Conflicting Security Laws and The Democratization of Credit: France’s Reform of the Napoleonic Code* Kevin Aretz Manchester Business School kevin.aretz@mbs.ac.uk Murillo Campello Cornell University & NBER campello@cornell.edu Maria-Teresa Marchica Manchester Business School maria.marchica@mbs.ac.uk This Draft: May 15, 2015 Abstract We exploit contradictions in a recent security law reform in France (Ordonnance 2006-343 ) to study how access to collateral shapes the composition of corporate debt, the distribution of debt across firms, and capital allocation efficiency. Ordonnance 2006-343 modernized France’s 200-year old security law system, in place since Napoléon Bonaparte, and it facilitated asset pledgeability in a country with a highly unequal access to corporate debt. Despite this, the reform was undermined by “non-codified laws,” introduced in the 1980s due to concerted lobbying by big firms wishing to pledge their liquid assets (receivables and inventory). Using a triple-differences strategy, we show that firms with the highest demand for fixed assets increased the use of debt financing the most after the reform (intensive margin). Also, the fraction of “zero-leverage” firms among them dropped from 88% to 32% (extensive margin). Using contract-level data, we further find that fixed assets allowed for significant reductions in loan mark-ups and increases in debt maturity after the reform. These effects stem from the use of secured debt and affect only private firms. Small, profitable, low risk-firms and start-ups benefitted the most from the reform. Our geographical analysis shows that the reform benefitted firms in rural areas, and that it led to a pronounced decline in the Gini-index of debt concentration across the whole of France. Finally, we show that the reform increased capital allocation efficiency. All together, our results suggest that the 2006 reform led to a “corporate debt democratization.” Key words: Security Laws, Contractibility, Collateral, Capital Structure, Social Welfare. JEL classification: G32, K22, O16. *We are grateful to Marie-Elodie Ancel from Université Paris Est Créteil, Rod Cork from Allen & Overy (Paris), and Philip Wood from Allen & Overy (London) for insightful discussions about the French legal system. Conflicting Security Laws and The Democratization of Credit: France’s Reform of the Napoleonic Code Abstract We exploit contradictions in a recent security law reform in France (Ordonnance 2006-343 ) to study how access to collateral shapes the composition of corporate debt, the distribution of debt across firms, and capital allocation efficiency. Ordonnance 2006-343 modernized France’s 200-year old security law system, in place since Napoléon Bonaparte, and it facilitated asset pledgeability in a country with a highly unequal access to corporate debt. Despite this, the reform was undermined by “non-codified laws,” introduced in the 1980s due to concerted lobbying by big firms wishing to pledge their liquid assets (receivables and inventory). Using a triple-differences strategy, we show that firms with the highest demand for fixed assets increased the use of debt financing the most after the reform (intensive margin). Also, the fraction of “zero-leverage” firms among them dropped from 88% to 32% (extensive margin). Using contract-level data, we further find that fixed assets allowed for significant reductions in loan mark-ups and increases in debt maturity after the reform. These effects stem from the use of secured debt and affect only private firms. Small, profitable, low risk-firms and start-ups benefitted the most from the reform. Our geographical analysis shows that the reform benefitted firms in rural areas, and that it led to a pronounced decline in the Gini-index of debt concentration across the whole of France. Finally, we show that the reform increased capital allocation efficiency. All together, our results suggest that the 2006 reform led to a “corporate debt democratization.” Key words: Security Laws, Contractibility, Collateral, Capital Structure, Social Welfare. JEL classification: G32, K22, O16. 1 Introduction Over the last years, many countries have reformed their security laws, a trend that is expected to continue as more economies converge to the global capital market. Economic policy-makers promote these reforms under the banner of “enhancing access to credit,” and they view them as a means to the ultimate goal of reducing discriminations in access to credit. Thus, the current reforms are less about the ability of large, established firms to tap debt markets — and more about the growth and value that can be unlocked by easing credit access across a larger number of firms (“debt democratization”). In particular, the current reforms are targeted at profitable, but financially-constrained firms, firms located outside of large urban areas, younger firms, and firms in less established but growing sectors of the economy. Reducing discriminations in access to corporate debt is a key factor since in most countries smaller companies employ more than half of all employees and create more than half of all value-added (Infelise (2014)). Despite the comprehensive goals of most security law reforms, recent academic work has primarily focused on whether these reforms have raised total corporate debt (see Lilienfeld-Toal et al. (2012), Vig (2013), and Campello and Larrain (2014)). A possible reason for this narrow focus may be the limited availability of information on access to credit in those countries analyzed in other studies: namely developing countries. In this paper, we circumvent this problem by studying a security law reform recently implemented by a developed country. This reform is Ordonnance 2006-346, enacted in France in 2006. For France, we have data on important dimensions related to credit access and economic outcomes that are unavailable for developing countries. Another advantage of studying France is that, despite the country’s motto promoting equality (“Liberté, Égalité, Fraternité”), access to credit in France was heavily concentrated in the hands of a few large firms before the reform. For example, before the reform, 90% of the private firms in our sample held no long-term debt and the average Gini index of long-term debt concentration was around 0.95. Given these numbers, it is hardly surprising that an explicit goal of the reform was to enable financially-constrained firms to borrow.1 Ordonnance 2006-346 derogated the “possessory” nature of asset ownership, in place since Napoléon Bonaparte. In addition, it also allowed for the creation of security interests over types 1 Indeed, in association with Ordonnance 2006-343, the Bank of France declared that “reforms in this area could have a significant effect on the distribution of credit, especially in case of borrowers that are currently excluded [from credit markets].” 1 of assets which could hitherto not be pledged, as, for example, fungible or future assets, and of security interests that can be charged to more than one creditor. Despite the fundamental changes made by Ordonnance 2006-346, several “non-codified” security laws already existed in France long before the reform. Of these non-codified laws, the two most popular were the Dailly Law and the Pledge of Ready Money. The Dailly Law and the Pledge of Ready Money allowed borrowers to pledge cash, cash equivalents, and receivables to financial institutions to secure credit. Because of the existence and modern flavor of these two laws, we reason that Ordonnance 2006-346 had a more pronounced effect on those firms whose asset structures are tilted more towards fixed rather than current assets. Thus, the coexistence of these “contradictory” laws allows us to define an identification strategy in which the treated firms are those heavily using fixed assets, and the controls are those heavily using current assets. Our empirical tests analyze both firm-level (quantity) and contract-level (price) information within a triple-differences framework. As a starting point, our tests show that the 2006 collateral reform profoundly changed the magnitude and composition of the corporate debt held by private French firms. Following the reform, the mean total debt-to-assets ratio increases from a pre-reform average of 9.1% to a post-reform average of 12.3%. Notably, this increase is only pronounced among those firms that are expected to be most affected by the reform: firms whose assets are mostly composed of land, buildings, machinery, and equipment. In particular, while firms in the highest fixed asset quartile observe an average increase in total leverage of 7.2%, the lowest fixed asset quartile-firms observe no significant changes. We look at debt maturity to better identify our results. Short-term debt is less likely to be secured than long-term debt (see Diamond (1991) for theoretical arguments). Crucially, even if short-term debt is secured, it tends to be secured with current — and not fixed — assets (Ross et al. (2012)). Conforming with these observations, our evidence shows that the total leverage increases are entirely driven by increases in long-term leverage. Specifically, while the long-term leverage of high-fixed asset usage firms rises by 7.8%, the long-term leverage of low-fixed asset usage firms rises by only 1.2%. By comparison, high- and low-fixed asset usage firms observe statistically indistinguishable drops in their short-term leverage. A triple differences (DIDID) estimate of the local effect of Ordonnance 2006-346 on firm debt can be computed as the spread in the long-term debt change between high- and low-fixed asset firms (7.8%–1.2%) minus the 2 corresponding spread in the short-term debt change (–4.0%–(–2.7%)), which is 7.9%. In addition to debt levels (the intensive margin), we also study the propensity of firms to take out any long-term debt at all as a result of the reform (the extensive margin). As stated before, before 2006, a whooping 90% of all firms have no long-term leverage (“zero-leverage firms”). However, during the reform year alone, this number drops to 47%. It falls further to 38% at the end of our sample period. Notably, these changes are mostly driven by high-fixed asset firms. In particular, while the proportion of zero leverage firms in the highest fixed asset quartile declines from 88% to 32% (a 56% drop), the proportion of zero leverage firms in the lowest fixed asset quartile declines from 91% to a mere 64% (a 27% drop). We also identify changes in debt contract terms that facilitated the credit expansion. To do so, we look at detailed loan contract data. In these tests, we use a more direct identification strategy. We assign to the treatment group those loans that are directly affected by the reform (secured loans) and to the control group those loans that are not directly affected by it (unsecured loans). We show that secured loans become both cheaper and more long-term compared to unsecured loans after the reform. More specifically, we find that the mark-ups of secured loans decrease by around 1.2% relative to those of unsecured loans. In addition, the times-to-maturity of secured loans increase by around 41 months relative to those of unsecured loans. As a next step, we examine what type of firms were able to enter the corporate debt market due to the reform. In doing so, we first look at firms that never held any long-term debt before the reform. We split these firms into those that had a positive long-term leverage ratio in every post-reform year (“debt switchers”) and those that continued to have zero long-term leverage in every post-reform year (“non-switchers”). Looking at pre-reform firm characteristics, the debt switchers are, on average, smaller, more mature firms with more employees than the non-switchers. Notably, the debt switchers have higher and more stable profits, hold more cash, and are located further away from the largest French cities than the non-switchers. Moreover, we report that the reform eased start-up’s access to credit. The mean long-term leverage ratio of start-ups in their year of incorporation more than doubles from a pre-reform average of 2.7% to a post-reform average of 5.8%. Also, the proportion of start-ups with no long-term debt in their year of incorporation decreases from an average of 90% to an average of 74%. We use data at the Departmental level to investigate the geographic effects of the collateral 3 reform.2 Our evidence shows that the greatest increases in long-term borrowing occurred in rural areas in the lower center of the country (close to Limousin). Consistent with these findings, landrich high fixed asset-usage firms observed significantly higher increases in long-term leverage than other high fixed asset-usage firms. We use the Gini index to more formally study whether the reform reduced inequalities in long-term debt usage. Before the reform, more than nine out of ten French Departments have a Gini index of long-term debt concentration exceeding 0.90, indicating that debt markets are heavily dominated by a handful of major players. After the reform, the Gini index drops to an average of around 0.70. Critically, the largest drops are again observed in the rural Departments close to the lower center of the country. We use court data at the Departmental level to link the effects of the 2006 reform to court efficiency. Because the French reform does not allow creditors to bypass inefficient courts (e.g., by seizing collateral out-of-court), we anticipate that the reform had more pronounced effects in Departments with more efficient courts. Our evidence partially supports this hypothesis. While the length of insolvency proceedings does not condition the magnitude with which high fixed asset-usage firms increase their long-term borrowing, it significantly lowers the proportion of firms that start to use long-term debt as a result of the collateral reform. Looking at the real-side effects of the reform, our tests show that high-fixed asset firms raised their investment levels more significantly than low-fixed asset firms after the reform. They also hired more people. Our evidence also suggests that the greater increases in capital and labor enabled high-fixed asset firms to pursue profitable investments with low risk. Critically, the changes in profitability and risk experienced by high-fixed assets firms had a more profound effect on their failure rates than their higher leverage ratios. To be exact, while high-fixed asset firms were more likely to fail than low-fixed asset firms before 2006, this relationship reversed after 2006. Thus, taking on more debt seems to be ultimately welfare-enhancing. Finally, we use industry-level data to study the elasticity between capital growth and valueadded. Our tests show that this elasticity increases significantly from a pre-reform level of 0.43 to a post-reform level of 0.70. Following Wurgler’s (2000) interpretation of this elasticity, our evidence suggests that the reform increased capital allocation efficiency in France. Consistent with this evidence, we also provide evidence suggesting that the collateral reform raised long2 The Department is the administrative division between the region and the commune. 