When Credit Dries Up: Job Losses in the Great Recession · Motivation Policymakers in Europe and...

When Credit Dries Up:Job Losses in the Great Recession

Samuel Bentolila

CEMFI

Marcel Jansen

U. Autónoma de Madrid

Gabriel Jiménez

Banco de España

Sonia Ruano

Banco de España

Presentation at the 1st CSEF Conference on Finance and Labor,Center for Studies in Economics and Finance, Capri, 26 August 2013

This paper is the sole responsibility of its authors. The views presented here do not necessarily reflect those

of either the Banco de España or the Eurosystem

1 / 64

Motivation

Policymakers in Europe and the US are concerned about

I The economic implications of the current lack of credit, andI The economic benefits from the bailout of banks

Yet, so far there is scant recent evidence on both issues. Why?

I Lack of good credit dataI Tricky identification issues

In this paper we contribute to the literature that explores the realeffects of credit supply shocks

2 / 64

The Spanish experience

The Spanish experience in the Great Recession offers an idealsetting to explore the real effects of shocks to credit supply

I Bank lending to non-financial firms contracted significantly:New credit growing at 27% p.a. in 2007 v. -25% in 2010

I There was an unprecedented decline in employment (in 201010% from peak)

I Boom-bust cycle in housing prices: trebled 1996:4 to 2007:4,then fell by 11% to 2010 (interesting parallels with US,Ireland)

I Further advantage for identification: extraordinarily highquality of the data

3 / 64

The basic challenge

How to disentangle credit supply and credit demand shocks

I A large macro shock may force banks to reduce credit supply,but it may also induce firms to reduce demand for credit

I Reverse causality: the economic troubles of firms may causethe hardship of banks rather than the other way around

I Selection: bad firms may select bad banks or vice versa

4 / 64

Recent related empirical literature

Country-industry panel (1970-2003): Pagano and Pica (2012). Inbanking crises in developed economies, negative shocks hurtemployment more in financially dependent industries

Quasi-experimental techniques:

I Large external shocks to the banking sector: Chava andPurnanandam (2011)

I Cross-sectional differences in firm’s financial vulnerability atthe onset of the Great Recession: Almeida et al. (2011),Benmelech et al. (2011), Garicano and Steinwender (2013),Boeri et al. (2013)

I Cross-sectional differences in the health of banks in the GR:Greenstone and Mas (2012), Chodorow-Reich (2013)

All find sizeable effects on real variables, but none can replicatecomplete banking relations and nor have such high-quality controls

5 / 64

Our approach

Our identification strategy exploits the large cross-sectionaldifferences in the health of lenders at the onset of the crisis

I Some banks, all but one of them savings banks (Cajas deAhorros), have been bailed out by the State while the restsurvived without financial assistance

I We show that the bailed-out or weak banks reduced creditmore than the other banks

I To identify real effects, we compare the change in employmentfrom the end of the boom (2006) to well within the recession(2010) at two sets of firms with different exposure (loans/totalassets) to weak banks using quasi-experimental techniques

6 / 64

Identification

Underlying assumptions: necessary conditions for the existence ofreal effects and their correct identification

I Financial market frictions: client companies of weak banksmust not be able to readily switch to healthier banks

I Firms must be unable to forecast the solvency problems ofweak banks (despite publicly available information on buildupof excessive risks)→ Both conditions are shown to hold

I Nonetheless, we find significant differences between firms inour treatment and control groups. Weak banks granted loansto worse firms. Hence, we need to deal with selection

7 / 64

Data

Merge 5 data sets: (a) all banks’balance sheets (226), (b)accounts of 217,025 non-real estate non-financial firms, (c) allbank loans above € 6,000, (d) all loan applications to non-currentbanks, (e) firm entry and exit

Vis-à-vis most closely related recent literature:

I v. Chodorow-Reich (2013): More than 200,000 firms v.2,000-3,000. Complete banking relations plus credit history ofall firms v. imputing within syndicated loans. Employment:Average 23, median 6 v. 3,000 and 620, respectively

I v. Greenstone and Mas (2012): Firm level v. county level.Actual credit v. bank-specific imputation of nationwidereduction to counties based on market share

