When Credit Dries Up: Job Losses in the Great Recession by ** Samuel Bentolila^*, Marcel Jansen Gabriel Jiménez^*** and Sonia Ruano^***

Documento de Trabajo 2013-20

December 2013

* CEMFI, CEPR and CESifo. ** Universidad Autónoma de Madrid, Fedea e IZA. *** Banco de España

Los Documentos de Trabajo se distribuyen gratuitamente a las Universidades e Instituciones de Investigación que lo solicitan. No obstante están disponibles en texto completo a través de Internet: http://www.fedea.es. These Working Paper are distributed free of charge to University Department and other Research Centres. They are also available through Internet: http://www.fedea.es. ISSN:1696-750

When Credit Dries Up: Job Losses in the Great Recession

Samuel Bentolila CEMFI

Gabriel Jiménez Banco de España

Marcel Jansen Universidad Autónoma de Madrid

Sonia Ruano Banco de España

November 2013

Abstract

We use a unique dataset to estimate the impact of a large credit supply shock on employment in Spain. We exploit marked diferences in banks’ health at the onset of the Great Recession. Several weak banks were rescued by the State and they reduced credit more than other banks. We compare employment changes from 2006 to 2010 at firms heavily indebted to weak banks before the crisis and the rest. Our estimates imply that these firms sufered an additional employment drop between 3 and 13.5 percentage points due to weak-bank attachment, representing between 8% and 36% of aggregate job losses.

Keywords: Job losses, Great Recession, credit constraints. JEL codes: D92, G33, J23.

∗Bentolila is also afiliated with CEPR and CESifo, Jansen is also afiliated with Fedea and IZA. This paper is the sole responsibility of its authors. The views presented here do not necessarily reflect those of the Banco de España or the Eurosystem. We are very grateful to Joshua Angrist, Manuel Arellano, Stéphane Bonhomme, Gabriel Chodorow-Reich, Juan J. Dolado, Markus Demary, David Dorn, Pietro Garibaldi, Tullio Jappelli, Juan F. Jimeno, Pedro Mira, Rafael Repullo, Javier Suarez, Etienne Wasmer, and seminar audiences at Banco de España, CEMFI, European Central Bank, IMT Lucca, Universidad Autónoma de Madrid, Universidad de Alcalá de Henares, Universitat Autònoma de Barcelona, and Universität Mainz, and at conferences at Banco de Portugal, CSEF (Naples), De Nederlandsche Bank, IZA/CEPR European Summer Symposium on Labor Economics, and Kiel Institute for the World Economy for their helpful comments. We also wish to thank Ana Esteban and José I. González-Pascual, from the Statistics Ofice of the Banco de España, for help with the data. All errors are our own. Jansen acknowledges funding from the Ministerio de Economía y Competitividad Grant ECO2012-37742.

1 Introduction

Policymakers in both Europe and the US are concerned about the economic implications of the current shortage of credit. As the International Monetary Fund puts it in a recent review of advanced economies’ eforts to revive their credit markets, “policymakers want to support markets because the decline in lending is seen to be a primary factor in the slow recovery” (IMF, 2013). The available evidence on the economic implications of the recent decline in lending is however still rather limited. For some of the most afected countries, like the US, there is a lack of good data on bank credit to firms and this poses a problem for identification.

In this study we contribute to the literature by providing new data that help to resolve the identification problems. We aim at estimating the impact of the fall in bank lending on employment in Spain during the Great Recession. The Spanish economy ofers an ideal setting to explore this question. To start with, Spanish firms rely heavily on bank credit and the high leverage ratio of many firms, mostly small and medium-sized enterprises (SMEs), made them vulnerable to the contraction in bank lending that took place during the recession. This credit shock originated in a boom-bust cycle in domestic housing prices that had a large impact on bank solvency. Thus, the Spanish example may help us to draw lessons that are applicable to other countries, like Ireland or the US, which also experienced a collapse of their housing markets and a strong rise in unemployment. Last but not least, the extraordinary quality of our data allows us to address the challenge of disentangling credit demand from credit supply shocks.

Our dataset draws from several sources. We have access to the oficial credit register of the Bank of Spain, which contains detailed information on virtually all existing and newly-granted loans to non-financial firms. Using these data we are able to reconstruct the complete banking relationships of over 217,000 companies working with almost 230 banks. We also have information on loan demand through loan applications from non-current customers of banks and on whether the applications are granted or not. All this information is linked to the balance sheets of all banks operating in Spain and to the balance sheets and income statements of the firms in our sample. The result is, as far as we know, the most comprehensive matched firm-loan-bank data set ever assembled to estimate the real efects of shocks to the banking system.

Our empirical strategy exploits large diferences in lenders’ health at the onset of the crisis. The collapse of the housing bubble afected all Spanish banks, but the impact on their solvency was far from uniform. Only a subset of the banks, all but one of them savings banks (called Cajas de Ahorros in Spanish), needed to be rescued by the State. The rescue entailed either a merger of banks or a solvent bank taking over an ailing bank, usually with loans and guarantees from the public sector, or a bank’s nationalization and recapitalization, sometimes followed by reprivatization via auction. Before the crisis, these bailed-out or weak banks accumulated a very large share of the loans to the real estate industry and, between the outbreak of the crisis and the end of 2010, these same banks reduced credit more than the other banks. To capture the real efects of this credit supply shock, we compare the changes in employment from 2006 to 2010 at two sets of firms: those with a high and those with a low exposure to weak banks, where exposure is measured as the pre-crisis ratio between a firm’s loans from weak banks and its asset value.

The underlying assumption is that the client firms of weak banks could not predict this credit shock when they chose their banking relations. In addition, they must not have been able to readily switch to healthier banks after the outbreak of the crisis. We will provide evidence to corroborate both claims. Finally, we avoid the risk of reverse causality by removing from our sample all firms belonging to the real estate industry as well as those selling a significant share of their output to this industry.

Our final goal is to replicate as closely as possible the conditions of a natural experiment in which some of the firms are randomly assigned to weak banks and others to healthy banks. This strategy requires the possibility of comparing firms in many dimensions, in order to achieve homogeneity between treated and control firms. In our benchmark, we estimate the impact of weak-bank attachment on employment —the so-called average treatment efect on the treated— using a diference-in-diferences specification with a large set of firm controls. But in order to minimize the risk of selection, we also make use of matching estimators. Furthermore, we show that there is a causal link between the differences in employment growth between firms in the treatment and the control group and their access to credit during the crisis. In these exercises, weak-bank attachment is used as an instrument for observed changes in credit and the predicted changes are subsequently used to explain the diferences in employment growth at the firm level. Lastly, in what is our most ambitious test, we check on the potential endogeneity of banking relationships by exploiting a change in banking regulation in 1988. This legal change liberalized the location decisions of savings banks, allowing them to expand freely beyond their region of origin. Thus we use the share of bank branches at the municipal level that belonged to weak banks right before this legal change as an instrument for weak-bank attachment in 2006.

Regardless of the approach followed, we find the same qualitative result. Firms with a relatively large exposure to weak banks at the start of the crisis destroyed a larger share of their jobs between 2006 and 2010 than other firms. Once selection efects are controled for, our estimates indicate that they destroyed an additional 3.0 to 13.5 percentage points, which in our sample would ceteris paribus represent between 8% and 36% of aggregate job losses. We also find large diferences across firms belonging to diferent industries and across firms with diferent credit histories. Moreover, credit constraints are shown to have operated mostly through firm closures than through employment adjustment at surviving firms.

The rest of the paper is organized as follows. In Section 2 we review previous theoretical and empirical work on our topic and in Section 3 we provide some institutional and background information on the Spanish economy before and during the financial crisis. Section 4 describes our data, Section 5 presents our empirical strategy and key results, and Section 6 presents an extensive battery of robustness checks. Section 7 contains our conclusions. The Appendix provides a detailed description of the variables used.

2 Literature review

Our identification strategy requires the existence of financial market frictions. In particular, firms attached to weak banks must not have been able to readily switch to healthier banks. Two strands of the literature provide the theoretical basis for this result.

The literature on financial accelerator mechanisms has shown that endogenous changes in credit market conditions may amplify shocks to the real economy. In these models, asymmetric information drives a wedge between the cost of internal and external funds that depends negatively on a firm’s net worth. Negative shocks to net worth are therefore associated with a rise in the external finance premium and this may force firms to cut back on their scale of operation (the net-worth efect).¹ Furthermore, the theory suggests that in recessions firms with weak balance sheets should be the main victims of this tightening of credit constraints (the flight to quality).

The bulk of the financial accelerator literature treats financial intermediation as a veil, but Gertler and Kiyotaki (2009) illustrate how the theory can be adapted to incorporate agency problems between capital-constrained banks and their lenders. In their setup, a negative shock to the bank’s net worth may generate a disruption in both the interbank market and the credit flow from banks to firms. The result is an ineficient allocation of capital and a drop in investment. We should stress that the fall in credit supply is not necessarily uniform. Banks with a relatively high leverage ratio are more vulnerable to shocks and this negatively afects their clients.

¹The initial studies focused on investment in physical capital. Early contributions considering the efect of financial constraints on employment are Greenwald and Stiglitz (1993) and Sharpe (1994).

