Fl

Ja

b

a

ARRAA

JGGG

KFLFCN

1

MUhbosuvtcI

r

M2h

h1

Journal of Financial Stability 18 (2015) 33–54

Contents lists available at ScienceDirect

Journal of Financial Stability

journal homepage: www.elsevier.com/locate/jfstabil

inance companies in Mexico: Unexpected victims of the globaliquidity crunch

ose M. Berrospidea,∗, Renata Herreriasb

Federal Reserve Board, Mailstop 153, 20th and C Streets N.W., Washington, DC 20551, USADepartment of Business Administration, ITAM, Av. Camino a Sta. Teresa 930, México, D.F. 10700, Mexico

r t i c l e i n f o

rticle history:eceived 17 December 2013eceived in revised form 11 June 2014ccepted 24 February 2015vailable online 5 March 2015

EL classification:012124

a b s t r a c t

We study the connection between the global liquidity crisis and the severe credit crunch experiencedby finance companies (SOFOLES) in Mexico using firm-level data between 2001 and 2011. Our resultsprovide supporting evidence that, as a result of the liquidity shock, SOFOLES faced severely restrictedaccess to their main funding sources (commercial bank loans, loans from other organizations, and publicdebt markets). After controlling for the potential endogeneity of their funding, we find that the liquidityshock explains 64 percent of SOFOLES’ credit contraction during the recent financial crisis (2008–2009).We use our estimates to disentangle supply from demand factors as determinants of the credit contraction.After controlling for the large decline in loan demand during the financial crisis, our findings suggest thatsupply factors (such as nonperforming loans and lower liquidity buffers) also played a significant role.

eywords:inancial crisisiquidity shockunding shockontagionon-bank finance companies

Finally, we find that financial deregulation implemented in 2006 may have amplified the effects of theglobal liquidity shock.

Published by Elsevier B.V.

tatM

tuEhtt

. Introduction

By the end of 2012, the three largest homebuilders inexico—Corporación Geo SAB, Desarrolladora Homex SAB, andrbi Desarrollos Urbanos SAB—were near financial collapse andad to undertake major debt restructuring programs to avoidankruptcy. One of the factors behind the harsh financial conditionsf these companies was the significant decline in planned homeales caused by a lack of household financing opportunities, partic-larly in the mid- and low-income sectors. This situation made itery difficult for homebuilders to honor their debts in time. Poten-

ial losses associated with homebuilder defaults were likely to hitommercial banks, some of which are owned by large U.S. banks.1

n this paper, we examine how the situation in 2012 was related

∗ Corresponding author. Tel.: +1 202 452 3590.E-mail addresses: jose.m.berrospide@frb.gov (J.M. Berrospide),

enata.herrerias@itam.mx (R. Herrerias).1 See Bloomberg article, “Citigroup’s Losses on Loans to Mexico Homebuildersay Increase”, October 15, 2013, available at http://www.bloomberg.com/news/

013-10-15/citigroup-s-losses-on-loans-to-mexico-homebuilders-may-increase.tml.

ippnchMMmac

ttp://dx.doi.org/10.1016/j.jfs.2015.02.004572-3089/Published by Elsevier B.V.

o the failure of Lehman Brothers during the fall of 2008, and wergue that financial contagion from this global liquidity shock inhe U.S. may have had long-lasting effects on the financial sector in

exico.The causes and consequences of the 2007–2008 liquidity crunch

hat followed the Lehman default in the U.S. have been widely doc-mented (see for example, Brunnermeier, 2009; Longstaff, 2010;ichengreen et al., 2012). So far, the narrative of the financial crisisas been centered on the large and visible events or institutionshat were often highlighted in the news, such as the collapse ofhe banking sector in Iceland or the bankruptcy of shadow bankingnstitutions that were heavily exposed to low-quality structuredroducts. In this paper we argue that, from a financial contagionerspective, there are other, less visible repercussions that haveot been explained because the links are more subtle and lesslear. In our example above, we argue that the financial troubles ofomebuilders in Mexico are linked to events that unfolded in theexican financial sector following the failure of Lehman Brothers.

ore specifically, the global liquidity shock affected the capitalarkets and the banking sector in Mexico, which in turn led to

severe credit crunch and the collapse of the nonbank finan-ial corporations of limited purpose (Sociedades Financieras de

3 al of F

OdsSiiMt

milccblmbscbfibc

yclifmeAmdtabMasts

tmtpeSwtsa

stAwcTsAt(

ptabt

wdecgtmitonfidp

cOiocatBcfio2(saffcciP

lldcailgmclvsosevere credit contraction.