4 term borrowing more strongly in industries that are highly dependent on external financing, possibly implying that the reform helped to lower financial constraints. We run a number of robustness tests to verify our results. First, we show that high- and low-fixed asset firms display similar underlying trends in total and long-term leverage before the reform; the same applies to covariates such as size, profitability, or cash flow volatility. Because public firms only borrow unsecured, we focus on private firms in our main tests. As such, as a falsification, we conduct tests using public firms. We confirm that public firms fail to produce any meaningful patterns in debt taking behavior across firms with different levels of fixed asset usage. Finally, we repeat our tests on “placebo countries.” We choose as placebo countries France’s most important civil law neighbors, Belgium, Italy, Portugal, and Spain. Neither of these countries passed a law similar to Ordonnance 2006-343 in our sample period. Likewise, neither of them experienced effects similar to those found in France.3 Our study adds to a large literature investigating the effects of creditor rights on economic outcomes. Since LaPorta et al. (1998), many cross-country studies show that stronger creditor rights raise capital availability and enhance economic and financial development (e.g., Djankov et al. (2007), Qian and Strahan (2007), and LaPorta et al. (2008)). Despite these findings, studies using recent security law reforms as quasi-natural experiments find contradicting evidence. For example, while Haselmann et al. (2009) and Campello and Larrain (2014) show that reforms in Eastern Europe increased lending to high quality borrowers, reforms in India and Brazil either decreased lending or increased lending to the wrong type of borrowers (see Lilienfeld-Toal et al. (2012), Vig (2013), and Assunção et al. (2014)). Our study adds to the above studies in two ways. First, it provides further evidence that security law reforms can lead to an expansion of credit. More importantly, however, it shows that security law reforms can have important distributional implications. In particular, the French reform democratized access to credit by allowing young, profitable, previously financiallyconstrained firms in rural areas to borrow long-term, in this way reducing discriminations in access to debt financing. It also led to an increase in capital allocation efficiency. Our article is organized as follows. Section 2 gives details about the 2006 French security law reform. In Section 3, we outline our data and our identification strategy. Section 4 analyzes 3 We also collapse the pre-reform and post-reform data to correct standard errors for autocorrelation in outcome variables (Bertrand et al. (2004)). 5 the effect of the reform on debt financing. Section 5 identifies those firms that most benefitted from the reform. Section 6 looks at geographical effects of the reform. In Section 7, we ask to what use the firms that most strongly increased their borrowing put the newly available debt capital. Section 8 offers robustness tests. Section 9 concludes. 2 Institutional Setting: Security Laws in France 2.1 The Napoleonic Code of 1804 Prior to the recent reform, security interests in France were governed by the 1804 Napoleonic Civil Law Code, whose sections dealing with security interests remained essentially unchanged over the last 200 years. The Napoleonic Code was introduced during a time when France was recovering from the tumults of the French Revolution. The Revolution transformed France back into an agricultural country, in which 90% of all property consisted of land. The French legal system recognizes two forms of security interests, the hypothec and the charge. A hypothec can only be taken out over real property (land and fixtures), while a charge can be taken out over all types of assets.4 Given its great importance, security laws at the time focused on security interests over land — security interests over other assets were deemed as of little importance (Omar (2007)). Also, the chaos of the revolutionary years convinced regulators that both borrowers and creditors had to be put into strong positions. In particular, borrowers had to be protected from exploitive creditors, while creditors had to be protected from borrowers refusing the surrender of collateral in default (Ancel (2008)). To this end, regulators established a highly formal, costly procedure to create security rights. Also, they refused to recognize rechargeable hypothecs (which could lead to conflicts about priority) or non-possessory charges (which could make it easier for borrowers to not surrender collateral). Ruling out non-possessory charges implied that security interests over fungible or future assets could also not be allowed because such assets cannot be physically transferred to creditors. To further protect borrowers, regulators explicitly forbade out-of-court seizures of collateral in default. The Napoleonic Code did not promote the creation of a security interest registry covering assets other than land. To compensate for this deficiency, various local registries emerged over 4 See Boughida et al. (2011) for more details about French security interests. 6 Figure 1. Characteristics of the Old and the New French Security Law Regime This table compares the old French security law regime based on the 1804 Napoleonic Civil Law Code with the new regime based on Ordonnance no. 2006-346, enacted on 23 March 2006. The asterisk indicates that the Napoleonic Civil Law Code allowed for non-possessory security rights over real property (land and fixtures). time, each covering a narrow range of assets and each restricted to local borrowers. Creditors had to travel to these registries in person to make inquires (World Bank (2007)). 2.2 Reforming the Napoleonic Code: Ordonnance 2006-346 In the early 2000s, French politicians became increasingly aware of the fact that French security laws were less competitive than those of other countries, especially those of the so-called common law countries (Renaudin (2013)). As a result, in March 2005, President Jacques Chirac called upon the Minister of Justice to reform the country’s security laws. In April 2005, a working group set up by the Minster of Justice handed in a set of recommendations about reforming security rights, and in March 2006 parliament enacted a reform based on these. In addition to combining security laws into a new book in the Napoleonic Code, the 2006 reform also made several more substantial changes. First, it expanded the scope of assets over which security interests could be taken out by establishing non-possessory charges and security interests over fungible and future assets (Herbet and Sabbah (2006) and Ancel (2008)). Second, the reform established an electronic public registry covering all types of assets. Third, the reform lowered the costs of creating security interests. For example, it reduced the notarial fees, taxes, and registration costs associated with creating hypothecs. The reform further lowered transaction costs by allowing security interests to be rechargeable to the same or different creditors, either simultaneously or consecutively (Herbet and Sabbah (2006)). In Figure 1, we give an overview of the main differences between the old security law regime from 1804 and the security law regime introduced in March 2006. 7 2.3 Undermining the Reform: Non-Codified Security Laws French managers recognized that the 1804 security laws were outdated long before politicians did. Thus, in the 1970s and 80s, concerted lobbying by them spurred new security laws, which were not added to the Napoleonic Code. Of these “non-codified laws,” the most popular were the Dailly Law and the Pledge of Ready Money. The Dailly Law allowed firms to assign their receivables to financial institutions for security purposes, while the Pledge of Ready Money allowed them to assign their cash and cash equivalents for similar purposes. Both laws were remarkably modern insofar as they permitted security to be taken out over fungible and future assets. Also, they granted super-priority rights to secured creditors in official bankruptcy.5 Their major limitation was that they were restricted to liquid corporate assets. Davydenko and Franks (2008) show that the Dailly Law and the Pledge of Ready Money were popular in France until the 2006 reform. They report that around three quarters of the collateral used by French firms consisted of “cash collateral” (i.e., cash, guarantees, and accounts receivables); while three quarters of the collateral used by German and UK firms consisted of “hard collateral” (i.e., real estate, buildings, and machinery and equipments). 3 Methodology and Data 3.1 Methodology We employ a DID (alternatively DIDID) methodology to identify the effects of the reform. We analyze both firm-level and debt contract-level data. In the firm-level analysis, we exploit the insight that, in contrast to current assets, fixed assets could not be efficiently pledged before the reform. Consequently, firms operating a large portion of fixed assets would be more significantly affected by the 2006 reform than firms operating a larger portion of current assets. Accordingly, we compare the mean of an outcome variable across high- and low-fixed asset firms over the pre-reform and the post-reform periods. We first calculate the simple mean of the outcome variable by fixed asset quartile and year. For each fixed asset quartile, we average the yearly 5 Super-priority rights allow secured creditors to extract assets out of the borrowers’ estate before bankruptcy proceedings begin. Secured creditors are able to do so because, in legal terms, assets covered by super-priority rights are not property of the bankrupt party, but instead of them. 8 mean of the outcome variable over the pre-reform and the post-reform period and compute the difference. Subtracting the difference of the lowest fixed asset quartile from that of the highest quartile gives us the DID estimate for the reform’s effect on the outcome variable. While the above analysis is intuitive, the resulting estimate is possibly biased due to firms entering and leaving the sample and firm unobservables. Thus, we also estimate multivariate models that include firm and year-fixed effects. These models can be written as: Yi,t = α + αi + αt + βP ostt × T reatedi + Xi,t γ + εi,t , (1) where Yi,t is the value of the outcome variable for firm i in year t, with Yi,t {TotalLeverage, ShortTermLeverage, LongTermLeverage, DummyNoTotalLeverage, DummyNoShortTermLeverage, DummyNoLongTermLeverage, Growth, Employees, Sales, Profitability, ProfitVolatility}. We define these variables in the next subsection. P ostt is a dummy variable equal to one during or after the reform year (2006) and otherwise zero; T reatedi is a dummy variable equal to one if firm i’s fixed assets-to-total assets ratio averaged over our sample period is in the top quartile and otherwise zero; and X is a vector of firm-specific controls. αi and αt are firm- and year-fixed effects, respectively. We cluster standard errors at the firm level. Note that β in Equation (1) can be interpreted as the regression-based DID estimate after accounting for the effects of the control variables and firm- and time-invariant effects. We use an analogous model for the loan-level analysis: Zi,j,t = α + αk + αt + βP ostt × T reatedi,j,t + Xi,t γ + Wi,j,t δ + εi,j,t , (2) where Zi,j,t is the value of the outcome variable for loan j taken out by firm i in year t, with Zi,j,t {LoanCost, LoanMaturity}, P ostt is a dummy variable equal to one for secured loans and otherwise zero, Wi,j,t is a vector of loan-contract specific controls, and αk is an industryfixed effect. We do not use firm-fixed effects in this analysis because roughly 70% of the loan observations correspond to firms that contribute only one observation to the sample. 