When Credit Dries Up: Job Losses in the Great Recession

Transcription

1 When Credit Dries Up: Job Losses in the Great Recession Samuel Bentolila CEMF Gabriel Jiménez Banco de España Marcel Jansen U. Autónoma de Madrid Sonia Ruano Banco de España Presentation at MéRA-AMSE Conference on Growth in Europe?, Marseille 24 September 2015 This paper is the sole responsibility of its authors. The views presented here do not necessarily re ect those of either the Banco de España or the Eurosystem 1 / 29

2 Motivation Do shocks to the banking system have real e ects and, if so, do they give rise to employment losses? Policymakers in Europe and the US are concerned because the decline the lending during the Great Recession is seen as a primary factor in the slow recovery (MF, 2013) These concerns led to massive injections of capital in banks with solvency problems (over e 600 Bn in Europe) However, recent evidence on the real e ects of credit supply shocks and the economic bene ts of bailouts is scarce. Why? Lack of good data on bank credit to rms (in the US) Challenging identi cation issues 2 / 29

3 The basic challenge Disentangle credit supply and credit demand shocks: A nancial crisis may force banks to reduce credit supply, but it may also induce rms to reduce credit demand The troubles of rms may cause the hardship of banks, inducing reverse causality Firms may have alternative sources of funding Selection, with an over-representation of vulnerable rms in weak banks n recent years most studies exploit quasi-experimental techniques to overcome these identi cation problems 3 / 29

4 The Spanish experience in the Great Recession (GR) The Spanish economy o ers an ideal setting to explore how shocks to credit supply spillover to the real economy: A large employment fall: 9% over Spanish rms rely heavily on bank credit and the high leverage ratios of many rms, mostly SMEs, made them vulnerable to the reduction in credit supply during the GR This credit supply shock, reducing credit by 40% in real terms from 2007 to 2010, was originated by a boom-bust cycle in housing prices with catastrophic e ects on bank solvency 4 / 29

5 Our approach We exploit large cross-sectional di erences in lender health at the onset of the crisis The weakest banks, all of them cajas de ahorros (savings banks), were bailed out by the State mostly after The rest survived without nancial assistance Bailed-out or weak banks overinvested in loans to real estate during the boom Weak banks reduced credit more during the recession We compare the change in employment from 2006 to 2010 at two sets of Spanish rms: those with a pre-crisis exposure to weak banks in 2006 and those with no exposure 5 / 29

6 Our strength Data set We have access to a unique dataset with con dential information about all bank loans to non- nancial rms and loan applications to non-current banks These data are linked to rm and bank balance sheet data The data allow us to reconstruct the entire banking relations and credit histories of close to 170,000 non- nancial rms To the best of our knowledge, it is the most extensive database ever assembled for the analysis of credit supply shocks n our empirical analysis we include an exhaustive set of rm controls to account for selection and have instruments generating exogenous variation in weak-bank exposure 6 / 29

7 Summary of results Controling for selection, weak-bank exposure caused an extra employment fall of around 2.2 pp from 2006 to 2010 This corresponds to about one-quarter of aggregate job losses in rms exposed to weak banks in our sample The results are very robust and reveal sizable di erences depending on rm nancial vulnerability and rm size Di erences in employment destruction and rm exit are concentrated at multi-bank rms 7 / 29

8 Plan of the talk The nancial crisis in Spain Data and treatment variable Empirical strategy Empirical results: Baseline (DD) Transmission mechanism and exogenous exposure Treatment heterogeneity Probability of exit Job loss estimates Conclusions 8 / 29

9 The nancial crisis in Spain The Spanish cycle Expansion, : GDP 3.7%; employment 4.2% (p.a.) Recession, : GDP -3.0%; employment -9.0% Bank credit boom-bust: Real annual ow of new credit to non- nancial rms by deposit institutions : 23%, : -38% Euro membership + Expansionary monetary policy (ECB): Sharp drop in real interest rates and easy loans to real estate developers and construction companies (RE): 14.8% of GDP 2002 to 43% in 2007! Housing bubble: 59% rise in real housing prices over (-15% over ) 9 / 29

10 Savings banks and the credit collapse Savings banks: same regulation and supervision, di erent ownership and governance Market shares and exposure to RE (%): Credit to Non- Loans to RE/ Fin. Firms Loans to NFF Weak banks Healthy banks Di erential credit growth: Expansion ( ): Weak 40% v. healthy 12% Recession ( ): Weak -46% v. healthy -35% Both at intensive and extensive margins 10 / 29

