TY - JOUR
AU1 - Giacomelli,, Silvia
AU2 - Menon,, Carlo
AB - Abstract Poor contract enforcement can importantly affect firms’ incentives to grow. We investigate the causal effect of the weakness of contract enforcement on average firm size across Italian municipalities, exploiting spatial discontinuities in court jurisdictions for identification. Italy provides an ideal environment for this exercise, as it displays wide variation in judicial efficiency across courts, while the allocation of municipalities to jurisdictions is a historical legacy and does not overlap with other political or economic discontinuities. Our estimates indicate that reducing the length of judicial proceedings (i.e. improving contract enforceability) by 10% at court level leads to a 2% increase in average size of local firms. The outcome on turnover growth is of the same magnitude, suggesting that the effect operates at the intensive margin. 1. Introduction A well-functioning judicial system that ensures the enforcement of contracts is essential to the working of a market economy. It influences both real and financial outcomes, affecting firms’ capability and incentives to grow.1 This is relevant for economic growth, as enabling productive firms to scale up is one of the most important drivers of differences in aggregate cross-country productivity (Disney et al., 2003; Bartelsman et al., 2013) and there is robust evidence of strong correlation of firm size and productivity, both at firm and at aggregate level (Hsieh and Klenow, 2014; García-Santana and Ramos, 2015). Institutional characteristics of countries—among which the quality of the legal and judicial system has a prominent role—are relevant factors in explaining international differences in average firm size and the efficiency of resource allocation toward the best performing firms (Davis and Henrekson, 1997; Braguinsky et al., 2011; Andrews and Cingano, 2014; Garicano et al., 2016). From a theoretical point of view, the functioning of the courts, by shaping the contractual environment in which firms operate, may affect firms’ decisions over investment, employment, organization and financial structure, and through them their incentives to grow. A positive association between the quality of the judicial system and firm size is found in various studies that exploit either cross-country variation (Kumar et al., 1999; Beck et al., 2006) or within-country variation (Laeven and Woodruff, 2007; Fabbri, 2010).2 However, the existing literature has failed to establish a causal link due to the difficulties to isolate the effect of contract enforceability from other institutions, both formal and informal (unwritten norms and rules that affect the behavior of individuals and organizations). In this article, we analyze the causal effect of contract enforcement on firm size and firm growth. Our main original contribution consists in developing an identification strategy based on a spatial discontinuity design that exploits within-country variation in courts efficiency in Italy.3 More specifically, we compare the average size and growth of manufacturing firms located in contiguous municipalities on either side of several hundreds of court jurisdiction boundaries. At the boundaries, contract enforceability displays a discrete variation, reflecting differences in court efficiency, while all other relevant factors vary smoothly. This allows us to neatly identify the effects of contract enforcement on firm size. Using census data on the number of firms and employment at the municipal level, we find that less contract enforceability, measured by the length of judicial proceedings, leads to a smaller average firm size across Italian municipalities. The economic impact of our results is sizable: reducing the length of judicial proceedings by 10% would lead to a 2% increase in average firm size. Using sampled firm-level balance sheet data that allow us to compute a proxy for firm growth based on turnover, we also find that contract enforceability has a negative effect on growth; the magnitude of the effect is in the same ballpark of that on firm size, suggesting that the effect comes from the intensive margin. We also assess the relative importance of the different channels through which contract enforceability affects firm size employing a difference-in-difference approach. We consider three different channels: (i) relationship-specific investment, (ii) credit, and (iii) labor. The first channel builds on the well-established insights from contract theory. When investment is relationship-specific and contracts cannot be enforced ex post, parties will under invest to avoid holdup situations, thus leading to smaller firms. However, firms may be induced to adopt organizational structures that mitigate holdup problems, for instance, vertically integrating the production process (Williamson, 1979). This would lead to larger firm size.4 The second and the third channels are related to the enforcement of specific contracts, namely credit and labor contracts. An effective enforcement of credit contracts, providing stronger creditor protection, increases the availability of credit and improves the contractual terms for prospective borrowers (Qian and Strahan, 2007; Djankov et al., 2008; Bae and Goyal, 2009), thus loosening financial constraints to firm growth.5 Yet, the overall effect of a more efficient enforcement of credit contracts on average firm size is ambiguous. Prospective entrepreneurs would also have better access to external finance and given that new firms are usually smaller, this could reduce the average size. As for labor contracts, the literature has mainly focused on litigation of dismissal cases. Court inefficiency may add significantly to the costs of worker dismissal for firms, reducing ex ante their willingness to employ workers.6 To summarize, the impact of the first two channels is ambiguous, while the third channel implies that effective contract enforcement increases average firm size. Our results suggest that the negative effects of poor contract enforcement on firm size mainly come through the disincentives of relationship-specific investments. The reason why we chose Italy for our analysis is 2-fold. First, Italy provides an ideal environment for exploiting spatial discontinuities for identification: it is a centralized nation, with a high degree of legislative uniformity across the national territory (for instance, regarding contracts, labor, corporations, judicial procedure), yet it displays wide variation in judicial efficiency across courts (World Bank, 2013). Since disputes are assigned to local courts on a territorial basis, the variation in court efficiency leads to a variation in contract enforceability for local firms. Moreover, court jurisdiction boundaries do not systematically match other administrative boundaries (beyond the municipality ones) and their definition dates back to the 19th century, therefore, it can be argued that they do not introduce discontinuities other than in contract enforcement. Second, Italy is characterized both by very small average firm size (equal to eight employees in our sample) and by poor contract enforcement, as compared world’s largest economies. Compared with other European countries (EU-15), the average size of Italian firms is 40% smaller; significant disparities persist even if the differences in sectoral specialization are accounted for. Regarding contract enforcement, Italy ranks 147th out of 189 countries in the World Bank’s enforcing contracts indicator. This is largely due to the extreme length of judicial proceedings: in Italy, it takes on average 1120 days to resolve a commercial dispute through courts; it is more than twice the number of days needed on average in OECD countries (World Bank, 2015). The article is organized as follows: Section 2 illustrates the empirical methodology, Section 3 describes the data sources and the variables, Section 4 presents the results and Section 5 concludes. 2. Identification strategy 2.1. Spatial discontinuity design Our identification strategy is based on a spatial discontinuity design, similar to that employed by Black (1999), Holmes (1998) and Duranton et al. (2011) among others.7 The methodology consists in restricting the sample to observations that are located near a spatial discontinuity affecting only the variable of interest (in our case contract enforceability), and in mean differentiating all the variables within the group of observations that share the same discontinuity. As compared to the standard regression discontinuity design (RDD) with a dichotomous treatment variable and its geographic applications (Dell, 2010; Basten and Betz, 2013; Becker et al., 2016), the methodology adopted here presents some parallelisms, but there are also important differences. Most notably, in the spatial discontinuity approach there are generally several discontinuities, which are potentially distributed along the full distribution of the ‘treatment’ variable; in a standard RDD setting, there is often just one discontinuity at a specific value of the treatment variable. This implies that, in our setting, there could be treated (i.e. a municipality at the side of the jurisdiction boundary where contract enforcement is more effective) and control (i.e. a municipality at the side where contract enforcement is less effective) observations that share the same value of the treatment variable. This also implies that a correct specification of the regression functional form is less critical than it is in the case of RDD, as the ‘treatment effect’ is estimated on average over the full distribution of the dependent and treatment variables. It also implies that the unconfoundness tests that are generally discussed in RDD applications cannot be exactly replicated in this context, as they should compare the full distribution of the relevant variables, rather than the distribution at a specific threshold value (see the related discussion in Section 2.2). We apply this methodology to a sample of Italian municipalities located on either side of court jurisdiction boundaries (each municipality belongs entirely to a single jurisdiction area), the outcome variable being firm size. Figures 1 and 2 show municipality boundaries (gray lines) and court jurisdiction boundaries (bold black lines) in Northern Italy, while colored areas indicate municipalities that share common jurisdiction boundaries. In a nutshell, our approach consists in restricting the sample to colored municipalities and mean differentiating all the variables among municipalities of the same color (by including a group dummy). The fact that our identification only exploits mean differences among municipalities that are very close to each other implies that our estimates are not biased by omitted local factors which vary smoothly over space. Figure 1 Open in new tabDownload slide Court jurisdiction and municipality boundaries, Northern Italy. Notes: The darker bold lines in the map correspond to jurisdiction boundaries, while the thinner gray lines correspond to municipality boundaries. The map shows the Italian northern regions. Different gradations correspond to different boundary groups. Source: Based on ISTAT and Italian Ministry of Justice data. Figure 1 Open in new tabDownload slide Court jurisdiction and municipality boundaries, Northern Italy. Notes: The darker bold lines in the map correspond to jurisdiction boundaries, while the thinner gray lines correspond to municipality boundaries. The map shows the Italian northern regions. Different gradations correspond to different boundary groups. Source: Based on ISTAT and Italian Ministry of Justice data. Figure 2 Open in new tabDownload slide Court jurisdiction and municipality boundaries, Northern Italy, detail. Notes: The darker bold lines in the map correspond to jurisdiction boundaries, while the thinner gray lines correspond to municipality boundaries. The map is centred on the Veneto region. Different gradations correspond to different boundary groups. Source: Based on ISTAT and Italian Ministry of Justice data. Figure 2 Open in new tabDownload slide Court jurisdiction and municipality boundaries, Northern Italy, detail. Notes: The darker bold lines in the map correspond to jurisdiction boundaries, while the thinner gray lines correspond to municipality boundaries. The map is centred on the Veneto region. Different gradations correspond to different boundary groups. Source: Based on ISTAT and Italian Ministry of Justice data. This identification strategy requires that the following two assumptions are satisfied. First, spatial discontinuities affecting contract enforcement should not introduce any sharp discontinuity in other variables correlated with the output of interest (in short, they should be exogenous). Second, the spatial border should introduce a sharp discontinuity in contract enforcement. As regards the first assumption, at the time of our analysis (2001–2009) the territorial organization of the Italian judicial system was based on 165 court jurisdiction areas; within these areas, the courts of first instance administered both civil and criminal justice. This territorial distribution of courts was mainly determined by historical factors and largely resembled the one shaped in 1865, immediately after the unification of Italy. This, in turn, was based on the judicial systems of the previous states. Since then and until 2013, no existing court has ever been removed, although some new courts have been established in the 1960s and in the 1990s.8 Court jurisdiction boundaries do not systematically match other administrative boundaries, although in some cases they coincide with regional and provincial boundaries.9 In these circumstances, spatial discontinuities unrelated to contract enforcement might be introduced. This would be a cause of concern only if these discontinuities were correlated with court efficiency. However, this is unlikely to occur since the judicial system is fully autonomous from local administrative bodies and regions and provinces play no role in the functioning of local courts. Nonetheless, since regions have important regulatory powers over economic activity that may have an influence on firm size, we control for regional differences on opposite sides of court jurisdiction boundaries by adding regional fixed effects to the regressions. We are less concerned about the matching of court boundaries with provincial boundaries as provinces have very limited autonomous power over matters related to business activities.10 Assumption 1 can be empirically tested: in Section 4.1, we show that socioeconomic variables that should not be directly affected by local contract enforceability are uncorrelated with our measure of the