Our paper is also related to Mendoza (2012), who studies thefactors that increased the liquidity risk of debt securities issued

4 J.M. Berrospide, R. Herrerias / Journ

bjeto Limitado, SOFOLES). In this paper we use balance sheetata at the firm level for 68 SOFOLES from 2001 through 2011 totudy the credit crunch that followed the global liquidity shock.OFOLES were the key niche credit providers to mid- and low-ncome households, and their financial downfall has removed anmportant source of housing and construction finance from the

exican economy, which continues to drag down home sales andhe recovery of the housing market five years later.

Shortly after their creation in 1994, SOFOLES became one of theost important providers of mortgage loans in Mexico. As hous-

ng markets expanded considerably, SOFOLES funding began to relyess on government funding received through the federal mortgageompany (Sociedad Hipotecaria Federal, SHF) and more on bankredit and short-term debt from capital markets. Thus, SOFOLESecame typical shadow banks in Mexico, originating long-term

oans funded with significant amounts of short-term debt. In theid-2000s, this fragile funding model left them highly vulnera-

le to a liquidity shock—that is, to the breakdown of their fundingources. When the financial crisis erupted in the fall of 2008, theollapse of interbank and short-term funding markets for globalanks was transmitted internationally and had a severe impact onnancial markets in Mexico. With a fragile funding model, SOFOLESecame the easy prey and unexpected victims of the global liquidityrunch.

In providing support for the liquidity shock hypothesis, our anal-sis considers two main transmission channels through capital andredit markets by which the global liquidity crunch affected theending of SOFOLES. The first and most direct channel became man-fest immediately after the collapse of short-term funding marketsollowing the failure of Lehman Brothers. Credit spreads in Mexico

irrored credit spreads in the U.S. (e.g., the TED spread, the differ-nce in yield between the 3-month LIBOR and the 3-month T-Bill).lso, the abrupt collapse of Asset-Backed Commercial Paper (ABCP)arkets in the U.S. had an immediate counterpart in the crash of

ebt markets in Mexico, which were an important source of short-erm funding for SOFOLES. The second channel was more indirectnd involved the loans that SOFOLES obtained from commercialanks. The global liquidity shock affected the banking sector inexico—dominated by global banks—by reversing capital flows

nd restricting access to debt markets. Facing their own liquidityhock, banks immediately cut back on their lending to SOFOLES,hus creating a contagion channel within the Mexican financialector.

Our study attempts to link the collapse of the SOFOL sectoro the international financial crisis through contagion in financial

arkets. It also serves as a natural experiment to study a par-icular credit crunch within the financial sector of a country. Torovide empirical support to our liquidity shock hypothesis, ourconometric approach considers whether the large contraction inOFOL lending was driven primarily by supply factors. Specifically,e consider the impact on SOFOL loan growth of a liquidity shock

hat takes the form of a severe cutback in their traditional fundingources: commercial bank loans, loans from other organizations,nd public debt markets.

We find that SOFOLES did indeed experience a severe liquidityhock from commercial banks and the securities market that wasriggered by the collapse of Lehman Brothers in the fall of 2008.fter controlling for the potential endogeneity of their funding,e find that the liquidity shock explains 64 percent of SOFOLES’

redit contraction during the recent financial crisis (2008–2009).o further disentangle supply from demand factors, we use regres-

ion estimates to gauge the importance of individual determinants.fter controlling for the large contraction in loan demand during

he financial crisis, our findings suggest that other supply factorssuch as nonperforming loans and lower liquidity buffers) also

oc

inancial Stability 18 (2015) 33–54

layed a significant role. We explicitly account for other factorshat contributed to SOFOLES’ financial problems. Between 2006nd 2008, the sector lost a significant market share to commercialanks as the latter began expanding their mortgage credit businesso previously unattended sectors.