11 Savings banks and the credit collapse New credit to non- nancial rms by bank type (12-month backward moving average, 2007:10=100) 11 / 29

12 Savings banks and the credit collapse Acceptance rates of loan applications by non-current clients, by bank type (%) 12 / 29

13 The bank restructuring process Steps: So: 1. Nationalization and reprivatization (2 WBs, 3/2009-7/2010): 0.44% GDP 2. Mergers (26 WBs) and takeovers (5 WBs), from 3/2010: 1.1% GDP by 12/ Further consolidations and nationalizations (since 2011). Loan from European Financial Stability Facility to nance weak-bank recapitalization (6/2012, 4% GDP) Weak bank de nition: nationalized, merged with State support, or taken over by another bank : Run by own managers (exc. 2 weak banks in 1.) nstitutional Protection System: separate legal entities 13 / 29

14 Data: Six di erent databases 1. Central Credit Register of the Bank of Spain (CR): All loans above e 6,000: identity of bank and borrower, collateral, maturity, etc. Firms credit history: non-performing loans and potentially problematic loans 2. CR: Loan applications by non-current borrowers 3. Annual balance sheets and income statements of rms from Spanish Mercantile Registers via SAB Exclude construction, real estate, and related industries: 169,295 rms Coverage: 21% rms, 32% value added, 48% private employees 4. Firm entry and exit from Central Business Register 5. Bank balance sheets from supervisory Bank of Spain database 6. Bank location database 14 / 29

15 The treatment variable WB i : Dummy variable that takes the value 1 if the rm had any loans with weak banks in 2006 Alternative continuous treatment: WB intensity i : loans from weak banks to asset value Sample data: Employment fell by 8.1% 39% of rms had credit from weak banks 15 / 29

16 Firm characteristics There are signi cant di erences in the characteristics of rms in the treatment and control groups. Treated rms are on average: Older and larger Financially worse: less capitalized, liquid, and pro table, more indebted with banks More loan applications to non-current banks, more frequent defaults These di erences lead to di erent unconditional trends. We rely on random selection conditional on controls 16 / 29

17 Empirical strategy Di erences in di erences (DD), estimated in di erences: 4 log 1 + n ijkt = α + βwbi + X i γ + d j δ + d k λ + u ijkt 4 =4-year di erence, n ijkt =employment at rm i in municipality j, industry k, year t=2010! Trends in all control variables n ijkt set to zero for rms in the sample in 2006 but not in 2010 because they closed down! log(1 + n ijkt )! Surviving and closing rms β measures Average Treatment e ect on the Treated (ATT) 17 / 29

18 Threats to identi cation Demand e ects (Mian-Su, 2014): Lending in RE / certain areas! Larger drop in household demand and higher density of (non-re) rms exposed to WB! Job losses from consumption rather than credit! Dummies by: Municipality (3,697) and ndustry (80) Non-random matching: laxer loan-approval criteria at WB may cause bias in risk pro le and reduce access to credit. And aggregate shocks may di erentially a ect productivity (Paravisini et al., 2015)! Firm controls: Productivity: age, age 2, size, ROA, share of temp contracts Finance: bank debt, short-term and long-term bank debt, liquidity, own funds, no. past loan applications to non-current banks and whether all accepted, past loan defaults, current loan defaults, credit lines, no. banking relationships and square, share of uncollateralized loans 18 / 29

19 Baseline: Di erence in di erences Dependent variable: 4 log 1 + n ijkt No rm Some rm Baseline Placebo controls controls ( 02-06) WB i (0.010) (0.010) (0.007) (0.008) R No. obs. 169, , , ,997 Municipality and industry xed e ects, and rm controls included (No controls: -7.7% di erence.) 19 / 29

20 Alternative speci cations Selection: Exact matching within municipality, industry, and rm control cells (0-1 dummies) Selection: Panel with rm-speci c trends (t=2007,...,2010): -3.3 pp log (1 + n ijkt )=α 0 i +WB id t β 0 +X i d t γ 0 +d j d t δ 0 +d k d t λ 0 +d t φ + v ijkt Demand: traded-goods sectors, based on high concentration - highest quartile Her ndahl (Mian-Su, 2014) Alternative de nition of weak banks: 2006 loan exposure to rms in RE within upper quartile E ect is expected to increase with the degree of exposure WB ntensity i = Loans from weak banks / Asset Value At average intensity of exposure, the e ect is 1.9 pp 20 / 29