The second line of research focuses on the role of relationship banking (e.g. Freixas, 2005). Repeated interaction with a client often provides soft information that allows a better assessment of the firm’s future profitability. This explains why banks may give a preferential treatment to their incumbent clients when capital is scarce and why a profitable firm may not be able to find alternative sources of funding.

The same literature is less clear-cut on the optimal number of banking relations. A strong relationship with a single bank reduces transaction costs and makes it easier to restructure the firm’s debt in case of financial distress. But attachment to a single bank may also impede the firm from undertaking a profitable project due to financial distress on the part of the bank, as the presence of asymmetric information may impose high switching costs for borrowers. Firms may therefore prefer to establish relationships with several banks to insure themselves against this type of liquidity risk (Detragiache et al., 1990).

Finally, firms that are more prone to sufer from credit constraints can use several strategies to reduce the impact of these future constraints. One option is to maintain a bufer stock of liquid assets. Another option is to maintain a fringe of flexible workers on fixed-term contracts. Ex ante this makes firms less vulnerable to financial shocks, but ex post it may also provoke quick and sizeable adjustments in employment levels (Caggese and Cuñat, 2009).

Moving now to the empirical literature, in recent years there has been a surge of studies exploiting quasi-experimental techniques to estimate the real efects of credit supply shocks. Broadly speaking, we can divide them into three groups depending on their identification strategy. A first strand of papers exploits the heterogeneous impact of large external shocks to banks in the US (e.g. Chava and Purnanandam, 2011, or Benmelech et al., 2012). A second line of work exploits cross-sectional diferences in the financial vulnerability of firms at the start of the Great Recession. Almeida et al. (2011), Benmelech et al. (2011), and Boeri et al. (2013) exploit diferences in the debt maturity structure of firms. Since this maturity is often determined years in advance, it leads to fairly exogenous diferences in firms’ refinancing needs at a time when capital becomes very scarce. Similarly, Garicano and Steinwender (2013) try to elicit the impact of credit constraints in Spain by comparing the evolution of investment and employment at nationally-based manufacturing firms with foreign-owned ones, which have better access to credit.

The third route, which is the one adopted here, is to exploit cross-sectional diferences in bank health. Greenstone and Mas (2012) construct a countylevel credit supply shock from the product of the change in US banks’ smallbusiness lending at the national level and their predetermined credit market share at the county level. They find that this measure is highly predictive of the considerable reduction in county-level credit to small, standalone firms and in their employment levels in the period going from 2008 to 2010. Similarly, Chodorow-Reich (2013) uses data from the Dealscan syndicated loan database and measures the relative health of a firm’s lenders using the reduction in lending to other borrowers during the crisis by the firm’s pre-crisis syndicate. This data is matched to confidential data from the Bureau of Labor Statistics Longitudinal Database for a sample of just over 2,000 firms. In line with Greenstone and Mas (2012), he finds that relatively smaller firms that had pre-crisis relationships with less healthy banks faced stronger credit constraints after the fall of Lehman Brothers and reduced their employment more compared to clients of healthier banks. By contrast, for larger companies there are no significant efects.

It should be stressed that none of the papers above have access to a credit register. Nor do they have access to information about the loan applications or the credit history of firms. As explained in the Introduction, the access to loan level data with detailed financial information about lenders and borrowers is crucial for identifying shocks to credit supply. Moreover, our loan application data are a unique source of information about the extent of the credit constraints faced by Spanish firms. They allow us not only to control for cross-sectional diferences in the financial vulnerability of both borrowers and lenders, but also to perform a wide range of robustness tests that cannot be replicated with the available data for the US. Finally, our sample of firms is roughly one-hundred times bigger than the one in the closest-related study of Chodorow-Reich (2012) and it predominantly contains SMEs that according to the theory are most susceptible to changes in credit market conditions.

3 The financial crisis in Spain

The Spanish economy has experienced an acute credit crunch in the Great Recession. In this section we briefly document its magnitude and origins, focusing on the role played by weak banks. We end with some evidence showing that financial markets failed to anticipate the economic troubles of these weak banks.

3.1 The credit collapse

Spain provides an ideal setting to study credit constraints arising unanticipated reductions in bank lending. To start with, Spanish firms rely more on bank credit than their counterparts in most other developed countries. For example, in 2006 the stock of loans from credit institutions to non-financial corporations represented 86% of GDP vis-à-vis 62% on average in the EU.² Moreover, alternative sources of funding are hard to come by. In particular, corporate debt issue is not an option: over the period 2002-2010 on average only five very large companies issued debt in the market each year. And very few firms are quoted in the stock market. Indeed, our sample only contains 28 listed firms (i.e. 0.01% of our sample).

Secondly, the latest Spanish business cycle coincided with a boom-bust cycle in the credit market. The Spanish economy experienced an expansion from 1996 to 2007, with GDP and employment respectively growing at 3.7% and 4.1% per annum. By contrast, GDP fell by 1.1% per annum over 2008-2010 and by the end of 2010 employment had fallen by 10%, while the unemployment rate had soared from 8.6% to 20.3%. At the same time, credit grew very rapidly during the boom and fell precipitously in the bust. The annual average flow of new credit to non-financial firms by deposit institutions increased in real terms by 23% from 2003 to its peak in 2007, subsequently falling by 38% in the period to 2010.

²Source: European Central Bank (2010), Annex Tables 4 and 14.

The credit crunch resulted from the interaction of the international financial crisis and domestic events. During the boom, the expansionary monetary policy pursued by the European Central Bank (ECB) induced Spanish banks to take on more risks (the risk-taking channel of monetary policy). In particular, they fueled a housing market bubble with cheap loans to real estate developers and construction companies —real estate industry or REI, hereafter—, as well as homeowners (Jiménez et al., 2013). The stock of loans to the REI grew from 14.8% of GDP 2002 to 43% in 2007. As a result, housing prices rose by 56% in real terms over 2003-2007, while by the end of 2010 they had fallen by 15%.

Two features of the Spanish banking regulation helped to protect banks in the initial stage of the crisis (Jiménez et al., 2012b). The Spanish regulator, namely the Bank of Spain, forced banks to keep securitized assets on their balance sheets, and in 2000 it implemented a system of dynamic provisioning, which obliged banks to build up provisions against unrealized loan losses. The banks had however funded a significant fraction of their new lending by issuing debt abroad and were therefore acutely hit by the freezing of wholesale Eurozone markets in 2008. The European Central Bank ofered relief to Euro area banks, but the losses at REI firms increasingly threatened the solvency of many banks and this induced them to curtail lending.

What happened to interest rates? The average interest rate on loans to non-financial companies rose from 3.3% in 2005:11 to 5.9% in 2008:09 —closely following the path of the ECB’s policy rate. However, it steadily fell thereafter to 2.4% in 2010:5, rising again to 3% by the year’s end. Thus, while there was tightening at the beginning of the recession, it was sharply reversed upon Lehman Brothers’ bankruptcy. Moreover, while in the recession weak banks started charging firms a higher interest rate for their loans than healthy banks, the diference was quite small, on average less than 30 basis points. For this reason, we focus on the volume of credit rather than on interest rates.

3.2 The demise of the savings banks

The buildup of risks was not uniform across banks, with major risks being concentrated in the savings banks. One factor that may have contributed to the diferential buildup of risks is the peculiar governance of savings banks. These banks were not exposed to the same market discipline as private banks, as they were not listed on the stock market and de facto they were controled by the corresponding regional government (see Cuñat and Garicano, 2010).

Solvency problems at savings banks eventually had to be dealt with through State bailouts. These entailed either a merger of banks or a solvent bank taking over an ailing bank, usually with loans and guarantees from the public sector, or a bank’s nationalization and recapitalization, sometimes followed by reprivatization via auction. Between 2006 and 2011 the number of savings banks went down from 47 to 11, but over our sample period (2006-2010) nationalization only afected two very small savings banks (International Monetary Fund, 2012). Throughout the rest of the analysis we define weak banks as those banks that obtained funding from the State in order to survive (this set only includes one private, non-savings bank). We refer to the remainder as healthy banks, including those that received funds to acquire ailing banks. To fund the recapitalization of weak banks, the Spanish Government obtained a loan of 41.4 billion euros (around 4% of GDP) from the European Financial Stability Facility in June 2012.

Our empirical strategy exploits the diferences between weak and healthy banks. In 2006 the former accounted for about one-third of outstanding credit to the non-financial sector. While the REI represented one-third of loans at healthy banks, it comprised almost two-thirds at weak banks (Panel A of Table 1). This explains the considerable diferences in lender health at the onset of

the crisis.

Furthermore, credit grew more at weak than at healthy banks during the boom —in real terms, 60% v. 12% from 2003 to 2007— and it fell more during the slump —46% v. 35% from 2007 to 2010. New credit to non-financial firms fell by 46% from 2006 to 2010 for weak banks and 5% for healthy banks (Panel B of Table 1).

These evolutions stemmed from changes at both the extensive margin (credit to new customers) and the intensive margin (new credit to current clients). Fig ure 1 depicts acceptance rates for loan applications by non-client firms. As a rough control for firm quality, we report acceptance rates for firms applying simultaneously to at least one weak and one healthy bank. During 2002-2004 acceptance rates were 6.5 percentage points (hereafter pp) higher for weak than for healthy banks, then both rates fell precipitously during 2007-2008, and subsequently acceptance rates switched to being 6.3 pp lower for weak banks in 2009-2010.