9 3.2 Data To perform our analyses, we use data from several different sources. Our primary source of financial data is Bureau van Dyck’s AMADEUS Top 250,000. AMADEUS contains comprehensive data on both public and private firms from 35 European countries that satisfy a certain size threshold.6 We collect data on French firms as well as on firms in other important European civil law countries (Belgium, Italy, Spain, and Portugal). We use the non-French data to run placebo tests. In France, annual accounts and other corporate documents are collected and made available by local commercial court registries (“Greffes des Tribunaux de Commerce”). On average, Bureau van Dyck requires only one single month to collect the company data and to include it in AMADEUS once these data become available from the registries. We exclude firms in the financial (SIC Code 600–699), services (700–899), utilities (490–494), and public administration (900–999) sectors. Our main tests exclude public firms because these hardly ever borrow secured. We later run a falsification test on public firms. To perform the loan contract-level analysis, we merge the AMADEUS database with the LPC-Dealscan database. We manually match the Dealscan name of each French borrower who took out a loan in the 2001-2009 period (the “BorrowerIssuerName”) with the AMADEUS firm name (the “CompanyName”). We succeed in doing so for about 90% of all borrowers. We only keep revolver and term loans in this analysis because they should be most reflective of a bank’s pricing terms and its credit restrictiveness (Campello and Gao (2014)). We use COMPUSTAT and WorldScope data to construct industry-level land, buildings, and machinery and equipment indexes, and also an industry-level external financial dependence index. We collect jurisdiction-level data on the duration of insolvency procedures from the French Ministry of Justice. Finally, we use the United Nations’ General Industrial Statistics (UNIDO) database (INDSTAT-4) to gather information on gross fixed capital formation and value added for 124 4 digit-ISIC manufacturing industries.7 6 For France, the database includes all companies that meet at least one of the following criteria: (1) revenues of at least e15 million, (2) total assets of at least e30 million, and (3) at least 200 employees. 7 See http://www.justice.gouv.fr/statistiques.html. 10 3.3 3.3.1 Variable Construction Firm-Level Analysis Variables To examine the effects of the reform on firm-level leverage, we use as outcome variables total, long-term, and short-term leverage. TotalLeverage is the sum of short-term debt (loan) and longterm debt (ltdb) divided by total assets (toas).8 ShortTermLeverage is defined as short-term debt divided by total assets, and LongTermLeverage as long-term debt divided by total assets. We further define three dummy variables, DummyNoTotalLeverage, DummyNoShortTermLeverage, and DummyNoLongTermLeverage; these are equal to one if the corresponding leverage variable is equal to zero. We define FixedAssets as fixed assets (fias) divided by total assets. To study real-side effects, we use as outcomes variables growth, employees, sales, profitability, and risk. Growth is the change in intangible fixed assets (ifas) plus the change in inventories (stok), both calculated from the prior fiscal year end to the current one, divided by the average of total assets over the two fiscal year ends.9 Employees is the log of the number of employees (emp). Sales is the log of sales. Profitability is earnings before interest and taxes (ebta) divided by total assets. ProfitVolatility is the standard deviation of Profitability over the four latest fiscal years. We set ProfitVolatility to zero if it is based on less than three observations. We use standard controls, including Size and Age. Size is the log of total assets. Age is the log of the difference between the current year and the year of incorporation.10 To compare the attributes of treated and control firms, we analyze firms’ pre-reform size, age, number of employees, profitability, risk, cash reserves, capital-to-labor ratio, and levels of tangible and fixed assets. We also look at their distance to the biggest French cities and to the Paris Bourse. Cash is cash reserves (cash) divided by total assets. CapitalToLabor is defined as the log of the ratio of tangible fixed assets to the number of employees. To calculate the distance variables, we use resources from geonames.org to find the longitude and latitude associated with each French postal code, including those of the city centers of the ten largest cities. Using the AMADEUS postal code variable, we then assign longitudes and latitudes to each firm in our sample. Finally, we use the formula for the spherical distance 8 The items in parentheses are the AMADEUS variable mnemonics. In France, R&D and formation costs are capitalized under intangible fixed assets. 10 Continuous variables are winsorized at the 1st and 99th percentiles. 9 11 between two points to create the distance variables: DISTi,j = arccos(deglatlon) × r, (3) where DISTi,j is the spherical distance between points i and j, deglatlon is given by: deglatlon = cos(LATi )cos(LGTi )cos(LATj )cos(LGTj ) + sin(LGTi )sin(LGTj ) + cos(LATi )sin(LGTi )cos(LATj )sin(LGTj ), LGTx and LATx are the longitude and latitude of point x, respectively, and r is the radius of the earth (see Coval and Moskowitz (2001)). We define Distance5 and Distance10 as a firm’s distance to the city center of the nearest of the five or ten biggest French cities, respectively. Finally, DistanceBourse is defined as a firm’s distance to the Paris Bourse. 3.3.2 Loan-Level Variables In the loan-level analysis, we use loan mark-ups and maturity as outcome variables. Spread is the log of the sum of a loan’s coupon and annual fees scaled by its nominal value minus the six month LIBOR rate, in basis points. Maturity is the log of the difference between the loan’s maturity date and its initiation date, in months. We use Secured, a dummy variable equal to one for secured loans and zero otherwise, to assign firms to treatment. In line with Chava and Roberts (2008), Roberts and Sufi (2009), and Campello et al. (2011), our control variables are size, age, profitability, total leverage, the existence of a credit rating, the credit spread, the term spread, loan size, loan type, and the other endogenous variable. Rating is a dummy variable equal to one if the firm taking out the loan is rated, else zero. CreditSpread is the mean of the monthly difference between the yields of a corporate bond index and a long-term government bond index. TermSpread is the mean of the monthly difference between the yields of a long-term and a short-term government bond index. Both means are taken over data from the current year. LoanSize is the log of the notional loan value; and LoanType is a dummy variable equal to one for term loans and zero for revolver loans. 12 3.3.3 Industry-Level Variables To identify the types of fixed assets that were used to raise long-term financing, we condition our tests on the availability of land, buildings, or machinery and equipment. Because we lack firm-level data on these items, we construct time-invariant indexes at the 4-digit SIC-industry level using public firms with available data in Compustat and Worldscope. To calculate the index values, we scale the relevant item (land, buildings, or machinery and equipment) by total assets, take the mean by industry and fiscal year, and then by industry alone. Finally, to study capital allocation efficiency, we first use 4-digit ISIC industry-level data for French manufacturing sectors. InvestmentGrowth is the natural logarithm of the ratio of current gross fixed capital formation to its one year-lagged value; ValueAddedGrowth is the natural logarithm of the ratio of current value added to its one year-lagged value.11 Both variables are deflated using the US GDP deflator. Second, we construct an external financial dependence index in the spirit of Rajan and Zingales (1998). Following Larrain (2014), we define the index as the ratio of capital expenditures less cash flow from operations to capital expenditures for the median public US firm in each 3-digit SIC industry over the 1975-2005 period. 3.3.4 Court Efficiency Variable We use data from the French Ministry of Justice to measure court efficiency. We proxy for court efficiency using the average duration of insolvency procedures (restructurings and liquidations), aggregated up to the Departmental level. HighCourtEfficiency is a dummy variable equal to one for firms headquartered in Departments in which the average duration is below the country-wide median, else zero. Because the data required to create HighCourtEfficiency are only available from 2008, we base the dummy variable on data from this year alone. 11 Gross fixed capital formation is defined as the cost of new and used fixed assets minus the value of sales of used fixed assets, where fixed assets include land, buildings, and machinery and equipment. Value added is defined as the value of shipments of goods produced (output) minus the cost of intermediate goods and required services (but not including labor), with appropriate adjustments made for inventories of finished goods, work-in-progress, and raw materials. 13 4 Financial Outcomes 4.1 Debt Usage We first study how the reform affected debt levels and the propensity to have any debt at all. 4.1.1 Total Debt Table 1 shows that the pre-reform mean total leverage ratio of the sample firms hovers around 9%, without trending upward or downward. However, during the reform year alone, this magnitude shoots up to nearly 12% and continues to rise to 13% in 2009. Comparing mean total leverage before and after the reform, we find an increase of 3.2%. This increase is equivalent to 35.2% of mean total leverage before the reform and thus economically meaningful. As a result, the collateral reform had a significant effect on overall borrowing in France. Ta b l e 1 A b o u t H e r e Sorting the universe of French firms into fixed asset quartiles, we find that the above increase in total leverage is entirely driven by firms that make heavy use of fixed assets. Particularly, while neither high- nor low-fixed asset firms display systematic changes in total leverage before the 2006 reform, the high-fixed asset firms exhibit far greater jumps in total leverage in the year of the reform. Comparing mean total leverage before and after the reform, the high-fixed asset firms observe a 7.2% increase in total leverage, while the low-fixed asset firms observe only a 0.8% increase. The 6.4% difference, the DID estimate of the collateral reform on total leverage, is highly statistically significant. Because the difference amounts to 70% of average pre-reform total leverage, it is economically meaningful as well. Figure 2 is useful in depicting the effect of the collateral reform on total leverage across different fixed asset-level firms. 4.1.2 Short-term versus Long-term Debt Next, we split total leverage into short-term and long-term leverage. We do so because shortterm debt is largely unsecured in France, and it is thus not expected to be particularly affected by the reform. Indeed, while Table 2 shows that short-term leverage declines by around 3.5% 14 Figure 2. Evolution of Mean Total Leverage This figure shows the evolution of the mean total leverage ratio by fixed asset quartile (Q) over the 2001–2009 period. Q1, Q2, Q3, and Q4 indicate the first, second, third, and fourth fixed asset quartile, respectively. over the 2001–2009 period, this decrease is not concentrated in specific years. Notably, we also find that high- and low-fixed asset firms experience similar declines in short-term debt from the pre-reform to the post-reform period. In particular, high-fixed asset firms experience a drop of 4.0%, while low-fixed asset firms experience a 2.7% drop. The difference between these two numbers, which can be interpreted as the DID estimate for the effect of the collateral reform on short-term leverage, is not statistically significant. Ta b l e 2 A b o u t H e r e In contrast, the patterns observed in long-term leverage closely resemble those observed in total leverage. While mean long-term leverage remains almost flat at 1.5% until 2005, it climbs up to an average of 5.3% after the collateral reform. Conditioning on different fixed asset-levels, we find that high-fixed asset firms experience far larger increases in long-term debt than low-fixed asset firms. While the mean long-term leverage of firms in the highest fixed asset quartile rises by 7.8%, the lowest fixed asset quartile firms observe an increase of only 1.2%. The 6.6% spread, the DID estimate for the effect of the collateral reform on long-term debt, is highly statistically significant. Since this spread is a striking 440% of the average value of long-term leverage before the reform, it is also highly economically important. Figure 3 depicts the monotonic impact of the reform on the short-term and long-term leverage of firms within the different fixed asset quartiles. 15 Figure 3. Evolution of Mean Short-term vs.Mean Long-term Leverage This figure shows the evolution of mean short-term and mean long-term leverage by fixed asset quartile (Q) over the 2001–2009 period. Q1, Q2, Q3, and Q4 indicate the first, second, third, and fourth fixed asset quartile, respectively. 4.1.3 Zero Leverage Firms We next study the effects of the 2006 reform on the distribution of short-term and long-term debt. To do so, we analyze the proportion of firms with no short-term or long-term debt. Table 3 shows that the proportion of zero short-term debt firms is relatively stable over time for the whole sample, and also all four fixed asset-subsamples. As a result, the difference in the changes observed by the highest- and lowest fixed asset firms, the DID estimate of the effect of the reform on the proportion of zero short-term leverage firms, is an insignificant 0.3%. In sharp contrast, the proportion of zero long-term debt firms is around 90% before the reform, but drops to 47% in the reform year and continues to decrease to 38% in 2009. As before, high-fixed asset firms are most responsible for this pattern. In particular, while the proportion of zero long-term debt firms in the lowest-fixed asset quartile drops by only 27%, the corresponding drop in the highest-fixed asset quartile is more than double, 56%. The DID estimate for the effect of the reform on the proportion of zero long-term leverage firms is statistically and economically significant (–56%–(–27%)=–29%). Ta b l e 3 A b o u t H e r e In Figure 4, we graphically show the impact of the reform on the proportion of zero shortterm or long-term leverage firms across firms with different levels of fixed asset-usage. 16 Figure 4. Evolution of the Proportion of Zero Short-term vs. Zero Long-term Leverage Firms This figure shows the evolution of the proportion of zero short-term and zero long-term leverage firms by fixed asset quartile (Q) over the 2001–2009 period. Q1, Q2, Q3, and Q4 indicate the first, second, third, and fourth fixed asset quartile, respectively. 