As an alternative to our liquidity shock hypothesis, we explorehether the contraction in SOFOL lending was also the result oferegulation which, in the view of industry observers, ended upxacerbating the risk perception among investors and commer-ial banks. In mid-2006, deregulation arose as an effort by theovernment to ease the financial strains in the SOFOL sector ando help SOFOLES withstand the increased competition from com-

ercial banks. As a result, many SOFOLES turned into deregulatednstitutions that stopped reporting their financial statements tohe supervisory authorities. Deregulation facilitated the entrancef thousands of new market participants into the lending busi-ess, with no single regulatory authority having control over theirnancial activities. Our results, however, suggest that financialeregulation had only a minor role in causing the credit crunch,erhaps by amplifying the impact of the financial crisis.

Our study relates to a large literature on liquidity shocks, creditrunches, and their contagion in different markets and regions.ur findings highlight the importance of spillover effects of liquid-

ty shocks from troubled banks in one region to the banks inther regions through different channels, such as valuation losses,ontractual links between banks, and loss spirals resulting fromsset fire sales. Examples of theoretical work modeling those fea-ures are Allen and Gale (2000), Diamond and Rajan (2005) andrunnermeier and Pedersen (2009), respectively. From an empiri-al perspective, our results also provide evidence consistent withnancial contagion channels such as the panic that took the formf a “run” in short-term funding markets for U.S. banks (Gorton,009; Cornett et al., 2011; Covitz et al., 2013), and for global banksBeltratti and Stulz, 2012). Similarly, our empirical evidence is con-istent with the international transmission of liquidity shocks fromdvanced economies to emerging markets through the lending oforeign banks (Cetorelli and Goldberg, 2011). Moreover, our resultsor Mexico complement the existing evidence on the transmissionhannels through which international liquidity shocks affect bankredit across borders, such as Khwaja and Mian (2008) for banksn Pakistan, Aiyar (2011) for U.K. banks, and Schnabl (2012) foreruvian banks.

Our paper builds on this previous work and contributes to theiterature on credit crunches and financial contagion through theending channel in several ways. First, we use financial micro-levelata for a developing country exposed to international financialontagion, and emphasize contagion channels across countries butlso within the financial sector of a country. Most of the empir-cal literature focuses mainly on the cross-border contagion of aiquidity shock whereas empirical evidence on the effects of conta-ion within a country’s financial sector, like the one in our paper, isore scant.2 Second, we document the main sources of the credit

runch for a particular type of financial institution, which is simi-ar in nature to finance companies in the U.S. but for which there isery little empirical evidence. Third, our results from the decompo-ition of the factors affecting the loan growth of SOFOLES shed lightn how important supply and demand factors are in explaining a

2 Ramcharan et al. (2015) illustrate the effects within the U.S. financial industryf a liquidity shock transmitted from asset-backed securities (ABS) markets to theredit union sector.

al of Financial Stability 18 (2015) 33–54 35

bapgwsail

ltieitSitbaamreHvcoet

pStct

2

tsUthmSBsmbr

assis

ipo

Fl

bape

gbslacmHom

J.M. Berrospide, R. Herrerias / Journ

y SOFOLES that specialized in mortgage lending. Mendoza (2012)rgues that the rating agency measures of credit risk used in thericing of those securities were distorted because they consideredovernment funding in their assessment. Like Mendoza (2012),e also highlight the importance of the liquidity risk in SOFOLES

hort-term borrowing, but we study the whole universe of SOFOLESnd emphasize the significant contraction in their lending result-ng from restricted access to their main funding sources when theiquidity risk materialized.3