3.3 Were weak bank troubles anticipated?

The diferential buildup in risks at the two sets of banks is striking. But could firms anticipate the solvency problems of weak banks? If so, our identification procedure would be invalid. To study this issue, we analyze the risk premia charged to Spanish banks’ securitization issues prior to the recession. We employ data on tranches of mortgage backed securities (MBS) and asset backed securities (ABS) in 2006. By the end of this year, the ratio of securitization to total assets was significantly higher for weak than for healthy banks, 16.7% and 13.5% respectively.

We group the ratings into three categories: prime (AAA), investment grade (AA+ to BBB-), and speculative (BB+ to D). In total we have 303 observations (deal-tranches) with a floating rate, quarterly coupon frequency, and referenced to the 3-month Euribor, from 24 issuer parents (source: Dealogic).

Without any controls, weak banks actually paid 7 basis points less than healthy banks. To control for issue characteristics, we regress coupon diferentials in basis points on variables capturing the type of securitization, risk category, month of issue, years to maturity, collateral type, and guarantor type. Standard errors are clustered by issuer parent. The estimated coeficient associated with a weak bank dummy is positive but not significant (2.8 basis points, with a p-value of 0.55). Hence, we cannot reject the hypothesis that financial markets failed to recognize the buildup of diferential risk at weak banks as late as 2006. It seems safe to assume that private firms, with a lower capacity to process available information than financial markets, could not possibly have predicted them either.

4 Data

In this section we describe the variables included in our matched firm-loan-bank data set. We end this section with a description of our treatment and control variables. Further details appear in the Appendix.

4.1 Data set construction

As noted, a negative aggregate shock may reduce both credit supply and demand. To disentangle them, it is essential to observe both bank and firm characteristics and, in particular, to have exogenous measures of firms’ vulnerability to bank credit shortages. Our data set combines six separate sources and contains such information.

We gather economic and financial information for more than 300,000 private, non-financial firms from balance sheets and income statements that Spanish corporations must submit yearly to the Spanish Mercantile Registers.³ In particular, it contains information on employment, measured as a yearly average, as well as on variables like the firm’s age, size or indebtedness, which are

³The source is the Iberian Balance sheet Analysis System produced by INFORMA D&B in collaboration with Bureau Van Dijk and the Central Balance Sheet Data Ofice of the Bank of Spain.

used as controls in our analysis.

To avoid the risk of reverse causality —so that the troubles of firms drive the solvency problems of banks—, we exclude the REI, as well as industries that in the year 2000 sold at least 20% of their output to it (see the Appendix). The date is chosen to minimize potential endogeneity through credit decisions taken in the later part of the expansion. We are left with a sample of 217,025 firms, representing 27% of firms, 37% of value added, and 61% of private sector employees, in the industries included in our analysis, in 2006. We complement this information with data from the Central Business Register on firm entry and exit, so as to disentangle job destruction at surviving firms from that due to firms closing down.

We match these data sets with loan and bank information. The loan information is obtained from the Central Credit Register of the Bank of Spain (CIR), a proprietary database with information on all loans above 6,000 euros (around 8,100 dollars) granted to companies by all banks operating in Spain. Given the low threshold, this data set can be taken as a census. From the CIR we construct exhaustive information on the banking relationships of the firms in our sample and we compute the ratio of loans from weak banks to the firm’s asset value, which is our key treatment variable. We also observe the number of bank relationships, collateralized loans, and credit lines, as well as a measure of loan maturity, so that we can control for firms’ refinancing needs at the onset of the crisis. Since we are interested in bank credit, we exclude firms with no loans in 2006. We also identify each firm’s main bank, defined as the one that accounts for the largest share of a firm’s outstanding loans. Though firms’ creditworthiness is typically unobservable, in our case information on non-performing loans and potentially problematic loans is available.

We also use information on loan applications. All banks receive monthlyupdated information from the CIR on their borrowers’ credit exposure and defaults vis-à-vis all banks in Spain. But banks can also costlessly obtain this information on “any firm that seriously approaches the bank to obtain credit”.

By matching the loan application data set with the CIR we can observe, for each application, whether the loan is granted or not. If not, either the bank denied it or else the firm obtained funding elsewhere (see Jiménez et al., 2012a). Since the loan application data set only gives information on whether a firm borrowed from its bank(s) if it has a credit history, we exclude loan applications from entering firms.

Lastly, we enlarge our information with two data sets on banks. The first one records their financial statements and it is used by the Bank of Spain for regulatory and supervisory purposes. It includes 226 banks, comprising commercial banks, savings banks, and credit cooperatives. The second dataset contains historical data on the location of bank branches at the municipal level and it has never been used for research purposes before.

4.2 The treatment variable and the sample

As already explained, we aim at measuring the employment losses caused by the diferential efect of the financial crisis on the lending capacity of banks due to the heterogeneity in their financial health. We do so by comparing the evolution of employment in firms with high and low exposure to weak banks. Exposure is captured by the ratio of the firm’s pre-crisis level of debt with weak banks to its asset value. This ratio jointly reflects the overall leverage ratio and the relative importance of weak banks in the firm’s bank debt.⁴

About one-third of firms had no credit from weak banks. In our benchmark treated firms are defined as those above the third decile of the cross-sectional distribution of firms with positive exposure, which takes a value of 6.3%. This figure corresponds to a share of weak banks in total bank credit of 51.4%, so that above half of bank credit comes from weak banks. On average, their ratio of credit with weak banks to assets is 25% and their share of bank credit with weak banks equals 71%. We will also show that our results hold qualitatively

⁴We focus on drawn credit, but we also check the robustness of our results to the inclusion of undrawn credit in several of our empirical specifications.

for all other deciles of the distribution.

Table 2 provides descriptive statistics for our treatment and control groups, revealing diferent characteristics across groups. Compared with the control group, firms in the treatment group are on average younger and smaller in terms of both employment and assets, and they have a worse financial profile: they are less capitalized and profitable, they have less liquidity, and they are more indebted with banks. They work with three banks on average and between 2002 and 2005 they defaulted more often on their bank loans. Treated firms also worked with banks that were smaller, less capitalized, less profitable, with less liquidity, with more mortgages as a share of loans, and with a larger ratio of non-performing loans. These diferences are not always large, although they are statistically significant. We must therefore exhaustively control for firm-level characteristics in any empirical exercise, since weak banks were more likely to grant loans to less profitable and potentially more vulnerable firms than healthy banks.

5 Empirical strategy and results

In this Section we discuss our empirical strategy and show the estimation results, both for standard diference in diferences and for two instrumental variables models.

5.1 Diference in diferences

We start by estimating the following standard diference-in-diferences (DD) equation:

\[\log (1 + n _ {i t}) = \alpha + \delta W B _ {i} + \gamma P o s t W B _ {i} + \beta P o s t + \eta d _ {s} + \theta P o s t d _ {s} + X _ {i} ^ {\prime} \phi + u _ {i t}\tag{1}\]

where is employment at firm i in year t (t=2006 and 2010), is a dummy variable for treated firms, P ost is a dummy variable for 2010, is a joint vector of 50 province and nine industry dummy variables, is a set of control variables, and denotes random shocks.

Our sample is an unbalanced panel: though most firms are present in both periods, some firms are only observed in 2006 and others only in We keep all observations so as to increase eficiency. For firms that are observed in 2006 but not in 2010 because they closed down we set to zero in 2010 —and therefore use as the dependent variable—, so that we can jointly measure employment changes both at surviving firms and due to firm closures. Below we will also study them separately.

Our main hypothesis is that firms working more intensely with weak banks in the expansion sufer more stringent credit constraints during the crisis, which translates into larger job losses. We do not intend to estimate all potential efects of credit constraints on employment, but only a partial efect that can be identified as being causal, namely the diferential impact of those credit constraints stemming from being attached to a weak bank, as opposed to other banks, measured by in equation (1).

We aim at isolating the impact of credit constraints on observationally identical firms choosing ex ante to borrow from an ex-post insolvent bank vis-à-vis a solvent one, so that selection efects which may bias our estimates are absent. The group controls and other characteristics are intended to achieve such ex-ante homogeneity across firms, allowing allows us to estimate the average treatment efect on the treated (ATT) by comparing firms in the treatment group to similar firms in the control group. The list of variables in is discussed next.

5.2 Threats to identification

The two main challenges for identification are the non-random assignment of firms to banks prior to the crisis and the possibility of firms avoiding treatment through a successful application for loans at healthy banks. The relevance of the first threat is highlighted by the diferent characteristics of the firms working with weak and healthy banks before the crisis. As shown in Section 4, the firms in the treatment group have worse financial statistics. It therefore seems that laxer loan-approval criteria at weak banks may have caused a systematic bias in the risk profiles of the companies in the treatment group or, alternatively, that they may have been a motive for self-selection of firms into weak banks.

⁵The total number of firms is 217,025. The breakdown is as follows: both in 2006 and 2010, 153,369; in 2006 but closed down by 2010, 17,088; in 2006 but not observed in 2010, 45,570, and observed only in 2010, 998.