8 / 64

Summary of results

I Controlling for selection, attachment to weak banks caused anextra employment reduction of 3 to 6 percentage points, i.e.18% and 35% of job losses, between 2006 and 2010

I The results are very robust: various estimation techniques(differences in differences, instrumental variables, exactmatching) and many robustness checks

I Sizeable differences across industries and firms with differentcredit histories

I Novel result: Single-bank firms did not suffer from attachmentto weak banks, due to better treatment

9 / 64

Plan of the talk

I Theoretical background

I The financial crisis in Spain

I Data

I Empirical strategy: DD, IV, Matching

I Estimation results

I Conclusions

10 / 64

Theoretical background

Two different theoretical literatures have dealt with thepropagation of shocks to the financial sector and their effects onthe real economy

I Financial accelerators

I Relationship banking

11 / 64


Financial acceleratorsEndogenous changes in credit markets may amplify, propagate oreven initiate shocks to the real economy

I Agency costs drive a wedge between cost of internal andexternal funds; it depends negatively on borrower’s net worth

I Pro-cyclical fluctuations in borrower’s net worth lead to a risein cost of funding during recessions

I Capital-weak borrowers are the first to suffer credit restrictions(“flight to quality”) → may have to cut back on activity

Kiyotaki and Moore (1997), Bernanke and Gertler (1998): focuson firms’financial health and their relationship with banks

Holmstrom and Tirole (1997), Gertler and Kiyotaki (2010): agencyproblems between banks and their funders → shocks to banks’networth can lead to a credit crunch spilling over to real sector

12 / 64


Relationship banking

I Long-lasting banking relationships may help to reduce agencycosts because banks acquire valuable information about theirclient firms (Freixas, 2005)

I This information is lost when the relationship ends and cannot(easily) be transmitted to other financial institutions

I Firms subject to credit restrictions may have trouble findingalternative sources of finance:

I Agency costs are high at the outset of a new relationship

I Adverse selection: many of those firms looking for credit wereamong the weakest clients of the banks that reduced theiraccess to credit

13 / 64


Credit market imperfections and employment

Literature focuses on capital, but it is also relevant to employment:

I Turnover costs (Greenwald and Stiglitz, 1993; Sharpe, 1994)I Search and matching frictions (Wasmer and Weil, 2004;Boeri, Garibaldi, and Moen, 2013; Petrosky-Nadeau, 2013)

→ Countercyclical fluctuations in cost of funding may depresshiring or cause firing in recessions

Other results:

I Temporary jobs serve as a buffer stock against expectedfuture credit restrictions (Cuñat and Caggese, 2008).

I Financial market disruptions may have a stronger impact onemployment in countries with more advanced financial sectors(Pagano and Pica, 2012)

14 / 64

The demise of savings banks

The Spanish cycle

I Great Expansion (1996-2007): GDP 3.7%, employment 4.1%I Great Recession (2008-2010): GDP -1.1%, employment -3.2%

The boom-bust cycle in the housing market

I Spanish banks fueled the bubble. Loans to construction firmsand real estate developers (real estate industry, REI): 10%GDP 1992 to 43% 2007

Bank solvency and supervision

I Regulation forced banks to keep most securitized assets onbalance sheets and introduced dynamic provisioning (2000):initial cushion, but banks fueled a local housing boom andfunded lending from external sources → a big problem whenEurozone money markets froze (2008)

15 / 64

The demise of savings banks

I Heterogeneity: Main risks were at savings banksI Initially mergers/takeovers: 47 in 2006 to 11 in 2011I Then increased capital requirements, large capital shortfall →€41 billion loan ('4% GDP) from European FinancialStability Facility

I Unlisted, peculiar governance (Cuñat and Garicano, 2010)I Market shares and exposure to Real Estate Industry (%):

Credit to Non- Loans to REI/Fin. Sector Loans to NFS

Weak banks 32 64Other banks 67 34

I Non-performing loans of weak v. other banks, 0.01 ppend-2006, 1.6 pp end-2010 (7.6 pp June 2012)

16 / 64

The credit collapse

I Exceptionally high reliance of Spanish firms on bank credit:loans from credit institutions to non-financial corporations,86% GDP v. 62% in EU (2006)