The full extent of the consequences of the SOFOL sector’s col-apse is yet to be seen. SOFOLES filled the gap in providing credito households and small businesses created by commercial banksn the aftermath of Mexico’s 1995 financial crisis. Most of the gov-rnment housing programs in the late 1990s and early 2000s, fornstance, were successfully implemented thanks to the existence ofhese nonbank financial intermediaries. At their peak in mid-2005,OFOLES underwrote more than 65 percent of the mortgages orig-nated in the country and granted 20 percent of the total credito households and businesses (about 2 percent of GDP). A num-er of SOFOLES specialized in consumer and commercial creditnd served informal micro- and small businesses by providing autond commercial loans for working capital and equipment—anotherarket left unattended by banks. Our findings suggest too that the

eturn of commercial banks to the housing finance market is a keyxplanatory factor for the SOFOL sector’s severe credit contraction.owever, micro-evidence that considers the type of financing pro-ided to each social sector suggests that, despite their aggressiveomeback, commercial banks have not filled the gap that the exitf SOFOLES left in credit markets. This has an important social andconomic impact as a notable fraction of the population continueso be excluded from the traditional banking sector.

The rest of the paper is organized as follows. The next sectionresents the history and evolution of the SOFOL sector in Mexico.ection 3 describes our methodology and presents the data athe institution level. Section 4 reports and discusses the empiri-al results, Section 5 considers the economic and social impact ofhe exit of SOFOLES, and Section 6 concludes.

. SOFOLES: the rise and demise of an industry

SOFOLES were created in 1994 as part of the NAFTA negotiationso promote the development of nonbank financial intermediariesimilar to finance companies that were already operating in the.S. and Canada. One of the main objectives of the SOFOL sec-

or was to increase competition in the financial sector, which wasistorically dominated by banks, and to allow credit to reach seg-ents of the population unattended by banks. From their creation,

OFOLES were financial intermediaries regulated by the Nationalanking and Securities Commission (CNBV). By the end of 1994, theevere financial and banking crisis in Mexico brought a substantialarket opportunity for the newly created SOFOLES as commercial

anks retrenched from credit markets and focused primarily onenegotiating distressed loans.

SOFOLES operated in good financial shape for several yearsfter their creation. Their success was attributed in part to theirpecialization in niches, with largest and most dominant players

erving the mortgage sector (Table 1). They held several compet-tive advantages over other financial institutions, such as greatertandardization in loan contracts and innovative servicing systems,

3 The literature on Mexican SOFOLES is also limited. When the sector was a dom-nant player in mortgage markets, Pickering (2000) pointed out the new challengesosed by increasing competition in the financial sector and questioned the durationf their privileged position in the mortgage business.

Bas

eaAit

ig. 1. The evolution of SOFOLES lending activity (total loans and growth rate of totaloans) and the evolution of the market share of SOFOLES and commercial banks.

oth of which reduced operating costs. Mortgage SOFOLES hadlso strong links with housing developers and government housingrograms. Subsidized government funds were channeled almostxclusively through them in the early 2000s.

SOFOLES played a key role in the economy by providing mort-ages and consumer credit to low- and middle-income sectors andy supporting government housing policies. As panel A in Fig. 1hows, by 2006 their outstanding loan balance was close to 200 bil-ion pesos (about 2.5 percent of GDP). Panel B in Fig. 1 shows that,t its peak in mid-2005, these institutions were providing 20 per-ent of the total credit to companies and families and underwritingore than 65 percent of the mortgages originated in the country.owever, as economic conditions improved and more profitablepportunities came about in consumer and mortgage loans, com-ercial banks started to regain their market share in these sectors.

y 2005, commercial banks fully returned to the credit businessnd began originating consumer and mortgage loans in marketegments they had previously left unattended.

In an effort to defend their market position, SOFOLES opted toxpand their client base to riskier customers: the self-employednd consumers with low credit scores and no access to bank credit.

s expected, these practices led to loosened credit standards and

nadequate servicing procedures, all of which contributed latero the quality deterioration of their credit portfolios. Eventually,

36 J.M. Berrospide, R. Herrerias / Journal of Financial Stability 18 (2015) 33–54

Table 1SOFOLES, firm characteristics as of 2011.