The exceptionally rich contents of our data set helps us avoid many threats to identification. To start with, our data go back four years before the outbreak of the recession, so that we can test for diferences in pre-existing trends in employment at attached and non-attached firms after conditioning on controls. Secondly, potential biases arising from a diferent geographical or sectoral concentration of the activities of either borrowers or banks are dealt with by including province and industry dummies in all specifications. Similarly, diferences across industries or provinces in the impact of the recession are absorbed by interactions between and the crisis dummy.

We also introduce a set of covariates controling for firm characteristics exante (2006, unless otherwise indicated) that could lead to diferential employment outcomes, like the firm’s age and its square (to capture nonlinear efects), its size (in terms of assets), and its rate of return on assets. A second set of variables is linked to financial health, such as a firm’s indebtedness with banks and its shares of short-term (up to one year) and long-term bank debt (above 5 years), intermediate terms being captured by the reference firm. A third set of variables captures the firm’s financial vulnerability, several of them serving as direct proxies for expected credit constraints: liquidity and own-funds ratios, the number of past loan applications to non-current banks (where “past” refers to 2002-2005) and an indicator for whether all were accepted, indicators for having any past loan defaults, any current loan defaults, and any credit lines, the number of banking relationships and its square, and the share of loan amounts that are uncollateralized. Lastly, a full set of dummies (226) captures synthetically the characteristics of the main bank that a firm works with.

It is also vital to control for diferences in the share of temporary contracts, which represent almost one-fourth of jobs in our sample. These contracts can be terminated at much lower costs than permanent ones and therefore, other things equal, we expect larger employment adjustments at firms with a larger temporary rate. Moreover, firms expecting to face financing constraints in future have an incentive to maintain a bufer stock of temporary contracts (Caggese and Cuñat, 2008).

This rich set of controls allows much better identification than is typical in the literature. The breakdown into 50 provinces afords a more accurate control of firms’ location than in research work that uses regions or states (in the US) instead. Moreover, most of the firm characteristics we introduce are simply unavailable in standard data sets. In particular, what makes our exercise exceptional is the use of firms’ banking relationships, in terms of the number and identity of the banks they work with, and the proxies for the banks’ assessment of a firm’s creditworthiness via its credit history: its decisions to apply for loans and its success in such applications, as well as its ability to meet repayment obligations. Lacking this information, researchers have resorted to proxying firms’ access to credit either by responses to questions about past loan denials (e.g. Caggese and Cuñat, 2008) or from actual credit balances.

Moreover, whereas typical sample sizes in the literature are around 2,000- 3,000, our data on more than 217,000 firms allows us to both attain very high precision and to apply matching methods using many controls, so that very similar firms, attached and non-attached to weak banks, are being compared. Self-selection through unobservables is however still possible, and we therefore need to rely on the assumption of randomness of the assignment of firms to the control and treatment groups conditional on observables.

Our approach would still be incorrect if treated firms could easily find alternative funding from healthy banks or other sources. As highlighted by the relationship banking framework, banks usually obtain information on firms’ profitability and solvency through long-standing relationships. This makes switching banks very costly for firms, since it takes time for other banks to acquire such knowledge. Thus, when the Great Recession arrived, obtaining loans from new banks became harder and many firms were largely limited to the funding provided by banks with which they had long-established relationships. As previously shown, acceptance rates of loan applications at all banks from non-current customers sharply fell starting in early 2007. Below we will also check whether it was even lower for firms with a high exposure to weak banks. Moreover, as shown below in the context of an instrumental variables model, committed credit fell significantly more for firms attached to weak banks and neither did bank nor non-bank credit sources allow firms to replace bank lending.

Lastly, it may be objected that the treatment is defined in terms of an outcome, namely bank bailout, that is realized several years after the outbreak of the crisis. The use of an ex-post criterion does not invalidate our results as long as the outcome was unforeseen. And we have shown in Section 3 that expected diferences in bank default risk were insignificant, since financial markets did not recognize them in the runup to the crisis. Nevertheless, in one of our robustness exercises we also experiment with an alternative definition of weak bank that relies on the pre-crisis exposure of banks to firms in the REI.

5.3 DD estimates

Table 3 presents the estimation results for our diference-in-diferences equation (1). We report robust standard errors corrected for clustering at firm and main bank level, unless otherwise stated. The raw mean diference between the proportional loss of employment at firms in the treatment and control group is equal to 8.5 percentage points. This figure remains unaltered after the inclusion of province and industry dummies, while it falls to 7.4 pp when firm characteristics are controled for (first and second columns). Adding main bank dummies and controling for diferential trends by province and industry reduces the treatment efect to 6.2 pp (third column).⁶ We take this as the baseline specification, in particular with respect to the set of control variables. We can perform a very rough estimate of the aggregate efect in our sample. Restricting ourselves to those firms in our sample whose employment is known for both 2006 and 2010, since firms attached to weak banks comprise 21.7% of employment in 2006, the estimated efect would explain an aggregate 1.3% employment fall. Given that the overall employment reduction at those firms from 2006 to 2010 was equal to 8%,⁷ weak-bank attachment would then account for 16.7% of aggregate job losses. Obviously this estimate cannot be taken as an approximation to the macroeconomic efect, since we are completely abstracting from any generalequilibrium efects.

We test for diferences in the pre-crisis trends for the treatment and control groups by running a placebo equation where we have chosen 2002 and 2006 as initial and final dates. As required, this specification test delivers a coeficient that is not significantly diferent from zero (last column in Table 3).

5.4 Instrumental variables estimates

We wish to ascertain that the impact of weak-bank attachment on employment is driven by credit constraints as opposed to other potential avenues. To this end we estimate the following instrumental variable (IV) model for the proportional change in employment:

\[\begin{array}{r c l} \Delta \log (1 + n _ {i t}) & = & \alpha^ {\prime} + \delta^ {\prime} \Delta \log (1 + C r e d i t _ {i t}) + \beta^ {\prime} P o s t _ {t} + \eta^ {\prime} d _ {s} + \sigma^ {\prime} d _ {i} + u _ {i t} ^ {\prime} \\ \Delta \log (1 + C r e d i t _ {i t}) & = & \pi + \mu P o s t _ {t} W B _ {i} + \omega P o s t _ {t} + \rho d _ {s} + \psi d _ {i} + v _ {i t} \end{array} (2)\]

where all variables are defined as in equation (1), except that is total credit committed by banks to firm i in year t —both drawn and undrawn, so as to minimize potential endogeneity—, is a vector of year dummies for , and is a firm fixed efect. Coeficient captures the diferential impact of weak banks on credit committed during the crisis, whereas captures the passthrough from credit to employment. The exclusion restriction is that working with a weak bank alters employment growth only through credit changes, as opposed to other channels.

⁶The estimated coeficient did not change when main bank dummies were replaced by either main bank characteristics or main, secondary, and tertiary bank dummies.

⁷This decline is very close to the fall in the number of private sector employees in Spain over the period, which was equal to 8.3%.

This model difers from the DD equation in several ways. First, it is estimated in first diferences because —in keeping with the literature— we are better able to explain credit changes than levels. Second, it is a panel of four rather than two periods, so that we exploit information for each recession year and we capture all firm-specific characteristics via fixed efects rather than through initial-year control variables.

As shown in the lower panel of Table 4, the instrumental variable is signif icantly and negatively correlated with credit, increasingly so as the recession lengthens (first column). Credit is also found to be a significant determinant of employment changes, so that a one pp increase in credit raises employment by 0.42 points. The product of this second-stage coeficient and the weak-bank efect on credit for 2010 (-0.154) yields an employment reduction of 6.5 pp in 2010 with respect to 2007 (the omitted year).⁸ This is very close to our baseline DD estimate of 6.2 pp, in spite of the diferent nature of the two models. This result supports the idea that credit is the key channel through which the weak-bank attachment operates.

In the second column we replace credit growth with an alternative measure of credit constraints, namely an indicator for having a loan application rejected. The efect of weak-bank attachment again increases over time in the first stage, and now the causal efect of a loan rejection is a very large reduction in employment, of about 90% . Note that here we are measuring a local average treatment efect for firms on the margin of having a loan approved (Imbens and Angrist, 1994).⁹ This finding is not so surprising once we realize that these losses stem from firms closing down, which represent 77% of aggregate job losses in our sample. In any event, this underlines the need to examine the efect of weak-bank attachment on the probability of exit, as is done below.

⁸The efect vis-à-vis 2006 is unidentified, but an IV estimation for 2007 alone gave a nonsignificant coeficient with respect to 2006, so that the former efect is likely to be around 6.5 pp as well.

5.4.1 Exogenous variation in exposure to weak banks

Firms choosing a weak bank may have been driven by motives, such as laxer credit standards, that subsequently contributed to the demise of the savings banks. In other words, to convincingly rule out selection efects we need an exogenous source of variation in firms’ attachment to weak banks. We exploit two variants.