latter; furthermore, the distribution of these variables is similar at both sides of the jurisdiction border. The second assumption is visually supported by Figure 3, which maps the average estimated length of civil proceedings, our proxy for contract enforceability, by court jurisdiction in the period 2002–2007 (Section 4). The figure shows that, although there is a clear geographical pattern (southern courts on average are twice as slow as northern ones), there is a fair amount of variation also within regions and between neighbouring courts. Figure 3 Open in new tabDownload slide Court jurisdiction and average length of judicial proceedings, 2002–2007. Notes: The polygons in the map correspond to the Italian court jurisdictions. The average length of judicial proceedings is estimated by an index based on caseflow data, as explained in Section 4.5. Source: Based on ISTAT and Italian Ministry of Justice data. Figure 3 Open in new tabDownload slide Court jurisdiction and average length of judicial proceedings, 2002–2007. Notes: The polygons in the map correspond to the Italian court jurisdictions. The average length of judicial proceedings is estimated by an index based on caseflow data, as explained in Section 4.5. Source: Based on ISTAT and Italian Ministry of Justice data. There is no clear evidence on the determinants of such a large variation in the length of civil proceedings; several factors seem to play a role (Bianco et al., 2007). First, there is a large variation in the demand for court services across the country with much higher litigation rates (number of disputes brought to courts over population) in the southern regions. Second, the decisions on the allocation of resources among courts are highly centralized with limited degree of flexibility to adjust to changes in local demand. Third, there is also significant variation on the supply side. The impossibility for local courts to select and appoint judges, together with the absence of effective court management systems and of incentive mechanisms for judges to expedite proceedings, results in some randomness in the distribution of judges’ ability and effort among courts. To the extent that demand side factors should not change discontinuously at the jurisdiction boundary and are thus absorbed by the boundary dummies, our estimation strategy mainly exploits variation on the supply side. Since civil proceedings are assigned to courts on a territorial basis, the variation in court efficiency leads to variation in the effectiveness of contract enforcement for local firms. As a general rule, the Italian code of civil procedure provides that cases should be assigned to a court according to the residence of the defendant, unless parties agree otherwise in a contract.11 This implies that if a firm is sued by other firms or customers and there is no previous agreement as to a different forum, the residence of the firm determines the jurisdiction. Furthermore, in certain matters, some of which are very important for business activity such as employment proceedings, the court’s jurisdiction is always determined by the residence of the firm, irrespective of who initiates legal action; similarly, the residence of the firm determines which court is in charge for the final enforcement of a sentence against the same firm. Nevertheless, because not all the proceedings in which a given firm is party are dealt with in the local court, it is worth stressing that our analysis assesses the effect of contract enforceability in the local court jurisdiction. 2.2. Econometric specification A simple formalization may help understand the econometric properties of the methodology. We are interested in assessing the effect of contract enforceability on average firm size at the municipality level. Consider the following model: yi,p=Ek,pβ+Xiγ+f(p)δ+εi (1) that is, the average firm size y in municipality i in the ‘place’ p (i.e. an abstract unique point in the space) is a function of contract enforceability E in the jurisdiction area of the court k in the same place p and of a vector X of observable characteristics of municipality i. The function f(p) represents unobserved factors influencing the outcome variables that vary across space and are potentially correlated with both E and X. In such a setting, estimates of β may be biased due to omitted variables. However, we can take a reasonably small variation of p next to a court jurisdiction border, from p1 to p2, which leads to a change in contract enforceability, from Ek to Eq. If the following condition holds: Corr((f(p1)−f(p2)),(Ek−Eq))=0 (2) that is, if the difference in the local unobservables of the two contiguous places is uncorrelated with the difference in contract enforceability, the estimate of β is unbiased. Another implication of Assumption 1 is that the distribution of the matrix of municipality-level covariates Z is independent on the variable of interest E: E[Z|E]=E[Z] (3) which means that, on average, municipalities on one side of the border are similar to municipalities on the other side. This also implies that adding additional controls to our specifications may increase the efficiency of the estimates of the court effect, but does not affect their consistency (and the value of the point estimates). Operationally, the main specification is based on the following OLS estimation, restricted to the sample of municipalities contiguous to a court boundary b: ymjb=Ejβ+Xmγ+∑b=1Bδb+εmjb (4) where ymjb is a measure of average firm size in municipality m,in court jurisdiction j, and belonging to the boundary group b. Ej is a measure of contract enforcement in court jurisdiction j. Xm is a set of municipality controls. δb is a set of boundary group dummies (each dummy is equal to one for all municipalities on both sides of a common court jurisdiction boundary b). We identify 571 different boundary groups (the colored areas in Figure 1), of which 332 contain municipalities on either side of the jurisdiction border and can, therefore, be used for identification. In order to minimize arbitrariness and to allow replicability, their composition is determined by a completely automated procedure using a Geographic Information System. More specifically, for each municipality we calculate the total number of jurisdiction polygons at zero distance. Non-border municipalities have only one jurisdiction at zero distance (the one they belong to). Boundary municipalities are those with two (or more) jurisdictions at zero distance (the one the municipality belongs to, and the contiguous one). If the contiguous jurisdictions are more than one, the one with the shortest distance between the municipality and jurisdiction centroids is taken. Each boundary group is composed of all municipalities sharing the same pair of zero-distance jurisdictions. As it can be seen in Figures 1 and 2, the procedure also produces 239 single-municipality border groups: these municipalities are located at the junction of more than two jurisdiction borders; they are contiguous to one or more borders, but no other municipality gets assigned to the same border group. These singleton border groups are dropped from the sample of 332 border groups used in the main analysis, as at least two municipalities for each border group are required for identification because of the mean differencing approach. On average, the 315 border groups that contain municipalities on either side of the jurisdiction border comprise 13.6 municipalities each, with a minimum of 2 and a maximum of 64. However, given that the analysis is limited to municipalities with more than 5000 inhabitants due to data availability, the average number of municipalities in the sample per border group is 3.2, with a minimum of 2 and a maximum of 24. All the results we present are robust to the exclusion of boundary groups with more than 10 municipalities. It is also worth stressing that, as compared to RDD applications (standard or geographic), using a distance-weighted specification (possibly based on polynomial forms) based on the full sample is not of easy implementation in this case, as municipalities are spatial polygons and there is not an univocal method to calculate their distance from a line (the border). Furthermore, in our setting around half of the municipalities in the full sample are located along the jurisdiction borders and are used for identification. Therefore, contrary to RDD applications, the sample reduction due to restricting only to observations contiguous to the discontinuities is not a critical issue here. 2.2.1. Other identification challenges There are a few additional challenges to our identification strategy. The first relates to the sorting of firms. If firms choose their location after observing the efficiency of the local court, and if larger firms expect to benefit more from more effective contract enforcement, then part of the effect we find is not due to a growth-enhancing effect of contract enforcement, but rather to an attraction effect. However, if this kind of sorting were driving our results, we would not find any significant effect on firm growth, but only on their average size. We anticipate that this is not the case, as we also find that contract enforceability positively affects firm growth. This does not rule out the possibility that firms with better growth prospects may decide ex ante to locate within jurisdictions where contract enforcement is more effective. This channel, however, is part of the substantial effect of contract enforceability on firm size that we discussed in Section 2, and which we aim to capture with our reduced-form equation. Nevertheless, the fact that the best performing firms may ‘vote with their feet’ and sort into the most efficient jurisdictions may have important policy implications, as the aggregate effects of nation-wide improvements of the judicial system could not be directly extrapolated from our local estimates. In the robustness section, we address a related concern: the possibility that boundary municipalities on the side where contract enforcement is more effective host a higher number of large firms than non-boundary municipalities within the same court jurisdiction, due to firm sorting across contiguous jurisdictions within the same boundary group. This would lead to an overestimate of the real effect. However, our test suggests that this is not the case. 3. Data sources and variable definition We assemble a dataset with data on contract enforcement at the jurisdiction level, and firm size in the manufacturing sector (employment and accounting-based measures) and other control variables at the municipal level. 3.1. Contract enforcement We use as a proxy for contract enforcement at court level a measure of the average length of first instance civil proceedings in each court. The idea behind the use of this proxy is that the longer it takes to resolve a dispute over a contract, the less effective is the enforcement of that contract. To be sure, the timely resolution of disputes is just one of the dimensions on which the performance of a judicial system, hence its ability to ensure contract enforcement, can be measured (other dimensions include fairness and predictability of judicial decisions). The choice of this measure is mainly motivated by data availability; moreover, it can be argued that the speediness of dispute resolution is a necessary condition for good performance in other dimensions.12 More in detail, we use caseflow data provided by the Italian Ministry of Justice to construct an index (Length civil) that provides an estimate of the average lifetime of court proceedings (in days), calculated as follows: Dt=Pt+Pt+1Et+Ft*365 (5) where P corresponds to pending cases at the beginning of the year t, F to the new cases filed during the year and E to the cases that ended with a judicial decision or were withdrawn by the parties during the year.13 The index is calculated with reference to ordinary litigious civil cases that include disputes on contracts, but also on other subjects like property and tort.14 In the robustness section, we will also consider the length of judicial proceedings related to labor disputes (Length labor) and criminal cases (Length criminal). Our data cover the period 2002–2007 and we take the average value across the 6 years. 3.2. Firm size We measure firm size using data on employment and number of units. Our source of data is the ASIA database produced by the Italian National Institute of Statistics (Istat) that contains information, at the municipal level, by aggregate industries (manufacturing, services, construction), on the number of enterprises, the number of plants, the number of employees and the distribution of enterprises and plants by size bin. The main advantage of this source is that it covers the universe of firms and employees in each municipality, though the database does not cover municipalities with less than 5000 inhabitants (the sample is restricted accordingly). The data refer to the year 2008. We restrict our analysis to the manufacturing sector.15 Originally, we considered both enterprise- and plant-level data, since from an institutional point of view there is not a clear indication on which kind of unit would be most suitable for our analysis. We found that while the two sets of results are very similar, plant-level data produce slightly more precise results. This seems to suggest that the enterprise records are a noisier source of information than plant records. In the light of this, and considering that only a small share of Italian firms are multiplant firms, in what follows we use plant-level data.16 Our main dependent variable is the ratio between the total number of employees and the total number of plants (Av. firm size). We also estimate separately the effects of contract enforcement on total employment (Employment) and on the number of firms (No. of firms), in order to assess whether the effect on average firm size is driven by specific dynamics in the numerator or in the denominator. We also consider an alternative proxy for firm size based on accounting data. Our source is the CERVED database, maintained by the private company Cerved Group, containing balance sheet data for corporations and partnerships.17 This database has two limitations. First, large firms are overrepresented. This is mainly due to the fact that it does not contain information on sole proprietorships, which are generally smaller than corporations and partnerships. Second, account-based measures are more sensitive than employment-based measures to short-term disturbances. However, there is an important advantage in using this database: since it contains information at firm level it allows us to measure both firm size and firm growth.18 From this database, we compute two variables: the average value of firms’ turnover at the municipal level over 2 years for the periods 2001–2002 and 2008–2009.19 The average turnover value for the period 2008–2009 (Average turnover 2008/2009) is our measure of firm size, while the growth rate between the two periods is our measure of firm growth (Av. turnover growth 2001/2009). The sample only retains single-plant firms20 that survived for the whole period. In addition, as the data are noisy, we also drop the first and last centiles of the firm distribution of total turnover growth. 