ID SOFOL Domestic/foreign

Type Sector Status as of 2011 Avg. assetsmill. of pesos

Structureddebt

1 Hipotecaria Su Casita Domestic Independent Mortgage Bankrupt $24,006 Yes2 Ford Credit de Mexico Foreign Industrial group Automobile Deregulated $21,9373 GMAC Mexicana Foreign Industrial group Automobile SOFOL $21,0584 Hipotecaria Nacional Domestic Bank affiliate Mortgage Acquired by BBVA,

Deregulated$17,026

5 Hipotecaria Credito y Casa Domestic Financial group Bridge loans SOFOL $13,986 Yes6 Dexia Credito Local Mexico Foreign Bank affiliate Commercial Closed $12,8217 Credito Inmobilario Foreign Bank affiliate Mortgage Bankrupt $12,170 Yes8 GE Money (General Hipotecaria) Foreign Industrial group Mortgage Closed $10,5049 NR Finance Mexico Foreign Industrial group Automobile Deregulated $8745

10 Patrimonio Domestic Independent Mortgage SOFOL $7784 Yes11 ING Hipotecaria Foreign Financial group Mortgage SOFOL $7451 Yes12 Metrofinanciera Domestic Independent Mortgage Bankrupt $7131 Yes13 Hipotecaria Mexicana Domestic Independent Mortgage Merged to H.

Credito y casa$4531

14 Servicios Financieros Navistar Foreign Financial group Commercial Deregulated $423815 GMAC Financiera Foreign Industrial group Bridge loans Merged to GMAC

Hipotecaria$4072 Yes

16 BNP Paribas Personal Finance(Cetelem)

Foreign Bank affiliate Consumer SOFOL $3981

17 BMW Financial Services de Mexico Foreign Industrial group Automobile Deregulated $348518 Financiera Independencia Domestic Independent Microfinance Deregulated $346919 Fincasa Hipotecaria Domestic Bank affiliate Mortgage Acquired by IXE,

Deregulated$3378 Yes

20 Consupago Domestic Industrial group Consumer SOFOL $322221 Credito Familiar Domestic Independent Microfinance Acquired by

Banamex$2738

22 Hipotecaria Mexico Domestic Independent Mortgage Closed $208123 Financiamiento Azteca Domestic Industrial group Consumer Converted into

bank$1781

24 Hipotecaria Bajio (HipotecariaVanguardia)

Domestic Bank affiliate Mortgage Acquired by Bajio,Deregulated

$1689

25 Hipotecaria Vertice (TerrasHipotecaria)

Domestic Independent Mortgage Deregulated $1651

26 GMAC Hipotecaria Foreign Industrial group Mortgage Merged to GMACFinanciera

$1533

27 Caterpillar Credito Foreign Industrial group Agriculture Deregulated $130928 Hipotecaria Casa Mexicana Domestic Independent Mortgage Converted into

bank$1272

29 Financiera Compartamos Domestic Independent Microfinance Converted intobank

$999

30 Sociedad de Fomento a la Educacion Domestic Independent Student loans SOFOL $97431 BPF Finance Mexico Foreign Industrial group Automobile Deregulated $92732 Finpatria Domestic Independent Mortgage SOFOL $81133 Sociedad Financiera Agropecuaria Domestic Independent Agriculture SOFOL $77934 Agrofinanzas Domestic Industrial group Agriculture SOFOL $61635 Cemex Capital Domestic Industrial group Commercial Converted into

bank$594

36 Sociedad Financiera Associates Domestic Financial group Consumer Acquired byBanamex

$565

37 Credito Familiar (Serv. de CreditoAssociates)

Domestic Financial group Microfinance Acquired byBanamex

$530

38 Financiera Finsol Domestic Independent Microfinance Deregulated $49039 Corporativo Financiero Vimifos Domestic Industrial group Agriculture SOFOL $48540 Creditos Pronegocio Domestic Independent Commercial Merged to Banorte $45541 Operaciones Hipotecarias de Mexico Domestic Independent Mortgage Merged to

Financiera Bajio$444

42 FinTerra Domestic Independent Agriculture Deregulated $40943 Hipotecaria Independiente Domestic Independent Mortgage Deregulated $31944 Corporacion Financiera de Occidente Domestic Independent Commercial SOFOL $31345 Agropecuaria Financiera Domestic Independent Agriculture SOFOL $29646 CNH Servicios Comerciales Foreign Industrial group Commercial Deregulated $28947 Agrofinanciera del Noroeste Domestic Independent Agriculture SOFOL $28848 Finarmex Domestic Independent Commercial Closed $22849 Ficen Domestic Independent Commercial SOFOL $22850 Hir Pyme Domestic Industrial group Commercial SOFOL $22651 Corporacion Hipotecaria Domestic Independent Mortgage SOFOL $20252 Corporacion Financiera America del

