of all new entrants in the manufacturing sectors exit from an isolated location. In contrast, 20 percent of all entrants into strong clustering fail after 10 years. In order to investigate whether the differences between the survival rates are statistically significant, we test for equality using Wilcoxon and Log-Rank tests. Table 2.1 shows that according to both tests, firms operating in different levels of clustering confront statistically different hazard rates. Note that these differences may result from entry selection processes as well as survival enhancement offered in clusters. We disentangle these effects in our econometric models by including initial resource endowments. The results confirm the existence of survival benefits in clusters. The reason why in equilibrium firms may still enter isolation is relation to entry barriers that resource poor companies may face or the idiosyncratic preferences of entrepreneurs (i.e., utility rather than profit maximization). TABLE 2.1 - Life-Table Estimates of Survival Rates year Isolation Moderate clustering Strong clustering 1 95.0 96.2 97.2 2 89.5 91.3 93.0 3 85.0 87.2 90.2 4 82.0 84.6 87.5 5 78.5 82.0 85.2 6 75.5 79.0 83.8 7 73.0 77.0 82.6 8 71.0 75.0 81.9 9 69.4 74.0 80.7 10 68.4 73.0 79.3 11 66.2 72.0 78.0 12 64.7 70.7 77.0 13 63.5 69.8 75.5 14 62.2 68.4 75.2 15 61.0 67.8 74.8 Test of equality of survival fuctions across samples Isolation versus Moderate clustering Wilcoxon (DF=1) Chi-Square=67.17 Pr>Chi-Square 0.0002 Log-Rank (DF=1) Chi-Square=40.07 Pr>Chi-Square 0.0004 Isolation versus Strong clustering Wilcoxon (DF=1) Chi-Square=77.52 Pr>Chi-Square 0.0000 Log-Rank (DF=1) Chi-Square=49.32 Pr>Chi-Square 0.0000 Moderate clustering versus Strong clustering Wilcoxon (DF=1) Chi-Square=32.04 Pr>Chi-Square 0.0005 Log-Rank (DF=1) Chi-Square=21.76 Pr>Chi-Square 0.0007 FIGURE 2.1: Survival of New Entrants Operating in Locations with Different Levels of Industry Clustering 0.65 0.6 —! ! ! ! ! ! ! ^ ^ ( ^ { { (— 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 analysis time isolation strong clustering moderate clustering Sample correlations between the independent variables are reported in Appendix B . Our empirical strategy includes two types of models. First, we estimate a simple model of survival where local industry clustering is included as a dummy variable. Second, we model the impact of clustering intensity by including a set of interactive terms between the covariates and the level of industry clustering. We use the second model to test our hypotheses. Exit of New Firms The results of our first model are presented in Table 2.2. These specifications assume that the probability of failure of isolated and clustered firms differ only by proportional factor. The reported standard errors in all of our specifications are robust- having been adjusted for clustering by firm. Since our primary interest lies in the differences between the determinants of failure among firms operating in various levels of industry clustering we use the specification presented in this table as our base model and w i l l provide only succinct discussion of the results. Column (1) represents the unconditional impact of local industry clustering on the probability of exit. The coefficient estimate of C L U S T E R (dummy variable which assumes the value of 1 i f a firm operates from locations with moderate or strong industry clustering and 0 otherwise) is negative and highly significant; indicating that being located in a cluster reduces the hazard rate facing new entrants. Columns (2) - (4) add firm-specific current attributes. The coefficients of the firm level variables are consistent with previous findings. Specifically, firm's relative size, relative quality of human capital, leverage and productivity - are all significant determinants of survival. Larger firms, with higher quality of human capital, lower leverage and higher productivity, experience lower hazards. Consistent with Zingales (1998), leverage increases the hazard facing new entrants even when controlling for efficiency or productivity. Specifications (3) and (4) add past growth and quadratic growth term to account for a possible nonlinear effect. Due to uncertainty, financial constraints and adjustment costs, most firms operate in less than optimal size during their initial stages (Penrose 1959). Therefore, past growth has an important positive impact on survival as it reflects the ability of new small firms to overcome initial constraints of size. Often, size and growth are considered approximations for efficiency as more efficient firms are likely to survive and grow. Here, size and growth contribute positively to survival even when controlling for productivity. The survival advantage of growing firms has a diminishing effect as observed by the positive coefficient on the quadratic growth term. This may suggest the importance of the pace of growth and the well documented negative relationship between size and growth (Geroski 1995, Cabral 1995). Column (5) adds firm's initial characteristics profile. The significant coefficients imply that apart from current conditions, initial endowment of resources and capabilities has an impact on survival. Specifically, entrants with higher initial quality of human capital, higher initial productivity, and lower initial leverage have survival advantage. A likelihood ratio test for the unrestricted model (column 5) versus the restricted model (column 4) supports the unrestricted model as a more appropriate specification. Note that after including the growth, initial relative size is not a significant explanatory variable and therefore wi l l be excluded from later parsimonious specifications. Column (6) adds industry characteristics. The coefficients suggest that increase in L O C A L M A R K E T C O M P E T I T I V E N E S S and in L O C A L E N T R Y R A T E are negatively associated with survival. A s local markets become populated by a larger number of firms, market competitiveness increases and the hazard rate increases. High entry rates are associated with lower entry barriers and high turnover which may increase local competition and decrease the threshold for voluntary exit. Growing industry is an environment which increases survival prospects. The primary effect of interest, the cluster dummy, has not changed and remained statistically significant and negative while controlling for firm-specific and industry characteristics. TABLE 2.2 - Cox Hazard Model - unconditional impact of clustering Cluster (dummy) relative size (Iag1) initial relative size growth (growth) A2 relative quality of HC (Iag1) initial relative quality of HC leverage (Iag1) initial leverage productivity (Iag1) initial productivity local market competitiveness industry growth (2YA) local entry rate (2YA) Coef. robust std. err. Coef. robust std. err. Coef. robust std. err. Coef. robust std. err. Coef. robust std. err. Coef. robust std. err. Coef. robust std. err. Coef. robust std. err. Coef. robust std. err. Coef. robust std. err. Coef. robust std. err. Coef. robust std. err. Coef. robust std. err. Coef. robust std. err. 1 -0.124— 0.012 2 -0.094\"* 0.013 •0.689*'* 0.051 •0.189*\" 0.084 0.657\"' 0.203 -0.045*\" 0.011 0.012 •0.478'\" 0.041 -1.301'\" 0.192 -0.178— 0.066 0.585'\" 0.021 -0.028\" 0.013 0.012 -0.474\"' 0.041 -1.137— 0.038 0.129\"' 0.023 -0.182\"' 0.066 0.586*\" 0.021 -0.027\" 0.013 5 -0.086\"* 0.013 -0.471*\" 0.044 -0.019 0.016 -1.159— 0.039 0.124\"* 0.024 -0.433*\" 0.080 -0.485*\" 0.080 0.279*\" 0.029 0.469*\" 0.029 -0.060'\" 0.018 -0.116\"' 0.014 48,406 48,406 48,406 256,632 256,632 256,632 11,217 11,217 11,217 260,121 260,121 260,121 9,680 12,201 11,519 -107,971 -107,887 -106,584 Comments: standard errors adjusted for clustering on the firm all specifications include time, economic region and 2 digit SIC dummies •:p<0.10, \"p<0.05, —p<0.01 No. of subjects 48,406 48,406 No. of observations 256,632 256,632 No.offailures 11,217 11,217 Time at risk 260,121 260,121 Wals chi2 (9) 99 2,093 LogLikelihood -115,222 110,714 6 -0.086*\" 0.012 -0.471 — 0.042 -0.015 0.015 -1.160*'* 0.039 0.124*\" 0.024 -0.427\"' 0.079 -0.489'\" 0.079 0.280\"* 0.029 0.469\"* 0.029 -0.061\"* 0.018 -0.116*\" 0.014 0.337\" 0.163 -0.785\" 0.293 0.328*\" 0.067 48,406 256,632 11,217 260,121 11,456 -106,406 Testing for Structural Differences between Levels of Local Industry Clustering The models tested in Table 2.3 add interaction terms between levels of local clustering and the other covariates to the specifications presented in Table 2.2. This formulation allows us to test for structural differences between the impacts of the covariates. Firm-specific attributes (columns 1, 2) and local industry characteristics (column 3) are added sequentially to the model. The likelihood ratio test suggests that model 2 in Table 2.3 is more appropriate than model 5 in Table 2.2. We start by presenting the empirical results of each step followed by a discussion. The coefficients of the first-order covariates provide estimates of the factors influencing the hazard rate of firms in isolation. The specifications in column (1) include current firm specific characteristics; column (2) adds initial characteristics. The story told by the first-order covariates is straightforward. Firms in isolation that have larger initial size, higher growth, better initial and current quality of human capital, lower leverage, and higher initial and current productivity have survival advantages. Our main interest lies in the differences (if any) between survival patterns of firms in various levels of local industry clustering. The hypothesis that each of the first-order coefficients is identical to its counterparts (the second-order interaction terms) is tested by means of a Mest on the corresponding interaction terms. The second-order interaction term of I N I T I A L R E L A T I V E SIZE with moderate clustering level (column 2) is not statistically significant suggesting that the positive impact of initial size on survival is identical in isolation and in moderate clustering locations. The significant and negative coefficient of the interaction term between I N I T I A L R E L A T I V E SIZE and S T R O N G C L U S T E R I N G suggests that an increase in initial relative size has a stronger positive effect on the likelihood of survival for firms in strong clustering locations than on firms operating in isolation or moderate clustering. The first and second order coefficients of G R O W T H (column 1) imply that the magnitude of growth has a higher positive impact on the probability of survival of firms in locations with lower levels of clustering. However, after controlling for initial relative size (column 2) this difference vanishes. This is consistent with findings that founding strategies (size) have 'almost permanent' impact on survival (Geroski et al. 2002: 18). In order to examine the possibility of nonlinear effect of growth on the probability of failure we include a quadratic term ( G R O W T H A 2 ) . The positive coefficient on the quadratic growth terms suggests that the survival advantage of growing firms has a diminishing effect. This effect is more pronounced for firms in locations with lower levels of clustering. The next firm-level characteristic included in our estimation is the relative quality of human capital. Here also we investigate the effects of current and initial attributes on firm survival. A s expected, the coefficients confirm that both effects decrease the probability of failure. The impact of I N I T I A L R E L A T I V E Q U A L I T Y O F H U M A N C A P I T A L seems to be important and permanent. The analysis reveals noticeable structural differences in the impacts of the quality of human capital on various levels of local clustering. A n increase in the initial quality of human capital has a stronger positive effect on the likelihood of survival for firms in higher levels of clustering. On the other hand, current quality of human capital ( R E L A T I V E Q U A L I T Y O F H C , lagl) seems to have a stronger positive effect on survival prospects of firms in lower levels of clustering. Both initial and current levels of leverage ( INITIAL L E V E R A G E , L E V E R A G E lagl) have negative effects on the survival of entrants. There are no significant differences with respect to the effects of leverage among the various levels of clustering, according to the coefficients of the interaction terms. TABLE 2.3 - The Determinants of Firm Exit: Semi-Proportional Cox Hazard Model with Interaction Terms between Levels of Local Industry Clustering and Other Covariates 1 2 3 initial relative size Coef. -0.206*\" -0.198*** robust std. err. 0.046 0.040 initial relative size x moderate clustering Coef. -0.011 -0.082 robust std. err. 0.078 0.075 initial relative size x strong clustering Coef. -0.172** -0.160\" robust std. err. 0.070 0.066 growth Coef. -1.264*\" -1.271*** -1.236*** robust std. err. 0.043 0.044 0.044 growth x moderate clustering Coef. 0.236** 0.019 -0.246 robust std. err. 0.123 0.034 0.034 grwoth x strong clustering Coef. 0.485\" -0.046 -0.047 robust std. err. 0.235 0.035 0.035 (growth)A2 Coef. 0.160\"* 0.175\"* 0.176*** robust std. err. 0.027 0.028 0.028 (growth)A2 x moderate clustering Coef. -0.044** -0.042* -0.044* robust std. err. 0.019 0.020 0.024 (growth)A2 x strong clustering Coef. 0.016 -0.070** -0.063\" robust std. err. 0.019 0.030 0.030 initial relative quality of HC Coef. -0.506\"* -0.492*** robust std. err. 0.150 0.152 initial relative quality of HC x moderate clustering Coef. -0.070\" -0.068\" robust std. err. 0.030 0.029 initial relative quality of HC x strong clustering Coef. -0.115\" -0.106\" robust std. err. 0.050 0.045 relative quality of HC (lagl) Coef. -0.582*** -0.578*** -0.565*** robust std. err. 0.090 0.149 0.150 relative quality of HC (lagl) x moderate clustering Coef. 0.136* 0.117* 0.089* robust std. err. 0.062 0.062 0.046 relative quality of HC (lagl) x strong clustering Coef. 0.193\" 0.156\"* 0.133\" robust std. err. 0.080 0.064 0.054 initial leverage Coef. 0.481*** 0.474** robust std. err. 0.052 0.050 initial leverage x moderate clustering Coef. -0.035 -0.022 robust std. err. 0.070 0.069 initial leverage x strong clustering Coef. 0.015 0.029 robust std. err. 0.073 0.071 leverage (lagl) Coef. 0.638*\" 0.287\"* 0.287*** robust std. err. 0.039 0.054 0.054 leverage (lagl) x moderate clustering Coef. -0.042 -0.007 -0.004 robust std. err. 0.052 0.072 0.072 leverage (lagl) x strong clustering Coef. 0.016 0.001 0.040 robust std. err. 0.052 0.074 0.073 initial productivity Coef. -0.134\"* -0.136*** robust std. err. 0.025 0.025 initial productivity x moderate clustering Coef. 0.058\" 0.052* robust std. err. 0.028 0.028 initial productivity x strong clustering Coef. 0.086*** 0.084\"* robust std. err. 0.036 0.036 productivity (Iag1) Coef. -0.163*\" -0.052\" -0.051\" robust std. err. 0.047 0.026 0.028 productivity (Iag1) x moderate clustering Coef. 0.035** 0.008 0.010 robust std. err. 0.017 0.039 0.040 productivity (Iag1) x strong clustering Coef. 0.081\"* 0.015* 0.014* robust std. err. 0.030 0.009 0.007 local market competitiveness Coef. 0.000 robust std. err. 0.001 local market competitiveness x moderate clustering Coef. 0.226*** robust std. err. 0.049 local market competitiveness x strong clustering Coef. 0.495*** robust std. err. 0.282 industry growth (2YA) Coef. -0.702* robust