3.3. Local controls In addition, we use information provided mostly by Istat to construct a number of control variables at the municipal level. As a proxy for the size of the municipality, we use the municipal population as recorded in the 2001 Census data (Population). High crime rates may discourage economic activity and thus the birth and growth of firms, and at the same time congest local courts. To take these factors into account, we include the ratio of reported crimes to population as a proxy for crime rates (Crime). To proxy for the functioning of informal institution and for the local civic capital endowment, we use the average measure of cheating in the math test of the nationwide school quality assessment exercise (INVALSI), taken by all pupils at the age of 11 (in the fifth year of primary school) (Paccagnella and Sestito, 2014).21 As financial development is an important determinant of firm size, we include the number of retail banking branches in our dataset. In order to further control for court congestion, we include a measure of litigation intensity within the court’s jurisdiction (Litigation rate), expressed as the ratio of the number of filed proceedings in the period 2002–2007 to the total population of the jurisdiction in the year 2001. Although the variable is defined by a flow, it may be a proxy for a more general litigation propensity in the area. In principle, if we were to consider the hypothesis that larger firms are characterized by more litigation, this variable would also control for a possible reverse causality channel. However, to the extent that jurisdiction borders do not introduce any discontinuity in the litigation rate, the possible reverse causality channel is neutralized by the spatial discontinuity approach, in the same way as for the other unobserved factors. Finally, we include a measure of local taxation on business real estate (Imposta Comunale sugli Immobili), since this is surely the most important policy tool at the municipality level that may affect firms’ location choices and growth opportunities (Local tax rate). In Table 1, we report the main descriptive statistics of the variables used in the empirical analysis, both in the full and the restricted sample. Most differences between the two groups are not statistically significant, and the differences that are statistically significant are relatively small. This suggests that the restricted sample of border municipalities is broadly representative of the full sample of Italian municipalities, and that the external validity of the findings for the whole country should not be an issue. Table 1 Summary statistics Variables Sample Mean St. dev. Mean St. dev. t-test Full (N=2185) Border (N=1019) Av. firm size 8.3 5.5 8.2 5.6 0.47 Firms/pop. 0.011 0.006 0.011 0.007 0.00 Empl./pop. 0.1 0.095 0.1 0.094 0.00 EWAS 88 249 92 296 −0.37 Average turnover 2008–2009 6291 19,687 6114 12,221 0.31 Turnover growth 2001–2009 1.4 0.74 1.4 0.85 0.00 Length civil 931 307 914 301 1.48 Length labor 718 297 725 309 −0.60 Length criminal 299 152 311 161 −2.00 Population 20,577 73,029 24,167 97,320 −1.05 Bank branches 10 42 12 55 −1.03 Crime 0.15 0.86 0.15 0.96 0.00 Cheating 0.053 0.107 0.057 0.113 −0.95 Litigation rate 0.0062 0.0019 0.0063 0.002 −1.34 Local tax rate 6.2 0.68 6.3 0.68 −3.88 Variables Sample Mean St. dev. Mean St. dev. t-test Full (N=2185) Border (N=1019) Av. firm size 8.3 5.5 8.2 5.6 0.47 Firms/pop. 0.011 0.006 0.011 0.007 0.00 Empl./pop. 0.1 0.095 0.1 0.094 0.00 EWAS 88 249 92 296 −0.37 Average turnover 2008–2009 6291 19,687 6114 12,221 0.31 Turnover growth 2001–2009 1.4 0.74 1.4 0.85 0.00 Length civil 931 307 914 301 1.48 Length labor 718 297 725 309 −0.60 Length criminal 299 152 311 161 −2.00 Population 20,577 73,029 24,167 97,320 −1.05 Bank branches 10 42 12 55 −1.03 Crime 0.15 0.86 0.15 0.96 0.00 Cheating 0.053 0.107 0.057 0.113 −0.95 Litigation rate 0.0062 0.0019 0.0063 0.002 −1.34 Local tax rate 6.2 0.68 6.3 0.68 −3.88 Notes: The full sample is composed of all municipalities with over 5000 inhabitants for which data are available. The border sample contains only municipalities which are included to the full sample and which are contiguous to a jurisdiction border. Source: See Table 3. Table 1 Summary statistics Variables Sample Mean St. dev. Mean St. dev. t-test Full (N=2185) Border (N=1019) Av. firm size 8.3 5.5 8.2 5.6 0.47 Firms/pop. 0.011 0.006 0.011 0.007 0.00 Empl./pop. 0.1 0.095 0.1 0.094 0.00 EWAS 88 249 92 296 −0.37 Average turnover 2008–2009 6291 19,687 6114 12,221 0.31 Turnover growth 2001–2009 1.4 0.74 1.4 0.85 0.00 Length civil 931 307 914 301 1.48 Length labor 718 297 725 309 −0.60 Length criminal 299 152 311 161 −2.00 Population 20,577 73,029 24,167 97,320 −1.05 Bank branches 10 42 12 55 −1.03 Crime 0.15 0.86 0.15 0.96 0.00 Cheating 0.053 0.107 0.057 0.113 −0.95 Litigation rate 0.0062 0.0019 0.0063 0.002 −1.34 Local tax rate 6.2 0.68 6.3 0.68 −3.88 Variables Sample Mean St. dev. Mean St. dev. t-test Full (N=2185) Border (N=1019) Av. firm size 8.3 5.5 8.2 5.6 0.47 Firms/pop. 0.011 0.006 0.011 0.007 0.00 Empl./pop. 0.1 0.095 0.1 0.094 0.00 EWAS 88 249 92 296 −0.37 Average turnover 2008–2009 6291 19,687 6114 12,221 0.31 Turnover growth 2001–2009 1.4 0.74 1.4 0.85 0.00 Length civil 931 307 914 301 1.48 Length labor 718 297 725 309 −0.60 Length criminal 299 152 311 161 −2.00 Population 20,577 73,029 24,167 97,320 −1.05 Bank branches 10 42 12 55 −1.03 Crime 0.15 0.86 0.15 0.96 0.00 Cheating 0.053 0.107 0.057 0.113 −0.95 Litigation rate 0.0062 0.0019 0.0063 0.002 −1.34 Local tax rate 6.2 0.68 6.3 0.68 −3.88 Notes: The full sample is composed of all municipalities with over 5000 inhabitants for which data are available. The border sample contains only municipalities which are included to the full sample and which are contiguous to a jurisdiction border. Source: See Table 3. There are 2267 municipalities with a population of 5000 or more in the year 2001. For 82 of those, data on local tax rate or number of retail bank branches are not available; this leaves us with 2185 municipalities, which constitute the main sample when control variables are included. In the regression analysis, all variables are expressed in logarithmic form and standard errors are clustered at the court level. 4. Results 4.1. Testing Assumption 1 In order to test Assumption 1 and Equation (3), we estimate Equation (4) using as dependent variables the socioeconomic control variables at the municipality level that should not be affected—at least in the short/medium term—by average contract enforceability over the 2002–2007 period: the number of bank branches located in the municipality, the crime rate and the average measure of cheating in the math tests. We also mirror the same estimation at the firm level, using firm age as dependent variable (calculated as the number of days elapsed since the incorporation date on 31 December 2008). This variable would be affected by contract enforceability if the latter had a significant effect on firms at the extensive margin, i.e. if it were to affect firms’ entry and exit. However, we can anticipate that our results point to a significant effect only on the intensive margin. Therefore, firm age should reflect the long-term local structural composition of the business environment and, under Assumption 1, should be independent from contract enforceability. The results are reported in Table 2: the coefficient on the contract enforceability variable is never significant and always close to zero. This implies that, once local factors are controlled for, on average there are not statistically significant differences between neighbouring municipalities that are correlated with local contract enforceability. Table 2 Test of unconfoundness Dep. var. Unit of obs. Sample (1) Bank branches Municipality Border (2) Crime Municipality Border (3) Cheating Municipality Border (4) Age (in days) Firm Border Length civil 0.048 0.008 −0.110 −0.052 (0.086) (0.083) (0.252) (0.039) Population 0.074*** 0.234*** 0.154*** (0.018) (0.017) (0.054) Border FE Yes Yes Yes Yes Region FE Yes Yes Yes No Observations 1019 1019 1019 387,041 R2 0.713 0.591 0.625 0.015 Dep. var. Unit of obs. Sample (1) Bank branches Municipality Border (2) Crime Municipality Border (3) Cheating Municipality Border (4) Age (in days) Firm Border Length civil 0.048 0.008 −0.110 −0.052 (0.086) (0.083) (0.252) (0.039) Population 0.074*** 0.234*** 0.154*** (0.018) (0.017) (0.054) Border FE Yes Yes Yes Yes Region FE Yes Yes Yes No Observations 1019 1019 1019 387,041 R2 0.713 0.591 0.625 0.015 Notes: Robust standard errors clustered at court level in parentheses. All variables are in logarithms. ***p < 0.01; **p < 0.05; *p < 0.1. Source: See Table 3. Table 2 Test of unconfoundness Dep. var. Unit of obs. Sample (1) Bank branches Municipality Border (2) Crime Municipality Border (3) Cheating Municipality Border (4) Age (in days) Firm Border Length civil 0.048 0.008 −0.110 −0.052 (0.086) (0.083) (0.252) (0.039) Population 0.074*** 0.234*** 0.154*** (0.018) (0.017) (0.054) Border FE Yes Yes Yes Yes Region FE Yes Yes Yes No Observations 1019 1019 1019 387,041 R2 0.713 0.591 0.625 0.015 Dep. var. Unit of obs. Sample (1) Bank branches Municipality Border (2) Crime Municipality Border (3) Cheating Municipality Border (4) Age (in days) Firm Border Length civil 0.048 0.008 −0.110 −0.052 (0.086) (0.083) (0.252) (0.039) Population 0.074*** 0.234*** 0.154*** (0.018) (0.017) (0.054) Border FE Yes Yes Yes Yes Region FE Yes Yes Yes No Observations 1019 1019 1019 387,041 R2 0.713 0.591 0.625 0.015 Notes: Robust standard errors clustered at court level in parentheses. All variables are in logarithms. ***p < 0.01; **p < 0.05; *p < 0.1. Source: See Table 3. Table 3 Definitions of variables Variable Definition Level Period Source Av. firm size Employment over number of plants Municip. 2008 ASIA-ISTAT Firms/pop. Number of plants over population Municip. 2008/2001 ASIA-ISTAT Empl/pop. Employment over population Municip. 2008/2001 ASIA-ISTAT EWAS Av. plant size with greater weight on large plants Municip. 2008 ASIA-ISTAT Av. turnover 2008–2009 Average plant turnover Municip. 2008–2009 CERVED Turnover growth 2001–2009 Plant turnover in 2008–2009 over plant turnover Municip. 2001–2009 CERVED in 2001–2002 divided by the number of surviving plants Firm age Age in days calculated based on the incorporation date Firm 2008 CERVED Cheating Average measure of cheating in INVALSI math tests Municip. 2009/2010 INVALSI Length civil Estimated length in days of civil cases Court jur. 2002-2007 Italian Ministry of Justice Length labor Estimated length in days of labor cases Court jur. 2002–2006 Italian Ministry of Justice Length criminal Estimated length in days of criminal cases Court jur. 2002–2007 Italian Ministry of Justice Population Total population residing in the municipality Municip. 2001 Population Census 2001, ISTAT Litigation rate Number of new cases over total population Court jur. 2002–2007 Italian Ministry of Justice Bank branches Number of bank branches Municip. 2001 Atlante Stastico Comunale, ISTAT Crime Number of reported crimes over total population Municip. 2004–2009 Italian Interior Ministry Local tax rate Average municipal tax rate on real estate (ICI) Municip. 2002–2007 Bank of Italy Variable Definition Level Period Source Av. firm size Employment over number of plants Municip. 2008 ASIA-ISTAT Firms/pop. Number of plants over population Municip. 2008/2001 ASIA-ISTAT Empl/pop. Employment over population Municip. 2008/2001 ASIA-ISTAT EWAS Av. plant size with greater weight on large plants Municip. 2008 ASIA-ISTAT Av. turnover 2008–2009 Average plant turnover Municip. 2008–2009 CERVED Turnover growth 2001–2009 Plant turnover in 2008–2009 over plant turnover Municip. 2001–2009 CERVED in 2001–2002 divided by the number of surviving plants Firm age Age in days calculated based on the incorporation date Firm 2008 CERVED Cheating Average measure of cheating in INVALSI math tests Municip. 2009/2010 INVALSI Length civil Estimated length in days of civil cases Court jur. 2002-2007 Italian Ministry of Justice Length labor Estimated length in days of labor cases Court jur. 2002–2006 Italian Ministry of Justice Length criminal Estimated length in days of criminal cases Court jur. 2002–2007 Italian Ministry of Justice Population Total population residing in the municipality Municip. 2001 Population Census 2001, ISTAT Litigation rate Number of new cases over total population Court jur. 2002–2007 Italian Ministry of Justice Bank branches Number of bank branches Municip. 2001 Atlante Stastico Comunale, ISTAT Crime Number of reported crimes over total population Municip. 2004–2009 Italian Interior Ministry Local tax rate Average municipal tax rate on real estate (ICI) Municip. 2002–2007 Bank of Italy Table 3 Definitions of variables Variable Definition Level Period Source Av. firm size Employment over number of plants Municip. 2008 ASIA-ISTAT Firms/pop. Number of plants over population Municip. 2008/2001 ASIA-ISTAT Empl/pop. Employment over population Municip. 2008/2001 ASIA-ISTAT EWAS Av. plant size with greater weight on large plants Municip. 2008 ASIA-ISTAT Av. turnover 2008–2009 Average plant turnover Municip. 2008–2009 CERVED Turnover growth 2001–2009 Plant turnover in 2008–2009 over plant turnover Municip. 2001–2009 CERVED in 2001–2002 divided by the number of surviving plants Firm age Age in days calculated based on the incorporation date Firm 2008 CERVED Cheating Average measure of cheating in INVALSI math tests Municip. 2009/2010 INVALSI Length civil Estimated length in days of civil cases Court jur. 2002-2007 Italian Ministry of Justice Length labor Estimated length in days of labor cases Court jur. 2002–2006 Italian Ministry of Justice Length criminal Estimated length in days of criminal cases Court jur. 2002–2007 Italian Ministry of Justice Population Total population residing in the municipality Municip. 2001 Population Census 2001, ISTAT Litigation rate Number of new cases over total population Court jur. 2002–2007 Italian Ministry of Justice Bank branches Number of bank branches Municip. 2001 Atlante Stastico Comunale, ISTAT Crime Number of reported crimes over total population Municip. 