First, we use a regulation-based instrumental variable. Up until 1988 savings banks could open at most 12 branches outside their region of origin, but in December 1988 a new law removed all location restrictions (Real Decreto-ley 1582/1988). In order to better exploit this variation, we compute for each municipality the share of bank branches that belonged to our set of weak banks in December 1988 (6,101 municipalities with bank branches). Our instrumental variable is this weak-bank density in the municipality where the firm is located. This variable should capture exogenous variation in the probability of weakbank attachment, since it is more likely that a firm will work with a bank if it is located in a municipality where the bank traditionally operates. In Table 5 we see that high weak-bank density in 1988 significantly predicts weak-bank attachment 18 years later (first column). The associated employment efect amounts to 8.4 pp, which is higher than the DD baseline value of 6.2 pp, though not significantly so. According to this estimate, 22.7% of aggregate job losses would be attributed to weak-bank attachment.

Alternatively, we use traditional bank ties to real estate firms to make sure that credit restrictions faced by firms indebted with weak banks do not simply result from poor bank management. Our instrumental variable is now bank exposure to the REI in 2000, well before the beginning of the bubble, which is commonly thought to have started around mid-2003 (see Ayuso and Restoy, 2006). The second column reveals that the instrument is very powerful. The employment efect of predicted weak-bank attachment of 13.5 pp is now significantly higher than our DD baseline. This finding suggests that to some extent weak banks got into trouble because of their historical ties to real estate firms and not only because they aggressively pursued real estate lending just before the crisis. Repeating the simple extrapolation exercise performed above, this estimate would imply that 36.4% of job losses would be attributed to weak-bank attachment.

⁹Under the monotonicity assumption that access to credit always improves with lender health.

6 Robustness checks

In this section we check the robustness of our baseline estimates in many ways. The checks are presented in terms of the dimension of variation: timing, treatment variable and reference sample, firms’ financial vulnerability, level of exposure to weak banks, probability of exit, measure of credit, and estimation method.

6.1 Timing, treatment variable, and reference sample

We first explore the timing of the impact of the credit constraint on firms by choosing alternative ex-post periods. Our estimates are as follows (in pp, s.e. between parentheses): 0.5 (0.3) for 2007, -0.6 (0.5) for 2008, and -3.1 (0.7) for 2009. Thus the weak-bank efect does not become significant until 2009. Secondly, to avoid potential anticipation efects, we progressively restrict the analysis to firms with long-run banking relationships, established years before the outbreak of the crisis. In Table 6 we report the efect of shifting back the year at which the firm control variables are measured (first two columns). This restriction moderates the efect to 5.9 pp when 2002 is used and to 6.1 pp for 2005.

We also shift back the date of treatment. The assignment of firms to the treatment is now based on their weak-bank exposure in either 2000 or 2002. This approach involves a tradeof: it potentially weakens endogeneity concerns but it also brings us farther away from the conditions faced by firms just before the crisis —which are likely to be more relevant. As Table 6 shows, the corresponding estimated efects of attachment to weak banks for benchmark years 2000 and 2002 are respectively 3.5 pp and 4.9 pp, which are still sizeable and with the former being significantly diferent from the baseline (third and fourth columns).

To check whether using a treatment defined by an ex-post event, i.e. the bailout, may be biasing our results, we re-classify banks on the basis of their exposure to the REI in 2006. This is measured as the share of a bank’s loans that are granted to REI-firms, and all banks with an exposure above the third decile of the distribution are classified as weak banks. This specification leads to an estimated employment efect of 6.2 pp (fifth column), which is identical to the baseline, confirming that REI exposure drives weak-bank troubles.¹⁰

Lastly, we estimate the efect only for surviving firms. The estimated job loss for treated firms is equal to 1.3 pp, which is significantly lower than for the full sample. As indicated, firms that close down comprise 77% of overall job losses and, compared to continuing firms, they are smaller, younger, less capitalized, and less profitable. This finding indicates that credit constraints have been more important in driving firm closures than in leading surviving firms to cut jobs. We return to this issue below, when we examine the probability of firm exit.

6.2 Firms’ financial vulnerability

In this section we allow for heterogeneity of the treatment efect across firms with diferent characteristics. To this efect we interact the product of the P ost dummy and the weak bank dummy with the firm characteristic of interest. We begin with nine industry dummies. For five out of the nine we find significant efects, as follows: Manufacturing (8.4 pp), Machinery, Renting, Computing, and R&D (6.8), Trade (6.1), Transport, Storage, and Communications (4.6), and Hotels and Catering (4.3). These results are quite reasonable, since the efects tend to be larger for industries requiring more capital and therefore typically more credit.

¹⁰ Including committed but undrawn loans in the treatment yields a job loss of 5.7 pp (s.e. 0.8 pp).

A second set of triple diference specifications examines whether the employment cost of weak-bank attachment depends on a firm’s financial vulnerability.¹¹ The first measure is an indicator for whether the firm had a loan application rejected over 2002-2006. Table 7 reveals that these firms sufer an additional loss of jobs of 6.4 pp in the recession but no extra loss if they were attached to weak banks. Similarly, firms that defaulted on a loan over the same period experience an additional loss in employment of 22.9 pp, though working with a weak bank again does not add to it. Note that, at face value, this implies an impact that is almost five times higher than for other treated firms.

Next, firms with a share of short-term bank debt in total debt above onehalf in 2006 —implying that they subsequently had to renew a sizeable fraction of it— sufer an additional job loss of 9.4 pp, and another 7.1 pp if attached to weak banks. Further, small firms (defined as those with assets below 10 million euros) sufer an extra 12 pp job losses, but only if they were attached to weak banks (the interaction with P ost is only significant at the 10% level). These findings are in accordance with standard theoretical predictions that smaller, less transparent, and financially weaker firms should be more vulnerable to changes in credit market conditions. In contrast to the literature, however, our results are based on direct measures of credit constraints and credit records. A noteworthy result is that we find no significant diferences in the penalties that weak banks impose on firms with a bad credit record.

We next examine whether the impact of credit constraints varies with the number of banking relations. We distinguish between firms working with only one bank from the rest. The empirical literature has not reached a robust conclusion on this issue. According to Hoshi et al. (1990), in Japan firms whose debt is concentrated with a single bank, within a group of firms or keyretsu, have better access to credit in periods of distress. On the contrary, in the case of the US, Houston and James (2001) find that cash-flow sensitivity is larger at firms with an exclusive banking relationship. For Spain we find that job losses at single-bank firms are 3.8 pp lower than at multi-bank firms, as shown in Table 7, and still 2.9 pp lower if the single bank is weak. In the next subsection we further explore the latter finding.

¹¹To avoid having to weigh coeficients by the variables’ average values, regressors are in deviations from their means.

6.3 Degree of exposure to weak banks

We have so far presented results for the degree of exposure to weak banks using the third decile of the cross-sectional distribution in 2006 as the threshold. To check the sensitivity of our results to this choice, we reestimate equation (1) for firms with any positive loan balances with weak banks and for exposure levels above each decile of the distribution of firms in that set. Figure 2 shows that the weak-bank efect is present at all deciles and that there is relatively little variation in the estimates for any exposure up to the sixth decile, ranging from 5.2 to 6.3 pp (continuous line). The magnitude however falls for higher deciles.

This decline may reflect a composition efect since the share of firms with a single banking relationship grows from 29% above the first decile of exposure to weak banks to 50% above the ninth decile and we know that these firms sufered smaller employment losses. To check this hypothesis, we reestimate equation (1) separately for single- and multiple-bank firms. The estimates are now stable and significantly diferent: job losses from weak-bank attachment are on average 8.8 pp higher for multi-bank firms but 2.8 pp lower for single-bank firms.

It may be that single-bank firms are better borrowers. In our sample, they have better ratios of capitalization, liquidity, return on assets, and bank debt, and they are less likely to have defaulted on their debt obligations. Thus they may also be better along other dimensions we have not controled for. Alternatively, there may be an advantage for a firm in working with a single bank, which acquires more information about it and has a stronger stake in its economic success. Some evidence on this hypothesis is given by Frazzoni et al. (2012), who study a set of Italian firms over 2004-2009 and find that the strength of a firm’s relationship with its main bank —measured by the ratio of loans from that bank to the firm’s asset value— has a positive impact on its propensity to innovate and export. This result suggests that relationship banking helps with funding of innovation and in accessing foreign markets.

To make progress on this issue, we test whether banks favored firms that concentrated their loans with them. Using our firm-bank-loan database, we regress the yearly change in credit committed in the recession (2007-2010) on the share of loans of the firm with each bank in 2006, including firm and bankyear fixed efects to control for unobservable demand and supply factors.¹² This data set has 3.75 million observations on 509,800 firms. We find that in general banks did not extend more credit to firms whose credit was more concentrated with them, except for weak banks: each pp of loans with the bank raises credit growth by 0.34 pp (s.e. 0.09). This result hints at an evergreening of loans by these banks.

Why would only weak banks behave in this way? During this period, weak banks were more closely monitored by the markets than healthy banks, due to their large exposure to real estate, so that they might have been more eager to avoid increases in their non-performing loan rate. Alternatively, while obtaining credit became harder for all firms, it may have been more dificult for firms heavily depending on weak banks. We test this stigma hypothesis using our data on loan applications from non-current customers. We previously found no significant efects from weak-bank attachment for 2007, so we employ monthly loan (firm-bank) data from 2008:01 to 2010:12.¹³ We estimate a linear probability model for the event that a loan is requested and granted on the share of bank credit that a firm had with weak banks over 2002-2006, including the same control variables as in equation (1) and bank-time dummies to control for time-varying unobservable supply factors. The result is a decrease in that probability of 3% for each pp increase in that share, as long as the firm has a share with weak banks above 80%.