std. err. 0.328 industry growth (2YA) x moderate clustering Coef. 0.002 robust std. err. 0.110 industry growth (2YA) x strong clustering Coef. -0.057 robust std. err. 0.113 local entry rate (2YA) Coef. 0.263*** robust std. err. 0.085 local entry rate (2YA) x moderate clustering Coef. 0.152*** robust std. err. 0.008 local entry rate (2YA) x strong clustering Coef. 0.181\"* robust std. err. 0.007 No. of subjects 48,406 48,406 48,406 No. of observations 256,632 256,632 256,632 No. of failures 11,217 11,217 11,217 Time at risk 260,121 260,121 260,121 Wals chi2 12,184 11,532 12,090 LogLikelihood -108,383 -106,379 -106,287 Comments: standard errors adjusted for clustering on the firm all specifications include time, economic region and 2 digit SIC dummies ':p<0.10, \"p<0.05, \"'p<0.01 The last firm-level characteristics in our model are current and initial productivities ( INITIAL P R O D U C T I V I T Y , P R O D U C T I V I T Y lagl) . Here, again our estimates reveal structural differences with respect to the positive effect of productivity on survival prospects of entrants in different levels of local industry clustering. A n increase in productivity has a stronger positive effect on the likelihood of survival for entrants in isolation than for entrants in moderate or strong clustering. Moreover, initial productivity has much lower an impact on survival in strong clustering than it has in isolation. Our results indicate that the main differences in the determinants of survival between firms in locations with higher levels of industry clustering and firms in locations with lower clustering levels are in initial size, the effect of growth, initial and current quality of human capital, and initial and current productivity. Specifically, survival prospects of firms operating in clusters are more positively affected by initial relative size, and initial quality of human capital than for firms in less clustered locations. Larger entrants most likely enjoy economies to scale in identifying, accessing, and exploiting localization externalities. The ability of new firms to screen, assess, absorb, and internalize knowledge spillovers is largely determined by the initial relative quality of their employees. Access to valuable and up-to-date information about trends in an industry (Cohen and Levinthal 1994), activities, offerings and suppliers of competitors should be realized in high level of performance. Moreover, entrants who are capable to assimilate this knowledge and align their products or services early in their lives with the demand of the market, are likely to overcome issues of 'limited legitimacy' (Aldrich and F io l 1994, Stinchombe 1965) and 'liability of newness'. Thus, larger new ventures and those with higher initial quality of human resources operating from locations with higher levels of industry clustering have survival advantages. On the other hand, the likelihood of survival of entrants to clusters is affected positively, though to a lesser extent, by current quality of human capital, initial and current productivity than the survival of entrants into moderate clusters or relative isolation. Critical mass of industry activity in clusters implies the existence of \"firms in downstream industries; producers of complementary products; specialized infrastructure providers; government and other institutions providing specialized training, education, information, research and technical support\" (Porter 1998: 199). These externalities suggest that the likelihood of survival of firms in clusters is not solely determined by firm's internal resources and capabilities (such as current quality of employees or productivity) but also by their ability to utilize external resources and capabilities which reside in the cluster. Since the survival prospects of firms in isolation are determined to a large extent by the firm's internal resources and capabilities an increase in productivity and quality of employees have higher positive effects. We find a positive effect of past G R O W T H on the probability of survival across the three cluster levels. Past growth has the largest effect on survival of new entrants in isolation. Recent growth is a sign of good past performance and may also suggest optimism about expected performance and thus, should be positively related to survival (Jovanovic 1982, Mata and Portugal 2002). Nevertheless, the results suggest that the positive effect of growth has a diminishing impact on survival with a stronger declining impact on hazard rates of firms operating from locations with lower levels of industry clustering. In other words, excessive growth has higher negative consequences for isolated firms than for firms in clusters. The explanation for this result may stem from the ability of new ventures operating from locations with high levels of industry clustering to better optimize or gradually adjust their internal growth process while minimizing the adjustment costs and damages associated with excessive growth (Penrose 1959, Bogner et al. 1996). Growth for firms in isolation may involve investment in infrastructure, production facilities, logistics, specialized training programs, and acquiring new technologies. On the other hand, firms in clusters may limit their internal growth to some activities or aspects and outsource other activities to specialized firms operating in the region and thus achieve higher efficiency. Moreover, firms in clusters may choose to imitate a capability that exists in other local organizations rather than developing it from scratch (Helfat and Peteraf 2003) an alternative that may not be available for firms operating in locations with lower level of industry concentration. Column 3 adds industry characteristics. The likelihood ratio test statistic for the unrestricted model (column 3) versus the restricted model (column 2) is well above the critical value suggesting that the unrestricted model should be considered having more appropriate specification of the two. Local market competitiveness as measured by establishments per workers has a negative impact on survival. While the effect of market competitiveness is insignificant for firms operating in locations with low level of industry concentration, the negative impact of market competitiveness on survival is increasing with the level of clustering. The existence of many small firms rather than a few large ones may result in pressures on prices and entrepreneurial profit margins. Industries which are growing offer an environment that supports survival. I N D U S T R Y G R O W T H has similar positive effect on the survival of firms operating in various levels of clustering. Finally, higher L O C A L E N T R Y R A T E increases the hazard rate. This effect is stronger with the increase in the level of clustering. The impact of entry on competition for local resources is higher in denser markets as further increases in density challenge their carrying capacity. Higher entry signals lower entry and exit costs allowing weaker firms to enter and experiment, and reduce the threshold for voluntary exit. In isolation increases in density may contribute to the creation of resources (e.g., shared infrastructure, and attracting employees from outside the region with the increase in job opportunities). The Honeymoon Period Figure 2.2 compares the variation in the length of the 'honeymoon' period (i.e., the period over which initial endowments of resources of new entrants provide a safety net form external shocks) among entrants operating from locations with different levels of industry clustering controlling for initial firm characteristics 6 7. While the honeymoon lasts about two years for new entrants in isolation, the period is stretched up to about four years for entrants in high clustering 6 8. There are several factors that may explain the observed differences. Clusters provide a sheltered environment where new entrants can outsource more easily by forging local strategic alliances (Almeida, Dokko and Resenkopf 2003). Furthermore, cluster externalities may help conserve scarce resources offering a better infrastructure and more efficient end results. Clusters may facilitate accelerated learning processes through knowledge spillovers and imitation of incumbents' best practices. We used STATA 8 for illustrating hazard functions with a kernel smooth of the hazard for the Cox model (equation 4) (http://www.stata.com/stb/stb60/dm90). FIGURE 2.2: Smoothed Hazard Function of New Entrants Operating in Locations with Different Levels of Industry Clustering Robustness In this subsection we investigate the robustness of the estimates to different assumptions. For our cluster classification we conducted three types of robustness checks. First, we checked the sensitivity of our results to the choice of the radius of the concentric rings. Following Rosenthal and Strange (2003) we experimented with rings of 10, 30, and 40 miles. Second, we grouped firms into clusters by minimizing within-group average distance. In each year, we compared each firm's mean within-cluster distance to the overall cluster mean, and considered as cluster members only firms whose average distance was two or less standard deviations below the cluster average. We excluded from the cluster all firms whose average distance was above the cutoff (Aharonson, Baum and Feldman 2004). Third, we classified a location as a cluster i f its density (or L Q ) is larger than the median of the sample. Our estimates are not sensitive to these various cluster classifications. A benefit of using longitudinal, firm-level data is that we can control for permanent unobserved location and industry characteristics that might bias a simpler cross - sectional specification. In all of our models we control for many permanent and time-varying factors using location, industry, and time fixed effects. Economic region (ER) specific effects ^ E R are permanent factors common to all firms operating in the economic region. Industry fixed effects i^') at the 2-digit SIC-E absorb industry permanent heterogeneity. Time fixed effect (^ ' ) for each of the years picks up the time-varying trend in the error which may come from business cycles (aggregate instability in the macroeconomic environment), as well as general trends in technology. In addition to the fixed effects we also experimented with the following interactions: E R x year effects to absorb any E R specific time varying shocks that are shared by all firms operating in the E R such as policies supporting new ventures, the construction of rail link or the opening of an airport; sector x year effects to capture sector specific time-varying shocks, such as the introduction of an industry-specific new technology, or industry- life cycle (Klepper 1996); and E R x year x industry effects. The coefficients retain the same level of statistical significance and approximate magnitude. Lastly, rather than defining the dependent variable as length of survival, we performed Probit estimations for all our models where survival is defined as a binary variable. Our results do not materially change across specifications. Conclusion Organizational ecologists, economists, and strategy scholars have long speculated that location externalities may have significant impact on survival of new heterogeneous firms. Yet, despite significant managerial and policy implications, systematic empirical evidence was lacking. In this paper, we take a more direct and comprehensive approach. We have asked whether the survival patterns of new entrants in cluster and non-cluster locations differ?; and what firm specific and environmental characteristics may explain the observed survival differences? Our modeling approach uses longitudinal firm-level data at a fine-grained, empirically determined classification of the level of industrial agglomeration. We control for endogeneity in the location choice, and the impact of initial endowments of resources and capabilities. Our results show that survival rates of de novo entrants operating in locations with high levels of industry clustering are higher than those of entrants in locations with lower levels of same industry concentration. Extending prior research, our findings reveal differences in firm specific survival enhancing strategies among firms located at different levels of clustering. Larger initial size, higher growth, better initial and current quality of human capital, lower leverage, and higher initial and current productivity have survival advantages in all locations irrespective of clustering levels. However, survival prospects of firms operating in clusters are more positively affected by initial relative size, and initial quality of human capital than those of firms in lower levels of clustering. Larger entrants most likely enjoy economies to scale in identifying, accessing, and exploiting localization externalities. The ability of new firms to screen, assess, absorb, and internalize knowledge spillovers is largely determined by the initial relative quality of their employees. In isolation, current resource endowments, productivity, and recent growth have higher impacts on the survival of firms. Survival of firms in isolation depends to a larger extent on the firm's internal resources and capabilities while firms operating in clusters also benefit from their ability to utilize externalities which reside in the cluster. B y outsourcing activities to specialized firms operating in the cluster, forging local strategic alliances, and limiting their internal growth to specific activities which are more likely to generate competitive advantage, firms in clusters optimize and gradually adjust their internal growth thus minimizing damages and adjustment costs associated with excessively rapid growth. Controlling for initial endowments of resources we found that clusters extend the adolescence period of new entrants. This finding suggests that clusters facilitate and accelerate learning processes leading to development of capabilities in particular productivity improvements. References Aharonson, B.S . , J . A . C , Baum and M . P . Feldman 2004a. Borrowing from neighbors: The location choice of entrepreneurs. Working Paper - Rotman School of Management. Aharonson, B.S . , J . A . C , Baum and M . P . Feldman 2004b. Industrial clustering and the returns to inventive activity: Canadian biotechnology firms, 1991-2000. Working Paper - Rotman School of Management. Agarwal, R. 1997. Survival of Firms over the Product Life Cycle. Southern Economic Journal, 63(3): 571-84. Agarwal, R. and M . Gort 1996. The evolution of markets and entry, exit and survival of firms. Review of Economics and Statistics, 78(3): 489-98. Almazan, A . , A . de Motta and S. Titman 2003. A theory of location choice and the utilization of human capital. Working Paper. Almeida, P., G . Dokko, and L . Rosenkopf 2003. Startup size and the mechanisms of external learning: increasing opportunity and decreasing ability? Research Policy 32: 301-315. Amit , R. and P. J. H . S. Schoemaker 1993. Strategic assets and organizational rent. Strategic Management Journal, 14:33-46. Amit , R. and S. Thornhill 1999. Why do young firms fail? Managerial capabilities, organizational assets, and the liability of newness. University of British Columbia working paper. Audretsch, D . B . 