2004–2009 Italian Interior Ministry Local tax rate Average municipal tax rate on real estate (ICI) Municip. 2002–2007 Bank of Italy Variable Definition Level Period Source Av. firm size Employment over number of plants Municip. 2008 ASIA-ISTAT Firms/pop. Number of plants over population Municip. 2008/2001 ASIA-ISTAT Empl/pop. Employment over population Municip. 2008/2001 ASIA-ISTAT EWAS Av. plant size with greater weight on large plants Municip. 2008 ASIA-ISTAT Av. turnover 2008–2009 Average plant turnover Municip. 2008–2009 CERVED Turnover growth 2001–2009 Plant turnover in 2008–2009 over plant turnover Municip. 2001–2009 CERVED in 2001–2002 divided by the number of surviving plants Firm age Age in days calculated based on the incorporation date Firm 2008 CERVED Cheating Average measure of cheating in INVALSI math tests Municip. 2009/2010 INVALSI Length civil Estimated length in days of civil cases Court jur. 2002-2007 Italian Ministry of Justice Length labor Estimated length in days of labor cases Court jur. 2002–2006 Italian Ministry of Justice Length criminal Estimated length in days of criminal cases Court jur. 2002–2007 Italian Ministry of Justice Population Total population residing in the municipality Municip. 2001 Population Census 2001, ISTAT Litigation rate Number of new cases over total population Court jur. 2002–2007 Italian Ministry of Justice Bank branches Number of bank branches Municip. 2001 Atlante Stastico Comunale, ISTAT Crime Number of reported crimes over total population Municip. 2004–2009 Italian Interior Ministry Local tax rate Average municipal tax rate on real estate (ICI) Municip. 2002–2007 Bank of Italy Assumption 1 can also be graphically tested. In Figure 4, we plot the kernel density distributions of the same four variables we regress in Table 2.22 We now convert the contract enforceability variable into a binary classification of municipalities contiguous to a jurisdiction boundary: ‘slow (fast) side’ municipalities are those that belong to a local jurisdiction where contract enforcement is less (more) effective than in the neighbouring jurisdiction. In the graphs, the continuous black lines plot the values of observations located in municipalities at the ‘slow’ side of a jurisdiction boundary. Symmetrically, the dotted red lines plot the distribution of the observations located in municipalities at the ‘fast’ side of a jurisdiction boundary. As the graphs show, there are no systematic differences between the two subsamples along the full distributions of the four variables. Figure 4 Open in new tabDownload slide Kernel distributions on either side of a jurisdiction boundary. Notes: The graphs show the Epanechnikov kernel distribution of the relative variables. The top 5% of the distribution is dropped to easy readability. Vertical lines show the mean values. The sample is restricted to municipalities over 5000 inhabitants located along a jurisdiction boundary (graphs in the top row). Source: Based on INVALSI data, Italian Ministry of Justice data and CERVED (graph on firm age). Figure 4 Open in new tabDownload slide Kernel distributions on either side of a jurisdiction boundary. Notes: The graphs show the Epanechnikov kernel distribution of the relative variables. The top 5% of the distribution is dropped to easy readability. Vertical lines show the mean values. The sample is restricted to municipalities over 5000 inhabitants located along a jurisdiction boundary (graphs in the top row). Source: Based on INVALSI data, Italian Ministry of Justice data and CERVED (graph on firm age). 4.2. Average and total employment and number of units In a similar vein, a graphical representation of the logic behind our analysis is shown in Figure 5, which plots the kernel density distributions of average firm size, employment over population and number of firms over population for municipalities located on the ‘slow side’ (continuous black line) of jurisdiction boundaries (i.e. the side where contract enforcement is less effective), and on the ‘fast side’ (dotted red line) of jurisdiction boundaries (i.e. the side where contract enforcement is more effective). For all the variables, the plots show that the distribution of municipalities on the slow side of the boundary stands to the left of the distribution of municipalities on the fast side of the boundary, meaning that variable values are lower. For average firm size and number of units, the effect is quite uniform along the full distribution. The econometric analysis that follows shows that the effect on average firm size is statistically significant, economically important and robust. Figure 5 Open in new tabDownload slide Kernel distributions of average firm size, employment and number of firms in municipalities on either side of a jurisdiction boundary. Notes: The graphs show the Epanechnikov kernel distribution of the relative variables. The top 5% of the distribution has been dropped to improve readability. Vertical lines show the mean values. The sample is restricted to the municipalities with over 5000 inhabitants located along a jurisdiction boundary. Source: Based on ISTAT and Italian Ministry of Justice data. Figure 5 Open in new tabDownload slide Kernel distributions of average firm size, employment and number of firms in municipalities on either side of a jurisdiction boundary. Notes: The graphs show the Epanechnikov kernel distribution of the relative variables. The top 5% of the distribution has been dropped to improve readability. Vertical lines show the mean values. The sample is restricted to the municipalities with over 5000 inhabitants located along a jurisdiction boundary. Source: Based on ISTAT and Italian Ministry of Justice data. We first estimate Equation 4 on our main dependent variable: average firm size. Table 4 reports in columns 1 and 2 the estimates based on the whole sample of municipalities, and in columns 3 and 4 the estimates on the sample restricted to those municipalities situated along court jurisdictions boundaries. In columns 3 and 4, we also introduce a set of fixed effects for all the groups of municipalities sharing the same boundary; those fixed effects control for a wide set of observable and unobservable factors, while still leaving within-group variability in the contract enforcement variables due to the change in court jurisdiction. Table 4 Effect on average firm size (1) (2) (3) (4) Dep. var. Sample Average firm size Full Full Border Border Length civil −0.680*** 0.083 −0.190** −0.214** (0.076) (0.059) (0.087) (0.098) Population 0.045** 0.093*** 0.072*** 0.063** (0.020) (0.017) (0.025) (0.027) Litigation rate 0.014 0.038 (0.041) (0.052) Bank branches 0.015 0.076 (0.035) (0.072) Crime −0.001 −0.003 (0.030) (0.056) Local tax rate −0.107 0.063 (0.099) (0.186) Cheating 0.000 0.001 (0.009) (0.016) Region FE No Yes No Yes Border FE No No Yes Yes Observations 2267 2185 1019 1019 R2 0.146 0.438 0.633 0.638 (1) (2) (3) (4) Dep. var. Sample Average firm size Full Full Border Border Length civil −0.680*** 0.083 −0.190** −0.214** (0.076) (0.059) (0.087) (0.098) Population 0.045** 0.093*** 0.072*** 0.063** (0.020) (0.017) (0.025) (0.027) Litigation rate 0.014 0.038 (0.041) (0.052) Bank branches 0.015 0.076 (0.035) (0.072) Crime −0.001 −0.003 (0.030) (0.056) Local tax rate −0.107 0.063 (0.099) (0.186) Cheating 0.000 0.001 (0.009) (0.016) Region FE No Yes No Yes Border FE No No Yes Yes Observations 2267 2185 1019 1019 R2 0.146 0.438 0.633 0.638 Notes: Robust standard errors clustered at court level in parentheses. All variables are in logarithms. The sample is composed of all municipalities with over 5000 inhabitants for which data are available. ***p < **p < 0.05; 0.01; *p < 0.1. Table 4 Effect on average firm size (1) (2) (3) (4) Dep. var. Sample Average firm size Full Full Border Border Length civil −0.680*** 0.083 −0.190** −0.214** (0.076) (0.059) (0.087) (0.098) Population 0.045** 0.093*** 0.072*** 0.063** (0.020) (0.017) (0.025) (0.027) Litigation rate 0.014 0.038 (0.041) (0.052) Bank branches 0.015 0.076 (0.035) (0.072) Crime −0.001 −0.003 (0.030) (0.056) Local tax rate −0.107 0.063 (0.099) (0.186) Cheating 0.000 0.001 (0.009) (0.016) Region FE No Yes No Yes Border FE No No Yes Yes Observations 2267 2185 1019 1019 R2 0.146 0.438 0.633 0.638 (1) (2) (3) (4) Dep. var. Sample Average firm size Full Full Border Border Length civil −0.680*** 0.083 −0.190** −0.214** (0.076) (0.059) (0.087) (0.098) Population 0.045** 0.093*** 0.072*** 0.063** (0.020) (0.017) (0.025) (0.027) Litigation rate 0.014 0.038 (0.041) (0.052) Bank branches 0.015 0.076 (0.035) (0.072) Crime −0.001 −0.003 (0.030) (0.056) Local tax rate −0.107 0.063 (0.099) (0.186) Cheating 0.000 0.001 (0.009) (0.016) Region FE No Yes No Yes Border FE No No Yes Yes Observations 2267 2185 1019 1019 R2 0.146 0.438 0.633 0.638 Notes: Robust standard errors clustered at court level in parentheses. All variables are in logarithms. The sample is composed of all municipalities with over 5000 inhabitants for which data are available. ***p < **p < 0.05; 0.01; *p < 0.1. In the estimates on the whole sample, the coefficient on the length of judicial civil proceedings is negative and significant when control variables at the municipal level and regional fixed effects are excluded (column 1), and is positive and significant otherwise. Unreported regressions including only regional fixed effects indicate that those fully absorb the negative correlation, suggesting that this correlation is entirely attributable to the North–South divide. As previously discussed, however, even results from the specification with regional fixed effects might be biased due to omitted variables that correlate both with firm size and the length of proceedings.23 When we restrict the sample and introduce a set of fixed effects for all the groups of municipalities sharing the same border, the effect of the length of civil proceedings on average firm size becomes negative, and is statistically significant. As expected, the inclusion of additional controls produces only minor changes in the point estimates of the coefficients of the variables of interest. Regional fixed effects are generally significant, but leave the coefficients of our variable of interest unaffected. This is particularly supportive of the robustness of our methodology, since, as already pointed out, regions in Italy are the local authorities with the strongest autonomy and powers on matters that are relevant to business activity. The other controls are all not significant. As for the magnitude of the effect, since regressions are log-linear and coefficients can be interpreted as elasticities, our estimates imply that, if the length of civil proceedings were reduced by 10%, average firm size would increase by around 2% (based on the results in column 3). A simple back-of-envelope calculation based on linear predictions suggests that moving from the court at the 90th percentile to the one at the 10th percentile in the distribution of the length of proceedings would lead to an average increase in firm size of one employee, from 6.5 to 7.5. Table 5 shows the results of the estimates of Equation 4 where the dependent variables are the number of firms (columns 1 and 2) and total employment (columns 3 and 4). The model is estimated on the restricted sample and includes boundary fixed effects. The length of civil proceedings has a negative effect both on the number of firms and on total employment, but the coefficient is statistically significant only for employment and only when control variables and regional fixed effects are included (column 4). The results do not provide clear-cut evidence, however, we interpret them as an indication of the fact that ineffective contract enforcement hinders firm growth, while does not influence firms’ net entry. It is also interesting to point out that our control for local taxation is not significant for firm size, while it is significant and has a negative sign for the number of firms and for total employment. This is consistent with previous findings indicating that local taxes are more effective at the extensive margin (thus on the number of firms), rather than at the intensive one (i.e. on firms’ size), because their effect tends to be capitalized into rent (Duranton et al., 2011). Table 5 Effect on number of firms and total employment (1) (2) (3) (4) Dep. var. Sample No. of firms Employment Border Border Border Border Length civil −0.076 −0.147 −0.265 −0.361* (0.105) (0.119) (0.171) (0.190) Population 0.928*** 0.922*** 1.000*** 0.985*** (0.017) (0.023) (0.036) (0.044) Litigation rate 0.047 0.085 (0.074) (0.099) Bank branches 0.112** 0.188* (0.049) (0.097) Crime 0.010 0.007 (0.057) (0.092) Local tax rate −0.557*** −0.494* (0.157) (0.252) Cheating 0.017 0.018 (0.015) (0.026) Region FE No Yes No Yes Border FE Yes Yes Yes Yes Observations 1019 1019 1019 1019 R2 0.877 0.884 0.822 0.827 (1) (2) (3) (4) Dep. var. Sample No. of firms Employment Border Border Border Border Length civil −0.076 −0.147 −0.265 −0.361* (0.105) (0.119) (0.171) (0.190) Population 0.928*** 0.922*** 1.000*** 0.985*** (0.017) (0.023) (0.036) (0.044) Litigation rate 0.047 0.085 (0.074) (0.099) Bank branches 0.112** 0.188* (0.049) (0.097) Crime 0.010 0.007 (0.057) (0.092) Local tax rate −0.557*** −0.494* (0.157) (0.252) Cheating 0.017 0.018 (0.015) (0.026) Region FE No Yes No Yes Border FE Yes Yes Yes Yes Observations 1019 1019 1019 1019 R2 0.877 0.884 0.822 0.827 Notes: Robust standard errors clustered at court level in parentheses. All variables are in logarithms. The sample is composed of municipalities with over 5000 inhabitants contiguous to a jurisdiction boundary for which data are available. ***p < 0.01; **p < 0.05; *p < 0.1. Table 5 Effect on number of firms and total employment (1) (2) (3) (4) Dep. var. Sample No. of firms Employment Border Border Border Border Length civil −0.076 −0.147 −0.265 −0.361* (0.105) (0.119) (0.171) (0.190) Population 0.928*** 0.922*** 1.000*** 0.985*** (0.017) (0.023) (0.036) (0.044) Litigation rate 0.047 0.085 (0.074) (0.099) Bank branches 0.112** 0.188* (0.049) (0.097) Crime 0.010 0.007 (0.057) (0.092) Local tax rate −0.557*** −0.494* (0.157) (0.252) Cheating 0.017 0.018 (0.015) (0.026) Region FE No Yes No Yes Border FE Yes Yes Yes Yes Observations 1019 1019 1019 1019 R2 0.877 0.884 0.822 0.827 (1) (2) (3) (4) Dep. var. Sample No. of firms Employment Border Border Border Border Length civil −0.076 −0.147 −0.265 −0.361* (0.105) (0.119) (0.171) (0.190) Population 0.928*** 0.922*** 1.000*** 0.985*** (0.017) (0.023) (0.036) (0.044) Litigation rate 0.047 0.085 (0.074) (0.099) Bank branches 0.112** 0.188* (0.049) (0.097) Crime 0.010 0.007 (0.057) (0.092) Local tax rate −0.557*** −0.494* (0.157) (0.252) Cheating 0.017 0.018 (0.015) (0.026) Region FE No Yes No Yes Border FE Yes Yes Yes Yes Observations 1019 1019 1019 1019 R2 0.877 0.884 0.822 0.827 Notes: Robust standard errors clustered at court level in parentheses. All variables are in logarithms. The sample is composed of municipalities with over 5000 inhabitants contiguous to a jurisdiction boundary for which data are available. ***p < 0.01; **p < 0.05; *p < 0.1. 