I Corporate debt issue is not an alternative: Over 2002-2010only 16 very big companies issued debt in the market (onaverage, 5 companies per year)

→ Spain is very well suited to study credit constraints arisingfrom bank credit supply shocks, since alternative sources offunding are hard to come by

I Real bank credit grew enormously during the boom and thencollapsed spectacularly:

17 / 64

The credit collapse

�New credit to non-‐financial firms by bank type (12-‐month backward moving average, 2007:10=100)

The credit collapse

Evolution

I Credit by weak banks grew more than by other banks duringthe boom and fell more during the slump

I Both intensive margin and extensive margin:

Difference between weak and healthy banks in acceptance rateof loan applications by non-current clients (pp):

All Applied to bothWBs & non-WBs

2002-2004 3.1 4.02009-2010 -6.5 -6.3

19 / 64

The credit collapse

��Acceptance rates of loan applicaFons by non-‐current clients, by bank type (%)

The credit collapse

��Acceptance rates of loan applicaFons by non-‐current clients, by bank type. Firms applying to at least one bank per type (%)

Not much acFon in interest rates

Difference in average annual interest rate for new loans to non-‐financial firms between bailed out and non-‐bailed out banks (basis points)

Anticipation of bank insolvency risk?

I Weak banks relied more on securitization (ratio to assets, %)

Weak banks Other banks2002 5.0 3.62006 16.7 13.5

I 2006 securitization: Floating-rate, quarterly coupons, ref. to3-month Euribor (303 deal-tranche obs., 24 issuers)

I Controls: type (MBS, ABS), risk category (AAA, AA+ toBBB-, BB+ to D), collateral type, guarantor type, years tomaturity, month

I Dummy=1 if weak bank: 2.8 basis points (p-value: 0.55)

→ Financial markets failed to recognize differential risk at weakbanks as late as 2006 → Private firms could not have predictedthem either

23 / 64

Data

I Annual balance sheets and income statements (2006) fromSpanish Mercantile Registers via SABI. Coverage: 27% firms,37% value added, 61% private employees

I Exclude construction, real estate, and related industries (20%output to REI in 2000 input-ouput tables): 217,025 firms

I Firm entry and exit from Central Business RegisterI Loan information from Central Credit Register (B. of Spain):

I All loans above € 6,000 ($ 8,100): identity of bank, sharefrom weak banks, collateral, maturity, etc.

I Firms’credit history: non-performing loans and potentiallyproblematic loans

I Loan applications by non-current borrowers and decisions

I Banks’balance sheets from regulatory and supervisory Bankof Spain database (226)

24 / 64

The treatment variable

Aim: Measure employment losses from differential effect offinancial crisis on lending capacity of banks due to differences intheir health at the onset of the crisis. How?

I Change in employment in 2010 (Post) v. 2006 (Pre-crisis)between firms with high and low weak bank loan shares

Treatment

I About one-third of firms had no exposure to weak banksI WBi: Pre-crisis share of firm’s debt with weak banks in totalasset value of the firm (overall leverage + relative importanceof WBs) above the third decile of the distribution, i.e. 6.3%→ share of WBs in bank credit at least 51.4%(averages: 25% assets and 71% bank credit, respectively)

25 / 64


�Histogram of the exposure of firms to weak banks (excluding no exposure) (%)


Are firms comparable in the treatment and control groups? No.Treated firms are on average:

I Younger and smallerI Financially worse: less capitalized, lower liquidity, lessprofitable, more indebted with banks, but loans with highermaturity and more collateralized

I Credit history (2002-2005): more loan applications to newbanks and more applications accepted, more defaults

I Their banks: smaller, less capitalized, lower liquidity, highershare of loans in mortgages, less profitable, and morenon-performing loans

I Their employment grew more in the boom and fell more in therecession → Different unconditional trends → Need to controlfor firm characteristics