1991. New-firm survival and the technological regime. Review of Economics and Statistics, 73(3): 441-450. Audretsch, D . B . 1995a. Innovation, Growth and Survival. International Journal of Industrial Organization, 13(4): 441-57. Audretsch, D . B . 1995b. The Propensity to Exi t and Innovation. Review of Industrial Organization, 10(5): 589-605. Audretsch, D . B . , 1995c, Innovation and Industry Evolution, Cambridge ( M A ) , M I T Press. Audretsch, D . B . and M . P. Feldman 1996. R & D spillovers and the geography of innovation and production. American Economic Review, 630-640. Audretsch, D . B . and T. Mahmood 1993. Entry, growth, and survival: The new learning on firm selection and industry evolution. Empirica, 20(1): 25-33. Audretsch, D . B . and T. Mahmood 1994a. F i rm selection and industry evolution: The post-entry performance of new firms. Journal of Evolutionary Economics, 4(3): 243-60. Audretsch, D . B . and T. Mahmood 1994b. The rate of hazard confronting new firms and plants in U.S . manufacturing. Review of Industrial Organization, 9(1): 41-56. Audretsch, D . B . and T. Mahmood 1995. New firm survival: New results using a hazard function. Review of Economics and Statistics, 77(1): 97-103. Baldwin, J.R. 1995. The Dynamics of Industrial Competition: A North American Perspective. Cambridge University Press: Baldwin, J.R. and M . Rafiquzzaman 1995. Selection versus evolutionary adaption: Learning and post-entry performance. International Journal of Industrial Organization, 13: 501-522. Baptista, R. and P. Swann 1998. Do Firms in Clusters Innovate More? Research Policy, 27(5): 525-540. Baum, J . A . C . and S.J. Mezias 1992. Localized competition and organizational failure in the Manhattan hotel industry, 1898-1990. Administrative Science Quarterly, 36:187-218. Baum, J . A . C . and C . Oliver 1991. Institutional linkages and organizational mortality. Administrative Science Quarterly, 36:187-218. Bernard, A . B . and J.B. Jensen 2003. F i rm Structure, multinationals, and manufacturing plant deaths. N B E R working paper #9026. Bogner, W . , H . Thomas and J. McGee, 1996. A longitudinal study of the competitive positions and entry paths of European firms in the U.S . pharmaceutical market', Strategic Management Journal 17, 85-107. Braconier, H . and K . Ekholm 2000. Swedish multinationals and competition from high- and low-wage locations. Review of International Economics 8(3): 448-461. Brander, J. , K . Hendricks, R. Amit and D . Whistler 1999. F i rm Size Dynamics and the Engine of Growth Hypothesis: The role of entry, exit and sectoral effects. UBC Working Paper, Caballero, R.J . and M . L . Hammour 1994. The cleansing effect of recessions. American Economic Review, 84(5): 1350-1368. Cable, J. and J. Schwalbach 1991. International comparison of entry and exit, In P. A . Geroski and J. Schwalbach (Ed.), Entry and Market Contestability: An International Comparison, Blackwell Press: Cambridge, 257-281. Cabral, L . M . B . 1993. Experience advantages and entry dynamics. Journal of Economic Theory 59(2): 403-416. Carroll, G . and M . Hannan 1989. Density delay in the evolution of organizational populations. Administrative Science Quarterly, 34:411-430. Carroll, G.R. 1983. A stochastic model of organizational mortality: Review and reanalysis. Social Science Research, 12(4): 303-329. Carroll, G.R. and M . T . Hannan 2000. The Demography of Corporations and Industries. Princeton University Press: Princeton. Caves, R . E . 1998. Industrial organization and new findings on the turnover and mobility of firms. Journal of Economic Literature, XXXVI(December) : 1947-1982. Caves, R. E . and M . E . Porter 1977. From entry barriers to mobility barriers: Conjectural decisions and contrived deterrence to new competition. Quarterly Journal of Economics, 91(2): 241-61. Christensen, C , F. Suarez and J. Utterback 1998. Strategies for survival in fast-changing industries. Management Science, 42(12): 207-220. Dixi t , A . K . and R.S. Pindyck 1994. Investment Under Uncertainty. Princeton University Press: Princeton. Doeringer, P .B . and D . G . Terkla 1999. Business Strategy and Cross-Industry Clusters. Sage Publications Inc.: Thousand Oaks, C A . Dunne, T., S.D. Kl imek and M . J Roberts 2004. Exit from regional manufacturing markets: The role of entrant experience. Working Paper. Dunne, T., Hughes, A . 1994. Ages, size, growth and survival: U K companies in the 1980s. The Journal of Industrial Economics, 2:115-140. Dunne, T., M . Roberts and L . Samuelson 1989. The growth and failure of U . S . manufacturing plants. Quarterly Journal of Economics, 104(4): 671-98. Dunne, T., M . J . Roberts and L . Samuelson 1988. Patterns of firm entry and exit in U .S . manufacturing industries. Rand Journal of Economics, 19(4): 495-515. Eide, G .E . , E . Omenaas and A . Gulsvik 1996. The semi-proportional hazards model revisited: Practical reparametrizations. Statistics in Medicine, 15:1771-1777. Ericson, R. and A . Pakes 1995. Markov-perfect industry dynamics: A framework for empirical work. Review of Economic Studies, 62(1): 53-82. Evans, D.S. 1987. The relationship between firm growth, size, and age: Estimates for 100 manufacturing industries. Journal of Industrial Economics, 35(4): 567-581. Feser, E . J . and E . M . Bergman 2000. National industry cluster templates: A framework for applied regional cluster analysis. Regional Studies. Fichman, M . and D . A . Levinthal 1991. Honeymoon and the liability of adolescence: A new perspective on duration dependence in social and organizational relationships. Academy of Management Review, 16:442-468. Fotopoulos, G . and H . Louri 2000. Location and survival of new entry. Small Business Economics, 14(4): 311-21. Frank, M . 1988. A n intertemporal model of industrial exit. Quarterly Journal of Economics, 103(2): 333-344. Freeman, J. , G.R. Carrol and M . T . Hannan 1983. The liability of newness: Age dependence in organization death rates. Amercan Sociological Review, 48:692-710. Geroski, P. A . 1991. Market Dynamics and Entry. Geroski, P . A . 1995. What do we know about entry? International Journal of Industrial Organization, 13): 421-440. Geroski, P .A . , J. Mata and P. Portugal 2002. Founding conditions and the survival of new firms. Working Paper, Gertler, M . , Gilchrist, S. 1994. Monetary Policy, Business Cycles, and The Behavior of Small Manufacturing Firms. Quarterly Journal of Economics, 109:309-340. Glaeser, E . L . 1999. Learning in cities. Journal of Urban Economics, 46(2): 254-77. Glassman, U . and H . Voelzkow 2001. The governance of local economic in Germany. Oxford University Press: Oxford. Gort, M . and S. Klepper 1982. Time paths in the diffusion of product innovations. Economic Journal, 92(367): 630-653. Hambrick, D . C . , MacMi l l l an , I.C. 1984. Asset Parsimony-Managing Assets to Manage Profits. Management Review, 25(2): 67-74. Hannan, M . 1998. Rethinking age dependence in organizational mortality: Logical formalizations. American Journal of Sociology, 104:126-164. Hannan, M . T . and G.R. Carroll 1992. Dynamics of Organizational Populations. Oxford University Press: Oxford. Hannan, M . T . and J. Freeman 1977. The population ecology of organizations. American Journal of Sociology, 82(5): 929-964. Helfat C E . and M . B . Lieberman 2002. The birth of capabilities: Market entry and the importance of pre-history. Industrial and Corporate Change, 11(4): 725-760. Helfat, C E , and M . Peteraf 2003. The dynamic resource-based view: Capability lifecycles. Strategic Management Journal, 24: 997-1010. Helsley, R . W . 2000. Urban and Real Estate Economics. U B C Real Estate Division. Helsley, R . W and W . C . Strange 2002. Innovation and input sharing. Journal of Urban Economics, 51(1): 22-45. Henderson, A . 1999. F i rm strategy and age dependence in organizational mortality: a Contingent view of the liability of newness, adolescence and obsolescence. Administrative Science Quarterly, 44:281-314. Henderson, J .V . 2000. Marshall's scale economies. NBER Working Paper #7358, Highfield, R. and R. Smiley 1987. New business starts and economic activity. International Journal of Industrial Organization, 5(1): 51-66. Hopenhayn, H . A . 1992. Entry, Exit, and Fi rm Dynamics in Long Run Equilibrium. Econometrica, 60(5): 1127-1150. Jaffee, J. 2003. Law firm office location and firm survival in Silicon Valley, 1969 to 1998. Jovanovic, B . 1982. Selection and the evolution of industry. Econometrica, 50(3): 649-70. Keasy, K . , Watson, R. 1991. The State of the Art of Small Business Failure Prediction: Achievements and Prognosis. International Small Business Journal, 4(9): 11-29. Keller, W . 2002. Geographic localization of international technology diffusion. American Economic Review, 92(1): 120-142. Kiefer, Nicholas M . 1988. Economic duration data and hazard functions. Journal of Economic Literature, 26(2): 646-79. Klepper, S. 2003. The geography of organizational knowledge. Carnegie Mellon University working paper. Krugman, P. 1991a. Geography and Trade. M I T Press: Cambridge. Krugman, P. 1991b. Increasing returns and economic geography. Journal of Political Economy, 99(3): 483-99. Lancaster, T. 1990. The Econometric Analysis of Transition Data. Cambridge University Press: Cambridge. Levinthal, D . and K . March 1981. A Model of adaptive organizational search. Journal of Economic Behavior and Organizations, 2:307-333. Levinthal, D . A . 1991. Random walks and organizational mortality. Administrative Science Quarterly, 36:397-420. L i , J. 1995. Foreign entry and survival: Effects of strategic choices on performance in international markets. Strategic Management Journal, 16(5): 333-351. Lippman, S. and R. Rumelt 1992. Uncertain immitability: A n analysis of interfirm differences in efficiency under competition. Bell Journal of Economics, 13, 418-438. L o m i , A . 2000. Density dependence and spatial duality in organizational founding rates: Danish commercial banks, 1846-1989. Organization Studies, 21(2): 433-461. Mahmood, T. 2000. Survival of Newly Founded Businesses: A Log-Logistic Model Approach. Small Business Economics, 14): 223-237. Malmberg, A . and P. Maskell 1997. Towards an explanation of regional specialization and industry agglomeration. European Planning Studies, 5(1): 25-42. March, J. G . and Z . Shapira 1987. Managerial perspectives on risk and risk taking. Management Science, 1404-1418. Mata, J. and P. Portugal 1994. Life duration of new firms. Journal of Industrial Economics, 42(3): 227-45. Mata, J. and P. Portugal 2002. The survival of new domestic and foreign-owned firms. Strategic Management Journal, 23:323-343. Mata, J. , P. Portugal and P. Guimaraes 1995. The survival of new plants: Start-up conditions and post-entry evolution. International Journal of Industrial Organization, 13(4): 459-81. Mitchel , W . , J. Shaver and B . Yeung 1994. Foreign entrant survival and foreign market share: Canadian companies' experience in United States medical sector markets. Strategic Management Journal, 15(7): 555-567. Moretti, E . 2004. Workers' education, spillovers and productivity: Evidence from plant-level production functions. UCLA andNBER working paper. Myers, S. 1977. Determinants of Corporate Borrowing. Journal of Financial Economics, 5:147-155. Olley, S.G., and A . Pakes 1996. The dynamic of productivity in the telecommunications equipment industry. Econometrica, 64(6), 1263-1297. Pakes, A . and R. Ericson 1998. Empirical implications of alternative models of firm dynamics. Journal of Economic Theory, 79(1): 1-45. Pfeiffer, F . and F. Reize 2000. Business start-ups by the unemployed - A n econometric analysis based on firm data. Labour Economics, 7:629-663. Porter, M . E . 1990. The Competitive Advantage of Nations. Free Press: New York. Porter, M . E . 1998a. Clusters and the new economics of competition. Harvard Business Review, Nov-Dec:77-90. Porter, M . E . 1998b. On Competition. Harvard Business School Press: Boston, M A . Porter, M . E . 2000. Location, competition, and economic development: Local clusters in a global economy. Economic Development Quarterly, 14(1): 15-34. Porter, M . E . 2001. Clusters of Innovation: Regional Foundations of U .S . Competitiveness, Council of Competitiveness. Ranger-Moore, J. 1997. Bigger may be better, but Is older wiser? American Sociological Review, 62:903-920. Richardson, H . W . 1979. Regional Economics. University of Illinois Press: Urbana. Romanelli, E . 1989. Environments and strategies of organization start-ups: Effects on early survival. Administrative Science Quarterly, 34:369-387. Rosenfeld, S. 1997. Bringing business clusters into the mainstream of economic development. European Planning Studies, 5:3-23. Rosenthal, S. and W . C . Strange 2001. The determinants of agglomeration. Journal of Urban Economics, 20(2): 191-229. Rosenthal, S. and W . C . Strange 2003. Evidence on the nature and sources of agglomeration economies, In J. V . Henderson and J. F. Thisse (Ed.), Handbook of Urban and Regional Economics. Schary, M . A . 1991. The probability of exit. Rand Journal of Economics, 22:339-353. Shane, S. and M . D . Foo 1999. New Fi rm Survival: Institutional Explanations for New Franchisor Mortality. Management Science, 45(2): 142-159. Shaver, J . M . and F. Flyer 2000. Agglomeration economies, firm heterogeneity, and foreign direct investment in the United States. Strategic Management Journal, 21(12): 1175-1193. Sorenson, O. and P .G . Audia 2000. The social structure of entrepreneurial activity: Geographic concentration of footwear production in the United States, 1940-1989. American Journal of Sociology, 106(2): 424-462. Stinchcombe, A . L . 1965. Social structure and organizations, In J. G . March (Ed.), Handbook of Organizations, Rand McNal ly : Chicago IL, 153-193. Stuart, T .E . and O. Sorenson 2003. The geography of opportunity: Spatial hetrogeneity in founding rates and the performance of biotechnology firms. Research Policy 32: 229-253. Sutton, J. 1997. Gibrat's legacy. Journal of Economic Literature, 35:40-59. Tirole, J. 1988. The Theory of Industrial Organization. The M I T Press: Cambridge, M A . Troske, K . R . 1996. The dynamics adjustment process of firm entry and exit in manufacturing and finance, insurance, and real estate. Journal of Law Economics, 39(2): 705-735. Tveteras, R., Eide, G . E . 2000. Survival of new plants in different industry environments in Norwegian manufacturing: A semi-proportional Cox model approach. Small Business Economics, 14:65-82. Wagner, J. 1994. The post-entry performance of new small firms in German manufacturing industries. Journal of Industrial Economics, XLII(2): 141-154. Winter, S.G. 1984. Schumpeterian competition in alternative technological regimes. Journal of Economic Behavior and Organization, 5:287-320. Zingales, L . 