4.3. Turnover Columns 1 and 2 of Table 6 report the estimates of Equation 4 on average turnover level in the years 2008–2009 (average value across the 2 years) on the restricted sample with the inclusion of boundary fixed effects, without and with local controls, respectively. The results on the level of turnover confirm a negative effect of the length of civil proceedings on firm size. However, although the coefficients on turnover levels are of similar magnitude as those on employment-based measure of firm size, these estimates are below the standard 10% significance level, being thus less precise than those reported in Table 4. These results suggest that the ASIA database is a better data source for this kind of analysis. As already pointed out, ASIA covers the whole population of firms in municipalities with more than 5000 inhabitants, while the coverage of the turnover data is less reliable for the very small firms. This may induce a selection bias, as smaller firms in inefficient jurisdictions are more likely to be excluded than relatively larger firms in more efficient jurisdictions. Table 6 Effect on the level and growth of turnover (1) (2) (3) (4) Dep. var. Sample Turnover level, av. 2008–2009 Turnover growth 2001–2009 Border Border Border Border Length civil −0.275 −0.353 −0.205** −0.216** (0.237) (0.254) (0.091) (0.103) Population 0.216*** 0.207*** 0.003 0.008 (0.039) (0.044) (0.018) (0.025) Av. turnover 2001/2002 −0.048* (0.028) Other controls No Yes No Yes Region FE No Yes No Yes Border FE Yes Yes YES Yes Observations 967 967 967 967 R2 0.528 0.543 0.305 0.321 (1) (2) (3) (4) Dep. var. Sample Turnover level, av. 2008–2009 Turnover growth 2001–2009 Border Border Border Border Length civil −0.275 −0.353 −0.205** −0.216** (0.237) (0.254) (0.091) (0.103) Population 0.216*** 0.207*** 0.003 0.008 (0.039) (0.044) (0.018) (0.025) Av. turnover 2001/2002 −0.048* (0.028) Other controls No Yes No Yes Region FE No Yes No Yes Border FE Yes Yes YES Yes Observations 967 967 967 967 R2 0.528 0.543 0.305 0.321 Notes: Robust standard errors clustered at court level in parentheses. All variables are in logarithms. The sample is composed of municipalities with over 5000 inhabitants contiguous to a jurisdiction boundary for which data are available. ***p < 0.01; **p < 0.05; *p < 0.1. Table 6 Effect on the level and growth of turnover (1) (2) (3) (4) Dep. var. Sample Turnover level, av. 2008–2009 Turnover growth 2001–2009 Border Border Border Border Length civil −0.275 −0.353 −0.205** −0.216** (0.237) (0.254) (0.091) (0.103) Population 0.216*** 0.207*** 0.003 0.008 (0.039) (0.044) (0.018) (0.025) Av. turnover 2001/2002 −0.048* (0.028) Other controls No Yes No Yes Region FE No Yes No Yes Border FE Yes Yes YES Yes Observations 967 967 967 967 R2 0.528 0.543 0.305 0.321 (1) (2) (3) (4) Dep. var. Sample Turnover level, av. 2008–2009 Turnover growth 2001–2009 Border Border Border Border Length civil −0.275 −0.353 −0.205** −0.216** (0.237) (0.254) (0.091) (0.103) Population 0.216*** 0.207*** 0.003 0.008 (0.039) (0.044) (0.018) (0.025) Av. turnover 2001/2002 −0.048* (0.028) Other controls No Yes No Yes Region FE No Yes No Yes Border FE Yes Yes YES Yes Observations 967 967 967 967 R2 0.528 0.543 0.305 0.321 Notes: Robust standard errors clustered at court level in parentheses. All variables are in logarithms. The sample is composed of municipalities with over 5000 inhabitants contiguous to a jurisdiction boundary for which data are available. ***p < 0.01; **p < 0.05; *p < 0.1. Columns 3 and 4 present the estimates for turnover growth between 2001 and 2009, calculated as the log of the ratio of the final level to the initial one. The specifications mirror those of columns 1 and 2, with the addition of a control for the turnover level at the beginning of the period in column 4. The coefficients on turnover growth are negative, statistically significant and remarkably similar in magnitude to those on average firm size. These results thus provide additional evidence that poor contract enforcement hinders firm growth. 4.4. Checking for sorting A first concern is related to the possibility that municipalities along jurisdiction boundaries with more efficient courts host larger firms than inner municipalities (i.e. municipalities in the middle of a jurisdiction rather than along the border) in the same court’s jurisdiction, due to the sorting of firms across municipalities belonging to the same boundary group. For instance, let us assume that an industrial district (composed of several municipalities) with a specialized business structure is split by a jurisdiction boundary and, therefore, extends over two jurisdictions with different degree of contract enforceability. We report a stylized example in Figure 6, showing 24 municipalities belonging to two different jurisdictions, one ‘fast’ and one ‘slow’; inner municipalities are reported in bold (11, 12, 23, 24), while the gray cell background (municipalities 4, 5, 21, 22) identifies a business district. Firms in municipalities 21 and 22 may decide to marginally change their location and move to municipalities 4 and 5, on the better side of the jurisdiction border, while still enjoying the positive district spillovers. Firms located in municipalities in the same inefficient jurisdiction but outside the district and further away from the jurisdiction boundary, instead, may decide not to relocate, since the distance is longer and the business environment may be substantially different. As a consequence, municipalities along jurisdiction boundaries would be systematically different from inner municipalities, as firms at the boundary have higher incentives to relocate. A similar reasoning may also apply to workers’ relocation. As firms at both sides of the jurisdiction border are likely to share the same labor market, the small average firm size of firms located at the slow side is not only due to the jurisdiction’s own poor efficiency, but it might also be due to the attraction effect of municipalities in the neighbouring, better performing jurisdiction.24 Figure 6 Open in new tabDownload slide Inner and border municipalities—stylized example. Figure 6 Open in new tabDownload slide Inner and border municipalities—stylized example. In our setting, this would lead to overestimating the effect of contract enforcement on firm size, as the effect at the jurisdiction border is amplified with respect to the jurisdiction core. Municipalities along jurisdiction boundaries (included in the analysis) in better jurisdictions would have more and/or larger firms than inner municipalities in the same court’s jurisdiction (excluded from the sample), and, vice versa, municipalities in worse jurisdictions would have less and/or smaller firms at the border than at the core. To test whether this occurs, we define two binary dummy variables taking value of one for municipalities located at the fast or slow sides of the jurisdiction boundaries, respectively, and zero otherwise. Using the full sample of municipalities (fast side municipalities, slow side municipalities and inner municipalities), we regress our set of independent variables on the two ‘fast side’ and ‘slow side’ dummies (the inner municipalities being the omitted group), on the controls and on a full set of court jurisdiction dummies. The results are reported in columns 1–3 of Table 7: only the ‘slow side’ coefficient of column 1 (average firm size) is significant at the 10% level, and negative. This could mean that some larger firms might relocate at the other side of the jurisdiction border, wherever this would imply benefiting from stronger contract enforcement. As a consequence, the baseline results can be partially amplified by the smaller average firm size of firms located at the border of the jurisdiction, as compared to those located in the core. However, the coefficient is rather small—being only around one-quarter of the overall effect—and only marginally significant; furthermore, the coefficient on the ‘fast side’ is positive, as expected, but not significant. A similar robustness test consists of running Equation (4) on the full sample including an additional interaction term between the measure of court enforcement and a dummy equal to one if the municipality is on the border, in order to directly check whether the court enforcement effect is somehow amplified for border municipalities.25 This would be reflected in a significant coefficient on the interaction term. As is possible to see in column 4 of Table 7, the coefficient is instead very close to zero and not significant. Therefore, the robustness test shows that some relocation across jurisdiction borders may actually happen, and this could imply that the main effect might be slightly overestimated. However, the size of the overestimation is likely to be extremely small, and it further supports the existence of a substantial effect of contract enforceability on firm activity. Table 7 Robustness: sorting within jurisdictions Variables (1) Av. firm size (2) Av. turnover level 2008–2009 (3) Av. turnover growth 2001–2009 (4) Av. firm size Slow-side dummy −0.048* −0.063 0.019 (0.026) (0.066) (0.027) Fast-side dummy 0.025 0.069 0.033 (0.027) (0.065) (0.027) Length civil −0.703*** (0.075) Length civil −0.008 X border dummy (0.005) Region FE Yes Yes Yes No Jurisdiction FE Yes Yes Yes No Observations 2185 1942 1942 2185 R2 0.495 0.373 0.129 0.147 Variables (1) Av. firm size (2) Av. turnover level 2008–2009 (3) Av. turnover growth 2001–2009 (4) Av. firm size Slow-side dummy −0.048* −0.063 0.019 (0.026) (0.066) (0.027) Fast-side dummy 0.025 0.069 0.033 (0.027) (0.065) (0.027) Length civil −0.703*** (0.075) Length civil −0.008 X border dummy (0.005) Region FE Yes Yes Yes No Jurisdiction FE Yes Yes Yes No Observations 2185 1942 1942 2185 R2 0.495 0.373 0.129 0.147 Notes: Robust standard errors clustered at court level in parentheses. All variables are in logarithms. The sample is composed of municipalities with over 5000 inhabitants for which data are available. ***p < 0.01; **p < 0.05; *p < 0.1. Table 7 Robustness: sorting within jurisdictions Variables (1) Av. firm size (2) Av. turnover level 2008–2009 (3) Av. turnover growth 2001–2009 (4) Av. firm size Slow-side dummy −0.048* −0.063 0.019 (0.026) (0.066) (0.027) Fast-side dummy 0.025 0.069 0.033 (0.027) (0.065) (0.027) Length civil −0.703*** (0.075) Length civil −0.008 X border dummy (0.005) Region FE Yes Yes Yes No Jurisdiction FE Yes Yes Yes No Observations 2185 1942 1942 2185 R2 0.495 0.373 0.129 0.147 Variables (1) Av. firm size (2) Av. turnover level 2008–2009 (3) Av. turnover growth 2001–2009 (4) Av. firm size Slow-side dummy −0.048* −0.063 0.019 (0.026) (0.066) (0.027) Fast-side dummy 0.025 0.069 0.033 (0.027) (0.065) (0.027) Length civil −0.703*** (0.075) Length civil −0.008 X border dummy (0.005) Region FE Yes Yes Yes No Jurisdiction FE Yes Yes Yes No Observations 2185 1942 1942 2185 R2 0.495 0.373 0.129 0.147 Notes: Robust standard errors clustered at court level in parentheses. All variables are in logarithms. The sample is composed of municipalities with over 5000 inhabitants for which data are available. ***p < 0.01; **p < 0.05; *p < 0.1. There is also a more general concern related to sorting due to the possibility that better performing firms could locate in more efficient jurisdictions. If this were the case, our results would reflect the matching of larger firms with more efficient jurisdictions, rather than a causal effect of contract enforcement on firm size. There are several pieces of evidence presented so far in this article that make such a hypothesis unlikely. First, the test discussed above shows that there is no evidence of sorting at short distances across jurisdiction borders, which in turn makes it also unlikely that there is sorting at longer distances. Second, the regression results show that the effect on firm growth is statistically correspondent to the effect on firm size. This suggests that most of the effect we find on level variables is due to the (lack of) growth of incumbent firms, rather than to the sorting of larger firms into more efficient jurisdictions. However, the sorting of firms could still happen for entrants: i.e. new firms may systematically choose to locate in faster jurisdictions. The aggregate effect of this would be unclear, as entrants are generally smaller than incumbents, but tend to grow much faster on average. The bottom-right graph of Figure 4, however, shows that the firm age distribution is extremely similar on the good and bad side of the jurisdiction borders, respectively. This provides evidence against sorting, because if entrants ‘voted with their feet’ by locating in the faster jurisdictions, average age would be lower in the ‘good side’ municipalities. We also estimate the regressions with firm turnover as the dependent variable limiting the sample only to young firms (aged 10 or less). There is indeed evidence (Michelacci and Silva, 2007) that the fraction of entrepreneurs working in the region where they were born is significantly higher than the corresponding fraction for dependent workers. To the extent that this is due to the advantage of locals in exploiting financial opportunities (as maintained by the authors), young firms should be more ‘location constrained’, which in turn implies that sorting should be less of an issue for the subset of young firms. The results, not shown for brevity but available upon request, are consistent with the main findings, with a negative and significant coefficient on the length of civil proceedings that is even larger in absolute magnitude (equal to −0.81 with a standard error of 0.41). As mentioned in Section 2, this still does not entirely rule out the possibility that firms with better growth prospects may decide ex ante to locate within jurisdictions where contract enforcement is more effective, while less ambitious firms would locate in less effective jurisdictions. We believe that this is unlikely to happen for a number of reasons, among which the fact that entrants are generally not fully informed about their ‘type’ and future growth prospects. Furthermore, we provide evidence that this is not happening at short distances across contiguous jurisdictions, as explained above. 