27 / 64

Variable (2006) Control Treated T-CNo. of Firms 155,167 60,860Share Loans Weak Banks 0.10 0.71 0.61Employment 24.63 18.73 -5.91Firm Size (million euros) 5.08 3,01 -2.07Firm Age (years) 12.16 11.01 -1.15Own Funds 0.33 0.24 -0.10Liquidity 0.12 0.09 -0.04Return on Assets 0.06 0.05 -0.01Bank Debt 0.32 0.50 0.19Short-Term Bank Debt 0.48 0.44 -0.04Long-Term Bank Debt 0.22 0.31 0.09Uncollateralized Loans 0.81 0.72 -0.09Banking Relationships 1.94 2.98 1.03Past Defaults (share) 0.02 0.03 0.01Loan Applications (share) 0.55 0.69 0.14All Applications Accepted 0.22 0.26 0.04

28 / 64

Three takes on credit constraints

I. Difference in differences (DD)

log(1+ nit) = α+ δWBi + γPost WBi + βPost+ ηds + θPost ds +X′iφ+ uit

I nit = employment at firm i in year t (2006, 2010), Post =2010, dS = full sets (minus one) of province (50) and industry(9) dummies, Xi = controls

I Unbalanced panel: some firms observed in 2006, others in2010. Firms in 2006 but not in 2010 because they closeddown have nit=0 (thus log(1+ nit)) → surviving and closingfirms (4.4% obs., 25% job losses)

I γ measures Average Treatment effect on the Treated (ATT)I Many robustness checks and DDD (interactions)

29 / 64


I. Difference in differences (DD). Control variables

I Province (50) and industry (10) dummies

I Main bank dummies (226)

I Other firm characteristics:

Firm Size, Firm Age, Firm Age Squared, Own Funds,Liquidity, Return on Assets, Temporary Employment, BankDebt, Short-Term Bank Debt, Long-Term Bank Debt,Uncollateralized Loans, Credit Line, Banking Relationships,Banking Relationships Squared, Current Defaults, PastDefaults, Loan Applications, All Applications Accepted

30 / 64


II. Instrumental variables: Is it credit?

∆ log(1+ nit) = α′ + δ′∆ log(1+ Creditit) + β′Postt

+ η′ds + σ′di + u′it∆ log(1+ Creditit) = π + µPosttWBi +ωPostt + ρds + ψdi + vit

I Creditit = total credit committed —drawn and undrawn—bybanks to firm i in year t, di = firm fixed effect, t = 2007,...,2010 (panel)

I Exclusion restriction: working with a weak bank affectsemployment changes only through credit changes

31 / 64


II. Instrumental variables: Exogenous variation in weak-bankattachment

I We exploit a legal change in December 1988 that allowedexpansion of savings banks beyond their region of origin

I We compute the share of bank branches belonging to weakbanks at the province level right before the legal change anduse a dummy for high-density of weak banks as an instrumentfor WBi

32 / 64


III. Exact matching

I Estimate DD equation using coarsened exact matchingmethod, i.e. within cells from discretizations of dS and Xi

I Cells defined by variables chosen according to theirsignificance in DD estimates. Each variable is split in twousing: 0/1 nature, sample median, province (East Coast plusIslands), industry (Agriculture, Farming, Mining)

I 14 variables: Defaults, Bank Debt, Credit Line, Firm Age,Firm Size, Industry, Long-Term Bank Debt, Short-Term BankDebt, No. of Banking Relationships, Own Funds, Province,Rejected Loan Application, Return on Assets, and TemporaryEmployment

33 / 64

Results: I. Difference in differences

Table 2 —Baseline

I Mean employment differential fall from treatment: 8.5 pp,same as controling for province and industry dummies

I Plus firm characteristics: 7.4 pp

I Plus employment trends by province and industry: 6.2 pp(baseline), with sum of Post terms: 17.7 pp → 35% extra joblosses

Note: ** = 1%, * = 5% significance.