1998. Survival of the Fittest? Exit and Financing in the Trucking Industry. Journal of Finance, 53(3): 905-938. Appendix 2.A: Variables Definition Variable Firm Attributes Relative size * Growth Relative quality of HC Leverage * Productivity * Definition Firm size measured by Average Labor Units (ALUs) divided by incumbents' average size for the sector operating within the critical chosen radius. Source: T2LEAP z - z t-2 ( Z , + Z , _ 2 ) / 2 where Zt represents some attribute of the firm in year t. Source: T2LEAP Average wages paid by a firm {in founding year} divided by the sector's average wages within the critical chosen radius. Source: T2LEAP ln(debtl_l I assets ). Source: T2LEAP \\n{ATFP() = \\n ^ sales; ^ alu aln ^ assets, — inventories ^ alu-where alu, is the average labor units or total employees of the firm, sales, is the total sales of the firm, assets, its total assets of that firm and inventories • is its closing inventories, all measured by the end of the first year of operation. Source: T2LEAP; Annual Survey of Manufacturing initial values of these covariates were calculated in founding year. Variable Definition Industry Industry growth ** • Average 3 digit SIC industry percentage sales growth over 2 years. • Average percentage change in employment over 2 years in the focal sector Source: T2LEAP Local market structure • Establishments per worker in the 3 digit SIC sector within the critical chosen radius. ** (competitiveness) • Inverse Herfindahl index in the 3 digit SIC sector within the critical chosen radius. Source: T2LEAP Local entry rate 2 year average entry rate into the focal sector (SIC3) within the critical chosen radius. Source: T2LEAP Fixed-effects Economic Region Economic region fixed-effects Industry 2 digit SIC Industry dummies Time Year dummies The first definition is used in the reported results. Alternative measures were used for robustness. Appendix 2.B: Correlation Table Average S.D. 1 2 1 Cluster (dummy) 0.560 2 initial relative size 0.668 1.215 -0.021 3 growth 0.413 0.885 -0.003 -0.009 4 (growth) A 2 0.955 1.546 -0.003 0.000 5 relative quality of HC (Iag1) 0.928 0.145 -0.214 0.169 6 initial relative quality of HC 0.913 0.149 -0.235 0.184 7 leverage (Iag1) 0.634 0.472 -0.007 -0.040 8 initial leverage 0.847 0.375 -0.014 -0.075 9 productivity (Iag1) 2.037 1.354 0.028 -0.023 10 initial productivity 2.759 0.875 0.057 -0.052 11 local market competitiveness 0.169 4.719 0.032 0.002 12 industry growth (2YA) 0.100 0.169 0.002 0.006 13 local entry rate (2YA) 0.106 0.146 0.125 0.011 3 4 5 6 7 8 9 10 0.811 -0.066 -0.023 0.027 0.031 0.600 -0.363 -0.374 0.002 -0.024 0.061 0.074 -0.043 -0.037 0.410 -0.415 -0.426 0.045 -0.003 0.386 -0.018 0.025 -0.002 0.011 0.046 0.015 0.043 0.313 -0.013 0.015 0.014 0.020 0.003 0.005 -0.007 -0.006 0.046 0.028 -0.016 -0.019 -0.281 0.005 -0.028 -0.031 0.307 0.384 0.026 0.049 -0.014 0.044 -0.277 -0.038 CHAPTER THREE Firm Failures as a Determinant of New Entry: Is There Evidence of Local Creative Destruction? ABSTRACT The study posits that the causal links between entry and failure rates also flow from failure to entry. A s older firms fail, resources are recycled by new entrants permitting local renewal. Failure levels affect both the intensity of the entrepreneurial search for opportunities and the opportunity landscape facing entrepreneurs. Persistent high failure rates, however, provide signals about the difficulties of succeeding in a particular location. The model is tested using a unique longitudinal panel database made available by Statistics Canada, which encompasses the entry and exit of all Canadian enterprises between 1984 and 1998. Introduction \"The market must clean itself out by taking resources away from the losers, so it creatively destroys the losing companies and reallocates resources to new companies\". Former House Majority Leader Dick Armey (March 2002) A process of renewal, where the destruction of old firms facilitates the entry of new ones, was identified by Schumpeter (1942) as a key driver of economic growth in capitalistic systems. This paper asks whether there is evidence that exits of older firms facilitate the entry of new firms at the local level, and how information about persistent failure rates affects this process of renewal. The existence of high correlation between entry and exit rates in many countries is well documented (e.g., Dunne, et al. 1989, Eaton and Lipsey 1980, Cable and Schwalbach 1991). These correlations are attributed in the organizational ecology and economic literatures to causal links flowing from entry to failure (Geroski 1995). These links include the 'liability of newness', the relationship between entry and failure barriers, and adverse selection. The contribution of this paper is in showing that exits stimulate entry and that the process of renewal takes place within relatively small spaces. We find that persistent failure rates provide entrepreneurs with signals they use to calibrate their location choices within the region so as to reduce the risks they face. This study also shows how signals about location characteristics from different neighborhoods within a region \"compete\" in influencing location entry strategies. B y defining location in terms of relatively smaller neighborhoods within regions and examining the impact of economic variables on new entry at different levels of geographical aggregation, we probe the spatial dimensions of clusters, and show how impacts attenuate with distance. Understanding both why firms choose specific locations and the dynamics of exit and entry and its spatial dimensions help in formulating effective regional growth policies, which tend to rely on incentives and promotion. Our study provides insights into who should be targeted and what signals are likely to have an impact on entrepreneurial entry. We estimate tobit models of entry events belonging to five manufacturing sectors in Canada between 1984 and 1998 at the census subdivision level, using a unique database made available by Statistics Canada that accurately identifies firm entries and exits. Theory Development Failures Geroski (1995: 423) observed that a common view among economists is that enterprises entry occurs when super-normal profits are positive, and exit when they are negative. He pointed out, however, that entry and exit rates are highly positively correlated. Indeed Baldwin and Goreski (1991) have shown that the correlation between entry and exit in Canada ranged between 0.5 and 0.7 during the 1970s. Dunne et al. (1988) found a consistent positive relationship between entry and exit for U . S . manufacturing industries in the 1963-82 period. Cable and Schwalbach (1991) provided the correlation results for other countries for varying periods. For example, Belgium 0.66, Norway 0.49, United Kingdom as high as 0.79, and Germany 0.34 6 9 . There are three economic explanations for this positive correlation. First, as entry increases, the number of firms may rise, leading to increased competition and, hence, more exit. Second, i f new entrants have lower survival rates than incumbent firms in the market, high entry should be associated with high exit (Agarwal and Gort 1996). Third, high entry barriers imply high exit barriers since i f firms cannot enter, incumbents who wish to exit may not receive sufficient compensation for 6 9 In contrast Klepper (1996) and Agarwal and Gort (1996) suggest that entry and exit rates need not be positively correlated, and that across stages of the industry life cycle, these correlations change from negative to positive to zero. specific immobile resources and therefore may choose to delay their exit (Dixit and Pindyck 1994, Eaton and Lipsey 1980). We argue that the causal links between entry and failure may flow also from failure to entry. Furthermore, these relationships are anchored in proximate locations. Failure levels affect the pool of potential entrepreneurs searching for opportunities in a particular location in the industries where the failure occurred. Employees of a given organizational form gain knowledge of how to operate its production technology (Sorenson and Audia 2000), including explicit knowledge learned through instruction and tacit knowledge learned through experience 7 0. Through their normal work, employees have low-cost access to a stock of knowledge that outsiders can obtain only at great cost (Greve 2000). American and European studies provide evidences that one third of new manufacturing firms were established by unemployed individuals (Storey 1985, Tveteras 2000). Since some of the knowledge and the social capital (such as local credibility, established relationships with venture capitalists, potential strategic partners, suppliers and customers) that entrepreneurs have accumulated are anchored in the location of their former employers, they may have an advantage in locating new enterprises in those locations (Shane and Stuart 2002). Furthermore, work practices, culture and technical terminology are often particular to a region and vary dramatically across regions (Saxenian 1994), creating additional incentives for potential entrepreneurs to search for opportunities in the location where they have been previously employed. More generally, the costs of search are lower when local social networks are used since spatial proximity greatly facilitates location idiosyncratic knowledge, relationship formation and the exchange of information (Krugman 1991, Porter 2000, Saxenian 1994). Similarly, entrepreneurs who plan to start an enterprise for other reasons are likely to establish organizations in which they can draw on their experience in a 7 0 Stickiness of knowledge (i.e., complexity, causal ambiguity) indicates that it is not easily transferable to recipients outside the organization (Szulanski 2000). particular industry and location (Cooper & Dunkelberg 1987, Klepper 2002, 2003, Dahlqvist, et al. 2000). There is ample evidence that entrepreneurs tend to stay in the immediate area where they have local connections and familiarity with local institutions (Reynolds 1997, Cooper 1984) 7 1. The majority of new entrants into the chip sub-sector in the Silicon Valley, for example, emerged from employees of spatially proximate firms (Boeker 1989); the same was true for new law firms (Jaffee 2003). More than seventy percent of the founders of biotechnology firms in the state of Washington founded their firms near their residence (Haug 1995). Valuable information concerning failures and closures and the opportunities that they present for establishing new firms may remain within regional or even local geographic boundaries. McGrath (1999:14) asserted that 'The initiative that fails may still improve knowledge or methods of production'. Thus, the consequences of failures contribute to the accumulation of knowledge in a location and may benefit not only incumbents but also new entrants. Failures also release resources. Some of these resources are immobile (e.g., plants, machinery) or partially immobile (e.g., skilled workers who prefer to stay in the same location). These immobile resources present opportunities as their prices may fall to clear the market. Since information about these opportunities usually travels faster through local social networks, local 72 potential entrepreneurs are more likely to take advantage of these opportunities than outsiders . A l l else equal, local potential entrepreneurs are more likely to use the released resources locally, even when these resources are mobile. We thus have the following hypothesis. 7 1 Delmar and Davidsson (2000: 14) found lack of mobility in the Swedish nascent entrepreneurs' population: \"People stay where they are, and do not move to where new jobs are (or could be) created.\" 7 2 Geographic proximity enables more face-to-face interaction that enhances trust and the utility of the mechanisms of knowledge flows (Porter 2000). Hypothesis 1. The likelihood of new entry to a location is positively related to the number of failures in that location and in neighboring locations. Failure levels in other parts of a region may also release resources. If these resources are mobile, their availability w i l l increase the attractiveness of neighboring locations. The availability of immobile resources may, however, attract entrepreneurs from other locations to establish new ventures in the location where the failure occurred. The net impact may vary from sector to sector, reflecting differences in asset mobility and the relative importance of location-specific knowledge and social ties among sectors. Since local entrepreneurs have an advantage in identifying and evaluating immobile resources released to the market after a failure and are more motivated than outsiders to use them in starting a new business, arguably the availability of such resources w i l l attract fewer entrepreneurs from outside the neighborhood. Moreover, entrepreneurs tend not to relocate to establish new firms but rather use local existing networks and contacts to start their firms (Feldman 2001). On the other hand, mobile resources are easier to evaluate without the benefit of local knowledge and thus may have a stronger effect on entry to other locations. Since the density of social networks tends to decrease with distance and noise increases with distance, we expect that the externalities of failure w i l l attenuate with distance. This may also occur because transferring resources from location to location involves transportation costs, which increase with distance. This leads us to the second hypothesis. Hypothesis 2. The impact of firm failure on entry in a specific location attenuates with distance. While failure levels generate opportunities that may encourage entry to the location, they may also signal risks associated with the location. High failure rates that are perceived to be temporary, for example, those associated with an external shock or a down period in the business cycle, or failure of older firms that are perceived to be misaligned with the market w i l l not damage the reputation of the location and w i l l not deter potential entry. However, failure rates which persist over time may send a strong signal that a location is risky and therefore detract potential entrants. We thus have the next hypothesis. Hypothesis 3. The likelihood of new entry in a particular industry is lower in locations with higher persistent failure rates in the same industry. Persistent same industry high failure rates in neighboring locations may spill-over and reduce the reputation of a location. This impact, however, should attenuate with distance. Reputation of a location may also be affected by persistent failure rates in other sectors within the region. Agglomeration Economies and Competition Other key factors that drive the location decision processes and birth of new enterprises are agglomeration economies and local competition. Agglomeration economies are a form of scale economies external to any one firm but internal to regions containing clusters of the same type of enterprises (Krugman 1991). Agglomeration economies imply positive returns to scale at the regional level such that the advantage to an organization of locating in a particular region increases with the number of other firms in the area. There are two types of externalities: localization economies which arise from the scale of the local focal industry and urbanization economies where firms benefit from local information spillovers from all firms and employees in a region outside their own industry through industrial cross-fertilization (Ellison and Glaeser 1997, Henderson 2000). New establishments located in areas with strong agglomeration economies w i l l benefit from lower costs and superior access to skilled workers (Helsley and Strange 1990), higher productivity, knowledge spillovers (Henderson 1994, Marshall 1920), reduced resource constraints, the existence of specialized input providers and business services all particularly tailored to the specific industry (Porter 1998), and the availability of financial and managerial support (Zackarakis 2001). Since clusters provide many opportunities for outsourcing, new establishments in clusters can increase their efficiency by focusing on their internal competencies. Firms in clusters are more exposed to external knowledge and collaboration opportunities with other firms within the same industry. Thus, they are more likely to adopt, implement and commercialize innovations than their more isolated peers. 