4.5. Other robustness checks We run a series of robustness checks on our estimates that leaves the main findings unaffected. The results are presented in Table 8. In all the regressions reported, with the exception of column 5, the dependent variable is average firm size computed as the ratio between the number of employees and the number of firms. Table 8 Robustness: alternative measure of justice efficiency, excluding courts coinciding with provinces, and EWAS (1) (2) (3) (4) (5) Variables Sample Average firm size EWAS No prov. Border Border Border Border Length civil −0.333** −0.208** −0.183* −0.412** (0.129) (0.098) (0.098) (0.204) Length labor −0.057 (0.046) Length criminal −0.136** (0.053) Length civil −0.216** (alternative index) (0.094) Other controls Yes Yes Yes Yes Yes Region FE Yes Yes Yes Yes Yes Border FE Yes Yes Yes Yes Yes Observations 646 1019 1019 1019 1019 R2 0.664 0.639 0.640 0.639 0.584 (1) (2) (3) (4) (5) Variables Sample Average firm size EWAS No prov. Border Border Border Border Length civil −0.333** −0.208** −0.183* −0.412** (0.129) (0.098) (0.098) (0.204) Length labor −0.057 (0.046) Length criminal −0.136** (0.053) Length civil −0.216** (alternative index) (0.094) Other controls Yes Yes Yes Yes Yes Region FE Yes Yes Yes Yes Yes Border FE Yes Yes Yes Yes Yes Observations 646 1019 1019 1019 1019 R2 0.664 0.639 0.640 0.639 0.584 Notes: Robust standard errors clustered at court level in parentheses. All variables are in logarithms. Other controls include regional fixed effects and variables listed in Table 5. EWAS stands for ‘employee-weighted average size indicator’ and is defined in Section 5.3. The sample is composed of municipalities with over 5000 inhabitants contiguous to a jurisdiction boundary (columns 2–5) and excluding jurisdiction coinciding with provinces (column 1) for which data are available. ***p < 0.01; **p < 0.05; *p < 0.1. Table 8 Robustness: alternative measure of justice efficiency, excluding courts coinciding with provinces, and EWAS (1) (2) (3) (4) (5) Variables Sample Average firm size EWAS No prov. Border Border Border Border Length civil −0.333** −0.208** −0.183* −0.412** (0.129) (0.098) (0.098) (0.204) Length labor −0.057 (0.046) Length criminal −0.136** (0.053) Length civil −0.216** (alternative index) (0.094) Other controls Yes Yes Yes Yes Yes Region FE Yes Yes Yes Yes Yes Border FE Yes Yes Yes Yes Yes Observations 646 1019 1019 1019 1019 R2 0.664 0.639 0.640 0.639 0.584 (1) (2) (3) (4) (5) Variables Sample Average firm size EWAS No prov. Border Border Border Border Length civil −0.333** −0.208** −0.183* −0.412** (0.129) (0.098) (0.098) (0.204) Length labor −0.057 (0.046) Length criminal −0.136** (0.053) Length civil −0.216** (alternative index) (0.094) Other controls Yes Yes Yes Yes Yes Region FE Yes Yes Yes Yes Yes Border FE Yes Yes Yes Yes Yes Observations 646 1019 1019 1019 1019 R2 0.664 0.639 0.640 0.639 0.584 Notes: Robust standard errors clustered at court level in parentheses. All variables are in logarithms. Other controls include regional fixed effects and variables listed in Table 5. EWAS stands for ‘employee-weighted average size indicator’ and is defined in Section 5.3. The sample is composed of municipalities with over 5000 inhabitants contiguous to a jurisdiction boundary (columns 2–5) and excluding jurisdiction coinciding with provinces (column 1) for which data are available. ***p < 0.01; **p < 0.05; *p < 0.1. The first cause of concern that we address is the partial overlapping of jurisdictional and provincial boundaries (Section 3.1). To account for this, we exclude from our sample the observations for which courts and provincial boundaries coincide. The results are presented in column 1. Our main findings on the effects of contract enforcement on average firm size are fully confirmed, as the main coefficient is now even bigger (0.33) and still significant at the 5% level. In columns 2 and 3, we include in the regressions the length of other judicial proceedings at court level, namely labor and criminal proceedings. The first variable is a direct measure of enforceability of labor contracts, while the second accounts for the fact that first instance courts also deal with criminal cases. The inclusion of the length of labor proceedings does not affect our main results. Moreover, the fact that the coefficient is not significant provides first evidence of the channels through which contract enforcement affects firm size. Although until the 2012 reform of labor legislation in Italy the length of judicial proceedings on worker dismissal directly translated into higher firing costs for larger firms,26 the ‘labour channel’ does not seem to be playing a relevant role.27 The coefficient of the length of criminal proceedings is instead significant and negative, however, the results on our main variable of interest are confirmed. In column 4, we measure the length of civil proceedings using an alternative index, originally suggested by Clark and Merryman (1976),28 based on the following formulation: D=Pt+FE−1 (6) where P are pending cases at the beginning of the year, F are the new cases filed during the year and E are the cases completed or withdrawn during the year. The index is averaged across the 6 years for which we have data (2002–2007).29 The new index is correlated at 97% with the previous one, and leads to almost identical results. Since an imprecise index may also introduce measurement error, and thus attenuation bias in the estimates, we also instrument the first index with the second. To the extent that the measurement error in the two indexes is uncorrelated, the IV strategy is consistent. The 2SLS coefficient, however, is only 10% larger (in absolute terms), suggesting that the component of the measurement error linked to the choice of the index is negligible. We also build an alternative measure based on the ratio of the number of judges in year 2005 over pending cases and we similarly use it as an instrumental variable to tackle a possible attenuation bias due to measurement error. The measure is also strongly (negatively) correlated with the main index of proceeding length, and again the second-stage results are just 10% bigger than in the baseline estimates (results are not reported for brevity). We also adopt an alternative measure of firm size based on employment, originally proposed by Davis and Henrekson (1997) and later adapted by Kumar et al. (1999) to data at the level of firm size bins, which places more emphasis on large firms. Following Kumar et al. (1999), we use an employee-weighted average size indicator (EWAS) that is calculated as follows: EWAS=∑bin=1n(empbinemptot)*(empbinfirmsbin) (7) where empbin and emptot refer to total employment in the plant size bin and in the sector, respectively, and firmsbin corresponds to the number of firms (or plants) in the size bin. The size bins we used are those originally defined in the ASIA archive: 1–9, 10–19, 20–49, 50–99, 100–199, 200–499, 500–999, 1000–4999 and more than 5000. This index calculates an ‘employee-weighted’ measure of average firm size that accounts for the fact that, in the light of the skewed firm size distribution, the share of employees working for large firms is disproportionately higher than the share of large firms. Results using the EWAS index are reported in Table 8, column 5 and show that the effect of the length of civil proceedings on large firms is even greater (the coefficient is twice as large as the one on average firm size). Another test relates to possible outliers due to the presence of a small number of extremely large firms in small municipalities (for instance, automotive industry plants in Italy are mainly located in small municipalities). We, therefore, exclude firms with more than 200 employees from the calculation of all the dependent variables. The results are similar to our main estimates, despite being less significant. A further point of concern is the variability in municipal populations. In our sample municipal population ranges from a minimum of 5062 inhabitants to a maximum of 2,545,860, with a standard deviation of 72,566 and a 95th percentile of 53,219. Dropping the municipalities in the top 5% of the population from the sample leaves the results unaffected. Weighting the restricted sample according to population gives very similar results, although they are less precise. A further test relates to newly established courts. As mentioned above, although the general shaping of court jurisdictions goes back to 1864, 11 small courts were added during the 1960s and the 1990s. One may worry that more recent courts would show endogenous boundaries: for instance, politically influential mayors may succeed in having their municipality included within the more efficient court, and their activism may also affect the growth of local firms. To test for this, we exclude from the sample all boundary groups involving a court created after 1960. Results are almost identical to those of the main regressions. Finally, the last test addresses the issue of different sectoral composition at the municipality level. The concern here is that part of the difference in average size across municipalities depends on the sectoral composition, which in turn could introduce noise into the estimates, or even a bias in the case in which the sectoral composition is systematically correlated with contract enforcement. The test is done by using as dependent variable a measure of firm average size at the municipality level that is ‘partialed-out’ of the effect of the sectoral composition. The measure is obtained by regressing the firm turnover aggregated at the municipality and three-digit industry level on a full set of municipality and three-digit industry dummies, and in retaining the values of the municipality dummies.30 The measure can be built for the turnover variable only, as the sectoral classification in the ASIA dataset is too coarse. The results obtained with the new dependent variable are reported in Table 9. The coefficients maintain their significance and are close in magnitude to those obtained from the baseline regressions. Table 9 Robustness: partialing-out the sector composition effect Dep. var. Sample (1) (2) (3) (4) Turnover level, av. 2008–2009 Turnover growth 2001–2009 Border Border Border Border Length civil −0.193* −0.232* −0.175** −0.144* (0.105) (0.123) (0.075) (0.085) Population 0.247*** 0.234*** −0.015 −0.024 (0.019) (0.025) (0.016) (0.020) Other controls No Yes No Yes Region FE No Yes No Yes Border FE Yes Yes Yes Yes Observations 967 967 967 967 R2 Dep. var. Sample (1) (2) (3) (4) Turnover level, av. 2008–2009 Turnover growth 2001–2009 Border Border Border Border Length civil −0.193* −0.232* −0.175** −0.144* (0.105) (0.123) (0.075) (0.085) Population 0.247*** 0.234*** −0.015 −0.024 (0.019) (0.025) (0.016) (0.020) Other controls No Yes No Yes Region FE No Yes No Yes Border FE Yes Yes Yes Yes Observations 967 967 967 967 R2 Notes: Robust standard errors clustered at court level in parentheses. All variables are in logarithms. The dependent variable is winsorized at 5% level on both tails before the logarithmic transformation. The sample is composed of municipalities with over 5000 inhabitants contiguous to a jurisdiction boundary for which data are available. ***p < 0.01; **p < 0.05; *p < 0.1. Table 9 Robustness: partialing-out the sector composition effect Dep. var. Sample (1) (2) (3) (4) Turnover level, av. 2008–2009 Turnover growth 2001–2009 Border Border Border Border Length civil −0.193* −0.232* −0.175** −0.144* (0.105) (0.123) (0.075) (0.085) Population 0.247*** 0.234*** −0.015 −0.024 (0.019) (0.025) (0.016) (0.020) Other controls No Yes No Yes Region FE No Yes No Yes Border FE Yes Yes Yes Yes Observations 967 967 967 967 R2 Dep. var. Sample (1) (2) (3) (4) Turnover level, av. 2008–2009 Turnover growth 2001–2009 Border Border Border Border Length civil −0.193* −0.232* −0.175** −0.144* (0.105) (0.123) (0.075) (0.085) Population 0.247*** 0.234*** −0.015 −0.024 (0.019) (0.025) (0.016) (0.020) Other controls No Yes No Yes Region FE No Yes No Yes Border FE Yes Yes Yes Yes Observations 967 967 967 967 R2 Notes: Robust standard errors clustered at court level in parentheses. All variables are in logarithms. The dependent variable is winsorized at 5% level on both tails before the logarithmic transformation. The sample is composed of municipalities with over 5000 inhabitants contiguous to a jurisdiction boundary for which data are available. ***p < 0.01; **p < 0.05; *p < 0.1. 4.6. Disentangling the different channels The results discussed so far provides a picture of the overall effects of contract enforcement on average firm size, but they are not informative on the relative importance of the different channels discussed in Section 1. In the following, we present some evidence on those, conditional on the limited data availability. Namely, we test the following channels: (i) relationship-specific investment, (ii) credit, and (iii) labor. The empirical strategy in this case is based on a difference-in-difference approach, close in spirit to the methodology proposed in a different context by Rajan and Zingales (1998), and widely used in many different empirical applications since then. The underlying intuition is that firms in different sectors are more or less exposed to the effects of specific channels in proportion to sector-specific characteristics. For instance, assuming that contract enforcement affects firm size through the credit channel, firms that are technologically more dependent on external finance would grow less in a jurisdiction with poor contract enforcement than other firms in the same jurisdiction—and less than firms that are equally dependent on external finance in jurisdictions with better contact enforcement. The comparison of these ‘differences in differences’ is made possible by the inclusion of sector and jurisdiction dummies, while the cross-sectional unit of observation varies at the industry-jurisdiction level. The model is estimated using balance sheet data, as they provide a more detailed sectoral classification. The data are aggregated to the level of jurisdiction and two-digit sector pair: ykj=α+β(IndVark*Ej)+κk+δj+εkj (8) where IndVar is a proxy of the industry exposure to a specific channel, E is the measure of contract enforcement, k is a sector fixed effect and δ is a jurisdiction fixed effect. The same model is estimated for two different dependent variables: the sector total turnover in 2008–2009, and the percentage turnover growth from 2001–2002 to 2008–2009; all values are expressed as 2-year averages and both the variables are expressed in logs. As with previous estimates, the analysis is limited to manufacturing. We include three industry variables (IndVar) in the analysis. The first one is the contract-intensity proxy developed in Nunn (2007), expressed as the fraction of input that is neither bought