NorteForeign Government Commercial Deregulated $185

53 Financiera Educativa de Mexico Domestic Independent Student loans SOFOL $18454 Sociedad Agroindustrial Sofihaa Domestic Industrial group Agriculture Deregulated $17455 Financiera TuEliges Domestic Independent Automobile Closed $17056 Financiera Sumate (Financiera

Mercurio)Domestic Independent Microfinance SOFOL $158

J.M. Berrospide, R. Herrerias / Journal of Financial Stability 18 (2015) 33–54 37

Table 1 (Continued)

ID SOFOL Domestic/foreign

Type Sector Status as of 2011 Avg. assetsmill. of pesos

Structureddebt

57 Fomento Hipotecario Domestic Independent Mortgage Closed $13058 De Lage Landen Agricredit Foreign Bank affiliate Agriculture Deregulated $12959 Monex Financiera Domestic Bank affiliate Commercial Merged to Banco

Monex$114

60 Ixe Sofol Domestic Bank affiliate Consumer Deregulated $9861 Credito Ideal Domestic Independent Consumer Deregulated $8562 Sociedad de Fomento Local Tepeyac Domestic Industrial group Agriculture Deregulated $7663 Financiera Alcanza Domestic Independent Consumer Deregulated $6264 Hipotecaria Associates Domestic Financial group Mortgage Acquired by

Banamex$54

65 Capital Plus Foreign Financial group Consumer Deregulated $4866 Unimex Financiera Domestic Independent Microfinance Deregulated $4167 Sociedad Financiera Equipate Domestic Independent Commercial Deregulated $4168 Prime Capital Domestic Independent Commercial Closed $39

F

amttt

cSyf

ofcceidp

kscccbmoitd2MlSnttMdtd

ipetition from banks. New rules allowed some of the nonbankintermediaries to operate without regulation or supervision. Insti-tutions opting for the new scheme were converted into SOFOMES

ig. 2. SOFOLES asset structure by type of credit and SOFOLES funding structure.

s shown in panel B of Fig. 1, they lost an important part of their

arket share to banks. Panel A of Fig. 2 shows the structure of

heir credit portfolio that accounted for almost 90 percent of theirotal assets. Until 2006, mortgage loans had the largest share ofheir loans (54 percent), followed by consumer (35 percent) and 2

ommercial (11 percent) loans. The average asset maturity forOFOLES specializing in mortgage loans was between 10 and 30ears for residential mortgage loans, and between 2 and 5 yearsor commercial real estate loans (loans to housing developers).4

Panel B of Fig. 2 shows the funding structure of SOFOLES. Duringur period of study, SOFOL funding shifted from subsidized loansrom development banks and government housing funds, espe-ially the federal mortgage company (SHF), to commercial bankredit and short-term debt from capital markets. The share of gov-rnment funding decreased from 56 percent in 2003 to 38 percentn 2010. In contrast, funding from commercial banks and marketebt rose from 24 percent to 33 percent and from 20 percent to 29ercent, respectively, between 2003 and 2010.

Loan securitizations were part of the innovations in the mar-et for housing finance in Mexico during the 2000s. Mortgageecuritizations were concentrated in a small number of finan-ial institutions: eight SOFOLES (see Table 1), the SHF and a fewommercial banks. Between 2003 and 2008, the SOFOL sector’sumulative amount of structured debt (RMBS issuances) was 67.5illion of pesos (about US$ 6 billion). Between 2003 and 2008,arket debt represented about 24 percent of SOFOLES’ funding

n average, one-third of which was short-term and uncollateral-zed debt. Before September 2008, mortgage SOFOLES were ableo continuously sell and renew commercial paper and short-termebt in a very liquid market and at low interest rates. By September008, total outstanding short-term debt used to finance housing inexico was about 65.2 billion pesos. Of that amount, about 20 bil-

ion (31 percent) was short-term debt issued by six of the largestOFOLES, with maturities between 3 and 6 months.5 The busi-ess model of SOFOLES was characterized by significant maturityransformation, originating long-term loans funded with short-erm debt. In practice, SOFOLES became typical shadow banks in

exico. Unlike banks, which rely on more stable funding such aseposits, SOFOLES operated with a very fragile funding structurehat by 2008 had left the sector highly exposed to the volatility ofebt markets and to a liquidity shock.