7 3 New establishments limited by their internal capabilities and by their access to available resources (Shane and Stuart 2002, Shane and Venkatataman 2000) prefer locations where entry barriers are lower. Indeed, locations with strong localization economies that are engendered by clusters act as nurseries or incubators that attract and support entrepreneurial activity. Demand externalities arise when the spatial concentration of sellers reduces consumers' search costs and thus increases the likelihood of visitation and purchase (Baum and Haveman 1997, Chung and Kalnins 2001). If an industry is subject more to urbanization economies, new establishments w i l l seek more diverse and hence usually a larger local economic environment (Henderson 2000). Jacobs (1969) asserts that the most fertile locations for entrepreneurial entry are locations with a diverse set of related industries. Thus, locations with strong localization economies may not exhibit high startup activity. Glaeser et al. (1992) find that spillovers across industries are more important for startups than spillovers within industries. Duranton and Puga (2001) assert that new experimental production at the development stage is more likely to be found in large diverse metropolitan areas due to externalities generated by cross-fertilization. A diversified environment allows entrepreneurs to experiment without costly relocation after each trial. 7 3 Note however that strong companies may choose to locate distantly to slow information spillovers that might erode their competitive advantage (Shaver and Flyer 2000). Concentration, however, amplifies rivalry and competition (Porter 2000). Competition makes a difference to the location choice of a new enterprise in several ways. A high level of competition may facilitate innovation and the adoption of 'best practices' and may thus enhance localization economies (spillover effects). The existence of many small firms rather than a few large ones may reduce entry costs, making a location more welcoming and open to new enterprises. Nevertheless, high levels of competition also represent pressures on prices and entrepreneurial profit margins. The importance of each of these impacts depends on the nature of the industry (i.e., the importance of innovation, knowledge spillovers, entry barriers and stage in the life cycle of its products) (Agarwal and Gort 1996, Gort and Klepper 1982). Indeed, the empirical evidence is mixed. Porter (1990) argues based on case evidence that business w i l l become more productive when consumers are demanding, and when competitive pressures compel sustained innovation. Glaeser, et al. (1992) and Rosenthal and Strange (2003b) in econometric studies show that competition is positively associated with growth and new births. In contrast, Baum and Mezias (1992) find that the benefits of agglomerations (demand externalities) diminish as local competition increases (see also Baum and Singh (1994), and L o m i (1995) for similar effects). Method Model To test the hypotheses presented earlier, we develop a basic econometric model that examines the location choice of entrepreneurs in various industries. Normalizing the price of output to one, an enterprise w i l l generate profit n{y) that can be expressed as: (1) n (y) = a{y)f{x)-c{x) where x is the vector of factor inputs, c(x) is the cost of inputs, f(x) is the production function, and a(y) is a vector that represents a shift in the production function as a result of the impact of locational characteristics. These include in our model agglomeration economies, resources availability, competition, and location specific risks. Entrepreneurs w i l l choose input quantities in order to maximize profits by satisfying the first order conditions as long as the profit generated realizing the investment opportunity is non-negative. We therefore assume that with a large pool of entrepreneurs searching for opportunities to establish new enterprises, all opportunities that are expected to yield non-negative profits w i l l be taken (c(x) provides for normal profits sufficient to compensate for entrepreneurial efforts). New ventures are heterogeneous in their potential profitability. We thus incorporate heterogeneity in the profit function by multiplying the production function by (1 + £), where £ is independent and identically distributed across enterprises: Suppose that £ is iid across new establishments according to the cumulative distribution We assume that the location choice set of entrepreneurs is all Census Subdivisions (CSDs) in Canada, j=1,... ,J 7 4 . A Location decision is made at time t-1 taking the existing economic ye environment as given; the establishment is born one period later at time t . Entrepreneurs seek 7 4 The hierarchy of geographic units in Canada is: Province, Economic Region (ER), Census Division (CD), and Census Subdivision (CSD). 7 5 The creation process of new organizations is not a discrete event (Sarason 1976). Following previous research and given our data we define the founding decision as a one stage event that incorporates the observed economic environment (e.g. Baum and Singh 1994). (2) x (y) = a{y)f(x)a + £)-c{x) function For any y, there is a critical level £ [y profit-maximizing locations and therefore wi l l be attracted to the most productive CSDs . Thus, our 'risk set' is all 5,260 census-subdivisions in Canada who are 'at risk' of event occurrence (births of new establishments) at each point in time (see F I G U R E 3.1). Aggregating over new enterprises in each location C S D j in year t for a given industrial sector i, yields the number of births 5 ; . . Each year there are births in many census subdivisions (the uncensored observations) 7 6, but zero births in the majority of the census subdivisions (the censored observations). The large number of censored observations suggests the use of a Tobit estimation which allows us to use all of the observations (Wooldridge 2003: 565) 7 7 7 8 . Thus, BCSDj,, is observed i f there are opportunities with positive profits in CSD. in sector i, and is not observed (censored) otherwise. The observed . is a function of the local characteristics y modeled CSDj,t,i y as follows (3) BCSD.U = 0 otherwise . - I N The estimation of this model is done using the maximum likelihood method (Maddala 2001). The maximum number of annual births in a CSD is 67 for the electronic sector. 7 7 However, this is fixed censoring (or Type I censoring) so it is unnecessary to make any further assumptions about the nature of the censoring process (Maddala 2001). This empirical strategy was used in Henderson et al. (1995), and Rosenthal and Strange (2003). 7 8 Other estimation methods are available for limited dependent variables which get the value zero for nontrivial fraction of the population but are essentially continuously distributed over strictly positive values. A Poisson estimation is a common procedure for count data. However, the model is very restrictive. Specifically it assumes that the econometrician observes all sources of heterogeneity, the events occur at a constant rate, and that the variance equals the mean. Negative Binomial and Zero Inflated Poisson (ZIP) regression models using maximum likelihood techniques cope with these assumptions and are alternative estimations. ZIP models are formed by a mixture of two distributions: a point mass distribution at zero, and a Poisson (or overdispersed Poisson) distribution. A zero-inflated model changes the mean structure of the pure Poisson model such that the predictive quality is improved. The key results of this paper are robust to alternative estimation methods and various measures of the covariates (see robustness section). FIGURE 3.1: Location Choice Set 8 In the theory section above, we suggested that locations are heterogeneous with respect to the number of births they generate. Specifically, the hazard rate of births for a given census-subdivision increases with failure levels, decreases with local failure rates and increases with existing economic activity in a sector. We incorporate the sources of that heterogeneity as explanatory variables in the model by partitioning the location characteristics of each C S D , y. f i . into two vectors, yCSDj M . and y . The elements of y t ] . vary by C S D and sector. They include covariates that may influence births in a specific C S D , as we discuss below. The elements of yER. vary by economic region (i.e., the heterogeneity component of the ER) and include attributes such as macroeconomic and political stability, fiscal policies, labor market policies, intellectual property rules, quality of workforce, public infrastructure systems, legal system, efficiency of local authorities, wage rates, and natural attributes of the E R such as climate. Given the wide range of economic-region attributes that affect productivity, risks, and new investments incentives, we believe that identification and measurement of them in our model could result in omitted variable bias. However, they cut across all industries and are common factors to the E R . Thus, in order to control for these effects, we introduce ER-specific fixed effects in our estimation 7 9. Accordingly, we can write the aggregated number births (B) in a census subdivision (CSD) j at time t in sector i as: (4) BCSDj,t,i = tycSW-H + dER +d<+ di + £j,1,i where y ( 1 . are the location characteristics of C S D j at time t-1 for sector i. The rest of the variables are dummies for various unmeasured effects, dER are the E R specific fixed effects that absorb permanent heterogeneity at the E R level, dt is a time fixed effect for each of the years which picks up the time-varying trend in the error which may come from business cycles or general trends in technology, dt is 2-digit SIC-E industry fixed effects that absorb industry permanent heterogeneity8 0, and e . is a location-contemporaneous error term. Fixed effect estimation assumes that the R H S variables are uncorrected with the unobserved error P in c j,t,i order to avoid biased estimators 8 1. Measures The explanatory variables of interest included in the vector y M . are the measures of failure levels and rates at the C S D j and neighboring CSDs within the same C D in the sector i, as well as 7 9 This empirical strategy has a principal drawback - it does not allow us to identify, test or measure the ER characteristics that affect a location choice. 8 0 Since some of the repressors vary by 3-digit SIC level we use 2-digit SIC level fixed effects. Moreover, industry fixed effects are expected to operate at higher aggregation level than 3-digit. 8 1 In the robustness section we describe interaction dummy terms. failure rates of other sectors in the C D in periods (t-1) and (7-2). Identification problem may arise from an exogenous shock that drives both failures and births at a location for a specific sector. Our approach to deal with this problem is the use of instruments. Failures at the C S D level are measured as the lagged number of older failed firms (> 4 years old) in the focal sector (CSD F A I L U R E ) . This measure is unlikely to pick up cyclical effects as the majority of entrants fail within the first three years. Moreover, the focus on failures of older firms allows us to examine Schumpeter's hypothesis of creative destruction. Because it may take some time to release assets and other resources from failed enterprises to the market we use in our estimation failure levels with both one and two-year lags, . , r=l,2. The influence of failures of firms that existed j o CSDj,t-r,i at least four years from the same focal sector in neighboring CSDs within the same C D (CD F A I L U R E - O T H E R CSDs) is represented by the vector FCD ,_r. , r=l,2: To control for unobserved local shocks we also include the vector F - - the lagged number older failed CD,t-l,l 0 0 firms (> 4 years old) from other sectors (CD F A I L U R E - O T H E R HMD). The local probability of failure (CSD F A I L U R E R A T E ) , p „ c n . ., is the sum of the number of all failures (young and older firms) over the three years prior to the observed entry in the focal sector at the C S D , divided by the relevant population at risk (i.e., C S D ' s incumbents in the sector at time t-4), 3 ^ FcSDj,t-r,i PCSD- t-i i ~ • This variable is introduced to capture the impact of persistent J' '' incumbents CSDjt_Ai failures in the location on its reputation and thus entry. We also include the persistent of failure rates for neighboring CSDs within the same C D in the focal sector (CD F A I L U R E R A T E -O T H E R CSDs) P C D T _ U , and the C D probability of failure in other sectors (CD F A I L U R E R A T E See alternative measures at the robustness section. - & O T H E R END.), P , = — , where / represents all other manufacturing ' '' incumbentsCD , 83 sectors . Other characteristics introduced as controls are represented by the vectors C S D • f_,, and C D , • To control for localization externalities, we employ measures of C S D N U M B E R O F P L A N T S (e.g., Head et al., 1995), or C S D E M P L O Y M E N T level (e.g., Henderson 2000). The A R E A of each C S D is included to control for differences in C S D sizes and land supply and therefore land prices (e.g. Coughlin et al. 1991, Head and Mayer 2003). To control for local competition, following Glaeser et al. (1992) and Rosenthal & Strange (2003b), we include the number of E S T A B L I S H M E N T S P E R W O R K E R at the C S D (and neighboring C S D s within the same C D ) level in the focal sector. A s this ratio decreases the local environment in a given industry in the C S D is thought to become more competitive. Urbanization externalities are measured at the C D level in order to account for CD-wide industrial cross-fertilization through information spillovers, social networks and other sources due to diversity of the census division base 8 4. Following Henderson (2000: 12) these are measured through the total number of plants or employment levels in other sectors ( N U M B E R O F P L A N T S / E M P L O Y M E N T - O T H E R I N D U S T R I E S ) . Since we employ E C O N O M I C R E G I O N F I X E D E F F E C T S in all of our models, the above covariates capture the influence of within economic region variation in the industrial environment. In our most robust models we included C D x time interaction effect. Accordingly, we can rewrite the estimated model as: 8 3 For both levels and persistent rates of failure in other sectors we experimented with excluding the focal CSD. The results do not materially change. 