4.1.4 Regression Results Table 4 reports the results of DID regressions that correspond to the ANOVA tests in Tables 1 through 3. In the first three columns, we employ as outcome variables total, short-term, and long-term leverage; in the next three columns, we employ dummy variables equal to one if total, short-term, or long-term debt is zero and else zero (“zero-leverage firms”). Even after including control variables and firm and year-fixed effects and keeping assignment to treatment fixed, our results are essentially identical to the differences in means in Tables 1–3. The control variables are also highly significant with the expected signs. Ta b l e 4 A b o u t H e r e The DID regressions confirm that Ordonnance 2006-346 led to a significant increase in long-term borrowing by firms that operate more fixed assets (intensive margin). In addition, the proportion of high fixed-asset usage firms with no long-term debt decreases sharply with that same reform (extensive margin). In contrast, low-fixed asset firms do not significantly increase their long-term borrowing, and the proportion of zero long-term debt firms among them drops far less significantly. Albeit statistically weak, our evidence suggests that high-fixed asset-usage firms even substitute short-term debt for long-term debt after the reform. 17 4.2 Debt Contracting Terms We investigate whether debt contract terms change following the reform. To this end, we focus on the pricing and maturity terms of secured (treated) and unsecured (control) loans before and after the reform. The DID regressions in Table 5 show that secured loans become significantly cheaper and more long-term compared to unsecured loans after the reform, independent of whether we control for firm characteristics. In particular, the log all-in-drawn spread of secured loans decreases by around 1.30 more than the same spread for unsecured loans across the two models. This decrease is equivalent to a 120 basis points greater reduction in loan costs. At the same time, the log time-to-maturity of secured loans increases by around 0.83 more than that of unsecured loans across the two models. This decrease is equivalent to a 41 month greater increase in the time-to-maturity. The estimates on the control variables are consistent with those found elsewhere (e.g., Chava and Roberts (2008)). Overall, our evidence suggests that the reform made banks more willing to extend secured credit on more favorable terms. Ta b l e 5 A b o u t H e r e 5 Which Firms Were Most Affected By the Reform? 5.1 Characteristics of Debt Switchers and Non-Switchers Previous work shows that collateral reforms sometimes only benefit certain types of firms, and not always those in need of a better access to capital (e.g., Assunção et al. (2014)). Thus, as a next step, we examine which type of firms experienced the greatest increases in borrowing due to the reform. We compare the attributes of those firms that most significantly increased their borrowing following the reform with the attributes of those firms that did not. To this end, we focus on the subsample of firms with zero long-term leverage in every pre-reform year. We split these firms into two groups, one including those firms that have a positive long-term leverage ratio in every year after the reform (“debt switchers”) and another one including firms that keep a zero long-term leverage ratio every year after the reform (“non-switchers”). Table 6 reveals that the debt switchers are smaller, but slightly more mature firms than the non-switchers. Debt switchers, on average, employ approximately 20 more people than the 18 non-switchers. Also interestingly, they have a 1.5% higher profitability and a 1.2% lower profit volatility than the non-switchers. Finally, they hoard less cash, have higher capital-to-labor ratios, and are more tangible and fixed asset-intensive than the non-switchers. Notably, the debt switchers tend to be located further away from the five or ten biggest French cities and from the Paris Bourse than the non-switchers. Ta b l e 6 A b o u t H e r e Overall, our evidence shows that the collateral reform incentivized creditors to extend longterm financing to young, profitable firms with a low risk. It also suggests that the firms that benefitted most from the reform were located far away from the big cities. 5.2 The Financing of Start-Ups A large literature shows that access to capital helps to foster entrepreneurship (Evans and Leighton (1989) and Holtz-Eakin et al. (1994)). For example, Bertrand et al. (2007), Kerr and Nanda (2009), and Chatterji and Seamans (2012) show that deregulation of the banking sector or credit card markets leads to a higher rate of new businesses formation. They argue that these findings are driven by the fact that deregulating leads to a reduction in financial constraints, making prospective entrepreneurs more willing to compete with incumbents firms. In Table 7, we show that entrepreneurs in France also benefitted from the 2006 collateral reform. Before the reform, French start-ups had an average long-term leverage between 1.9% and 3.6% in their year of incorporation. However, this figure increases to 4.8% in the reform year, and it continues to increase to 7.2% in 2009. Similarly, before the reform, 90% of all start-ups had no long-term debt in their year of incorporation. However, this figure drops to 74% after the reform. Critically, a firm’s fixed asset-usage again conditions these results. In particular, the firms in the highest-fixed asset quartile show a 5.2% greater increase in their initial long-term leverage than firms in the lowest-fixed asset quartile. Ta b l e 7 A b o u t H e r e 19 Figure 5. The Long-term Leverage Distribution of French Departments Before and After the Reform The histograms show the distributions of mean long-term leverage and the proportion of no long-term leverage firms of the 96 French Departments both for the pre-reform and the post-reform period. The histogram entries are the mean of long-term leverage and the proportion of zero long-term leverage firms by Department and year, separately averaged over the pre-reform and the post-reform years. 6 The Geography of the Collateral Reform 6.1 Financing Effects of the Reform Across Departments Lilienfeld-Toal et al. (2012) find evidence that security law reforms can redistribute credit from firms headquartered in rural areas to those headquartered in larger cities. These findings raise concerns about the distributional effects of security law reforms because firms in larger cities have more options to raise external capital than those in rural areas. To study the geographical effects of the reform, we proceed in two steps. First, we identify the Departments that benefitted most from the reform. To wit, we calculate mean long-term leverage and the proportion of zero long-term leverage firms by Department and year. We separately average these measures over the pre-reform and the post-reform years. Figure 5 shows histograms of the distributions of Departments’ mean long-term leverage (Panel A) and proportions of zero long-term debt firms (Panel B) before and after the reform. The histograms reveal that, before the reform, the vast majority of Departments have a mean long-term leverage 20 between 1% and 3% and a mean proportion of zero long-term leverage firms between 60% and 90%. However, these figures change significantly following the reform. In particular, the reform shifts the bulk of the mean long-term leverage distribution to the right, so that after the reform most Departments have a mean long-term leverage ratio between 4% and 9%. Similarly, it shifts the bulk of the mean proportion of zero long-term leverage firms distribution to the left, so that after the reform most Departments have a mean proportion of zero long-term leverage firms between 0% and 40%. Thus, our empirical evidence suggests that the 2006 reform benefitted firms all over France, and not just those located in specific areas. Second, we analyze how specific Departments were affected by the reform. In doing so, we create for each Department the change in long-term leverage from before to after the reform and the proportion of firms that never had any long-term leverage before the reform, but that reported a positive long-term leverage every year after the reform (”debt switchers”). Figure 6 plots these changes. The darker a Department’s color in map the more the Department is affected. The yellow-rimmed Departments are those in which the ten largest French cities are located. The maps show that the largest increases in long-term leverage and the largest decreases in the proportion of zero long-term leverage firms occurred far away from the big cities. Thus, not only did the reform abstain from favoring firms headquartered in big cities, it actually channeled the largest amount of new capital to firms in rural areas. 6.2 What Type of Fixed Assets Were Used As Collateral? Our results show that the collateral reform led to the largest increase in long-term borrowing in rural areas with a low population density. Because firms located in these areas own a great amount of land, we conjecture that the collateral reform enabled these firms to pledge their land and thereby to secure long-term debt financing. To test this hypothesis, we proceed as follows. Within each fixed-assets quartile, we further sort firms into either land-, buildings-, or machinery and equipment-index quartiles. Then, for each double-sorted subgroup of firms, we calculate the average long-term leverage ratio before and after the reform. Table 8 shows the mean long-term leverage ratio of the extreme fixed asset-fixed asset type-subgroups over the pre-reform and post-reform period. The table confirms that land-rich high fixed asset-usage firms were able to raise more long-term debt financing due to the reform 21 Figure 6. Change in Mean Long-term Leverage and Proportion of Debt Switchers by Department In Panel A, we show the change in mean long-term leverage from the pre-reform to the post-reform period for each French Department. Mean long-term debt is long-term leverage, first averaged by Department and year and then by Department and pre-reform or post-reform period. In Panel B, we show the proportion of firms that had no long-term debt in any year preceding the reform, but that had a positive long-term leverage ratio in every year proceeding the reform. The darker a Department’s color in the maps, the greater is either the change in mean long-term leverage (Panel A) or the proportion of debt switchers (Panel B). The yellow-rimmed Departments in the maps are those containing the ten largest French cities. than land-poor high fixed asset-usage firms. In particular, only looking at firms in the highest land index quartile, the difference in the increase in the mean long-term leverage ratio across the highest and lowest fixed asset quartiles, the DID effect of the reform among land-rich firms, is 10.0%. In contrast, among firms in the lowest land index quartile, the same difference is only 4.9%. Taken together, these figures imply a DIDID effect of 5.1% (10.0%–4.9%). If we instead use the buildings index or the machinery and equipment index as second sorting variable, we find DIDID estimates of 4.0% (9.1%–5.1%) and 1.2% (5.6%–4.4%), respectively. Thus, while firms also used their buildings to secure new debt, there is only very weak evidence to suggest that they used their machinery and equipment for the same purposes. Ta b l e 8 A b o u t H e r e 6.3 Inequalities in Access to Long-term Debt Our results show that the collateral reform produced the largest increase in long-term borrowing among firms that are often expected to have less access to capital markets, such as small firms or firms located in rural areas. To further support the idea that the 2006 reform spurred a democratization of credit, we use as an alternative measure the Gini index for long-term debt 22 Figure 7. Pre- and Post-Reform Long-term Debt-Gini Index By Department The maps show the long-term debt-Gini index of French Departments for the pre-reform (Panel A) and the post-reform (Panel B) period. To calculate the Gini index, we create the Gini index by Department and year and then average by Department and pre-reform or post-reform period. The darker a Department’s color in the maps, the greater is its Gini index. The yellow-rimmed Departments in the maps are those containing the ten largest French cities. usage at the Departmental level. We calculate the Gini index as follows: GiniIndexk,t Nk,t Nk,t X X 1 |LongT ermLeveragei,t − LongT ermLeveragej,t |, = 2 2µk,t Nk,t i=1 j=1 (4) where GiniIndexk,t is Department k’s Gini index in year t, µk,t is the average long-term leverage of the firms in Department k, i and j are firm indexes, and Nk,t is the number of firms. Next, we average GiniIndexk,t by Department and year, and then by Department and pre- or post-reform period. The calculated Gini index ranges from zero to one, and a higher value indicates a more unequal distribution of long-term debt usage across firms. Figure 7 shows each Department’s pre- and post-reform Gini index value. The darker a Department’s color in the maps, the higher is its Gini index. Before the reform, only six out of 96 Departments observe a Gini index value below 0.90, suggesting that long-term debt is heavily concentrated in the hands of a few firms. However, following the reform, the Gini index drops significantly to an average value of 0.72. While all Departments observe profound decreases in their Gini index values, it is again rural Departments that experience the most dramatic changes. Overall, the collateral reform greatly decreased inequalities in the access to credit, and the greatest improvements took place in the country-side. 23 6.4 Court Efficiency We investigate whether court efficiency conditions the debt financing effects of the collateral reform. Campello and Larrain (2014) show that a Romanian security law reform led to greater expansions in corporate credit-taking in