In 2006, the authorities modified the regulatory frameworkn an effort to help SOFOLES to manage the increased com-

4 See Mendoza (2012).5 See Financial Report on SOFOLES/SOFOMES by Ixe Grupo Financiero, December

009.

38 J.M. Berrospide, R. Herrerias / Journal of Financial Stability 18 (2015) 33–54

Fa

(lmsatld2doiopolii

2

sUsttwIBipaoi

mvp

Fig. 4. The TED spread (difference between 3 month interbank rate and 3 month T-Bill) and the equivalent Mexican credit spread. Shaded areas represent the quarterso

Ua

mmcyedtiaiwntwc

Mopcttat0abgt

ig. 3. The number of regulated and unregulated SOFOLES in Mexico between 2001nd 2011.

multiple purpose financial societies). SOFOMES capital levels,oan loss provisions, and internal processes would no longer be

onitored and they would not be required to report financialtatements to the CNBV unless they issued public debt or wereffiliated with banks or financial groups. Starting in 2006, 37 ofhe SOFOLES in our sample (Table 1) began to become deregu-ated by converting into SOFOMES. More than half of these wereeregulated as of 2007, including four of the largest ones. By008, only 30 SOFOLES retained their charter (Fig. 3). Financialeregulation had a number of adverse consequences in the viewf industry experts. First, all related parties—namely authorities,nvestors, and creditors—failed to monitor SOFOM operations andverlooked the lack of transparency and lower quality of theirortfolios. Second, market competition was distorted as hundredsf new players gained access to the lending business. The pro-iferation of unregulated institutions contributed to increasingnformation asymmetries in credit markets and thus to the rise innvestors’ risk perception.6

.1. Financial contagion

Until June 2008, Mexico seemed untouched by the breakdown ofecuritization markets and the failure of financial institutions in the.S. Although SOFOLES were struggling to maintain their market

hare, the sector was still profitable and credit portfolios continuedo grow. Their business model was relatively simple: they fundedheir mortgage, commercial, and consumer loans using bank credit,hile the largest institutions were also able to issue debt securities.

n the days following the announcement of the failure of Lehmanrothers, liquidity dried up and the Mexican peso devalued signif-

cantly in response to the expected capital outflows (the Mexicaneso was devaluated by nearly 40 percent in a 6-month period). We

rgue that the liquidity shock hitting SOFOLES, which took the formf restricted access to debt and credit markets, was the result of thenternational contagion from the subprime mortgage crisis in the

6 The Mexican Association of Finance Companies (AMFE) estimates that, to date,ore than 4000 unregulated entities operate in Mexico. In their opinion, regulators

iew these nonbank institutions as not “systemically risky” due to their “limited”articipation in the formal payment system. See AMFE, Plan Estrategico 2020.

bBefltplD

f the financial crisis.

.S., and thus was relatively exogenous to many of the SOFOLES’ctions.

Financial contagion became manifest in Mexico through twoain channels. The first and most direct channel was through debtarkets. Fig. 4 shows that a measure of the severity of the finan-

ial crisis in the U.S., such as the TED spread (the difference inield between a 3-month LIBOR and 3-month Treasury bill) and thequivalent Mexican Treasury-interbank spread, moved in tandemuring the turmoil (shaded area). This fact corroborates the theoryhat debt markets served as a direct channel of contagion. As seenn Fig. 4, both the TED spread and the Mexican credit spread show

remarkable spike in October 2008. SOFOLES were the first debtssuers to be hit by the collapse of debt markets. Fig. 5 confirms that