8 4 Note, our urbanization covariate pools across two different dimensions: all industrial sectors excluding the focal sector, all neighboring CSDs within the same CD. BcSD;,t,i = ^,(fll,rFcSDj,t-r,i + fi2,rFcD,t-r,i)+ P^CDj-U + Xl^CSDj ,t-\\,i + %2^CD,t-\\,i + XiPCD,t-\\,i r=l 8£SDht_Xi + S2CD,_X + dER +d,+ dt + ehU Under this specification, the estimates of /?,,,/?, 2 , /?21 > A 2 a n ( * Xi c a n ^e u s e c * t 0 t e s t o u r hypotheses. A l l covariates were updated annually and lagged one or two years in the analysis to avoid simultaneity problems. Since the variance of the unobservables changes across different clusters of the population (3-digit SIC-E industries) all regressions use the HuberAVhite method to correct for hetroskedasticity. Data Sources We use two different Canadian databases to estimate our model. The first database, the Longitudinal Employment Analysis Program ( L E A P ) , is used to identify new entrepreneurial entry and levels of firms' failure. The second database is the Annual Survey (Census) of Manufactures - the Longitudinal Manufactures Research File ( L R M F ) . This database is used to assess agglomeration covariates. L E A P is a unique, firm-level database that includes all employers in Canada, both incorporated and unincorporated. The database tracks the employment and payroll characteristics of individual firms from their year of entry to their year of exit. The employment record of each 85 firm is derived from administrative taxation records that each Canadian employer must file . The payroll data are associated with a Revenue Canada employer identification number. Accordingly, firms enter the L E A P database in the year they first hire employees, and record 8 5 Every employer in Canada is required to register a payroll deduction account (for the purpose of unemployment insurance), and issue a T4 slip to each employee that summarizes earnings received in a given fiscal year. The LEAP database includes every business that issues a T4 taxation slip. their last entry in the database in the last year they have employees. For each year, total payroll and employment are calculated. The latter is the average annual count of employees within the firm, or average labor unit ( A L U ) . This payroll and employment information is then organized longitudinally; each observation in the database corresponds to a particular firm whose employment, payroll and industry characteristics are recorded at different points in time. The longitudinal nature of L E A P allows entry and exit times to be measured with precision. Births (entrants) in any given year are firms that have current payroll data, but did not have payroll data in the previous year. In our empirical estimation, we include only entrepreneurial entry (also referred as 'de novo', independent, or new entry); we do not include births of establishments that are owned by a firm that had establishments in previous years (also referred as dependent, subsidiary), or firms that were classified as belonging to another industry at time t-1. Entrepreneurial entry accounts for 85% to 94% of all newly created establishments, depending on the sector. Similarly, deaths (exits) in any given year are identified by the absence of current payroll data, where such data had existed in the previous year (see Appendix 3.B for further description of a labor tracking method that allows accurate measure of entry and exit). The L E A P database tracks the postal-codes of firms; we transform this information into various geographic disaggregated levels: economic region, census division and census subdivision. Thus, we are not handicapped by a crude unit of observation which characterizes former analysis of spatial entrepreneurial location. Our database covers the years 1984 to 1998. The second database that is used to measure location of manufacturing activity is the Annual Survey (Census) of Manufacturing. It contains data about physical entities 8 6 The total number of employees in LEAP is slightly less than the number of full time equivalent workers in the Canadian economy as LEAP excludes individuals who are self-employed. (establishments, plants) , irrespective of who owns them or how they relate to each other. The data are is derived from a survey that is sent to large plants and from administrative tax data for small plants, combined with the full census of manufacturing every five years. The data covers the years 1983 to 1997. The Geographic Units Previous research on the geography of entrepreneurial entry found a rapid spatial decay of some location externalities (Rosenthal and Strange 2001). Employment activity in a plant's own county affects plant productivity. But employment activity in neighboring counties was not found to affect the plant's productivity (Henderson 2000). Knowledge spillovers impact highly localized agglomeration (zip code), while labor impacts agglomeration at all levels of geography (zip code, county, state) (Rosenthal and Strange 2001). Thus, forces that contribute to agglomeration operate at a more refined geographic scale than the region or state units used in most previous research (e.g., Audretsch and Mahmood 1993, Dunne, et al. 1989, Mata 1993, Sorenson and Audia 2000). Therefore, a disaggregated level of geographic information is vital to measure the effect of industrial environment externalities on births. We use detailed location information of new entries and existing establishments at the census 88 subdivision (CSD) level, which has about the same mean area as the U.S . zip code . We will use plants and establishments interchangeably throughout the paper. Note, the vast majority of firms are single establishments. Thus, as a robustness check we ran our models measuring the control variables using the LEAP database. Our results do not materially change. 8 8 Alternatively, we could based our analysis on the relative distance between individual firms as in Stuart and Sorenson (2003). We prefer using the smallest existing administrative geographic units while evaluating several environmental aspects of the unit (e.g., localization and urbanization economies, market competitiveness) rather than classifying it as a cluster or no-cluster. Moreover, using the CSD as a level of observation we ensure that boundaries at the local level are not crossing different regulatory regimes. Industries We selected industries with substantial turnover that are important enough to have been the focus of other studies. Specifically, we estimate the determinants of entrepreneurial location choice of the food, apparel, fabricated metal, machinery and electronic sectors at the 3-digit Standard Industrial Classification (SIC-E) level (the selection is detailed in Appendix C ) 8 9 . The industries are a mix of traditional industries with established products (fabricated metals and machine industries) and more innovative industries (fashion and food). Additionally, the industries studied are a mix of heavy and light industries. Results The data cover 5,260 census subdivisions, 289 census divisions, and 71 economic regions. The large number of census subdivisions makes many of our estimates quite precise. Because of the large number of economic region fixed effects, it is hoped that all regional attributes that affect productivity are controlled for. Since regulations and zoning restrictions prevent some census subdivisions from having industrial activity, we have excluded from our risk set those C S D s that did not exhibit any existing manufacturing activity over fifteen years (within all 3 digit SIC-E codes), or new birth 9 0 . Appendix 3.A presents descriptive statistics and for independent and control variables. Our first Tobit fixed-effect model (Table3.1 column 1) serves as a baseline for statistical tests and identifies the impact of agglomeration effects on the birth of new establishments using the annual levels of new establishments at a census subdivision as dependent variables. Our analysis based on within economic region variation in the data yields results that are broadly 8 9 SIC-E was replaced by the North American Industry Classification System (NAICS). NAICS identifies hundreds of new, emerging, and advanced technology industries and reorganizes industries into more meaningful sectors. Our databases do not report NAICS. Nevertheless, major differences between the classification systems exist especially in the service-producing segments of the economy which are excluded from our analysis. 9 0 We excluded 1,352 CSDs, which leave us with a choice set of 3,908 active CSDs or 58,620 location-year observations. The mean area of CSD in our sample is 376 [square kilometer]. consistent with prior research that was based on variation between economic regions. The control variables receive the expected signs and reasonable magnitudes. Specifically, localization effects (measured as own-sector number of plants at the C S D level) are more important than urbanization effects (measured as other sectors number of plants at the C D level). Column 2 represents alternative measures to localization and urbanization economies which are based on employment levels. The coefficients on the localization variable are at least two orders of magnitude larger than the coefficients on the corresponding urbanization variables. The result that localization economies are more important than urbanization economies is consistent with Henderson (2000), and Rosenthal & Strange (2003a). Localization economy effects are expected to be positive at small geographic unit as they reflect the ability of new establishments to share supply channels of intermediate inputs, supporting services developed by existing firms, and a labor market pool. In choosing a location, founders of new establishments would consider the externalities offered by different potential locations and, all else equal, would prefer locations that provide higher agglomerative benefit. Running the Tobit estimates for the separate sectors shows that urbanization effects are not all positive and that there is a significant variation among sectors. Urbanization effects include tradeoffs between the benefits of locating near densely developed areas and the economies and amenities that the urban environment offers and congestion costs. Industries differ in the net benefits they derive from proximity to large and diverse metropolitan areas. Some enterprises prefer more densely developed areas while others prefer outlying locations. Some benefit from information spillovers due to cross-industrial-fertilization while others benefit from sector specific externalities. We have found that more births are generated in CSDs within more diverse CDs in the apparel and electronic sectors; the opposite is true for food and machinery sectors, which generate more births in less diverse CDs . New entrants take into account the costs of land as reflected by the positive coefficient on C S D area - larger CSDs attract significantly more entry than small ones 9 1. Specification 3 adds covariates of local competition in two geographic levels. We find that E S T A B L I S H M E N T S P E R W O R K E R in a sector at the C S D has a significant and negative influence on arrivals. Higher competition for resources in census subdivisions is associated with lower expected profits and thus deters arrival in all sectors. The impact of competition in the census division is also negative apart from the apparel sector (where the coefficient equals 2.72384 with z statistics of 1.867). In the apparel sector competition attracts new entries. Clothing is a sector in which new products are important since fashions change constantly and thus competition may offer the dynamic effects described by Porter (1998, 2000) suggesting that its benefits to the entrepreneur may outweigh its costs. We examined the robustness of this relationship by using two alternative measures: land rents, and manufacturing density. Land rents were obtained from the Census of Population and were interpolated between census years. Nevertheless, land rents may be correlated with unobserved locational attributes. These alternative measures obtained insignificant coefficients. TABLE 3.1: Tobit Estimates of New Entry: Investigating the Impact of Failures Dependent variable: aggregated number births in a census subdivision j, at time t, belonging to sector i Specification Regressors CSD number of plants (t-1) CSD employment (t-1) CD number of plants (t-1) [other ind.] CD employment (t-1) [other ind.] CSD area CSD establishments per worker (t-1) CD establishments per worker (t-1) [other CSDs within the same CD] CSD failure (t-1) CSD failure (t-2) CD failure (t-1) [other CSDs within the same CD] CD failure (t-2) [other CSDs within the same CD] CD failure (t-1) [other ind.] CSD failure rate (t-1) CD failure rate (t-1) [other CSDs within the same CD] CD failure rate (t-1) [other ind.] 0.04279*** (7.379) 0.00009** (2.545) 0.00005* (1.823) Economic region effects Time effects 2-digit industry effects Number of Births Number of CSD-year Yes Yes Yes 34,449 58,620 0.00024*** (3.725) 0.00002** (2.739) 0.00005* (1.826) Yes Yes Yes 34,449 58,620 0.04235*** (7.306) 0.00009** (2.526) 0.00004* (1.824) -1.33385** (2.556) -1.10551** (2.207) Yes Yes Yes 34,449 58,620 0.04156*** (6.379) 0.00009** (2.468) 0.00004* (1.825) -1.30723** (2.492) -1.00657** (2.286) 0.67664*** (5.349) 0.58141*** (4.448) 0.08334** (2.247) 0.10382*** (3.341) -0.00027* (1.703) Yes Yes Yes 34,449 58,620 0.04137*** (5.379) 0.00004* (1.824) -1.30755** (2.304) -1.00502** (2.148) -1.06772*** (3.353) 0.64852** (2.509) -0.21753** (2.233) Yes Yes Yes 34,449 58,620 0.04107*** (4.801) 0.00009** 0.00009** (2.220) (2.445) 0.00005* (1.823) -1.26816** (2.374) -1.00034** (2.645) 0.69005*** (6.824) . 