or sold on an exchange nor reference priced in a trade publication, and it is calculated on US data; in a given industry, the higher the contract intensity index, the more important are relationship-specific investments. The second variable is a proxy of dependency from external finance, calculated as the ratio of total debt over capital stock; it is meant to capture the importance of the credit channel. Finally, the third variable relates to the importance of labor in the production function of firms, and is calculated as the ratio of labor costs over net sales. The interaction variable definition and their sources are reported in Table 10.31 Consistently with the standard Rajan and Zingales (1998) approach, the two industry variables based on firm-level Italian data are calculated with reference to the ‘frictionless’ economy, which in our case is represented by the 5% most efficient jurisdictions; accordingly, those jurisdictions are excluded from the sample used to estimate Equation (8). As is conventional in this kind of differences-in-differences estimations, the interaction variables are initially included one by one, and subsequently the results are tested with a multivariate ‘horse-race’ regression that jointly includes all variables. Table 10 List of interaction variables Variable Definition Source Channel Contract Log of fraction of input that are neither bought Nunn (2007) Contract intensity and sold on an exchange nor reference based on US enforcement priced as being relationship-specific input–output tables Debts/ Log of total debts over External capital stock capital stock (attivo) CERVED finance Labor cost/ Log of total cost of labor over (see Section 4.2) EPL net sales net sales (ricavi netti) Variable Definition Source Channel Contract Log of fraction of input that are neither bought Nunn (2007) Contract intensity and sold on an exchange nor reference based on US enforcement priced as being relationship-specific input–output tables Debts/ Log of total debts over External capital stock capital stock (attivo) CERVED finance Labor cost/ Log of total cost of labor over (see Section 4.2) EPL net sales net sales (ricavi netti) Table 10 List of interaction variables Variable Definition Source Channel Contract Log of fraction of input that are neither bought Nunn (2007) Contract intensity and sold on an exchange nor reference based on US enforcement priced as being relationship-specific input–output tables Debts/ Log of total debts over External capital stock capital stock (attivo) CERVED finance Labor cost/ Log of total cost of labor over (see Section 4.2) EPL net sales net sales (ricavi netti) Variable Definition Source Channel Contract Log of fraction of input that are neither bought Nunn (2007) Contract intensity and sold on an exchange nor reference based on US enforcement priced as being relationship-specific input–output tables Debts/ Log of total debts over External capital stock capital stock (attivo) CERVED finance Labor cost/ Log of total cost of labor over (see Section 4.2) EPL net sales net sales (ricavi netti) The results of the regression of the total turnover level (Table 11) show that the three channels might play a role in determining a negative effect of poor contract enforcement on firm size, as—when included one by one—all the interaction coefficients are significant, and have a negative sign; however, the coefficients on external finance dependence and labor costs are not significant when the dependent variable is the total turnover growth over the period (Table 12). Furthermore, only the significance of the contract intensity coefficient survives in the ‘horse-race’ regressions with all the three variables included, as reported in column 4 of Tables 11 and 12. Table 11 Exploring the different channels—turnover level (1) (2) (3) (4) Dep. var. Total turnover 2008–2009 (log) CE × contract intensity −1.428** −1.297** (0.623) (0.626) CE × debts/capital stock −0.301** −0.139 (0.129) (0.211) CE × labor cost/net sales −0.342** −0.171 (0.141) (0.220) Two-digit industry FE Yes Yes Yes Yes Jurisdiction FE Yes Yes Yes Yes Observations 2996 2996 2996 2996 R2 0.669 0.669 0.669 0.668 (1) (2) (3) (4) Dep. var. Total turnover 2008–2009 (log) CE × contract intensity −1.428** −1.297** (0.623) (0.626) CE × debts/capital stock −0.301** −0.139 (0.129) (0.211) CE × labor cost/net sales −0.342** −0.171 (0.141) (0.220) Two-digit industry FE Yes Yes Yes Yes Jurisdiction FE Yes Yes Yes Yes Observations 2996 2996 2996 2996 R2 0.669 0.669 0.669 0.668 Notes: The reference unit is the jurisdiction-2 digit industry cell. CE stands for contract enforceability. Robust standard errors clustered at court level in parentheses. All variables are expressed in logarithmic form. ***p < 0.01; **p < 0.05; *p < 0.1. Source: See Table 3 and Nunn (2007). Table 11 Exploring the different channels—turnover level (1) (2) (3) (4) Dep. var. Total turnover 2008–2009 (log) CE × contract intensity −1.428** −1.297** (0.623) (0.626) CE × debts/capital stock −0.301** −0.139 (0.129) (0.211) CE × labor cost/net sales −0.342** −0.171 (0.141) (0.220) Two-digit industry FE Yes Yes Yes Yes Jurisdiction FE Yes Yes Yes Yes Observations 2996 2996 2996 2996 R2 0.669 0.669 0.669 0.668 (1) (2) (3) (4) Dep. var. Total turnover 2008–2009 (log) CE × contract intensity −1.428** −1.297** (0.623) (0.626) CE × debts/capital stock −0.301** −0.139 (0.129) (0.211) CE × labor cost/net sales −0.342** −0.171 (0.141) (0.220) Two-digit industry FE Yes Yes Yes Yes Jurisdiction FE Yes Yes Yes Yes Observations 2996 2996 2996 2996 R2 0.669 0.669 0.669 0.668 Notes: The reference unit is the jurisdiction-2 digit industry cell. CE stands for contract enforceability. Robust standard errors clustered at court level in parentheses. All variables are expressed in logarithmic form. ***p < 0.01; **p < 0.05; *p < 0.1. Source: See Table 3 and Nunn (2007). Table 12 Exploring the different channels—turnover growth (1) (2) (3) (4) Dep. var. Turnover growth 2001–2009 (log) Tot. turnover 2001–2002 (log) −0.313*** −0.312*** −0.312*** −0.313*** (0.0202) (0.0203) (0.0203) (0.0202) CE × contract intensity −0.795* −0.813* (0.434) (0.443) CE × debts/capital stock −0.0772 0.116 (0.0936) (0.150) CE × labor cost/net sales −0.130 −0.238 (0.102) (0.149) Two-digit industry FE Yes Yes Yes Yes Jurisdiction FE Yes Yes Yes Yes Observations 2996 2996 2996 2996 R2 0.322 0.323 0.322 0.322 (1) (2) (3) (4) Dep. var. Turnover growth 2001–2009 (log) Tot. turnover 2001–2002 (log) −0.313*** −0.312*** −0.312*** −0.313*** (0.0202) (0.0203) (0.0203) (0.0202) CE × contract intensity −0.795* −0.813* (0.434) (0.443) CE × debts/capital stock −0.0772 0.116 (0.0936) (0.150) CE × labor cost/net sales −0.130 −0.238 (0.102) (0.149) Two-digit industry FE Yes Yes Yes Yes Jurisdiction FE Yes Yes Yes Yes Observations 2996 2996 2996 2996 R2 0.322 0.323 0.322 0.322 Notes: The reference unit is the jurisdiction-2 digit industry cell. CE stands for contract enforceability. Robust standard errors clustered at court level in parentheses. All variables are expressed in logarithmic form. ***p < 0.01; **p < 0.05; *p < 0.1. Source: See Table 3 and Nunn (2007). Table 12 Exploring the different channels—turnover growth (1) (2) (3) (4) Dep. var. Turnover growth 2001–2009 (log) Tot. turnover 2001–2002 (log) −0.313*** −0.312*** −0.312*** −0.313*** (0.0202) (0.0203) (0.0203) (0.0202) CE × contract intensity −0.795* −0.813* (0.434) (0.443) CE × debts/capital stock −0.0772 0.116 (0.0936) (0.150) CE × labor cost/net sales −0.130 −0.238 (0.102) (0.149) Two-digit industry FE Yes Yes Yes Yes Jurisdiction FE Yes Yes Yes Yes Observations 2996 2996 2996 2996 R2 0.322 0.323 0.322 0.322 (1) (2) (3) (4) Dep. var. Turnover growth 2001–2009 (log) Tot. turnover 2001–2002 (log) −0.313*** −0.312*** −0.312*** −0.313*** (0.0202) (0.0203) (0.0203) (0.0202) CE × contract intensity −0.795* −0.813* (0.434) (0.443) CE × debts/capital stock −0.0772 0.116 (0.0936) (0.150) CE × labor cost/net sales −0.130 −0.238 (0.102) (0.149) Two-digit industry FE Yes Yes Yes Yes Jurisdiction FE Yes Yes Yes Yes Observations 2996 2996 2996 2996 R2 0.322 0.323 0.322 0.322 Notes: The reference unit is the jurisdiction-2 digit industry cell. CE stands for contract enforceability. Robust standard errors clustered at court level in parentheses. All variables are expressed in logarithmic form. ***p < 0.01; **p < 0.05; *p < 0.1. Source: See Table 3 and Nunn (2007). Therefore, although this empirical approach is somewhat weaker in terms of identification of causal links, overall these results provide suggestive evidence that the effect of poor contract enforcement on firm size mainly stems from the discouragement of relationship-specific investments. 5. Conclusion We explore the effect of contract enforceability through courts on the size of firms. Since theory does not ultimately provide an answer regarding the expected sign of the relationship, we resort to empirics to shed light on the subject. Improving on the existing literature, we address the identification and causality issues by applying a spatial discontinuity design to Italian municipalities, exploiting the fact that court jurisdictions were shaped in the 19th century and do not systematically match political and administrative geography. More specifically, we compare average firm size across contiguous municipalities that are located on the boundaries of court jurisdictions. This allows us to isolate the effects of contract enforcement, as municipalities on opposite sides of court boundaries experience a discrete jump in this variable, but not in other unobserved factors. We find that in municipalities where contract enforcement is poor, average firm size in manufacturing industries is smaller. We also find that contract enforceability has a negative effect on firm growth. The economic impact on firm size and on firm growth have a remarkably similar magnitude, suggesting that most of the effect is at the intensive margin. This is consistent with previous evidence from Duranton et al. (2011), who found that local factors—in their case taxes—are capitalized into rents for the incumbents only marginally, due to mobility constraints. On the contrary, new entrants are perfectly mobile and need to negotiate rents; therefore, the capitalization of local factors is higher for them. Given that we were assessing the effect of contract enforcement in the local court jurisdiction, our results may be interpreted as a lower bound of the effect of contract enforceability warranted by the national judicial system as a whole. This is because, not all the disputes involving a firm in a given jurisdiction area are dealt with at the local court. Although our data do not allow for direct testing of the channels through which contract enforcement affects firm size, we provide evidence that the effect mainly comes through the disincentives of relationship-specific investments. Our results also inform the debate on industrial and local development policy. Indeed, we show that policies aimed at improving the functioning of the judicial system, ensuring better contract enforceability, can significantly boost average firm size and total employment at the local level in the manufacturing sector. This can, therefore, be an example of a potentially successful horizontal and centrally planned policy, having large effects on employment and economic growth at the local level. Footnotes 1 This has been confirmed by a large body of empirical literature. It has been shown that judicial efficiency affects the development of financial and credit markets (Djankov et al., 2008), the availability and cost of credit (e.g. Qian and Strahan, 2007; Bae and Goyal, 2009; Fabbri, 2010; Ponticelli and Alencar, 2016), the volume of trade (Berkowitz et al., 2006), sectoral specialization (Nunn, 2007) and competition in markets (Johnson et al., 2002). 2 Kumar et al. (1999), using data on firm size in Western European countries, found that better judicial systems are associated with larger average firm size; the effects are bigger for industries where physical assets are less important. Beck et al. (2006), using firm-level data on the largest industrial firms in 44 countries, found that firm size is positively associated with institutional development (including judicial efficiency) and with the development of financial intermediaries. Laeven and Woodruff (2007), using firm census data in Mexico, showed that judicial efficiency has a positive link with average firm size and that this effect is larger for proprietorships than for corporations. Similar results were obtained by Fabbri (2010) on Spanish data; she found that more efficient courts are associated with larger firms and less costly bank financing. More recent empirical papers on the same topic using, respectively, Spanish and Mexican data are Garcia-Posada and Mora-Sanguinetti (2015) and Dougherty (2014). 3 The methodology presents some important differences compared to the more commonly implemented Regression Discontinuity Design (RDD) and its geographic applications that are discussed in the text. For previous applications of the spatial discontinuity design, see, among others, Black (1999), Holmes (1998) and Duranton et al. (2011). 4 Another way to minimize the transaction costs determined by inadequate contract enforcement institutions identified in the literature is to rely on relational contracting. Where courts are weak, reputation embedded in long-term bilateral relationships may be used as a device to ensure that contracts are enforced. The more the parties rely on relational contracting, the less willing they are to work with new contractual partners. This reduces the demand for a given firm’s output and hinders its growth. Also in this case, the overall effect on average firm size is ambiguous as relational contracting also creates barriers to entry for new firms that are usually smaller than incumbent firms, thus reducing average firm size (Johnson et al., 2002). Our data do not allow us to investigate this channel. 5 A theoretical model in which courts efficiency influences firm size through the credit channel is in Fabbri (2010). 6 As pointed out by the literature on the strictness of employment protection, higher firing costs may influence firm size (Schivardi and Torrini, 2008; Braguinsky et al., 2011). 7 Black (1999) applies the methodology to housing prices as a function of school quality in the USA by comparing the difference in prices of similar neighbouring houses located on different sides of schools districts. Holmes (1998) exploits spatial variation in US state legislations to find that there is a large increase in manufacturing activity when one crosses a state border from a state that has a right-to-work legislation into a state without such legislation. Duranton et al. (2011) use municipality boundaries in the UK to investigate the effect of local taxation on firm location and growth. 