¹²Standard errors are clustered at firm and bank level.

¹³The dataset has 240,179 observations on 109,172 firms. The available variables in 2002, which are used in this estimation are: Industry and Province Dummies, Size, Age, Age Squared, Own Funds, Liquidity, Return on Assets, and Past Defaults. Standard errors are clustered at firm and bank level.

6.4 Probability of exit

Since the bulk of job losses stem from firm closures, it is natural to estimate the efect of weak bank attachment on the probability of firm exit. We start by estimating a linear probability model for exit in 2010 with respect to 2006, using the same specification as in our baseline DD equation (1). The sample consists of the 170,457 firms with either a positive employment level or a zero level of employment in 2010 because the firm is known not to have survived the crisis.

As shown in the first column of Table 8, the treatment efect is significant. Weak-bank attachment leads to a marginal exit probability which is 8.4% (0.8 pp) higher than the baseline exit rate of 10%. We also try a second specification in which we use the actual ratio of weak-bank credit to assets rather than the treatment dummy. The efect, presented in the second column, is again significant, with a coeficient of 5.8 pp. This implies that ceteris paribus, compared to a firm with a ratio of weak bank debt to assets at the first decile —which is roughly nil—, a firm at the ninth decile, i.e. with an exposure of one-quarter, has a 14.4% higher probability of closing down. The last column of the Table confirms the preceding results, in that single-bank firms benefit from this condition by having a lower probability of exit. At the average exposure ratio (0.17), they have an 8% lower probability of exit than multi-bank firms.

Firm destruction carries job losses with larger economic costs than downsizing at surviving firms, and it probably makes recessions more protracted. It is therefore worth asking why do credit constraints cause employment losses mostly through closures. This is likely linked to Spain ranking relatively high —in the ninth position out of 28 OECD countries in 2006— in the degree of stringency of employment protection legislation for permanent contracts (OECD Indicators of Employment Protection, 2013 release). Therefore, once temporary jobs have been destroyed at a relatively low cost, it is quite costly and dificult —due to labor court procedures— to dismiss regular employees, so that eventually many firms have to close down.

6.5 Non-bank credit

So far we have focused on bank credit, which is the major source of funding for firms in Spain. However, trade credit is an alternative source, and a firm’s suppliers may have advantages over banks as credit providers, in terms of acquiring information, monitoring, and eficiency in liquidation (Petersen and Rajan, 1997).

We cannot fully address the question of whether trade credit may have compensated for restrictions in bank credit. The reason is that we only have data on firms’ liability structure for a subsample, namely for 15,323 firms (7% of the total). These are the firms that provide more detailed public accounts, which not surprisingly tend to be the largest ones. For example, in 2006 their median assets were equal to 9.1 millions, vis-à-vis 0.58 millions in our full sample. For them, at the median, financial institutions and trade credit each represent 34% of their liabilities.

We then reestimate our instrumental variables model (2) for the credit channel for these firms. The weak bank dummy is significant in the first stage and the overall efect of weak-bank attachment on employment is 4.0 pp, lower than the full sample estimate of 6.2 pp, see Table 9 (first column). This is consistent with the larger efect found for small firms in our triple diference estimation. Estimating the IV model with total credit rather than bank credit (second column), we find again that the weak bank dummy is significant and, contrary to the case of bank credit, it reveals a credit contraction starting in 2008. The overall efect is slightly higher than for bank credit, 4.4 pp, but not significantly so. Thus we conclude that trade credit did not alleviate the credit constraint. We cannot directly check whether the same is true also for smaller firms, which are usually more dependent on trade credit. However, our finding is consistent with the results by Molina Pérez (2012), who finds no increase in trade credit taken by firms over the period 2008-2010 with a sample of 9,602 Spanish firms, 85% of which are small and medium-sized firms (below 250 employees).

6.6 Exact matching

To achieve ex-ante comparability across firms, so far we have controled for a long list of firm characteristics. More accurate control for selection may however be attained through the use of matching techniques. Here we apply the coarsened exact matching method (Iacus et al., 2011). Sample sizes typically found in the literature severely limit the number of cells that can be constructed, whereas in our case we can use cells defined by 14 control variables, which we choose according to their significance in the baseline DD regression.

The coarsening entails each variable becoming a 0-1 dummy. For variables that were not originally of this type, we use the sample median value as the cutof, except for the number of banks, where we distinguish between firms with one and with multiple banking relationships. Regarding industry, we separate the Primary sector and Mining from the others, and for provinces we diferentiate those in the East coast of the Spanish Peninsula plus the Balearic and Canary Islands from the rest (see the Appendix for a full list). Out of 16,384 potential strata, we end up with 4,822 strata with observations, 3,553 of which can be matched across treated and control firms. Reassuringly, we have ascertained that the matching method suppresses any potential preexisting trend diferences between treated and control firms.

Using weighted least squares, the estimated employment efect attached to the weak bank dummy variable is equal to -3.0 pp (s.e. 1.4 pp), which is about half the size of our baseline DD estimate.¹⁴ In this case, the aggregate impact of weak-bank attachment would also be halved, to 8.1%. The sizeable diference between these estimates suggests that a rigorous control for selection can significantly alter estimated efects.

We also check the stability of the estimate with respect to the degree of exposure to weak banks. Matching estimates are less stable than DD estimates, ranging from 5.2 pp above the third decile to nil above the ninth decile (not shown). However, estimating separately for single- and multi-bank firms, we find that the former do not sufer additional job losses from weak bank attachment, whereas multiple-bank firms sufer losses around 4.8 pp on average.¹⁵ As already indicated, these firms obtained relatively more credit, possibly as a result of the evergreening of their loans.

7 Conclusions

In this paper we aim at measuring the impact of credit constraints on employment during the Great Recession in Spain for firms outside the real estate sector. We achieve identification by exploiting diferences in lender health at start of the crisis, as evidenced by public bailouts of savings banks. We proceed by comparing employment changes from the expansion to the recession between firms that are heavily exposed to weak banks and less exposed firms. Our exceptionally large matched bank-loan-firm data set allows us to control exhaustively for ex-ante characteristics of firms and for potential endogeneity, as well as to perform a wide range of robustness checks. The estimated efects are sizeable. Controling for selection, attachment to weak banks caused a larger fall in employment from 2006 to 2010 ranging from 3.0 to 13.5 percentage points, i.e. 8% to 36% of aggregate job losses in our sample, though of course these estimates cannot be taken as an approximation to the macroeconomic efect, since we are ignoring any general-equilibrium efects. We also find significant heterogeneity according to the ex-ante financial vulnerability of firms.

¹⁴There are 377,498 observations on 211,284 firms. This sample is not exactly the same as for DD; for the same sample the DD estimated efect is -0.063 (s.e. 0.009).

¹⁵Estimate precision falls as exposure to weak banks grows, because finding matches for treated firms becomes increasingly harder within smaller groups, but coeficients are generally significant.

Our results are within the ranges found in the preceding US literature. Greenstone and Mas (2012) infer that the decline in lending from 2007 to 2009 accounted for up to 20% of the employment decline in US firms with less than 20 employees and for 16% of the total employment loss. On the other hand, Chodorow-Reich (2013) finds that the withdrawal of credit explains between one-third and one-half of job losses at small and medium-sized firms in the year following the Lehman Brothers bankruptcy. However, these estimates are not directly comparable with ours for a number of reasons. In particular, we have achieved identification by focusing on credit constraints arising from only one channel, exposure to weak banks, while controling for other firm characteristics that are also traditionally thought to capture credit restrictions. Our estimates reveal that the joint consideration of both sets of characteristics lead to a wider range of estimated employment efects. For example, the impact for firms which ever defaulted on a loan is almost five times larger. This finding suggests that, lacking information on firms’ financial histories, the existing literature may be overestimating the impact of credit on employment, due to the lack of suficient controls to attain homogeneity between treated and control firms.

We also contribute to the literature on the interaction between credit constraints and the number of banks that firms work with. Our results clearly show that in the Spanish case firms that relied on a single bank were not adversely afected by that bank being weak. Lastly, we have also found that credit constraints caused employment losses mainly by driving firms to close down rather than to just downsize. This channel had not been identified in the existing literature, as far as we know. And it has potentially important welfare implications, since job losses via firm destruction carry a larger economic cost than downsizing at surviving firms, and they probably make the recession more protracted.

We can also make a statement regarding eficiency. Assuming that our quasi-experimental approach is valid, the assignment of firms to weak banks, as opposed to healthy banks, is as good as random. In other words, given our controls, these firms could have been granted as much credit from healthy rather than weak banks. In this sense, while the total job losses sufered by firms attached to weak banks may or may not have been eficient, the estimated employment efects of the credit constraints we identify, once selection has been taken into account, were ineficient.