regions hosting less efficient courts. This arises because the Romanian reform enabled secured creditors to seize collateral out-of-court, thereby allowing them to bypass inefficient courts. While formally the French reform also removed the ban on out-of-court seizures, French insolvency laws continue to stay such rights in bankruptcy. Thus, secured creditors in France remain highly dependent on courts. This may imply that the effects of Ordonnance 2006-343 are stronger in areas hosting efficient courts. We use DIDID regressions to study the intermediating effect of court efficiency. The endogenous variables are the same as in Table 4. As exogenous variables, we use the same variables as in Table 4 plus: the double-interaction between Post and HighCourtEfficiency and the tripleinteraction between Post, Treated, and HighCourtEfficiency. The DIDID regressions contain firm- and year-fixed effects; standard errors are clustered at the firm-level. The coefficient on the triple interaction gives the effect of court efficiency on the reform’s debt effects. Table 9 offers partial support for our hypothesis. While court efficiency does not condition the increase in leverage levels, it has a significant impact on the decline in the proportion of zero long-term leverage firms. In particular, while the proportion of zero long-term leverage firms drops by 10.0% in Departments hosting less efficient courts, it drops by 13.4% in Departments hosting more efficient courts. The difference is highly statistically significant. Ta b l e 9 A b o u t H e r e 7 Real Outcomes We also assess how the firms that most increased their borrowing after the reform used the new credit that they were able to tap. This is interesting since the higher leverage ratios could simply lead to a higher default risk in the economy. On the other hand, if the firms that increased their borrowing invested the proceeds from the new debt into profitable low risk-projects, the leverage-induced increase in default risk could be dominated by profitability- and volatility- 24 induced declines in default risk. It is an empirical question how the reform affected the default risk of the firms that experienced the largest increases in access to collateral. 7.1 Performance To address the above question, Table 10 reports the results of DID regressions using as outcome variables investment, employees, sales, profitability, and profit volatility. Our evidence shows that the firms that most benefitted from the collateral reform (i.e., firms with a lot of fixed assets) increased investment more and hired more people after the reform. For example, highfixed asset firms experienced an around 4.9% higher employment growth than low-fixed asset firms. In addition, firms with a lot of fixed assets raised their output more and became more profitable and less risky than others after the reform. For example, those firms experienced a 4.5% higher sales growth than low-fixed asset firms after the reform. Ta b l e 1 0 A b o u t H e r e 7.2 Survival Analysis We use a Cox proportional hazard model to fit the number of years until a firm fails. The Cox proportional hazard model assumes that the hazard rate is given by: λi,t = φt exp (βP ostt × T reatedi + Xi,t γ) , (5) where the hazard rate, λi,t , is defined as the probability of firm i failing at time t conditional on surviving until then. We back out the year in which a firm fails using the “legal status” variable in AMADEUS. We set the failure year equal to the calendar year during which a firm’s legal status changes from one of the active statuses to one of the failure statuses (if any).12 We exclude firms from the hazard analysis for which it is unclear whether they became inactive for performance reasons. φt is the “baseline” hazard rate common to all firms, and the exponential function allows for cross-sectional variations in hazard. Using a partial likelihood estimator, we are able to obtain estimates of β and γ without imposing any structure on φt . The same 12 The failure statuses are: “default of payment,” “insolvency proceedings,” “receivership,” “bankruptcy,” “dissolved (bankruptcy),” “dissolved (liquidation),” and “in liquidation.” 25 estimator is also able to account for right-censoring of data (i.e., firms that do not fail within the sample period or that leave the sample for reasons unrelated to performance). In Table 11, we show the results from the Cox proportional hazard models. In the absence of controls, firms with a lot of fixed assets experience a 28.6% (e−0.338 ) greater decrease in their failure rates than other firms after the reform. Controlling for size and cash does not influence these findings. However, controlling for leverage and thus for the fact that high-fixed asset firms increased their debt levels more than others (implying that they could have become more distressed than others), the difference in the change in failure rates between high- and low-fixed asset firms widens. Finally, adding profitability and profit volatility more than halves this difference. The implication is that the high-fixed asset firms’ greater increases in profitability and greater declines in profit volatility partially explain why their failure rates decline relative to the failure rates of low-fixed asset firms after the reform. Ta b l e 1 1 A b o u t H e r e 7.3 Capital Allocation Efficiency One of the ultimate goals of financial markets in general, and banks in particular, is to allocate capital efficiently (Schumpeter (1934), Diamond (1984), and Morck et al. (2011)). Given their unique advantage in acquiring and processing information to assess growth opportunities, banks have a unique advantage in doing so. However, the institutional setting and legal framework can significantly affect how banks operate (La Porta et al. (1998)). Our previous findings show that the 2006 reform made banks less reluctant to lend to profitable, less risky firms. The ultimate question then is whether this lower reluctance brought about a more efficient allocation of capital in the economy. To answer this question, we use two different tests. First, we follow Wurgler (2000) in reasoning that for the capital allocation process to be efficient firm-level investments should increase in sectors with better growth opportunities and decrease in sectors with worse growth opportunities. Using value added-growth as a proxy for growth opportunities, we use the sensitivity of capital investment to the growth in value-added as a measure of the efficiency of capital allocation, and we study whether this sensitivity has increased after the reform. In particular, we run the following regression on industry-level 26 panel-data from before and from after the reform: InvestmentGrowthk,t = α + βV alueAddedGrowthk,t + εk,t , (6) where InvestmentGrowthk,t is gross fixed capital formation and V alueAddedGrowthk,t the value-added of industry k in year t. Using pre-reform data, Table 12 shows that the elasticity between investment and value-added (β) is 0.43. Using post-reform data, it becomes 0.70. The difference between the two numbers, 0.27, is statistically significant at the 10% level (t-statistic: 1.79). Based on Wurgler’s (2000) interpretation of these elasticities, the reform thus led to a statistically and economically significant increase in capital allocation efficiency. Ta b l e 1 2 A b o u t H e r e Second, we use the external financial dependence measure proposed by Rajan and Zingales (1998) as alternative proxy for capital allocation efficiency. This measure captures an industry’s technological demand for external financing. In an efficient economic system, capital should be directed towards those sectors that are more reliant on external financing. Thus, if the 2006 reform relaxed lending constraints, we should observe a significantly larger increase in borrowing among firms operating in sectors highly dependent on external financing. To test this hypothesis, we split our sample into two groups, one including those with a high dependence on external financing (proxy value above median) and one including those with a low dependence (proxy value below median). Next, we split the firms in the high and low dependence-subsamples into treated and control firms, where treatment status is assigned as before. Considering all firms, Table 13 shows that firms with a high dependence on external financing observed greater increases in long-term leverage from the pre-reform to the post-reform period than other firms. In particular, while the mean long-term leverage ratio of highly dependent firms increases by 4.8%, the mean long-term leverage ratio of firms with a weaker dependence increases by only 2.9%. The difference is highly statistically significant. Restricting our attention to treated firms, the difference widens. In particular, while highly dependent high fixed assetusage firms observe a mean increase in their long-term leverage of 8.2%, the mean long-term 27 leverage of less dependent high fixed asset-usage firms increases by only 5.3%. Ta b l e 1 3 A b o u t H e r e Overall, these two tests suggest that the collateral reform helped the French economic system move towards allocating more capital to high-value added sectors and to sectors that rely more on external financing. 8 Robustness 8.1 Self-Selection and Parallel Trends The validity of DID tests rests on a number of assumptions, of which the most important one is that firms cannot self-select into treatment. In our context, this implies that we need to rule out that firms manage their assets in such a way that they always benefit from the prevailing security law regime. Results in Table 14 suggest that it is unlikely that firms manage their assets in such a way. Over the 2001–2005 period, a mere 6.8% of firms moved from the lower three fixed asset quartiles into the highest one; indistinguishable from the 7.0% that moved into the other direction. Notably, the 2001–2005 migration rates do not differ much from migration rates calculated over other five year-periods within or before our sample period. Ta b l e 1 4 A b o u t H e r e If the random assignment to treatment-assumption is fulfilled, treated and control firms should display parallel trends in outcome variables before the reform. We have verified that this is the case in Figures 1 and 2, in which neither treated nor control firms observe any significant trends in total or long-term leverage during the period preceding the reform. Additional statistical tests confirm this result (available from the authors upon request). 28 Figure 8. Evolution of the Proportion of Zero Long-term Leverage Firms By Country The bar chart shows the proportion of zero long-term leverage firms in France and each of the placebo countries (Belgium, Italy, Spain, and Portugal) over the 2001–2009 period. 8.2 Autocorrelation in Outcome Variables Bertrand et al. (2004) show that autocorrelation in the outcome variable can create upwardbiased inference levels in DID regressions. Following their work, we separately average each analysis variable over the pre- and the post-reform period and then repeat the DID regressions using the collapsed data. Results are virtually identical to those reported in Table 4. 8.3 8.3.1 Placebo Tests Public Firms Wood (2007) argues that public firms only borrow secured under exceptional circumstances, for example, when they are distressed or borrow for project financing purposes. Thus, we also repeat our tests on public firms, expecting these to be less affected by the reform than private firms. Table 15 confirms this intuition. The table shows that, while public firms also experience a slight increase in long-term leverage from the pre-reform to the post-reform period, we find no evidence to suggest that this increase is driven by high-fixed asset firms. If anything, results work in the opposite direction. Specifically, while firms in the lowest fixed asset quartile increase their long-term leverage by an average of 4.2%, those in the highest fixed asset quartile slightly 29 decrease it by an average of 0.9%, with the spread, however, not attaining statistical significance. Running DID regressions on public firms confirms these conclusions (not reported). Ta b l e 1 5 A b o u t H e r e 8.3.2 Other Civil Law Countries As a final falsification test, we repeat our tests on European civil law countries that did not reform their security laws during the sample period. These countries are Belgium (a neighbor with the same language and close cultural and economic ties), Italy, Spain, and Portugal. Figure 8 uses bar charts to graphically illustrate the results of these falsification tests. The figure displays the proportion of zero long-term leverage firms for France and each placebo country over the 2001–2009 period. The most important conclusion we draw from Figure 8 is that none of the placebo countries experiences a decrease in zero long-term debt firms as dramatic as the one experienced by France. Also interesting is that the security law reform makes the proportion of zero long-term leverage firms in France far more similar to those in the other countries, which already had efficient security laws in place at the start of our sample period. In Figure 9, we offer line charts showing the mean value of long-term leverage and the proportion of zero long-term leverage firms per placebo country. We again conclude that none of the placebo countries displays effects similar to those found in France. DID regressions using data from the placebo countries confirm this conclusion (not reported). 