21 Alternative speci cations Dependent variable: 4 log 1 + n ijkt Exact Tradable Loans ntensity matching goods to RE WB i WB i ntensity (0.010) (0.020) (0.008) (0.028) R No. obs. 169,295 21, , ,295 Municipality and industry xed e ects, and rm controls included 21 / 29

22 s credit the channel? Bank- rm sample (281,016 observations, 98,754 rms) 4 log (1 + Credit ibt ) =η i +βwb b +γfb ib +ε ibt Credit ibt = total credit committed by bank b to rm i, t=2010, FB it = rm-bank controls [log(no. of years with the bank), past defaults with the bank] Fixed e ects leave only rms that work with more than 1 bank, but this speci cation controls perfectly for credit demand (Khwaja-Mian, 2008) Result: bβ= (0.040)! Weak banks cut credit more to rms that were their clients in / 29

23 nstrumental variables: s credit the channel? Transmission mechanism (t = 2010): 4 log 1 + n ijkt 4 log 1 + Credit ijkt = α 00 +β 00 4 log 1 + Credit ijkt +X i γ 00 +d j δ 00 +d k λ 00 +ε ijkt = ρ + µwb i +X i η + d j σ + d k ψ + ω ijkt Credit it = total credit committed by banks. Exclusion restriction: Working with a weak bank a ects Employment only through Credit Trade credit: Subsample of rms (8%), with nancial institutions (32% of liabilities) and trade credit (35%) Exogenous exposure to WB: regulatory change in 12/1988 allowing savings banks to expand beyond region of origin! density of weak-bank branches in 12/1988 by municipality. Exclusion restriction: local weak-bank density only a ects employment through exposure to weak banks 23 / 29

24 nstrumental variables: s credit the channel? Dependent variable: 4 log 1 + n ijkt nstrumented 4 log 4 log WB i variable (1+Credit it ) (1+Credit it ) a (0.127) (0.367) (0.026) First stage WB i Local weak-bank density i (0.010) (0.019) (0.071) Overall e ect F test / p value 23.1/ / /0.00 No. obs. 169,295 12, ,295 ndustry f.e. and rm controls included. Col. (1) has municipality f.e.; Col. (3) has coastal province f.e. a Col. (2) uses total credit 24 / 29

25 Treatment heterogeneity: Financial vulnerability Dependent variable: 4 log(1 + n ijkt ) Rejected applications i log(total Assets) i (0.008) (0.005) WB i Rejected applic. i WB i log(total Assets) i (0.013) (0.006) Defaults i Single bank i (0.031) (0.007) WB i Defaults i WB i Single bank i (0.033) (0.016) Short-term debt i WB i (0.013) (0.007) WB i Short-term debt i R (0.014) No. obs. 169,295 Municipality and industry xed e ects, and rm controls included 25 / 29

26 Probability of exit Lower WB employment e ect on survivors (DD): 1.1 pp Dependent variable: Probability of exit from 2006 to 2010 i WB i (0.004) WB ntensity i (0.015) (0.016) WB ntensity i Single bank i (0.017) R No. obs. 169, , ,295 Municipality and industry xed e ects, and rm controls included WB ntensity i : 9th v. 1st decile! 1.7% higher probability (17% of baseline rate) 26 / 29

27 Job loss estimates Caveat: These are not macro e ects (Chodorow-Reich, 2014) These are only di erential e ects Share of observed aggregate job losses in exposed rms: Baseline: 23.7% of job losses Survivors: 52% of job losses Closing rms: 19% of job losses 27 / 29

28 Conclusions Aim: measure the impact of credit constraints on employment during the Great Recession in Spain denti cation: We exploit di erences in lender health at the onset of the crisis, as evidenced by savings banks bailouts We nd that job losses from expansion to recession at rms exposed to weak banks are signi cantly larger than at similar non-exposed rms This explains around one-fourth of aggregate job losses at exposed rms 28 / 29

29 Conclusions The estimated e ects vary considerably with the rm s creditworthiness and the structure of its banking relationships, in particular how many banks it works with Credit constraints do not just force rms to purge jobs but also cause some of them to close down Given our controls, constrained rms would have received more credit had they not been attached to weak banks and, in this sense, these job losses are ine cient 29 / 29