34 / 64


Dependent variable: log(1+Employmentit)

Baseline

Post×WBi -0.085∗∗

-0.074∗∗

-0.062∗∗

(0.013) (0.013) (0.009)

Province and Industry Dummies yes yes yesFirm Controls no yes yesMain Bank Dummies no no yesPost×Province & Industry Dummies no no yes

R2 0.009 0.489 0.494No. of firms 217,025 217,025 217,025No. of observations 387,482 387,482 387,482

35 / 64


Tables 3-4: Robustness checks

I Placebo (2002-2006): no significant effectI Monotonic: 0 (2007), 0.6 pp (2008), 3.1 (2009), 6.2 (2010)I Anticipation via date of controls: 5.9 pp 2002, 6.1 pp 2005I Date back WBi variable (tradeoff): 3.5 pp 2002, 4.9 pp 2005I New Treatment = Banks with REI share>3rd decile: 6.2 ppI Committed loans (v. drawn only): 5.7 ppI WBi loan levels (v. share of loans): 5.4 ppI Surviving firms: 1.3 pp

36 / 64


Dependent variable: log(1 + Employmentit)

Placebo Post year2002 2007 2008 2009 2010

Post×WBi -0.001 0.005 -0.006 -0.031∗∗

-0.062∗∗

(0.001) (0.003) (0.005) (0.007) (0.010)

R2 0.003 0.553 0.534 0.513 0.494No. of firms 101,515 216,895 217,035 217,078 217,025No. of observations 191,948 411,345 398,819 400,573 387,482

37 / 64



Timing of controlsFirms Banks

2002 2005 2000 2002


-0.061∗∗

-0.035∗∗

-0.049∗∗

(0.010) (0.009) (0.006) (0.007)

R2 0.488 0.500 0.526 0.517No. of firms 106,122 150,690 99,869 136,280No. of observations 192,765 271,540 181,751 246,362

38 / 64



Treatment variable Surviving% loans Committed Bank firmsto REI loans loans sample


-0.057∗∗

-0.054∗∗

-0.013∗∗

(0.008) (0.008) (0.009) (0.004)

R2 0.494 0.494 0.494 0.546No. of firms 217,025 217,025 217,025 199,691No. of observations 387,482 387,482 387,482 353,060

39 / 64

Results: I. Triple differences

I Table 5: By industry. Manufacturing (8.4), Machinery renting,Computing, and R&D (6.8),... Hotels and Catering (4.3)

I Do financially vulnerable firms destroy more jobs? / Do weakbanks treat firms them worse?Reference: 6.2 pp

Table 6. Triple difference: Post × WBi × Dummies

I Stronger effects for rejected credit applications and defaults,short-term debt, small size

I But: Not for firms with only one banking relationship

40 / 64

Results: I. Triple differences (I)


(1) (2) (6)


-0.058∗∗

0.093∗∗

(0.007) (0.009) (0.034)Post× Rejected applicationsi -0.066

∗∗-0.060

∗∗

(0.004) (0.005)Post×WBi× Rejected applicationsi -0.032

∗∗-0.013

(0.010) (0.010)Post× Defaultsi -0.247

∗∗-0.227

∗∗

(0.026) (0.028)Post×WBi× Defaultsi -0.033 -0.006

(0.022) (0.027)

41 / 64

Results: I. Triple differences (II)


(3) (4) (6)

Post× Short-term debti -0.077∗∗

-0.074∗∗

(0.007) (0.009)Post×WBi× Short-term debti -0.080

∗∗-0.071

∗∗

(0.012) (0.013)Post× Small firmi 0.045

∗∗0.010

(0.013) (0.015)Post×WBi× Small firmi -0.055 -0.120

∗∗

(0.033) (0.033)

42 / 64

Results: I. Triple differences (III)


(5) (6)

Post×WBi -0.010 0.093∗∗

(0.033) (0.034)Post× Single banki 0.041

∗∗0.030

∗∗

(0.004) (0.004)Post×WBi× Single banki 0.056

∗∗0.029

∗∗

(0.010) (0.010)

43 / 64

Results: I. Differences in differences by exposure level

Facts:

I Single- and multi-bank firms show differential effectsI Share of single-bank firms increases with exposure to WBs

→ Estimate for different levels of exposure, separately for single-and multi-bank firms

I Non-monotonicity starting in seventh decile: positive forsingle-bank firms and negative for multi-bank firms

I Why? See below

44 / 64

Results: I. Differences in Differences

The employment effect of exposure to weak banks by decile and number of banks (DD esFmates with 2-‐s.e. bands)

Results: I. Strength of single-‐bank dependence

Share of single-‐bank firms by decile of exposure to weak banks (%)

Results: II. Instrumental variables

1. Is it credit?

I Proportional change in employment on credit committed bybanks (drawn and undrawn), instrumented by WBi