0.51071*** (5.037) 0.07936** (2.136) 0.11153*** (4.267) -0.00021* (1.755) -0.942107** (2.306) 0.58772** (2.353) -0.15823** (2.519) Yes Yes Yes 34,449 58.620 Likelihood Ratio Index 0.221 0.206 0.308 0.360 0.358 0.389 Absolute value of z-statistics in parenthses * significant at 10% level; ** significant at 5% level; \" 'signif icant at 1% level Specifications 4, 5, and 6 add the primary variables of interest, failure levels and rates. Adding those covariates significantly improve the explanatory power of the regression equation. The likelihood ratio test (five and three degrees of freedom respectively) rejects the hypothesis that the added terms have no explanatory power at the 0.005 level. Model 4 reports the effects of local levels of failures by firms that existed at least four years when controlling for agglomeration, competition, area, time, industry, and economic-region fixed effects. Lagged levels of failure of older firms (t-1 and t-2) are large, significant and positive, while the coefficients of the control variables are almost unaffected. This finding supports hypothesis 1 — local levels of failures create a transitioning trigger (or opportunities) for local births. This result is not consistent with Sorenson and Audia (2000), who found no relationship between lagged failures and future organization founding in the U .S . footwear production sector 9 2. The magnitude and significance of our measurement of the level of failures at t-2 suggests that the response of entry to opportunities created by failures might be slow as it takes time for some opportunities to be realized. The coefficients of the C D levels of failure imply that some mobile resources or laid-off workers that are available elsewhere within the C D provide positive externalities to the entry process at the C S D level. Note that the magnitude of coefficient for the number of failures in the CDs is approximately five times lower than for the CSDs . This finding supports hypothesis 2 - the impact of firm failure on entry in a specific location attenuates with distance. Failures in other sectors in the C D have negative impact on births. Model 5 adds the covariates of failure rates. The negative and significant sign of failure rate in the census subdivisions implies that entrepreneurs are less attracted to locations with high persistent failure rates, as was suggested in hypothesis 3. Surprisingly, we did not find that high 9 2 However, since the footwear manufacturing industry is characterized by lack of strong scale economies and barriers to entry, and limited importance of human capital it may not necessarily represent other manufacturing sectors. persistent risks, in the same sector within neighboring locations, damage its reputation as we expected. On the contrary, the impact of persistent risks in neighboring locations increased the location attractiveness. This suggests that entrepreneurs calibrate their location choices, choosing, ceteris paribus, the location with the lower relative risk. The persistent probability of failure in other sectors has negative and significant coefficient suggesting that failure rates in all sectors in the region damage its reputation and affect the perceived riskiness of locations within it. Note that this effect is much smaller than the effect of own sector's risk at the C S D . Model 6 adds simultaneously the covariates of failure levels and rates. A s we move from specification 5 to specification 6, the coefficient on the C S D and C D failure rates moved to the 0.05 significant level. When compared to either specification 4 or specification 5 other coefficients retain their statistical significance and are quite stable. Robustness of Results Our empirical methodology raises a technical issue. Imprecise estimates of the fixed effects in nonlinear models typically lead to inconsistent estimates of the slope coefficients (Haiso 1986). This may not be a significant problem since the bias resulting from noisy estimates of fixed effects in nonlinear models goes to zero as the number of observations per fixed effect becomes arbitrarily large. Given that our sample has over 80 census subdivisions per fixed effect, we expect that this inconsistency is small. As a robustness check, however, we ran an O L S fixed-effect specification in which all census subdivisions with zero births in a given year were omitted (Appendix 3.D). These regressions are otherwise directly comparable to those in Table 3.1. In linear fixed-effect models noisy estimates of the fixed effects do not bias estimates of the slope coefficients. O f course, these results have a potential sample selection problem since most of the census subdivisions are eliminated. Nevertheless, the qualitative nature of the results is similar to results from the Tobit models, suggesting that the results are robust with respect to various econometric specifications. We also experimented with other measures of the covariates. For failure levels we tested the lagged levels of failed firms in the focal sector at the C S D , and lagged levels of firms that exit after three and five years of operation. The first alternative may be picking up a cyclical effect (i.e., last period of high firm births and corresponding high deaths), and was not preferred for that reason. For the number of failure in other industries at the C D level we experimented with a measure that excludes the focal C S D . For failure rates we tested the one year lag number of failures in the focal sector at the C S D over the population of incumbents at (t-2), f PCSD t-\\ i ~ C S D ' ''' • This measure may capture information about recent risks ' ' incumbentsCSDt_2i associated with a location rather than persistent risks. For regional failure of older firms and persistent risk of failure in other industries we also experimented with measures that include only neighboring C S D . However, our results do not materially change when we use these alternative measures. For localization effects we included several alternative repressors: a measure of the 93 spatial concentration of industrial sectors using C S D ' s market share , G in i coefficients, measures of density such as the number of plants and level of employment per C S D ' s area, and Location Quotients 9 4 . For urbanization economies we experimented with alternative measures such as total employment per square kilometer (e.g., Bartik 1985, Figueiredo, et al. 2002), and 9 3 Theories of agglomeration economies have been based on the idea that an increase in the absolute scale of activity has a positive effect. However, this effect (measured as local number of plants and employment levels) does not make direct predictions regarding the impact of the industry's market share in a particular location relative to other locations. We generate this variable by dividing the aggregated manufacturing shipments of a given sector over all existing establishments within the CSD by the total (national) shipments for the sector. 9 4 Location Quotients are defined as the percentage of local employment in a particular industry divided by the percentage of national employment in that industry (e.g., Glassman and Voelzkow 2001). inverse Herfindahl index of employment share with 2-digit industries at the C D level (e.g., Henderson 1995). In the most robust models, we included industry x year effects (dit) to capture industry and year specific shocks such as the introduction of an industry-specific new technology, E R x year (dER t) or C D x year effects (dCDt) to absorb shock that are shared by all firms in the region such as construction of a rail or opening of an airport. The coefficients retain the same level of statistical significance and approximate magnitude. Limitations of the Study There are several limitations to this study. Our models of the location of new entry may not have controlled adequately for some unobserved heterogeneity within the economic region or supply shocks that may make a location more favorable to entry. For example, industrial real estate is cheaper in one C S D than in others (e.g., Brooklyn compared to Manhattan). Entrepreneurs with fewer resources could be forced to establish businesses in the cheaper C S D for this reason rather than agglomeration, lack of competition, or opportunity reasons. High failure levels in a C S D may result from repeated entry attempts by less competent entrepreneurs (serial failures) who prefer to re-establish new ventures in the same location. Thus failure levels may be a proxy for other phenomena not accounted for in our model such as lack of networks, capabilities, talent, and industry experience of failing entrepreneurs.95 Given that our population consists of all CSDs in Canada, we believe that the magnitude and significance of the results is not driven by some unobserved heterogeneity. More detailed economic environment measures would allow more specific tests of the local process through 9 5 Note, however, Dahlqvist, et al. (2000) found that the start-up of a firm from unemployment reasons does not affect the survival or the performance level of the firm. which failures trigger employment transition, search for opportunities and entrepreneurial entry to a location. The dynamics of manufacturing spin-offs represents another intriguing alternative explanation for the differential in entrepreneurial entry levels across locations. Klepper (2002, 2003) finds that an important determinant of geographic concentration in the U.S . automobile industry involves the role of spin-offs of automobile manufactures. We cannot differentiate between entrepreneurial entry through spin-off and unaffiliated entry. This remains for future research. Finally, our study examined only entry within five manufacturing industries. Immobility of some tangible assets and partial immobility of employment may better characterize the manufacturing sector than the service sector. Consequently, we cannot generalize the effects of failure on entry to all service industries. Conclusions There is general recognition that business creation is an important element of economic growth and regional innovation. In this paper we contribute to the understanding of the role that enterprise failure plays in the industrial regional renewal process. Entrepreneurs must recognize, interpret and take an action to follow an opportunity (Gartner 2001, Shane and Cable 2002). The probability of opportunity recognition depends on the intensity of the search, the attention and alertness to new opportunities (Hayek 1952, Kirzner 1979), the costs of such search, unemployment (Delmar and Davidsson 2000), and entrepreneur's prior experience (Shane 2000, Shane and Venkatataman 2000) 9 6 . Our study suggests that failures of firms in a location may 9 5 Entrepreneurs can be categorized as either opportunity (pull factors), or necessity motivated (push factors) (Reynolds 2003). Opportunity motivated entrepreneurs establish new ventures as a response to opportunities for creating new products and services (Baumol 1990). Necessity motivated entrepreneurs establish new ventures out of the need to find suitable work. From a risk perspective, being unemployed will influence people to take higher risks (Delmar and Davidsson 2000). intensify search for entrepreneurial opportunities and/or generate more opportunities in that location thus, attracting new entry. The study shows that the impact of failure levels attenuates with distance thus, highlighting the role of local knowledge and social networks. Our results are consistent with Schumpeter's hypothesis of creative destruction as we show that exit of older firms stimulate entry and renewal. We also show that this process is anchored in relatively small spaces. We cannot, however, test whether the new deployment of resources by entrants is indeed creative and leave this task for future research. Perceived risk is a factor in a location choice. The study indicates that the reputation of the region as a whole (i.e. risk levels in all sectors in the region) affects the attractiveness of a location within it. However, same sector perceived risk levels in neighboring locations are used to calibrate location choices within the region. Entrepreneurs make their choices within the region, in part, on the basis of relative risks. Local competitive markets imply pressure on prices and elimination of abnormal entrepreneurial profits, thus reducing the attractiveness of a place. Competition is also associated with innovation, and in some sectors may present a positive externality (Porter 1998, 2000). Our study found that local competition in micro clusters deters entrepreneurs from entry into a specific location in all the sectors included in our study. Prior research results showed that agglomeration economies matter to location selection between regions; our results suggest that they matter also to within region choices. We also show that localization economies are more important than urbanization economies. The results of our study have important policy implications. They suggest that regional governments may find it more effective to facilitate entry of new firms than to attempt to attract external entrants through subsidies and tax concessions. Local governments should view the failure of enterprises (especially old ones) as part of a renewal process that should not be slowed down by shoring up failing companies. However, regions that suffer from persistent general failures may develop reputation that detracts investments and thus require the employment of promotion strategies. The study also suggests that diversity in the metropolitan region and specialization within a sub-region create an attractive environment for new enterprises. Given the higher impact of localization economies on attracting new entrants, policies should emphasize the concentration of resources around one sector within the region rather than the dispersion of resources among a variety of sectors, while promoting diversity across regions. References Agarwal, R. and M . Gort 1996. The evolution of markets and entry, exit and survival of firms. The Review of Economics and Statistics, 489-498. Aldrich, H . 1999. Organizations Evolving. Sage Publications: London. Audretsch, D . B . and T. Mahmood 1993. Entry, growth, and survival: The new learning on firm selection and industry evolution. Empirica, 20(1): 25-33. Baldwin, J. and P. Goreski 1991. F i rm entry and exit in Canadian manufacturing sector. Canadian Journal of Economics, 24:300-323. Baum, J . A . C . and H . A . Haveman 1997. Love thy neighbor? Differentiation and agglomeration in the Manhattan hotel industry, 1898-1990. Administrative Science Quarterly, 42(June): 304-338. Baum, J . A . C . and S.J. Mezias 1992. Localized competition and organizational failure in the Manhattan hotel industry, 1898-1990. Administrative Science Quarterly, 36:187-218. Baum, J . A . C . and C. Oliver 1991. Institutional linkages and organizational mortality. Administrative Science Quarterly, 36:187-218. Baum, J . A . C . and J .V. Singh 1994. Organizational niches and the dynamics of organizational founding. Organization Science, 5(4): 483-501. Baumol, W . J . 1990. Entrepreneurship: Productive, unproductive, and destructive. Journal of Political Economy, 98(5): 893-921. Boeker, W . 1989. Strategic change: The effects of founding and history. Academy of Management Journal, 32:485-488. Cable, J. and J. Schwalbach 1991. International comparison of entry and exit, In P. A . Geroski and J. Schwalbach (Ed.), Entry and Market Contestability: An International Comparison, Blackwell Press: Cambridge, 257-281. Carroll, G .R. and M . T . Hannan 2000. The demography of corporations and industries. Princeton University Press: Princeton. Caves, R . E . 1998. Industrial organization and new findings on the turnover and mobility of firms. Journal of Economic Literature, XXXVI(December) : 1947-1982. Chung, W . and A . Kalnins 2001. Agglomeration effects and performance: A test of the Texas lodging industry. Strategic Management Journal, 22:969-988. Chandler, G . N . , Hanks, S.H. 1994. Market attractiveness, resource-based capabilities, venture strategies, and venture performance, Journal of Business Venturing 9(4): 331-349. Ciccone, A . and R . E . Ha l l 1996. Productivity and the density of economic activity. American Economic Review, 86(1): 54-70. Cooper, A . C . 1984. Contrasts in the role of incubator organizations in the founding of growth-oriented companies, Babson College, 159-174. Dahlqvist, J., P. Davidsson and J. Wiklund 2000. Initial conditions as predictors of new venture performance: A replication and extension of the Cooper et al. study. Enterprise & Innovation Management Studies, 1(1): 1-17. Dixi t , A . K . and R.S. Pindyck 1994. Investment Under Uncertainty. Princeton University Press: Princeton. Deeds, D . L . , DeCarolis, D . , Coombs, J .E. 1999. Dynamic capabilities and new product development in high technology ventures: A n empirical analysis of new biotechnology firms. Journal of Business Venturing 15: 211-229. Delmar, F. and P. Davidsson 2000. Where do they come from? Prevalence and characteristics of nascent entrepreneurs. Entrepreneurship & Regional Development, 12:1-23. Duchesneau, D. , Gartner, W . 