9 Conclusion In recent years, many countries have undertaken reforms of their security laws, often under the banner of “enhancing access to credit.” Consistent with this aim, prior research has shown that such reforms can raise the volume of corporate borrowing (Vig (2013), Assunção et al. (2014), and Campello and Larrain (2014)). However, raising the volume of corporate borrowing is only the tip of the iceberg. More than on how much new debt capital large, established companies can raise, the success of these reforms hinges on whether they open up credit for companies that were hitherto locked out of debt markets. Focusing on financially-constrained firms — small and young firms, start-ups, firms located in the country-side — is important because these 30 Figure 9. Evolution of Mean Long-term Leverage and the Proportion of Zero Long-term Leverage Firms in the Placebo Countries The figure shows the evolution of mean long-term leverage (upper panels) and the proportion of zero long-term leverage firms (lower panels) for each placebo country and fixed asset quartile (Q) over the 2001–2009 period. The placebo countries are Belgium, Italy, Spain, and Portugal. Q1, Q2, Q3, and Q4 indicate the first, second, third, and fourth fixed asset quartile, respectively. firms tend to generate the lion share of most countries’ value-added. Compared to prior studies, we offer a more comprehensive analysis of the effects of collateral reforms. In particular, we do not only study the impact that such reforms have on the magnitude of debt financing, but we also look at their impact on the composition of corporate debt, the distribution of corporate debt usage across firms, and capital allocation efficiency. To achieve these goals, we consider a security law reform that was recently implemented by a developed country. This reform is Ordonnance 2006-346, enacted in France in March 2006. Looking at a reform in a developed country has the advantage that we have more data at our disposal, implying that we can study the reform from more angles than other studies. Our focus on France is also helpful to shed light on the distributional effects of collateral reforms. Up to the recent collateral reform, access to credit was highly unequally distributed across French firms, with a handful of major companies holding all the long-term debt capital. Our evidence shows that the firms expected to be most affected by the reform raised their long-term leverage from a pre-reform average of 2.4% to a post-reform average of 10.2%. Also, the fraction of zero long-term debt firms among them dropped from a pre-reform average of 88% to a post-reform average of 32%. In contrast, we find no discernable patterns in these variables among the control firms. Loan contract data suggest that the above credit expansions were driven by cheaper and more long-term secured credit. Thus, we conclude that the reform 31 made debt capital more attractive and raised overall corporate borrowing. However, our evidence also speaks to how the reform changed the demographics of corporate debt usage. In particular, we show that a large amount of the newly available capital went to small, profitable firms that are located in rural areas. A large fraction of the newly-raised debt financing was secured using land, explaining why the increases in long-term debt financing are especially pronounced among firms located in rural areas. The reform also led to increases in long-term debt usage among start-ups. By opening up access to credit, the reform produced significant decreases in the Gini index of long-term debt concentration all over the country. Overall, our evidence suggests that the reform produced a “democratization of corporate debt.” Finally, firms used the new debt financing in social welfare-enhancing ways. In particular, we show that the firms expected to be most affected by the reform became more profitable, less risky, and less likely to fail than others after the reform. Also, the reform led to a significant increase in the elasticity between investment and value-added and to significantly more borrowing to firms operating in industries heavily reliant on external financing. Both these findings could point to reform-induced improvements in the efficiency with which capital is allocated. 32 References Ancel, M.-E., 2008. Recent reform in France: The renaissance of a civilian collateral regime. In: Dahan, F., Simpson, J. (Eds.), Secured transaction reform and access to credit, Edward Elgar Publishing, Cheltenham, 259-272. Assunção, J., Benmelech, E., Silva, F., 2014. “Repossession and the democratization of credit.” Review of Financial Studies 27, 2661-2689. Bertrand, M., Duflo, E., Mullainathan, S., 2004. “How much should we trust differences-indifferences estimates?” Quarterly Journal of Economics 119, 249-275. Bertrand, M., Schoar, A., Thesmar, D., 2007. “Banking deregulation and industry structure: Evidence from the French banking reforms of 1985.” Journal of Financial Economics 62, 597-628. Boughida, S., Levesque, G., Roux, J., 2011. “Lending and taking security in France: An overview.” Available from http://uk.practicallaw.com/7-501-4362. Campello, M., Gao, J., 2014. “Customer concentration and loan contract terms.” Working Paper, Cornell University. Campello, M., Larrain, M., 2014. “Enhancing the contracting space: Collateral menus, access to credit, and economic activity.” Working Paper, Cornell University. Campello, M., Lin, C., Ma, Y., Zou, H., 2011. “The real and financial implications of corporate hedging.” The Journal of Finance 66, 1615-1647. Chatterji, A. K., Seamans, R. C., 2012. “Entrepreneurial finance, credit cards, and race.” Journal of Financial Economics 106, 182-195. Chava, S., Roberts, M., 2008. “How does financing impact investment? The role of debt covenants.” Journal of Finance 63, 2085-2121. Coval, J. D., Moskowitz, T. J., 2001. “The geography of investment: Informed trading and asset prices.” Journal of Political Economy 109, 811-841. Cox, D. R., Oakes, D., 1984. “Analysis of Survival Data.”Chapman & Hall. Davydenko, S., Franks, J., 2008. “Do bankruptcy codes matter? A study of defaults in France, Germany, and the UK.” The Journal of Finance 63, 565-608. Diamond, D. W., 1991. “Debt maturity structure and liquidity risk.” The Quarterly Journal of Economics 106, 709-737. Diamond, D. W., 1984. “Financial intermediation and delegated monitoring.” Review of Economic Studies 51, 393-414. Djankov, S., McLiesh, C., Shleifer, A., 2007. “Private credit in 129 countries” Journal of Financial Economics 84, 299-329. Evans, D. S., Leighton, L. S., 1989. “Some empirical aspects of entrepreneurship.” American Economic Review 79, 519-535. Haselmann, R., Pistor, K., Vig, V., 2009. “How law affects lending.” Review of Financial Studies 23, 549-580. 33 Herbet, J., Sabbah, C., 2006. “Will secured lending in France benefit from the recent overhaul of civil code provisions relating to security interests?” International Business Law Journal 6, 853-859. Holtz-Eakin, D., Joulfaian, D., Rosen, H. S., 1994. “Sticking it out: Entrepreneurial survival and liquidity constraints.” Journal of Political Economy 102, 53-75. Infelise, F., 2014. “Supporting access to finance by SMEs: Mapping the initiatives in five EU countries.” ECMI Research Report 9/2014. Kerr, W. R., Nanda., R., 2009. “Democratizing entry: Banking deregulation, financing constraints, and entrepreneurship.” Journal of Financial Economics 94, 124-149. La Porta, R., Lopez-de-Silanes, F., Shleifer, A., 2008. “The economic consequences of legal origins.” Journal of Economic Literature 42, 285-332. La Porta, R., Lopez-de-Silanes, F., Shleifer, A., Vishny, R., 1998. “Law and finance.” Journal of Political Economy 106, 1113-1155. Larrain, M., 2014. “Capital account opening and wage inequality.” Forthcoming in the Review of Financial Studies. Lilienfeld-Toal, U., Mookherjee, D., Visaria, S., 2012. “The distributive impact of reforms in credit enforcement: Evidence from Indian debt recovery tribunals.” Econometrica 80, 497-558. Morck, R., Yavuz, V. M., Yeung, B., 2011. “Banking system control, capital allocation, and economy performance.” Journal of Financial Economics 100, 264-283. Omar, P., 2007. “Updating the framework for asset security in France: The reforms of 2006.” Journal of Comparative Law 2, 189-209. Qian, J., Strahan, P. E., 2007. “How laws and institutions shape financial contracts: The case of bank loans.” The Journal of Finance 62, 2803-2834. Rajan, R., Zingales, L., 1998. “Financial dependence and growth.” American Economic Review 88, 559-586. Renaudin, M., 2013. “The modernisation of French secured credit law: Law as a competitive tool in global markets.” International Company and Commercial Law Review 24, 385-392. Roberts, M. R., Sufi, A., 2009. “Renegotiation of financial contracts: Evidence from private credit agreements.” Journal of Financial Economics 93, 159-184. Ross, S., Westerfield, R.W., Jaffe, J., 2012. “Corporate finance.” McGraw-Hill/Irwin. Schumpeter, J. A., 1934. “The theory of economic development. An inquiry into profits, capital, credit, interest, and the business cycle.” Cambridge: Harvard University Press. Vig, V., 2013. “Access to collateral and corporate debt structure: Evidence from a natural experiment.” The Journal of Finance 68, 881-928. Wood, P., 2007. “Principles of international insolvency.” Sweet & Maxwell. World Bank, 2007. “Doing Business 2007: How to reform.” World Bank Group. Wurgler, J., 2000. “Financial markets and the allocation of capital.” Journal of Financial Economics 58, 187-214. 34 Table 1 Effect of the Reform on Total Leverage: ANOVA Tests This table shows mean total leverage by year and fixed asset quartile over the 2001–2009 period. Total leverage (“TotalLeverage”) is defined as the sum of short-term and long-term debt scaled by total assets. Fixed assets (“FixedAssets”) is defined as fixed assets scaled by total assets. We sort firms into fixed asset quartiles according to the values of fixed assets and the fixed asset quartile breakpoints in the current year. Quartile Q1 contains firms with a low-fixed asset ratio, while Quartile Q4 contains firms with a high ratio. At the bottom of the table, we report averages for the pre-reform period means (2001–2005) and averages for the post-reform period means (2006–2009), and their differences. The final column shows the differences in means across the highest (Q4) and lowest (Q1) fixed asset quartile. ∗∗∗ , ∗∗ , and ∗ indicate statistical significance at the 99%, 95%, and 90% confidence levels, respectively, based on standard errors clustered at the firm level. Fixed Asset Quartile Year All 2001 Diff. Q1 Q2 Q3 Q4 0.098 0.078 0.092 0.102 0.120 0.043*** 2002 0.095 0.073 0.089 0.098 0.118 0.044*** 2003 0.091 0.069 0.085 0.095 0.113 0.045*** 2004 0.086 0.063 0.078 0.089 0.113 0.050*** 2005 0.088 0.067 0.078 0.090 0.115 0.049*** 2006 0.116 0.077 0.096 0.119 0.172 0.095*** 2007 0.120 0.077 0.098 0.121 0.183 0.106*** 2008 0.128 0.080 0.106 0.128 0.197 0.116*** 2009 0.130 0.079 0.105 0.135 0.202 0.123*** Mean 2001–2005 (1) Mean 2006–2009 (2) 0.091 0.123 0.070 0.078 0.085 0.101 0.095 0.126 0.116 0.188 Diff. (2)–(1) 0.032*** 0.008*** 0.016*** 0.031*** 0.072*** 35 Q4–Q1 0.064*** Table 2 Effect of the Reform on Mean Short-term and Long-term Leverage: ANOVA Tests This table shows mean short-term and mean long-term leverage by year and fixed asset quartile over the 2001– 2009 period. The upper number in each row is short-term (ST) leverage, while the lower number is long-term (LT) leverage. Short-term leverage (“ShortTermLeverage”) is defined as short-term debt scaled by total assets, whereas long-term leverage (“LongTermLeverage”) is defined as long-term debt scaled by total assets. Fixed assets (“FixedAssets”) is defined as fixed assets scaled by total assets. We sort firms into fixed asset quartiles according to the values of fixed assets and the fixed asset quartile breakpoints in the current year. Quartile Q1 contains firms with a low-fixed asset ratio, while Quartile Q4 contains firms with a high ratio. At the bottom of the table, we report averages for the pre-reform period attributes (2001–2005) and averages for the post-reform period attributes (2006–2009), and their differences. The final column shows the differences in attributes across the highest (Q4) and the lowest (Q1) fixed asset quartile. ∗∗∗ , ∗∗ , and ∗ indicate statistical significance at the 99%, 95%, and 90% confidence levels, respectively, based on standard errors clustered at the firm level. Debt Year Type 2001 Fixed Asset Quartiles Diff. All Q1 Q2 Q3 Q4 Q4–Q1 ST LT 0.080 0.016 0.067 0.009 0.078 0.013 0.084 0.017 0.090 0.026 0.023*** 0.017*** 2002 ST LT 0.077 0.015 0.062 0.010 0.076 0.012 0.082 0.015 0.090 0.024 0.028*** 0.014*** 2003 ST LT 0.074 0.014 0.059 0.008 0.073 0.011 0.079 0.015 0.086 0.023 0.027*** 0.015*** 2004 ST LT 0.070 0.013 0.055 0.006 0.069 0.009 0.073 0.014 0.084 0.023 0.030*** 0.017*** 2005 ST LT 0.071 0.015 0.057 0.008 0.068 0.010 0.074 0.016 0.085 0.025 0.028*** 0.018*** 2006 ST LT 0.071 0.042 0.059 0.016 0.067 0.028 0.074 0.044 0.085 0.082 0.026*** 0.066*** 2007 ST LT 0.071 0.046 0.059 0.016 0.068 0.029 0.072 0.048 0.085 0.093 0.027*** 0.077*** 2008 ST LT 0.068 0.058 0.055 0.023 0.067 0.037 0.070 0.057 0.080 0.113 0.024*** 0.090*** 2009 ST LT 0.063 0.065 0.050 0.027 0.062 0.041 0.065 0.069 0.076 0.122 0.026*** 0.095*** Mean 2001–2005 (1) ST LT 0.103 0.015 0.083 0.008 0.101 0.011 0.109 0.015 0.121 0.024 Mean 2006–2009 (2) ST LT 0.068 0.053 0.056 0.020 0.066 0.034 0.070 0.055 0.082 0.102 Diff. (2)–(1) ST LT -0.035*** 0.038*** -0.027*** 0.012*** 36 -0.035*** 0.023*** -0.039*** 0.039*** -0.040*** 0.078*** -0.013 0.066*** Table 3 Effect of the Reform on the Proportions of Zero Short-term and Long-term Leverage Firms: ANOVA Tests This table shows the fraction of zero short-term and long-term leverage firms by year and fixed asset quartile over the 2001–2009 period. The upper number in each row is short-term (ST) leverage, the lower number long-term (LT) leverage. Short-term leverage (“ShortTermLeverage”) is defined as short-term debt scaled by total assets; long-term leverage (“LongTermLeverage”) is defined as long-term debt scaled by total assets. Fixed assets (“FixedAssets”) is defined as fixed assets scaled by total assets. We sort firms into fixed asset quartiles according to the values of fixed assets and the fixed asset quartile breakpoints in the current year. Quartile Q1 contains firms with a low-fixed asset ratio, while Quartile Q4 contains firms with a high ratio. At the bottom of the table, we report averages for the pre-reform period