In order to achieve better identification of a causal efect, we have focused on a single channel through which credit constraints operate, namely estimating the diferential efect on employment at firms attached to weak banks vis-à-vis the non-attached firms. For this reason, it would be incorrect to extrapolate our estimates to the aggregate economy. In general equilibrium there would be further efects (see Chodorow-Reich, 2013). A drop in aggregate demand generally reduces labor demand by both constrained and unconstrained firms, but product demand may be shifted from the former to the latter, thus inducing an increase in their labor demand. The microeconomic efects need therefore not coincide with the aggregate efect.

A Appendix. Definitions of variables

Employment. Computed as the average level over the year, weighing tem porary employees by their weeks of work. The Temporary Employment ratio divides the temporary by the total number of employees. Set to 1 above the median for matching.

Treatment variable. The Weak Bank treatment (0-1) is equal to 1 if the ratio between the total value of a firm’s loans from weak banks and its book value in 2006 is above the third decile of the cross-sectional distribution of firms with positive exposure to weak banks.

Province. There are 50 provinces. For matching the dummy is set to 1 for the East coast, namely Girona, Barcelona, Tarragona, Castellón, Valencia, Alicante, Murcia, Almería, Granada, Málaga, Cádiz, and Huelva, plus the islands: Baleares, Las Palmas, and Santa Cruz de Tenerife.

Industry. Firms belonging to the following industries are excluded (share of output sold to Construction and Real Estate in 2000 shown between parentheses): Extraction of Non-metallic Minerals (35.2%), Wood and Cork (21.1%), Cement, Lime, and Plaster (46.4%), Clay (60.1%), Non-metallic Mineral Products n.e.c. (85.4%), Fabricated Metal Products except Machinery and Equipment (23.3%), Machinery and Electric Materials (19.2%), and Rental of Machinery and Household Goods (26.2%).

There are nine Industry dummies, for: Agriculture, Farming, and Fishing; Mining; Manufacturing; Electricity, Gas, and Water; Trade; Hotels and Catering; Transport, Storage and Communications; Rental of Machinery, Computing and R&D; and Other Service Activities. For matching, the dummy takes on the value 1 for the first two.

Balance sheet and income statement control variables (flows are in nominal values and stocks in book values in December). Size (Total Assets), Own Funds (Own Funds/Total Assets), Liquidity (Liquid Assets/Total Assets), Return on Assets (Earnings before interest, taxes, depreciation and amortization/Assets), Bank Debt (Bank Debt/Total Debt), Short-Term Bank Debt (Debt up to one year/Total Bank Debt), Long-Term Bank Debt (Debt of five years or more/Total Bank Debt), and Uncollateralized Loans (Uncollateralized Loans/Total Bank Debt). Age is defined as current year minus year of creation. For matching they are set to 1 when above the median. For triple diferences, a Small Firm is defined as one with Total Assets below 10 million euros.

Credit-related control variables. Credit Line (at least one), Current Defaults (any nonperforming loan in 2006), Past Defaults (any nonperforming loan over 2002-2005), Loan Applications, All Applications Accepted (over 2002- 2005). For triple diferences the following composite variable is used: Defaults = Current Defaults + Past Defaults.

Banking relationship control variables. Banking Relationships (number of banks with outstanding loans) (for matching set to 1 for multiple-bank firms), Duration of Banking Relationship (with Main Bank, in years), and Main Bank (bank with the largest amount lent).

Further tables and figures can be found in an Online Apppendix at: https://dl.dropboxusercontent.com/u/15338248/creditjobsonlineappendix.pdf.

References

1. [1] Almeida, H., M. Campello, B. Laranjeira, and S. Weisbenner (2012), “Corporate Debt Maturity and the Real Efects of the 2007 Credit Crisis”, Critical Finance Review 1, 3-58.
2. [2] Ayuso, J. and F. Restoy (2006), “House prices and rents: An equilibrium asset pricing approach”, Journal of Empirical Finance 13, 371—388.
3. [3] Benmelech, E., N. K. Bergman, and A. Seru et al. (2011), “Financing Labor”, NBER Working Paper no. 17144.
4. [4] Bernanke, B. S. and M. Gertler, “Agency Costs, Net Worth, and Business Fluctuations”, American Economic Review 79, 14-31.
5. [5] Bernanke, B. S. and M. Gertler (1995), “Inside the Black Box: The Credit Channel of Monetary Transmission”, Journal of Economic Perspectives 9, 27-48.
6. [6] Bernanke, B. S., M. Gertler and S. Gilchrist (1996), “The Financial Accelerator and the Flight to Quality”, Review of Economics and Statistics 78, 1-15.
7. [7] Boeri, T., P. Garibaldi, and E. Moen (2013), “Labor and Finance: Mortensen and Pissarides meet Holmstrom and Tirole”, mimeo, Bocconi University.
8. [8] Caggese, A. and V. Cuñat (2008), “Financing Constraints and Fixed-Term Employment Contracts”, Economic Journal 118, 2013-2046.
9. [9] Chava, S. and A. Purnanandam (2011), “The Efect of Banking Crisis on Bank-dependent Borrowers”, Journal of Financial Economics 99, 116-135.
10. [10] Chodorow-Reich, G. (2013), “The Employment Efects of Credit Market Disruptions: Firm-level Evidence from the 2008-09 Financial Crisis”, Quarterly Journal of Economics (forthcoming).
11. [11] Cuñat, V. and L. Garicano, “Did Good Cajas Extend Bad Loans? Governance, Human Capital and Loan Portfolios”, mimeo, London School of Economics.
12. [12] European Central Bank (2010), EU Banking Structures, Frankfurt.
13. [13] Frazzoni, S., M. L. Mancusi, Z. Rotondi, M. Sobrero, and A. Vezzulli (2012), “Relationships with Banks and Access to Credit for Innovation and Internationalization of SME”, mimeo, Unicredit.
14. [14] Freixas, X. (2005), “Deconstructing Relationship Banking”, Investigaciones Económicas 29, 3-31.
15. [15] Garicano, L. and C. Steinwender (2013), “Survive Another Day: Does Uncertain Financing Afect the Composition of Investment?”, Centre for Economic Performance, Discussion Paper no. 1188.
16. [16] Gertler, M. and N. Kiyotaki (2010), “Financial Intermediation and Credit Policy in Business Cycle Analysis”, in B. M. Friedman and M. Woodford (eds.), Handbook of Monetary Economics, vol. 3, Amsterdam, North Holland, 547—599.
17. [17] Greenstone, M. and A. Mas (2012), “Do Credit Market Shocks Afect the Real Economy? Quasi-Experimental Evidence from the Great Recession and ‘Normal’ Economic Times”, MIT Department of Economics Working Paper no. 12-27.
18. [18] Greenwald, B. and J. Stiglitz (1993), “Financial Market Imperfections and Business Cycles”, Quarterly Journal of Economics 108, 77-114.
19. [19] Hoshi, T., A. Kashyap, and D. Scharfstein (1990), “The Role of Banks in Reducing the Costs of Financial Distress in Japan”, Journal of Financial Economics 27, 67-88.
20. [20] Houston, J. F. and C. M. James (2001), “Do Relationships Have Limits? Banking Relationships, Financial Constraints, and Investment”, The Journal of Business 74, 347-374.
21. [21] Iacus, S. M., G. King, and G. Porro (2011), “Multivariate Matching Methods that Are Monotonic Imbalance Bounding”, Journal of the American Statistical Association 106, 345-361.
22. [22] Imbens, G. W. and J. D. Angrist (1994), “Identification and Estimation of Local Average Treatment Efects”, Econometrica 62, 467-475.
23. [23] International Monetary Fund (2012), “Spain: The Reform of Spanish Savings Banks. Technical Note”, Monetary and Capital Markets Department.
24. [24] International Monetary Fund (2013), World Economic Outlook, Washington, DC, October.
25. [25] Jiménez, G., S. Ongena, J.-L. Peydró, and J. Saurina (2013), “Hazardous Times for Monetary Policy: What Do Twenty-Three Million Bank Loans Say about the Efects of Monetary Policy on Credit Risk-Taking?”, Econometrica (forthcoming).
26. [26] Jiménez, G., S. Ongena, J.-L. Peydró, and J. Saurina (2012a), “Credit Supply and Monetary Policy: Identifying the Bank Balance-Sheet Channel with Loan Applications”, American Economic Review 102, 2301-2326.
27. [27] Jiménez, G., S. Ongena, J.-L. Peydró, and J. Saurina (2012b), “Macroprudencial Policy, Countercyclical Capital Bufers and Credit Supply: Evidence form the Spanish Dynamic Provisioning Experiments”, European Banking Center Discussion Paper no. 2012-011.
28. [28] Molina Pérez, J. C. (2012), “Trade Credit and Credit Crunches: Evidence for Spanish Firms from the Global Banking Crisis”, Estabilidad Financiera 23, 57-70.
29. [29] Petersen, M. A. and R. G. Rajan (1997), “Trade Credit: Theories and Evidence”, Review of Financial Studies 10, 661-691.
30. [30] Sharpe, S. (1990), “Asymmetric Information, Bank Lending and Implicit Contracts: A Stylized Model of Customer Relationships”, Journal of Finance 45, 1069-1087.
31. [31] Sharpe, S. (1994), “Financial Market Imperfections, Firm Leverage, and the Cyclicality of Employment”, American Economic Review 84, 1060- 1074.