1990. A profile of new venture success and failure in an emerging industry. Journal of Business Venturing 5: 297-312. Dumais, G . , G . Ell ison and E . L . Glaeser 2002. Geographic concentration as a dynamic process. The Review of Economics and Statistics, 84(2): 193-204. Dunne, T., M . J . Roberts and L . Samuelson 1988. Patterns of firm entry and exit in U .S . manufacturing industries. Rand Journal of Economics, 19(4): 495-515. Duranton, G . and D . Puga 2001. Nursery Cities: Urban diversity, process innovation, and the life-cycle of products. American Economic Review, 91(5): 1454-1477. Eaton, B . and R. Lipsey 1980. Entry barriers are exit barriers: The durability of capital as a barrier to entry. Bell Journal of Economics, 11:721-729. Ell ison, G . and E . L . Glaeser 1997. Geographic concentration in U .S . manufacturing industries: A dartboard approach. Journal of Political Economy, 105(5): 889-927. Fotopoulos, G . and N . Spence 1999. Spatial variations in net entry rates of establishments in Greek manufacturing industries: A n application of the shift-share A N O V A model. Environment and Planning A, 31(10): 1731-55. Gartner, W . B . 2001. Is there an elephant in entrepreneurship? B l ind assumptions in theory development. Entrepreneurship Theory and Practice, 25(4): 27-39. Geroski, P. A . 1995. What do we know about entry? International Journal of Industrial Organization, 13): 421-440. Glaeser, E . L . , H . Kal la l , J. Scheinkman and A . Schleifer 1992. Growth in cities. Journal of Political Economy, 100(6): 1126-1152. Gort, M . and S. Klepper 1982. Time paths in the diffusion of product innovations. Economic Journal, 92(367): 630-653. Haiso, C. 1986. Analysis of Panel Data. Cambridge University Press: New York. Hayek, F . A . 1952. The Sensory Order. The University of Chicago Press: Chicago. Head, K . , J. Ries and D . Swenson 1995. Agglomeration benefits and location choice: Evidence from Japanese manufacturing investments in the United States. Journal of International Economics, 38:223-247. Hedstorm, P. 1994. Contagious collectivities: On the spatial diffusion of Swedish trade unions, 189-1940. American Journal of Sociology, 99(5): 1157-1179. Henderson, J .V. 1994. Where Does an Industry Locate? Journal of Urban Economics, 35(1): 83-104. Henderson, J .V. 2000. Marshall's scale economies. NBER Working Paper #7358. Holmes, T.J . and J.A.Jr. Schmitz 1990. A theory of entrepreneurship and its application to the study of business transfers. Journal of Political Economy, 38(3): 261-287. Hotelling, H . 1929. Stability in Competition. Economic Journal, 39(March): 41-57. Jaffee, J. 2003. Law firm office location and firm survival in Silicon Valley, 1969 to 1998, Working Paper. Kaplan, S., Stromberg, P. 2000. How do venture capitalists choose investments. Working Paper, University of Chicago, Kennedy, P. 1992. A Guide to Econometric Methods (3 r d edition). Cambridge, M A : M I T Press. K i m , S. 1995. Expansion of markets and the geographic distribution of economic activities: The trends in U . S . regional manufacturing structure, 1860-1987. Quarterly Journal of Economics, llO(November): 881-908. Kirzner, I . M . 1979. Perception, opportunity, and profit: Studies in the theory of entrepreneurship. Review of Austrian Economics, 11:5-17. Klepper, S. 1996. Entry, exit, growth, and innovation over the product life cycle. American Economic Review, 86(3): 562-83. Klepper, S. 2003. The geography of organizational knowledge. Carnegie Mellon University working paper. Klepper, S. and J .H. Mi l l e r 1995. Entry, exit, and shakeouts in the United States in new manufactured products. International Journal of Industrial Organization, 13(4): 567-91. Krugman, P. 1991. Increasing returns and economic geography. Journal of Political Economy, 99(3): 483-99. L o m i , A . 1995. The population ecology of organizational founding: Location dependence and unobserved heterogeneity. Administrative Science Quarterly, 40(March): 111-144. Louri , H . and V . Anagnostaki 1995. Entry in Greek manufacturing industry: Athens vs. the rest of Greece. Urban Studies, 32(7): 1127-33. Maddala, G.S. 2001. Introduction to Econometrics. John Wiley & Sons: Marshall, A . 1920. Principles of Economics. MacMi l l an : London. Mata, J. 1993. Fi rm Entry and F i rm Growth. Review of Industrial Organization, 8(5): 567-78. Mata, J. and P. Portugal 1994. Life Duration of New Firms. Journal of Industrial Economics, 42(3): 227-45. McCann, J .E. 1991. Patterns of growth, competitive technology, and financial strategies in young ventures. Journal of Business Venturing 6(3): 189-208. McGrath, R . G . 1999. Falling forward: Real options reasoning and entrepreneurial failure. Academy of Management Review, 24(1): 13-30. Porter, M . E . 1998. Clusters and the new economics of competition. Harvard Business Review, Nov-Dec:77-90. Porter, M . E . 2000. Location, competition, and economic development: Local clusters in a global economy. Economic Development Quarterly, 14(1): 15-34. Prescott, E . C . and M . Visscher 1977. Sequential location among firms with foresight. Bell Journal of Economics, 8(Autumn): 378-393. Reynolds, P .D. 1997. Who starts new firms? Preliminary explorations of firms-in-gestation. Small Business Economics, 9:449-462. Rosenthal, S. and W . C . Strange 2001. The determinants of agglomeration. Journal of Urban Economics, 20(2): 191-229. Rosenthal, S. and W . C . Strange 2003. Geography, industrial organization, and agglomeration. Review of Economics and Statistics, 85(2): 377-393. Sandberg, W.R. , Hofer, C . W . 1987. Improving new venture performance: The role of strategy, industry structure, and the entrepreneur. Journal of Business Venturing 2: 5-28. Saxenian, A . 1994. Regional Advantage: Culture and Competition in Silicon Valley and Route 128. Harvard University: Cambridge, M A . Shane, S. 2000. Prior knowledge and the discovery of entrepreneurial opportunities. Organization Science, 11(4): 448-469. Shane, S. and D . Cable 2002. Network ties, reputation, and the financing of new ventures. Management Science, 48(3): 364-381. Shane, S. and T. Stuart 2002. Organizational endowments and the performance of university start-ups. Management Science, 48(1): 154-170. Shane, S. and S. Venkatataman 2000. The promise of entrepreneurship as a field of research. Academy of Management Review, 25(1): 217-226. Shaver, J . M . and F. Flyer 2000. Agglomeration economies, firm heterogeneity, and foreign direct investment in the United States. Strategic Management Journal, 21(12): 1175-1193. | Soh, P . H . 2003. The role of networking alliances in information acquisition and its implications for new product performance. Journal of Business Venturing 18:121-144. Sorenson, O. and P . G . Audia 2000. The social structure of entrepreneurial activity: Geographic concentration of footwear production in the United States, 1940-1989. American Journal of Sociology, 106(2): 424-462. Sorenson, O. and T .E . Stuart 2001. Syndication networks and the spatial distribution of venture capital investments. American Journal of Sociology, 106(6): 1546-1588. Storey, D.J . 1985. The problems facing new firms. Journal of Management Studies, 22:327-345. Stuart, T .E . and O. Sorenson 2002. Liquidity events and the geographic distribution of entrepreneurial activity. Working Paper. Stuart, T. E . and O. Sorenson 2003. The geography of opportunity: Spatial heterogeneity in founding rates and the performance of biotechnology firms. Research Policy, 32: 229-253 Sutton, J. 1997. Gibrat's legacy. Journal of Economic Literature, 35:40-59. Szulanski, G . 2000. The process of knowledge Transfer: A diachronic analysis of stickiness. Organizational Behavior and Human Processes, 82(1): 9-27. Tveteras, R., Eide, G . E . 2000. Survival of new plants in different industry environments in Norwegian manufacturing: A semi-proportional Cox model approach. Small Business Economics, 14:65-82. Appendix 3.A: Descriptive Statistics and Definitions Mean Std Dev CSD number of plants (t-1) —4.81 17.36 CSD employment (t-1) 184.42 551.73 CD number of plants (t-1) [other ind.] 287.24 765.75 CD employment (t-1) [other ind.] 8485.32 25741.53 CSD area 376.13 1543.15 CSD establishments per worker (t-1) 0.185 0.423 CD establishments per worker (t-1) 0.09 0.205 [other CSDs within the same CD] CSD failure (t-1) 0.54 1.52 CSD failure (t-2) 0.53 1.52 CD failure (t-1) 2.13 3.98 [other CSDs within the same CD] CD failure (t-2) 2.15 4.11 [other CSDs within the same CD] CD failure (t-1) [other ind] 8.46 14.98 CSD failure rate (t-1) 0.12 0.04 CD failure rate (t-1) 0.19 0.11 [other CSDs within the same CD] CD failure rate (t-1) [other ind] 0.38 0.17 Definition Lagged number of plants in the focal sector operating in the CSD Lagged level of employment in the focal sector operating in the CSD Lagged number of plants in other manufacturing sectorsr operating in the CD Lagged level of employment in other manufacturing sectorsr operating in the CD Area measured in square kilometers Lagged number of plants in the focal sector devided by level of employment in the CSD Lagged number of plants in the focal sector devided by level of employment in the CD (excluding focal CSD) Lagged levels of failed firms within the focal sector that existed at least four years at the CSD level Two years lagged levels Lagged levels of failed firms within the focal sector that existed at least four years in neighboring CSDs within the same CD Two years lagged levels Lagged levels of failed firms within other sectors that existed at least four years in the CD Three years aggregated number of lagged failures in the CSD from the focal sector divided by the number of CSD's incumbents at t-4 Three years aggregated number of lagged failures in neighboring CSDs (within the from the focal sector divided by the number of CSD's incumbents at t-4 Three years aggregated number of lagged failures in the CD from other sectors divided by the number of CD's incumbents at t-4 Underlying Concept Localization economies Urbanization economies Rents Local competition Competition in neighboring locations Local failures of older firms Failures of older firms in Neighboring locations Regional failures of older firms in other sectors Persistent local risk of failure Persistent risk of failure in neighboring locations Persistent regional risk of failure in other sectors A3>S Appendix 3.B - Labor Tracking Procedure Obtaining accurate measurements of births and deaths is not trivial due to the need to distinguish between 'real' births and deaths from 'false' ones. In many files (e.g., the L R D at the U.S . Bureau of the Census, or Dun and Bradstreet records), entry is measured as the appearance of a new entity. Thus, when a merger or a change in control occurs, an ongoing entity falsely appears to die and then be born. Real births and deaths reflect actual entry and exit events (the creation of new forms and the failure of existing ones); false births and deaths may simply reflect organizational restructuring within a firm, merger or acquisition, or a change in its reporting practices. A 'labor tracking' method was used in our database. This procedure relies on tracking the workforce of firms over time. Employees are followed from one payroll account to another and the percentage of workers present in a firm in one year that can be found in another is calculated. Labor is tracked from one firm to another and this path is used to establish whether a business was potentially linked to another. If a firm is falsely identified as dying, a substantial number of its workers should be found in another unit the following year. This is referred to as the pass-through rate. The firm with the largest path through rate is chosen as the target firm with which the originating business unit had the closest affiliation - the dominant link. When no-longer-identified firm that had more than 10 employees had pass-through-rate of 75% or more, these cases are reclassified as continuing firms. Similarly, where more than 75% of a birth came from another firm, it is classified as a false birth. A more restrictive rule is employed for smaller firms since there is less difference in pass-through rates between small continuing and no-longer identified firms. For example, i f firm A merged with firm B in year t, then a new firm, C, is created and given a synthetic history aggregated from the histories of firms A and B . The individual histories of A and B disappear from the data and firm C represents their joint operations. Appendix 3.C - Industries Capital Goods or Machinery Industries Description SIC-ECode ... . •, Number of Births. Food Industries Apparel (Fashion) Fabricated Metal Products Industries (except Machinery and Transportation Equipment industries) 101,102,103,104,105,106,108,109 243,244,245,249 301,302,303,304,305,306,307,308,309 11,046 3,647 11,053 ICT / SCI industries Description . . . . S1C-E Code Number of Births Machinery Industries (expect Electrical Machinery) Electrical and Electronic Products Industries 311,312,319 334, 335,336 5,943 2,760 Appendix 3.D: OLS Fixed Effects Estimates of New Entry Dependent variable: aggregated number births in a census subdivision j, at time t, belonging to sector i Specification Regressors CSD number of plants (t-1) 0.04902*** (9.365) 0.00128** (2.027) CSD employment (t-1) CD number of plants (t-1) [other ind.] CD employment (t-1) [other ind.] CSD area CSD establishments per worker (t-1) CD establishments per worker (t-1) [other CSDs within the same CD] CSD failure (t-1) CSD failure (t-2) CD failure (t-1) [other CSDs within the same CD] CD failure (t-2) [other CSDs within the same CD] CD failure (t-1) [other ind.] CSD failure rate (t-1) CD failure rate (t-1) [other CSDs within the same CD] CD failure rate (t-1) [other ind.] 0.00045** (2.199) 0.04547*** 0.04501*** 0.04537*** 0.04324*** (3.688) (4.002) (4.175) (3.875) 0.00125** 0.00126** 0.00124** 0.00120** (2.114) (2.034) (2.301) (2.196) 0.00002** (2.366) 0.00007 * 0.00005 * 0.00004* 0.00006* 0.00005 * 0.00005* (1.671) (2.042) (1.756) (1.812) (1.756) (1.776) -0.35512** -0.30627* -0.30004* -0.26826* (2.318) (1.746) (1.808) (1.954) -0.21957* -0.18274* (1.884) (2.022) 0.05248*** (12.986) 0.03975** (2.210) 0.03372** (2.224) 0.04418* (1.994) -0.00814* (2.022) -1.225 -0.18064 (1.937) (0.937) 0.05881*** (11.262) 0.03751** (2.351) 0.03056* (1.804) 0.05816 (1.268) -0.00673* (1.974) -0.08652*** -0.07407*** (3.894) (3.202) 0.12861** 0.10308** (2.142) (2.304) -0.00613*** -0.00547** • (3.805) (2.155) Economic region effects Yes Yes Yes Yes Yes Yes Time effects Yes Yes Yes Yes Yes Yes 2-digit industry effects Yes Yes Yes Yes Yes Yes Number of births 34,449 34,449 34,449 34,449 34,449 34,449 adjusted R-Sq 0.082 0.065 0.124 0.238 0.241 0.243 Absolute value of t-statistics in parenthses * significant at the 10% level; ** significant at 5% level; ***siginificant at 1% level "@en ; edm:hasType "Thesis/Dissertation"@en ; vivo:dateIssued "2005-05"@en ; edm:isShownAt "10.14288/1.0092903"@en ; dcterms:language "eng"@en ; ns0:degreeDiscipline "Business Administration"@en ; edm:provider "Vancouver : University of British Columbia Library"@en ; dcterms:publisher "University of British Columbia"@en ; dcterms:rights "For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use."@en ; ns0:scholarLevel "Graduate"@en ; dcterms:title "Entry strategies and performane [sic] of new ventures in clusters and isolation"@en ; dcterms:type "Text"@en ; ns0:identifierURI "http://hdl.handle.net/2429/18286"@en .