I Panel with firm fixed effects to absorb firm characteristics andinteract WBi with committed lending for three year indicators

Table 10:

I First stage: IV negatively correlated with credit change, moreso as the recession lengthens

I ∆Credit affects ∆Employment, with a coeffi cient of 0.42.Overall effect: in 2010 v. 2007 (2006 not identified) of 6.5 pp

I Similar effects for I(Loan Rejection) (6.7 pp) and % LoansAccepted (7.4 pp)

47 / 64

Results: II. Instrumental variables: (I) Credit channel

First stageDependent variable: ∆log(1+Creditit) I(Rejection) % Accepted

d2008× WBi -0.022∗∗

0.014∗∗

-0.005∗∗

(0.006) (0.003) (0.001)d2009× WBi -0.095

∗∗0.024

∗∗-0.011

∗∗

(0.014) (0.004) (0.002)d2010× WBi -0.154

∗∗0.029

∗∗-0.014

∗∗

(0.016) (0.005) (0.003)

p− value of F test 0.00 0.00 0.00No. of firms 716,678 716,678 502,331No. of observations 716,678 716,678 502,331

48 / 64

Results: II. Instrumental variables: (I) Credit channel

Dependent variable: ∆log(1+Employmentit)Regressor: ∆log(1+Creditit) I(Rejection) % Accepted

0.424∗∗

-2.280∗∗

5.364∗∗

(0.098) (0.461) (1.193)

Overall effect -0.065 -0.067 -0.074

49 / 64

Results: II. Instrumental variables

2. Exogenous exposure to weak banks

I Exploit a 1988 legal change whereby savings banks could startoperating outside their region of origin

I Use high-density of weak banks at the province level (share ofbranches above a given threshold) as instrument for WBi

I Table 11: Effects: P50 (5 pp), P75 (6.8 pp), P90 (7 pp)

I Use exposure to real estate industry in 2000 as instrument forWBi: 3.4 pp

50 / 64

Results: II. Instrumental variables: (II) Backdated IVs

Dependent variable: ∆log(1+Employmentit)

High weak-bank density Exposureprovince (1988) to REI

P50 P75 P90 (2000)First stage

Dependent variable: WBiHigh weak-bank densityi 0.034

∗∗0.034

∗∗0.042

∗∗0.032

∗∗

(0.004) (0.004) (0.006) (0.004)Dependent variable: Post×WBiPost×High weak-bank densityi 0.104

∗0.132

∗0.145

∗0.141

∗∗

(0.043) (0.054) (0.063) (0.038)

p− value of F test 0.00 0.00 0.00 0.00No. of firms 217,025 217,025 217,025 217,025No. of observations 387,482 387,482 387,482 387,482

51 / 64

Results: II. Instrumental variables: (II) Backdated IVs

High weak-bank density Exposureprovince (1988) to REI

P50 P75 P90 (2000)Dependent variable: ∆log(1 + Employmentit)


-0.512∗∗

-0.485∗∗

-0.239∗∗

(0.188) (0.293) (0.202) (0.076)

Marginal effect -0.050 -0.068 -0.070 -0.034

52 / 64

Results. III. Exact matching

I Estimate within 14 × 14 cells of ex-ante very similar firms,with coarsened exact matching method (Iacus et al., 2011)

I 16,384 potential strata, 4,822 strata with observations, 3,553strata can be matched

I Check for different pre-treatment trends (Figure)I Table 5: Weak bank dummy estimated effect of 3.2 ppI Separate estimates for single-bank (nil) and multi-bank firms(4.8 pp)

53 / 64

Results: III. Exact matching

EvoluFon of employment at firms aSached to weak banks and non-‐aSached firms, weighted by matching (2006=0) (%)


Dependent variable: log(1+Employmentit)

Postt× WBi -0.032∗∗

(0.014)

No. of strata 4,822No. of matched strata 3,553

R2 0.488No. of firms 211,284No. of observations 377,498

Note: Estimate from DD in this sample: -0.0632 (0.009).

55 / 64


The employment effect of exposure to weak banks by decile and number of banks (matching esFmates with 2-‐s.e. bands)

Back to Results: I. Strength of single-bank dependence

DDD estimates differ for single-bank and multi-bank firms. Why?