attributes (2001–2005) and averages for the post-reform period attributes (2006–2009), and their differences. The final column shows the differences in attributes across the highest (Q4) and the lowest (Q1) quartile. ∗∗∗ , ∗∗ , and ∗ indicate statistical significance at the 99%, 95%, and 90% confidence levels, respectively, based on standard errors clustered at the firm level. Debt Year Type 2001 Fixed Asset Quartile Diff. All Q1 Q2 Q3 Q4 Q4-Q1 ST LT 0.169 0.886 0.264 0.898 0.141 0.883 0.126 0.883 0.145 0.878 -0.119*** -0.020*** 2002 ST LT 0.172 0.893 0.280 0.903 0.145 0.893 0.123 0.890 0.141 0.886 -0.139*** -0.017*** 2003 ST LT 0.176 0.898 0.278 0.913 0.154 0.893 0.122 0.895 0.149 0.890 -0.129*** -0.023*** 2004 ST LT 0.180 0.902 0.293 0.919 0.152 0.909 0.128 0.889 0.145 0.890 -0.149*** -0.029*** 2005 ST LT 0.177 0.875 0.287 0.910 0.149 0.874 0.130 0.859 0.140 0.858 -0.147*** -0.052*** 2006 ST LT 0.172 0.470 0.277 0.683 0.152 0.441 0.121 0.380 0.138 0.376 -0.139*** -0.306*** 2007 ST LT 0.173 0.421 0.276 0.661 0.149 0.396 0.128 0.319 0.137 0.310 -0.139*** -0.351*** 2008 ST LT 0.212 0.397 0.316 0.621 0.185 0.368 0.160 0.304 0.187 0.294 -0.129*** -0.327*** 2009 ST LT 0.240 0.384 0.338 0.603 0.213 0.362 0.196 0.286 0.212 0.283 -0.126*** -0.319*** Mean 2001–2005 (1) ST LT 0.175 0.891 0.281 0.909 0.148 0.890 0.126 0.883 0.144 0.880 Mean 2006–2009 (2) ST LT 0.199 0.418 0.302 0.642 0.175 0.392 0.151 0.322 0.168 0.316 Diff. (2)–(1) ST LT 0.024*** -0.473*** 0.021*** -0.267*** 37 0.026*** -0.499*** 0.025*** -0.561*** 0.024*** -0.564*** 0.003 -0.298*** Table 4 Effect of the Reform on Leverage: DID Regressions This table shows the results from the following regression: Yi,t = α + αi + αt + βP ostt × T reatedi + Xi,t γ + εi,t , where Yi,t is either total, short-term, or long-term leverage or dummy variables indicating whether the former quantities are zero. Total (short-term) [long-term] leverage is the sum of short-term and long-term (short-term) [long-term] debt scaled by total assets. Dummy no total (short-term) [long-term] leverage is a dummy variable equal to one if total (short-term) [long-term] leverage is zero and else zero. Post is a dummy variable equal to one for years greater or equal to 2006 and else zero. Treated is a dummy variable equal to one for firms whose mean fixed asset-to-total asset ratio is in the top quartile and else zero. Xi,t is a vector of control variables, including size, profitability, and age. Size is the log of total assets; Profitability is the ratio of earnings before interest and taxes to total assets; and Age is the log of the current year minus the year of incorporation. α, β, and γ are free parameters. αi and αt indicate firm- and year-fixed effects. εi,t is the residual. Parameter estimates are in normal letter, while t-statistics are in square parentheses. T-statistics are calculated from standard errors that are clustered at the firm-level. ∗∗∗ , ∗∗ , and ∗ indicate statistical significance at the 99%, 95%, and 90% confidence levels, respectively. The sample period is 2001 to 2009. Dependent Variable Intensive Margin Post × Treated Size Profitability Age R-squared Observations Extensive Margin Total Leverage Short-term Leverage Long-term Leverage Dummy No Total Leverage 0.040*** [22.43] 0.021*** [16.40] -0.189*** [-36.33] -0.002 [-1.02] -0.002* [-1.77] 0.013*** [13.53] -0.127*** [-32.79] 0.001 [0.63] 0.044*** [32.45] 0.008*** [11.89] -0.050*** [-16.39] -0.003** [-2.36] 0.004 [1.27] -0.071*** [-27.23] 0.077*** [7.14] -0.034*** [-7.14] 0.006 [1.60] -0.077*** [-26.87] 0.095*** [7.59] -0.044*** [-8.00] -0.119*** [-19.07] -0.047*** [-15.51] 0.104*** [7.41] 0.079*** [11.80] 0.040 229,298 0.016 229,298 0.087 229,298 0.010 229,298 0.007 229,298 0.232 229,298 38 Dummy No Short-term Leverage Dummy No Long-term Leverage Table 5 Effect of the Reform on Loan Contracting Terms: DID Regressions This table shows the results from the following regression: Zi,j,t = α + αk + αt + βP ostt × T reatedi,j,t + Xi,t γ + Wi,j,t δ + εi,j,t , where Zi,j,t is either the loan spread, defined as the log of the sum of a loan’s coupon and annual fees scaled by its nominal value minus the six month LIBOR rate (in basis points), or the loan time-to-maturity, defined as the log of the difference between the loan’s maturity date and its initiation date (in months). Post is a dummy variable equal to one for years greater or equal to 2006 and else zero. Treated is a dummy variable equal to one for secured loans and else zero. Xi,t is a vector of firm-specific control variables. Wi,j,t is a vector of loan contract-specific control variables. Size is the log of total assets; Age is the log of the current year minus the year of incorporation; Profitability is the ratio of earnings before interest and taxes to total assets; and TotalLeverage is the ratio of the sum of short-term and long-term debt to total assets. Rating is a dummy variable equal to one if the firm taking out the loan is rated, else zero. CreditSpread is the mean monthly difference between the yields of a corporate bond index and a long-term government bond index, and TermSpread that between the yield of a long-term and a short-term government bond index, where means are calculated over the current year. LoanSize is the log of the notional value of the loan; and LoanType is a dummy variable equal to one for term loans, else zero. We also add the other endogenous variable to the control variables. α, β, δ, and γ are free parameters. αk and αt indicate industry- and year-fixed effects. Industry effects are based on 2-digit SIC codes. εi,t is the residual. Parameter estimates are in normal letter, while t-statistics are in square parentheses. T-statistics are calculated from standard errors clustered at the industry-level. ∗∗∗ , ∗∗ , and ∗ indicate statistical significance at the 99%, 95%, and 90% confidence levels, respectively. The sample period is 2001 to 2009. Dependent Variable Loan Spread Post × Treated -1.291*** [-4.09] -1.257*** [-5.16] 0.032 [1.07] -0.031 [-1.24] 0.265 [0.67] -0.065 [-0.43] -0.035 [-0.15] 1.086*** [11.70] 0.183*** [4.29] -0.080*** [-5.25] Size Age Profitability TotalLeverage Rating CreditSpread TermSpread LoanAmount Loan Maturity 1.078*** [5.58] 0.195*** [5.98] -0.063*** [-2.87] 0.786*** [6.27] -0.298*** [-3.44] -0.079*** [-3.65] -0.020 [-0.96] 0.871*** [6.94] -0.023 [-1.47] 0.033*** [3.83] -0.082 [-0.38] 0.100 [0.75] 0.168** [1.99] -0.388*** [-7.02] -0.066*** [-2.73] 0.001 [0.06] (continued on next page) 39 Table 5 Effect of the Reform on Loan Contracting Terms: DID Regressions (continued) Dependent Variable Loan Spread LoanType LoanMaturity 0.304*** [11.51] 0.249*** [5.17] Loan Maturity 0.314*** [10.77] 0.333*** [6.05] LoanSpread Treated R-squared Observations 1.550*** [16.77] 1.537*** [8.58] 0.445 456 0.473 409 40 0.032 [1.07] 0.201*** [3.64] -0.422*** [-3.69] 0.329 456 0.023 [0.74] 0.236*** [4.88] -0.576*** [-3.15] 0.410 409 41 Assets (in million $) Firm Age (in years) Number of Employees Operating Profit Operating Profit Volatility Cash Reserve Capital-to-Labor Tangible Asssets Fixed Assets Distance to Top 5 City (in miles) Distance to Top 10 City (in miles) Distance to Capital Market (in miles) 26.160 22.752 132.330 0.105 0.044 0.121 2.481 0.131 0.228 76.394 42.692 165.103 All Companies without Long-term Debt Before 2006 (1) 25.121 23.929 142.752 0.111 0.039 0.107 2.734 0.165 0.270 89.711 51.036 190.631 Companies Without Longterm Debt Before 2006 but Positive Long-term Debt in All Post-Reform Years (“Debt Switchers”) (2) 28.956 22.084 123.199 0.097 0.050 0.137 2.161 0.095 0.179 58.066 31.814 128.040 Companies Without Longterm Debt Before 2006 and No Long-term Debt in All Post-Reform Years (“Non-Switchers”) (3) -3.835 1.845 19.553 0.015 -0.012 -0.030 0.574 0.070 0.091 31.645 19.222 62.591 Difference (2)-(3) (0.00) (0.00) (0.27) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) p-value Difference (2) - (3) Table 6 Comparison of Debt Switchers and Non-Switchers This table compares several firm characteristics across firms that strongly benefitted from the reform and those that did not. The table only considers firms that never had any long-term debt before the reform (column (2)). From the firms in column (2), the reform beneficiaries are then those firms that had a positive long-term leverage in every year after the reform (column (3)), while the non-beneficiaries are those firms that continued to never have any long-term debt after the reform (column (4)). The difference in firm characteristics across beneficiaries and non-beneficiaries is given in column (5) and its p-value in column (6). The p-value is based on standard errors assuming unequal variances. To calculate the table entries, we first average a variable’s pre-reform values by firm and then by group. Assets is total assets in million $; Firm Age is the current year minus the year of incorporation; and Employees is the number of employees. Profitability is the ratio of earnings before interest and taxes to total assets; Profit volatility is the standard deviation of profitability over the most recent four fiscal years, including the most recent one. We set profit volatility equal to missing if it is based on fewer than three observations. Cash is cash reserves divided by total assets. The capital-to-labor ratio is the log of the ratio of tangible fixed assets to the number of employees; tangibility is the ratio of tangible fixed assets to total assets; and fixed assets is the ratio of fixed assets to total assets. We calculate the three distance variables by applying the formula for the spherical distance between two points to the longitudes and latitudes associated with a firm’s postcode and those associated with the city centers of the five/ten biggest French cities or the Paris Bourse. Table 7 Effect of the Reform on the Leverage of Start-Ups: ANOVA Tests This table shows the long-term leverage of start-ups in their year of incorporation by year and fixed asset quartile over the 2001–2009 period. We examine two attributes of long-term leverage in the table: the average (in Panel A) and the proportion of zero long-term leverage firms (in Panel B). Long-term leverage (“LongTermLeverage”) is defined as long-term debt scaled by total assets. Fixed assets (“FixedAssets”) is defined as fixed assets scaled by total assets. We sort firms into fixed asset quartiles according to the values of fixed assets and the fixed asset quartile breakpoints in the current year. Quartile Q1 contains firms with a low-fixed asset ratio, while Quartile Q4 contains firms with a high ratio. At the bottom of the table, we report averages for the pre-reform period attributes (2001–2005) and averages for the post-reform period attributes (2006–2009), and their differences. The final column shows the differences in attributes across the highest (Q4) and lowest (Q1) fixed asset quartile. ∗∗∗ ∗∗ , , and ∗ indicate statistical significance at the 99%, 95%, and 90% confidence levels, respectively, with standard errors clustered at the firm level (whenever possible). Fixed Asset Quartile Year Q1 Q2 Q3 Q4 Q4–Q1 Panel A: Long-term Leverage 2001 0.023 0.006 0.014 0.021 0.050 0.043** 2002 0.019 0.000 0.001 0.015 0.061 0.061*** 2003 0.031 0.000 0.027 0.025 0.073 0.073*** 2004 0.027 0.004 0.027 0.041 0.039 0.036** 2005 0.036 0.018 0.022 0.023 0.081 0.063* 2006 0.048 0.015 0.029 0.040 0.111 0.096*** 2007 0.053 0.022 0.018 0.055 0.119 0.098*** 2008 0.057 0.017 0.025 0.073 0.116 0.099*** 2009 0.072 0.010 0.022 0.116 0.146 0.136*** Mean 2001–2005 (1) Mean 2006–2009 (2) 0.027 0.058 0.006 0.016 0.018 0.023 0.025 0.071 0.061 0.123 Diff. (2)–(1) 0.031*** 0.010 0.005 0.046*** 0.062*** Panel B: Proportion of Zero Long-term Leverage Firms 2001 0.903 0.926 0.907 0.907 0.870 -0.056 2002 0.931 1.000 0.952 0.935 0.839 -0.161*** 2003 0.886 1.000 0.873 0.813 0.857 -0.143*** 2004 0.892 0.963 0.927 0.866 0.806 -0.157*** 2005 0.904 0.981 0.900 0.936 0.783 -0.199*** 2006 0.751 0.954 0.780 0.561 0.684 -0.270*** 2007 0.742 0.938 0.877 0.644 0.500 -0.438*** 2008 0.714 0.918 0.821 0.482 0.607 -0.311*** 2009 0.741 0.953 0.778 0.600 0.588 -0.365*** Mean 2001–2005 (1) Mean 2006–2009 (2) 0.903 0.737 0.974 0.941 0.912 0.814 0.891 0.572 0.831 0.595 Diff. (2)–(1) All Diff. -0.166*** -0.033** -0.098*** 42 -0.320*** -0.236*** 0.052** -0.203*** 43 0.026 0.086 0.060 0.049*** Mean 2001–2005 (1) Mean 2006–2009 (2) Diff. (2)–(1) Mean 2001–2005 (3) Mean 2006–2009 (4) Diff. (3)–(4) Diff. [(3)–(4)]–[(1)–(2)] Q4 0.009 0.020 0.011 Q1 0.100*** 0.023 0.128 0.105 0.005 0.011 0.005 Q4 Land Index Quartiles Q1 Quartile Fixed Asset 0.051*** -0.002 0.042*** 0.045*** -0.003*** -0.009*** -0.006*** Q4–Q1 Diff. 0.051*** 0.027 0.093 0.066 0.008 0.022 0.015 Q1 0.091*** 0.023 0.123 0.100 0.010 0.018 0.009 Q4 Building Index Quartiles 0.040*** -0.004** 0.030*** 0.034*** 0.002 -0.004** -0.006*** Q4–Q1 Diff. 0.044*** 0.023 0.081 0.058 0.008 0.021 0.014 Q1 0.056*** 0.025 0.092 0.067 0.011 0.022 0.011 Q4 Machinery Index Quartiles 0.012** 0.002 0.011*** 0.009** 0.004*** 0.001 -0.003 Q4–Q1 Diff. Table 8 Effect of the Reform on Firms with Different Fixed Asset Types: ANOVA Tests This table shows mean long-term leverage by fixed asset quartile and one of three fixed asset-type indexes for the 2001–2005 period and the 2006–2009 period. The fixed asset-type indexes are a land, a building, and a machinery index. Long-term (“LongtermLeverage”) is defined as long-term debt scaled by total assets. Fixed assets (“FixedAssets”) is defined as fixed assets scaled by total assets. The LandIndex is the ratio of land-to-total assets, the BuildingIndex index the ratio of buildings-to-total assets, and the MachineryIndex the ratio of machinery and equipment-to-total assets, first averaged by 4-digit SIC code industry and month and then by 4-digit SIC code industry. The fixed asset-type indexes are constructed from global COMPUSTAT and WorldScope data. We double-sort firms into quartiles according to their fixed assets and fixed asset-type index values and the fixed asset and fixed asset-type index quartile breakpoints in the current year. Quartile Q1 contains firms with a low sorting variable value, while Quartile Q4 contains firms with a high sorting variable value. For the sake of brevity, the table only shows results for the highest (Q4) and lowest (Q1) quartiles. The table reports the time-series average of the annual numbers for the pre-reform period (2001–2005) and the post-reform (2006–2009) period, and their differences (Diff). Conditioning on the fixed asset (fixed asset-type index) quartile, it also reports the difference in the change in long-term leverage across the highest and lowest fixed asset-type index (fixed asset) quartile. ∗∗∗ , ∗∗ , and ∗ indicate statistical significance at the 99%, 95%, and 90% confidence levels, respectively, based on standard errors clustered at the firm level. 