Table 1. Heterogeneity in bank exposure to the real estate industry and in credit change (%)

Notes. Panel A gives shares in loans to non-financial firms. There are 201 healthy and 33 weak banks. Panel B reports values for 10 weak banks, which result from consolidation of the 33 banks existing in 2006. Source: Own computations on banks balance sheet data from the Bank of Spain.

Table 2. Descriptive statistics of control and treated firms (2006)

Notes. Observations: 155,167 control firms and 60,860 treated firms. WB denotes weak banks. Variables are ratios unless otherwise indicated. The share of loans with weak banks is in bank credit.

Table 3. The employment efect of weak-bank attachment. Diference in Diferences Dependent variable: log (1+Employment_it)

Notes. OLS estimates for 2006 and 2010; in the last column, 2002 and 2006. Firm controls: Size, Age, Age Squared, Own Funds, Liquidity, Return on Assets, Temporary Employment, Bank Debt, Short-Term Bank Debt, Long-Term Bank Debt, Uncollateralized Loans, Credit Line, Banking Relationships, Banking Relationships Squared, Current Defaults, Past Defaults, Loan Applications, All Applications Accepted. Robust standard errors corrected for clustering at the firm and main bank level between parentheses. 2

Table 4. The employment efect of weak-bank attachment. Instrumental Variables Dependent variable: ∆log(1+Employment_it)

Notes. IV estimates for 2007 to 2010. All specifications include Firm and Time Fixed efects. Robust standard errors corrected for clustering at firm and main bank level between parentheses. p-value of the F test for the exclusion restriction reported.

Table 5. The employment efect of weak-bank attachment. Instrumental Variables Dependent variable: log(1+Employment_it)

Notes. Instrumental variables estimates for 2006 and 2010. All specifications include Industry Dummies, their interaction with P ost, and Main Bank Dummies. Firm controls as in Table 3. Robust standard errors corrected for clustering at firm and main bank level between parentheses. p-value of the F test for the exclusion restriction reported. p<0.01=***, p<0.05=**, p<0.10=*.

Table 6. Robustness checks. Diference in Diferences Dependent variable: log(1+Employment_it)

Notes. OLS estimates for 2006 and 2010. The first four columns report results for deeper lags of Firm controls and Main Bank, the fifth changes the definition of weak bank, and the sixth changes the sample to surviving firms. All specifications include Industry and Province Dummies, their interaction with P ost, and Main Bank Dummies. Firm controls as in Table 3. Robust standard errors corrected for clustering at firm and main bank level between parentheses.

Table 7. The employment efect of weak-bank attachment. Triple Diferences Dependent variable: log(1+Employment_it)

Notes. OLS estimates for 2006 and 2010. All variables are in deviations from their means. All specifications include Industry and Province Dummies, their interaction with , and Main Bank Dummies. Firm controls as in Table 3. Robust standard errors corrected for clustering at firm and main bank level between parentheses.

Table 8. Efect of weak-bank attachment on the probability of exit Dependent variable: Probability of exit from 2006 to 2010_i

Notes. OLS estimates using firms with observations for 2006 and 2010. All specifications include Industry, Province, and Main Bank Dummies. Firm controls as in Table 3. Robust standard errors corrected for clustering at firm and main bank level between parentheses. 2 2

Table 9. The employment efect of weak-bank attachment. Diferences in Diferences Dependent variable: ∆log(1+Employment_it)

Notes. IV estimates for 2007 to 2010. All specifications include Firm and Time Fixed efects. Robust standard errors corrected for clustering at firm and main bank level between parentheses. p-value of the F test for the exclusion restriction reported.

Figure 1: Acceptance rates of loan applications by non-current clients, by bank type. Firms applying to at least one bank per type (%)

Figure 2: The employment efect of exposure to weak banks by decile and number of banks (DD estimates with 2-s.e. bands)

1. 2013-20: “When Credit Dries Up: Job Losses in the Great Recession”, Samuel Bentolila, Marcel Jansen, Gabriel Jiménez y Sonia Ruano.
2. 2013-19: “Efectos de género en las escuelas, un enfoque basado en cohortes de edad”, Antonio Ciccone y Walter Garcia-Fontes.
3. 2013-18: “Oil Price Shocks, Income, and Democracy“, Markus Brückner , Antonio Ciccone y Andrea Tesei.
4. 2013-17: “Rainfall Risk and Religious Membership in the Late Nineteenth-Century US”, Philipp Ager y Antonio Ciccone.
5. 2013-16: “Immigration in Europe: Trends, Policies and Empirical Evidence”, Sara de la Rica, Albrecht Glitz y Francesc Ortega.
6. 2013-15: “The impact of family-friendly policies on the labor market: Evidence from Spain and Austria”, Sara de la Rica y Lucía Gorjón García.
7. 2013-14: “Gender Gaps in Performance Pay: New Evidence from Spain”, Sara de la Rica, Juan J. Dolado y Raquel Vegas.
8. 2013-13: “On Gender Gaps and Self-Fulfilling Expectation: Alternative Implications of Paid-For Training”, Juan J. Dolado, Cecilia García-Peñalosa y Sara de la Rica.
9. 2013-12: “Financial incentives, health and retirement in Spain”, Pilar García‐Gómez, Sergi Jiménez‐Martín y Judit Vall Castelló.
10. 2013-11: “Gender quotas and the quality of politicians”, Audinga Baltrunaite, Piera Bello, Alessandra Casarico y Paola Profeta.
11. 2013-10: “Brechas de Género en los Resultados de PISA :El Impacto de las Normas Sociales y la Transmisión Intergeneracional de las Actitudes de Género”, Sara de la Rica y Ainara González de San Román.
12. 2013-09: “¿Cómo escogen los padres la escuela de sus hijos? Teoría y evidencia para España”, Caterina Calsamiglia, Maia Güell.
13. 2013-08: “Evaluación de un programa de educación bilingüe en España: El impacto más allá del aprendizaje del idioma extranjero”, Brindusa Anghel, Antonio Cabrales y Jesús M. Carro.
14. 2013-07: “Publicación de los resultados de las pruebas estandarizadas externas: ¿Tiene ello un efecto sobre los resultados escolares?”, Brindusa Anghel, Antonio Cabrales, Jorge Sainz e Ismael Sanz.
15. 2013-06: “DYPES: A Microsimulation model for the Spanish retirement pension system”, F. J. Fernández-Díaz, C. Patxot y G. Souto.
16. 2013-05: “Vertical differentiation, schedule delay and entry deterrence: Low cost vs. full service airlines”, Jorge Validoa, M. Pilar Socorroa y Francesca Medda.
17. 2013-04: “Dropout Trends and Educational Reforms: The Role of the LOGSE in Spain”, Florentino Felgueroso, María Gutiérrez‐Domènech y Sergi Jiménez‐Martín.
18. 2013-03: “Understanding Different Migrant Selection Patterns in Rural and Urban Mexico”, Simone Bertoli, Herbert Brücker y Jesús Fernández-Huertas Moraga.
19. 2013-02: “Understanding Different Migrant Selection Patterns in Rural and Urban Mexico”, Jesús Fernández-Huertas Moraga.
20. 2013-01: “Publicizing the results of standardized external tests: Does it have an effect on school outcomes?, Brindusa Anghel, Antonio Cabrales, Jorge Sainz y Ismael Sanz.
21. 2012-12: “Visa Policies, Networks and the Cliff at the Border”, Simone Bertoli, Jesús Fernández-Huertas Moraga.
22. 2012-11: “Intergenerational and Socioeconomic Gradients of Child Obesity”, Joan Costa-Fonta y Joan Gil.
23. 2012-10: “Subsidies for resident passengers in air transport markets”, Jorge Valido, M. Pilar Socorro, Aday Hernández y Ofelia Betancor.
24. 2012-09: “Dual Labour Markets and the Tenure Distribution: Reducing Severance Pay or Introducing a Single Contract?”, J. Ignacio García Pérez y Victoria Osuna.
25. 2012-08: “The Influence of BMI, Obesity and Overweight on Medical Costs: A Panel Data Approach”, Toni Mora, Joan Gil y Antoni Sicras-Mainar.
26. 2012-07: “Strategic behavior in regressions: an experimental”, Javier Perote, Juan Perote-Peña y Marc Vorsatz.
27. 2012-06: “Access pricing, infrastructure investment and intermodal competition”, Ginés de Rus y M. Pilar Socorro.
28. 2012-05: “Trade-offs between environmental regulation and market competition: airlines, emission trading systems and entry deterrence”, Cristina Barbot, Ofelia Betancor, M. Pilar Socorro y M. Fernanda Viecens.
29. 2012-04: “Labor Income and the Design of Default Portfolios in Mandatory Pension Systems: An Application to Chile”, A. Sánchez Martín, S. Jiménez Martín, D. Robalino y F. Todeschini.
30. 2012-03: “Spain 2011 Pension Reform”, J. Ignacio Conde-Ruiz y Clara I. Gonzalez.
31. 2012-02: “Study Time and Scholarly Achievement in PISA”, Zöe Kuehn y Pedro Landeras.
32. 2012-01: “Reforming an Insider-Outsider Labor Market: The Spanish Experience”, Samuel Bentolila, Juan J. Dolado y Juan F. Jimeno.