Preferential treatment?

I ∆log(1+Creditijt) during recession on share of loans with samebank in 2006

I Significant only for weak banks: Evergreening?I Why? Lower cost: cheaper, more information, stronger stake

Stigma?

I Loan requested and grantedijt on share of loans with weakbanks (70% to 100%)

I Significant from 80% (weakly, at 10%, afterwards)

57 / 64

Back to Results: I. Strength of single-bank dependence

Dependent variable: ∆log(1+Creditijt)

Share of loans with the bankij -0.057(0.056)

Share of loans with the bankij ×WBi 0.338∗∗

(0.096)

R2 0.207No. of firms 509,800No. of observations 3,753,140

58 / 64

Back to Results: I. Loan application acceptance

Dependent variable: Loan requested and grantedijt

Threshold: ≥80% ≥90% 100%

Loan share with weak banks in -0.009∗

-0.009 -0.0082002-2006 above the thresholdi (0.004) (0.005) (0.005)

R2 0.010 0.010 0.010No. of firms 109,172 109,172 109,172No. of observations 240,179 240,179 240,179

59 / 64

Non-bank credit sources?

I We only have data on the liability structure for a subsample(15,323 firms, i.e. 7% of firms). It is composed of larger firms(€ thousand, 2006):

Full Restrictedsample sample

Mean assets 4,498 40,040Median assets 576 9,137

I Liability structure (median, 2006): Financial institutions, 34%;Commercial credit, 34%, Other firms in the group: 0.1%.

I Commercial credit is unlikely to have provided extra fundingduring the recession: Molina (2012) shows for a sample of9,602 Spanish large firms (Central de Balances) no increase incommercial credit taken in 2008-2010

60 / 64


Total Credit v. Bank Credit

First stageDependent variable: ∆log(1+ Bank Creditit) ∆log(1+ Total Creditit)

d2008× WBi 0.015 -0.072∗∗

(0.012) (0.013)d2009× WBi -0.100

∗∗-0.118

∗∗

(0.020) (0.013)d2010× WBi -0.150

∗∗-0.147

∗∗

(0.025) (0.021)

p− value of F test 0.00 0.00No. of firms 15,323 15,323No. of observations 57,013 57,013

61 / 64


Total Credit v. Bank Credit

Dependent variable: ∆log(1+Employmentit)Regressor: ∆log(1+Bank Creditit) ∆log(1+Total Creditit)

0.266∗∗

0.301∗∗

(0.096) (0.082)

Overall effect -0.040 -0.044

62 / 64

Comparison with the literature

This paperControlling for selection, attachment to weak banks caused largerjob losses from 2006 to 2010: 3.2 (matching) to 6.2 pp(Diff-in-Diffs, IV), i.e. 18% to 35% (*)

Greenstone-Mas (2012)Upper bound: the 2007-2009 decline in small business lendingaccounted for up to 20% of the decline in employment in firmswith less than 20 employees and 16% of the total loss

Chodorow-Reich (2012)Withdrawal of credit accounts for between 33% and 50% of theemployment decline at small and medium firms in the yearfollowing the Lehman bankruptcy

(*) Garicano and Steinwender (2013) find for manufacturing anemployment fall of 5.6 pp in 2008-2010 due to Spanish ownership

63 / 64

Conclusions

I We aim at measuring the impact of credit constraints onemployment during the Great Recession in Spain (outside thereal estate industry)

I We exploit differences in lender health at onset of the crisis—savings banks bailed out by the State—, lack of anticipationof the bailout, and diffi culty in changing bank in the recession→ Compare change in employment pre- v. post-crisis at clientfirms of weak banks to other firms

I We construct a large dataset that allows us to controlexhaustively for ex-ante characteristics of firms and forpotential endogeneity

I Controlling for selection, attachment to weak banks causedlarger fall in employment from 2006 to 2010: 3.2 to 6.2 pp,i.e. 18% to 35%

64 / 64

When Credit Dries Up: Job Losses in the Great Recession · Motivation Policymakers in Europe and...

Documents

Transcript of When Credit Dries Up: Job Losses in the Great Recession · Motivation Policymakers in Europe and...