44 +γP ostt × HighCourtEf f iciencyk + δP ostt × T reatedi + Xi,t η + εi,t , = α + αi + αt + βP ostt × T reatedi × HighCourtEf f iciencyk 4.01% 227,428 R-squared Observations Age Profitability Size Post × Treated Post × HighCourtEfficiency 0.000 [-0.01] 0.000 [-0.05] 0.040*** [15.90] 0.021*** [16.27] -0.189*** [-36.20] -0.002 [-0.89] Post × Treated × HighCourtEfficiency Total Leverage 1.62% 227,428 0.002 [0.88] 0.002** [2.26] -0.003* [-1.89] 0.013*** [13.52] -0.128*** [-32.88] 0.001 [0.78] Short-term Leverage Intensive Margin 8.76% 227,428 -0.001 [-0.50] -0.003*** [-3.39] 0.044*** [23.01] 0.008*** [11.73] -0.050*** [-16.15] -0.003** [-2.34] Long-term Leverage 0.94% 227,428 0.64% 227,428 0.004 [0.54] -0.001 [-0.18] 0.004 [0.83] -0.077*** [-26.78] 0.096*** [7.64] -0.044*** [-7.95] Dummy Short-term Leverage Extensive Margin 0.003 [0.43] 0.001 [0.16] 0.003 [0.66] -0.071*** [-27.06] 0.076*** [7.06] -0.034*** [-7.13] Dummy Total Leverage Dependent Variable 23.51% 227,428 -0.034*** [-2.77] 0.037*** [5.85] -0.100*** [-11.45] -0.046*** [-15.23] 0.102*** [7.26] 0.078*** [11.59] Dummy Long-term Leverage where Yi,t is total, short-term, or long-term leverage or dummy variables indicating whether the former quantities are zero. Total (short-term) [longterm] leverage is the sum of short-term and long-term (short-term) [long-term] debt scaled by total assets. Dummy no total (short-term) [long-term] leverage is a dummy variable equal to one if total (short-term) [long-term] leverage is zero and else zero. Post is a dummy variable equal to one for years greater or equal to 2006 and else zero. Treated is a dummy variable equal to one for firms whose mean fixed asset-to-total asset ratio is in the top quartile and else zero. HighCourtEfficiency is a dummy variable equal to one for firms located in Departments for which the average length of insolvency proceedings is below the sample median and else zero. Xi,t is a vector of control variables, including size, profitability, and age. Size is the log of total assets; Profitability is the ratio of earnings before interest and taxes to total assets; and Age is the log of the current year minus the year of incorporation. α, β, γ, δ, and η are free parameters. αi and αt indicate firm- and year-fixed effects. εi,t is the residual. Parameter estimates are in normal letter, while t-statistics are in square parentheses. T-statistics are calculated from standard errors that are clustered at the firm-level. ∗∗∗ , ∗∗ , and ∗ indicate statistical significance at the 99%, 95%, and 90% confidence levels, respectively. The sample period is 2001 to 2009. Yi,t Table 9 Intermediating Effect of Court Efficiency on the Reform’s Financing Results: DIDID Regressions This table shows the results from the following regression: Table 10 Effect of the Reform on Firm Performance: DID Regressions This table shows the results from the following regression: Yi,t = α + αi + αt + βP ostt × T reatedi + Xi,t γ + εi,t , where Yi,t is growth, employees, sales, profitability, or profit volatility. Growth is the sum of the change in intangible fixed assets and the change in inventories, where both changes are calculated from the prior fiscal year end to the current one and the sum is scaled by the average of total assets over the two fiscal year ends. Employees is the log of the number of employees. Sales is the log of sales. Profitability is earnings before interest and taxes scaled by total assets. ProfitVolatility is the standard deviation of Profitability over the most recent four fiscal years, including the most recent one. We set ProfitVolatility equal to missing if it is based on fewer than three observations. Post is a dummy variable equal to one for years greater or equal to 2006 and else zero. Treated is a dummy variable equal to one for firms whose mean fixed asset-to-total asset ratio is in the top quartile and else zero. Xi,t is a vector of control variables, including size, total leverage, and age. Size is the log of total assets; TotalLeverage is total debt scaled by total assets; and Age is the log of the current year minus the year of incorporation. α, β, and γ are free parameters. αi and αt indicate firm- and year-fixed effects. εi,t is the residual. Parameter estimates are in normal letter, while t-statistics are in square parentheses. T-statistics are calculated from standard errors that are clustered at the firm-level. ∗∗∗ , ∗∗ , and ∗ indicate statistical significance at the 99%, 95%, and 90% confidence levels, respectively. The sample period is 2001 to 2009. Dependent Variable Growth Post × Treated Size TotalLeverage Age R-squared Observations 0.006*** [6.90] 0.045*** [49.11] -0.021*** [-7.57] -0.045*** [-21.99] 0.014 209,979 Employees 0.051*** [7.05] 0.503*** [59.58] -0.010 [-0.63] 0.171*** [18.61] 0.450 174,872 45 Sales Profitability Profit Volatility 0.041*** [5.69] 0.780*** [102.87] -0.223*** [-13.60] 0.211*** [24.59] 0.005*** [4.39] 0.012*** [12.54] -0.106*** [-36.50] 0.021*** [13.95] -0.002*** [-4.08] -0.010*** [-16.98] 0.012*** [8.26] -0.018*** [-10.79] 0.597 231,112 0.004 229,298 0.035 158,374 Table 11 Effect of the Reform on Failure Rates: DID Regressions This table shows the results from the following proportional Cox hazard model: λi,t = φt exp (βP ostt × T reatedi + Xi,t γ) , where the hazard rate, λi,t , is the probability of firm i failing at time t conditional on surviving until then, and φt is the “baseline” hazard rate common to all firms. We set a firm’s failure year to the calendar year in which its legal status changes from an active status to one of the failure statuses: “default of payment”, “insolvency proceedings,” “receivership,” “bankruptcy,” “dissolved (bankruptcy),” “dissolved (liquidation),” and “in liquidation.” We exclude a firm from the analysis if it is unclear whether it became inactive for performance reasons, that is, when legal status changes to: “inactive (no precision),” “unknown,” and “dissolved.” Post is a dummy variable equal to one for years greater or equal to 2006 and else zero. Treated is a dummy variable equal to one for firms whose mean fixed asset-to-total asset ratio is in the top quartile and else zero. Xi,t is a vector of control variables, including total leverage, profitability, profit volatility, size, and cash reserves. TotalLeverage is the sum of short-term and long-term debt to total assets. Profitability is the ratio of earnings before interest and taxes to total assets. ProfitVolatility is the standard deviation of Profitability over the most recent four fiscal years. We set ProfitVolatility equal to missing if it is based on fewer than three observations. Size is the log of total assets, and Cash is cash reserves over total assets. β and γ are free parameters. Parameter estimates are in normal letter, while t-statistics are in square parentheses. T-statistics are calculated from standard errors that are clustered at the firm-level. ∗∗∗ , ∗∗ , and ∗ indicate statistical significance at the 99%, 95%, and 90% confidence levels, respectively. The sample period is 2001 to 2009. Dependent Variable = Time-to-Failure Post × Treated -0.338*** [-4.36] -0.391*** [-5.01] -0.458*** [-5.74] 0.821*** [4.84] -0.112*** [-4.89] -3.497*** [-10.71] -0.108*** [-4.75] -3.264*** [-9.92] TotalLeverage Profitability ProfitVolatility Size Cash R-squared Observations 0.001 182,249 0.008 177,904 46 0.009 177,904 -0.206** [-2.13] 0.295 [1.61] -4.612*** [-12.09] 2.630*** [3.09] -0.090*** [-3.22] -2.690*** [-6.48] 0.037 127,767 Table 12 Effect of the Reform on Capital Allocation This table shows the results from the following panel-data regression: InvestmentGrowthk,t = α + βV alueAddedGrowthk,t + εk,t , where InvestmentGrowthk,t is gross fixed capital formation for industry k in year t, and V aluedAddedGrowthk,t is the value-added. We perform this regression using pre-reform (2001–2005) and post-reform (2006–2007) data. α and β are free parameters, and εk,t is the residual. In addition to the parameter estimates and their inference levels (T-Statistic), the table shows the number of observations (Obs) and the R-squared (R-squared) per regression, and it also reports the difference in the slope coefficient across the two periods. Year Obs Elasticity (β) T-Statistic R-Squared 2001-2005 (1) 2006-2007 (2) 566 237 0.434 0.697 6.15 5.34 0.062 0.104 0.263 1.79 Diff (2)–(1) 47 48 All (1) 0.016 0.015 0.015 0.014 0.016 0.051 0.057 0.069 0.079 0.015 0.063 0.048*** Year 2001 2002 2003 2004 2005 2006 2007 2008 2009 Mean 2001–2005 (1) Mean 2006–2009 (2) Diff (2)–(1) 0.082*** 0.023 0.104 0.023 0.022 0.021 0.022 0.026 0.085 0.095 0.115 0.125 Treated (2) High 0.030*** 0.012 0.042 0.013 0.012 0.011 0.010 0.012 0.033 0.037 0.045 0.054 Controls (3) 0.029*** 0.015 0.044 0.017 0.015 0.014 0.013 0.014 0.037 0.039 0.048 0.053 All (4) Low 0.053*** 0.024 0.077 0.027 0.022 0.024 0.023 0.024 0.065 0.071 0.082 0.091 Treated (5) External Financial Dependence 0.025*** 0.013 0.038 0.015 0.014 0.013 0.011 0.012 0.031 0.033 0.042 0.046 Controls (6) 0.019*** -0.001 0.000 0.000 0.001 0.003*** 0.014*** 0.018*** 0.021*** 0.026*** All (1)–(4) 0.029*** -0.004* 0.000 -0.003 -0.002 0.001 0.020*** 0.024*** 0.033*** 0.034*** Treated (2)–(5) Dependence Across High–Low External Financial Table 13 Effect of the Reform on Firms with a High and Low External Financial Dependence: ANOVA Tests This table shows the mean long-term leverage of all, high fixed asset-, and low fixed asset-firms with a high or low dependence on external financing, by year and over the 2001–2009 period. Long-term leverage (“TotalLeverage”) is defined as long-term debt scaled by total assets. Fixed assets (“FixedAssets”) is defined as fixed assets scaled by total assets. We obtain an industry-specific measure of external finance dependence (“ExternalFinancialDependence”) by calculating the median proportion of capital expenditure that are not financed by cash flows from operations for U.S. public firms. We perform these calculations using COMPUSTAT data over the period from 1975–2005. We classify as firms with a high (low) external financial dependence those operating in industries with an above (below) median ExternalFinancialDependence. The treated firms are those with a mean fixed asset-to-total asset ratio in the top quartile, while the control firms are those in the remaining three quartiles. At the bottom of the table, we report averages for the pre-reform period means (2001–2005) and averages for the post-reform period means (2006–2009), and their differences. The penultimate (final) column shows the differences in means across all firms (treated firms) with a high and those with a low external financial dependence. ∗∗∗ , ∗∗ , and ∗ indicate statistical significance at the 99%, 95%, and 90% confidence levels, respectively, based on standard errors clustered at the firm level. Table 14 Migration Rates Between the High and Low Fixed Asset Groups This table shows migration rates between the high- and the low-fixed asset group over various five year-periods in the 1996–2009 period. The migration rates are calculated using the first and the last year in the five year period. A firm belongs to the high-fixed asset group if its fixed asset-to-total asset ratio is in the top quartile in a given year. Otherwise, it belongs to the low-fixed asset group. Period Moved Into the High-Fixed Asset Group Stayed in the Same Group Moved Into the Low-Fixed Asset Group 1996-2000 0.077 0.852 0.071 1997-2001 0.084 0.845 0.071 1998-2002 0.078 0.853 0.069 1999-2003 0.073 0.857 0.070 2000-2004 0.075 0.860 0.065 2001-2005 0.068 0.862 0.070 2002-2006 0.065 0.867 0.068 2003-2007 0.066 0.869 0.065 2004-2008 0.064 0.866 0.070 2005-2009 0.060 0.869 0.071 49 Table 15 Effect of the Reform on the Long-term Leverage of Public Firms: ANOVA Tests This table shows the mean long-term leverage of public firms by year and fixed asset quartile over the 2001–2009 period. Long-term leverage (“LongTermLeverage”) is defined as long-term debt scaled by total assets. Fixed assets (“FixedAssets”) is defined as fixed assets scaled by total assets. We sort firms into fixed asset quartiles according to the values of fixed assets and the fixed asset quartile breakpoints in the current year. Quartile Q1 contains firms with a low-fixed asset ratio, while Quartile Q4 contains firms with a high ratio. At the bottom of the table, we report averages for the pre-reform period means (2001–2005) and averages for the post-reform period means (2006–2009), and their differences. The final column shows the differences in means across the highest (Q4) and lowest (Q1) fixed asset quartile. ∗∗∗ , ∗∗ , and ∗ indicate statistical significance at the 99%, 95%, and 90% confidence levels, respectively, based on standard errors clustered at the firm level. Fixed Asset Quartile Diff. Year All Q1 Q2 Q3 Q4 Q4–Q1 2001 0.092 0.047 0.054 0.089 0.179 0.132*** 2002 0.106 0.059 0.073 0.116 0.177 0.119*** 2003 0.098 0.051 0.057 0.095 0.186 0.135*** 2004 0.097 0.045 0.058 0.104 0.183 0.138*** 2005 0.085 0.029 0.064 0.091 0.156 0.128*** 2006 0.102 0.047 0.078 0.121 0.162 0.116*** 2007 0.112 0.093 0.074 0.118 0.162 0.069* 2008 0.134 0.144 0.070 0.151 0.171 0.027 2009 0.118 0.069 0.096 0.132 0.174 0.105*** Mean 2001–2005 (1) Mean 2006–2009 (2) 0.096 0.116 0.046 0.088 0.061 0.080 0.099 0.131 0.176 0.167 Diff. (2)–(1) 0.021** 0.042 0.019 0.032 -0.009 50 -0.051*