8 Over the last 50 years, 11 new small courts have been established (5 in the 1960s and 6 in the 1990s). As we will discuss later, our results are unaffected by excluding these courts. This system was widely recognized to be inefficient and anachronistic due to the presence of a multitude of very small courts that might have been necessary in the past when transport and communication infrastructures were underdeveloped to ensure access to justice, but which are now no longer justified. After several attempts to reform it, and despite strong opposition at local level, the system was finally fully revised in 2013 (Legislative Decree No. 155/2012 of 9 September provides for the reduction in the number of courts from 165 to 134 starting from September 2013). 9 Regions and provinces are the administrative territorial units that correspond, respectively, to level 2 and level 3 in the Eurostat NUTS classification. 10 Their main tasks are related to environmental protection, road maintenance, school buildings (construction and maintenance) and waste disposal. Nevertheless, in the robustness section, we exclude from the sample all court jurisdictions whose boundaries coincide with the provincial ones. 11 Anecdotal evidence suggests that this is seldom done in standard contracts in Italy. In the case in which the parties agreed to solve the dispute in a more efficient court, our estimates would be biased toward zero. 12 A more in depth discussion can be found in Palumbo et al. (2013). The length of proceeding is used as a proxy for court efficiency, for instance, in Fabbri (2010). 13 This index, first developed to estimate the average turnover of stock inventories, has been proposed in the literature (Clark and Merryman, 1976) and it is frequently used, especially in cross-country comparisons, to estimate the length of proceedings when actual data are not available (Palumbo et al., 2013). In the robustness section, we show that our results are consistent with the use of a different version of the index. 14 Unfortunately, disaggregated data for each of these subjects are not available. 15 The reason for this choice is that the average firm size in the service sector is very small (around three employees) and the variation of the variable is limited. 16 Enterprise-level results are available from the authors upon request. Using plant-level data rather than firm-level data might introduce some correlation structure in the error term, but given that the large majority of Italian firms are single-plant, the resulting bias in the computation of the standard errors should be negligible. 17 Moreover, the CERVED database contains detailed information on the industry in which firms operate that we will exploit in Section 5.4 to investigate the channels through which contact enforcement affects firm size. 18 It is worth stressing that—regrettably—the CERVED dataset does not allow calculating any meaningful measure of productivity, as it does not report the number of employees, nor prices or any measures of capital stock depreciation. Attempts to build productivity proxies based on ratios of value added over labor cost or turnover produced very imprecise estimates. Therefore, although assessing the effect of judicial efficiency on productivity would be an interesting extension of our analysis, available data unfortunately hinders our ability to do so. 19 We average the value of turnover over a 2-year period to smooth short-term disturbances. 20 In the CERVED archive there is not a plant identifier, only a firm identifier. Therefore, it is not possible to correctly calculate an individual contribution of different plants to firm turnover. 21 Given that this measure is the only control variable not available for 91 municipalities above 5000 inhabitants, we replace the missing values with the sample average in order not to restrict the estimation sample. Results based on the alternative sample are almost identical and only slightly less precise, and are available from the authors upon request. 22 The only difference is that we now fully exploit the class-level microdata on cheating, rather than municipality-level averages, in order to make the test more precise. 23 A speculative explanation of the positive bias of the full sample regression is the presence of industrial districts associated with a smaller average firm size. To the extent that industrial districts are more frequent in areas with higher endowments of civicness and better performing institutions, this may explain the bias toward zero. 24 We are grateful to an anonymous referee for pointing out this issue. 25 A similar test has been employed by Einiö and Overman (2016) in the context of a spatial discontinuity exercise assessing the effect of an area-based intervention in the United Kingdom. 26 The law provided that if a dismissal was ruled to be unfair, firms with more than 15 employees had to compensate employees for the forgone wages over the time elapsed between the dismissal and the court’s decision (besides the reinstatement of the employee, unless she or he opts for a further severance payment equal to 15 months salary). 27 Our results are consistent with Schivardi and Torrini (2008) who find that workers’ dismissal provisions that apply to firms with more than 15 employees have quantitatively modest effects on the size distribution of Italian firms. See also Section 4.6. 28 In proposing this variation of the index, the authors aimed to overcome the logical flaw of the one described in Equation (5): the length estimated according to that index decreases when the number of new cases filed increases. 29 More precisely, we summed all the components over the whole period and then we calculated the index on the aggregate figures. 30 Given the highly skewed distribution of these variables, they are winsorized at the 5% level on both tails. 31 The two industry measures are calculated as the within-industry average of firm-level values. Since the large majority of firms in the dataset are micro-firms for which balance sheet data are often noisy, to improve its precision the within-sector average is weighted by firm total sales, in order to give more importance to larger firms reporting cleaner information. Acknowledgments The opinions expressed in the article are those of the authors and do not necessarily reflect the views of the Bank of Italy and the OECD and its member countries. We are particularly grateful to the editor Kristian Behrens for a number of very good suggestions. We also thank Giuseppe Berlingieri, Francesco Caselli, Federico Cingano, Domenico Depalo, Juan Francisco Jimeno, Francesco Manaresi, Enrique Moral, Henry Overman, Miguel García Posada, Lucia Rizzica, Giacomo Rodano, Alfonso Rosolia, Juan Mora Sanguinetti, Paolo Sestito, Marco Tonello, two anonymous referees and participants to LSE-SERC (London), Bank of Italy (Rome), Bank of Spain (Madrid), and OFCE-SKEMA (Nice) seminars and to the Urban Economics Sessions of the ERSA conference in Barcelona (2011) and the EPCS conference in Zurich (2013) for their useful comments. References Andrews D. , Cingano F. ( 2014 ) Public policy and resource allocation: evidence from firms in OECD countries . Economic Policy , 29 : 253 – 296 . Google Scholar Crossref Search ADS WorldCat Bae K.-H. , Goyal V. K. ( 2009 ) Creditor rights, enforcement, and bank loans . Journal of Finance 64 : 823 – 860 . Google Scholar Crossref Search ADS WorldCat Bartelsman E. , Haltiwanger J. , Scarpetta S. ( 2013 ) Cross-country differences in productivity: the role of allocation and selection . American Economic Review , 103 : 305 – 334 . Google Scholar Crossref Search ADS WorldCat Basten C. , Betz F. ( 2013 ) Beyond work ethic: religion, individual, and political preferences . American Economic Journal: Economic Policy , 5 : 67 – 91 . Google Scholar Crossref Search ADS WorldCat Beck T. , Demirg-Kunt A. , Maksimovic V. ( 2006 ) The influence of financial and legal institutions on firm size . Journal of Banking and Finance , 30 : 2995 – 3015 . Google Scholar Crossref Search ADS WorldCat Becker S. O. , Boeckh K. , Hainz C. , Woessmann L. ( 2016 ) The empire is dead, long live the empire! Long-run persistence of trust and corruption in the bureaucracy . Economic Journal , 126 : 40 – 74 . Google Scholar Crossref Search ADS WorldCat Berkowitz D. , Moenius J. , Pistor K. ( 2006 ) Trade, law, and product complexity . Review of Economics and Statistics , 88 : 363 – 373 . Google Scholar Crossref Search ADS WorldCat Bianco M. , Giacomelli S. , Giorgiantonio C. , Palumbo G. , Szego B. ( 2007 ) La durata (eccessiva) dei procedimenti civili in italia: offerta, domanda o rito? Rivista di Politica Economica , 47 : 3 – 53 . WorldCat Black S. E. ( 1999 ) Do better schools matter? Parental valuation of elementary education . Quarterly Journal of Economics , 114 : 577 – 599 . Google Scholar Crossref Search ADS WorldCat Braguinsky S. , Branstetter L. G. , Regateiro A. ( 2011 ) The incredible shrinking Portuguese firm. Working Paper No. 17265, National Bureau of Economic Research. Clark D. , Merryman J. ( 1976 ) Measuring the duration of judicial and administrative proceedings . Michigan Law Review , 75 : 89 – 99 . Google Scholar Crossref Search ADS WorldCat Davis S. J. , Henrekson M. ( 1997 ) Explaining national differences in the size and industrial distribution of employment. Research Institute of Industrial Economics Working Papers No. 482, Research Institute of Industrial Economics (IFN). Dell M. ( 2010 ). The persistent effects of Peru’s mining mita . Econometrica , 78 : 1863 – 1903 . Google Scholar Crossref Search ADS WorldCat Disney R. , Haskel J. , Heden Y. ( 2003 ) Restructuring and productivity growth in UK manufacturing . Economic Journal , 113 : 666 – 694 . Google Scholar Crossref Search ADS WorldCat Djankov S. , Hart O. , McLiesh C. , Shleifer A. ( 2008 ) Debt enforcement around the world . Journal of Political Economy , 116 : 1105 – 1149 . Google Scholar Crossref Search ADS WorldCat Dougherty S. M. ( 2014 ) Legal reform, contract enforcement and firm size in Mexico . Review of International Economics , 22 : 825 – 844 . Google Scholar Crossref Search ADS WorldCat Duranton G. , Gobillon L. , Overman H. G. ( 2011 ) Assessing the effects of local taxation using microgeographic data . Economic Journal , 121 : 1017 – 1046 . Google Scholar Crossref Search ADS WorldCat Einiö E. , Overman H. G. ( 2016 ) The (displacement) effects of spatially targeted enterprise initiatives: evidence from UK LEGI. Discussion Paper No. 191, Spatial Economics Research Centre. Fabbri D. ( 2010 ). Law enforcement and firm financing: theory and evidence . Journal of the European Economic Association , 8 : 776 – 816 . Google Scholar Crossref Search ADS WorldCat Garcia-Posada M. , Mora-Sanguinetti J. S. ( 2015 ) Does (average) size matter? Court enforcement, business demography and firm growth . Small Business Economics , 44 : 639 – 669 . Google Scholar Crossref Search ADS WorldCat García-Santana M. , Ramos R. ( 2015 ) Distortions and the size distribution of plants: evidence from cross-country data . SERIEs , 6 : 279 – 312 . Google Scholar Crossref Search ADS WorldCat Garicano L. , LeLarge C. , Reenen J. V. ( 2016 ) Firm size distortions and the productivity distribution: evidence from France . American Economic Review , forthcoming. WorldCat Holmes T. J. ( 1998 ) The effect of state policies on the location of manufacturing: evidence from state borders . Journal of Political Economy , 106 : 667 – 705 . Google Scholar Crossref Search ADS WorldCat Hsieh C.-T. , Klenow P. J. ( 2014 ) The life cycle of plants in India and Mexico . Quarterly Journal of Economics , 129 : 1035 – 1084 . Google Scholar Crossref Search ADS WorldCat Johnson S. , McMillan J. , Woodruff C. ( 2002 ) Courts and relational contracts . Journal of Law, Economics, and Organization , 18 : 221 – 277 . Google Scholar Crossref Search ADS WorldCat Kumar K. B. , Rajan R. G. , Zingales L. ( 1999 ) What determines firm size? Working Paper No. 7208, National Bureau of Economic Research. Laeven L. , Woodruff C. ( 2007 ) The quality of the legal system, firm ownership, and firm size . Review of Economics and Statistics , 89 : 601 – 614 . Google Scholar Crossref Search ADS WorldCat Michelacci C. , Silva O. ( 2007 ) Why so many local entrepreneurs? Review of Economics and Statistics , 89 : 615 – 633 . Google Scholar Crossref Search ADS WorldCat Nunn N. ( 2007 ) Relationship-specificity, incomplete contracts, and the pattern of trade . Quarterly Journal of Economics , 122 : 569 – 600 . Google Scholar Crossref Search ADS WorldCat Paccagnella M. , Sestito P. ( 2014 ) School cheating and social capital. Temi di discussione (Economic Working Papers) 952, Bank of Italy, Economic Research and International Relations Area. Palumbo G. , Giupponi G. , Nunziata L. , Sanguinetti J. S. M. ( 2013 ) The economics of civil justice: new cross-country data and empirics. Economics Department Working Paper No. 1060, OECD. Ponticelli J. , Alencar L. S. ( 2016 ) Court enforcement, bank loans, and firm investment: evidence from a bankruptcy reform in Brazil . The Quarterly Journal of Economics , 131 : 1365 – 1413 . Google Scholar Crossref Search ADS WorldCat Qian J. , Strahan P. E. ( 2007 ) How laws and institutions shape financial contracts: the case of bank loans . Journal of Finance , 62 : 2803 – 2834 . Google Scholar Crossref Search ADS WorldCat Rajan R. G. , Zingales L. ( 1998 ) Financial dependence and growth . American Economic Review , 88 : 559 – 586 . WorldCat Schivardi F. , Torrini R. ( 2008 ) Identifying the effects of firing restrictions through size-contingent differences in regulation . Labour Economics , 15 : 482 – 511 . Google Scholar Crossref Search ADS WorldCat Williamson O. E. ( 1979 ) Transaction-cost economics: the governance of contractual relations . Journal of Law and Economics , 22 : 233 – 261 . Google Scholar Crossref Search ADS WorldCat World Bank ( 2013 ) Doing Business in Italy 2013 . Washington, DC : World Bank Group . WorldCat COPAC World Bank ( 2015 ) Doing Business 2016. Measuring Regulatory Quality and Efficiency . Washington, DC : World Bank Group . WorldCat COPAC © The Author (2016). Published by Oxford University Press. All rights reserved. For Permissions, please email: journals.permissions@oup.com
TI - Does weak contract enforcement affect firm size? Evidence from the neighbour’s court
JF - Journal of Economic Geography
DO - 10.1093/jeg/lbw030
DA - 2017-11-01
UR - https://www.deepdyve.com/lp/oxford-university-press/does-weak-contract-enforcement-affect-firm-size-evidence-from-the-yUh49JG07v
SP - 1251
VL - 17
IS - 6
DP - DeepDyve
ER -