hen the rate spreads widened, SOFOL bond issuance decreasedotably (Panel A) and their structured debt (RMBS) issuances cameo a complete halt (Panel B). The effect on Mexican debt marketsas very similar to the collapse of the worldwide asset-backed

ommercial paper (ABCP) and interbank markets.The second and more indirect channel of contagion was through

exican commercial banks, most of which are internationallywned. Banks struggled with their own liquidity shock during theeak of the international financial crisis and were forced to cut theirredit supply, including loans to SOFOLES. Fig. 6 depicts the contrac-ion in the funding sources of commercial banks that resulted fromhe global liquidity crunch. Measured as a percentage of their totalssets, from the second half of 2008 to the end of 2009 foreign loanso commercial banks dropped from about 1.2 percent to less than.5 percent (Panel A). Similarly, the percentage of deposits to totalssets dropped more than 10 percent (Panel B). Foreign-ownedanks in Mexico played a significant role in the transmission of thelobal liquidity shock. Before the financial crisis, about 70 percent ofhe commercial bank loans to SOFOLES came from foreign-ownedanks (mainly from Banamex, owned by Citigroup; HSBC Mexico;BVA Bancomer; Santander; and Scotiabank). As shown in Pan-ls C and D, due to significant liquidity pressures during the crisis,oreign-owned banks cut their total lending (Panel C) and hoardediquid assets more than domestic banks. The ratio of their liquid tootal assets increased from 10 percent in mid-2007 to more than 16ercent in December 2008, while domestic banks increased their

iquid to total assets ratio from 7.5 percent to 11.5 percent (Panel).

J.M. Berrospide, R. Herrerias / Journal of F

2

lmmnBhucimbcicdt

fsoS

tifts

awpdttc

3

chmcustmb

�

tt�iafivta(f

stakSzmthis purpose. We manually collected detailed information on eachof the RMBS issuances for the eight SOFOLES that securitized their

Fig. 5. Annual amounts of debt and structured debt issuances by SOFOLES.

.2. The demise of the industry

In the middle of the financial storm, Metrofinanciera, the thirdargest SOFOL, was severely hit by fraud allegations.7 Widespread

istrust of the SOFOL industry quickly spread out through financialarkets and precipitated a large fund withdrawal. Their busi-

ess model and their maturity gap were no longer sustainable.y early 2009, nonperforming loan (NPL) ratios on mortgage loansit double digits and loan demand dropped as a result of risingnemployment rates and a severe recession.8 Given the severity ofonditions, government authorities, banks, and SOFOLES enterednto agreements to restructure the liabilities of institutions in the

ortgage sector. The government provided credit enhancementsy guaranteeing 65 percent of new debt issuances and offeredredit lines to SOFOLES to help them alleviate immediate liquid-ty needs. Between 2009 and 2010, SOFOLES were facing multiple

hallenges: a large and sudden drop in external funding, risingelinquencies in their loan portfolios (particularly in the construc-ion pools), small asset valuations, and weak capital levels. Despite

7 Highly reliant on new RMBS issuances, the company hoped that the misuse ofunds would not be revealed. However, those issuances came to a halt as the cri-is erupted and fraud was put in evidence. See Factiva, Housing Pessimists Chewn Metrofinanciera; Company puts figure on money owed to the trusts,” Assetecuritization Report, December 2008.8 See Herrerias (2011).

lt

aoit

w

inancial Stability 18 (2015) 33–54 39

he government aid, a vicious circle emerged directing the sector tots collapse: the deterioration of loans led to decreasing income andurther needs for loan loss reserves; at the same time, risk percep-ion abruptly jumped and, as a result, funding was cut even moreeverely and borrowing costs became almost unaffordable.

During the following months, several SOFOLES merged or werecquired by financial groups to be able to continue their operations,hile others closed because of funding cuts from their parent com-anies (see Table 1). With the global liquidity crunch disruptingebt and credit markets in Mexico—that is, without enough short-erm funds—SOFOLES were unable to support daily operations ando originate new loans. Only a few SOFOLES managed to survive therisis.

. Methodology and data

Based on the two transmission channels of the global liquidityrisis discussed above, in this section we test our liquidity shockypothesis by looking more closely at how the breakdown of theain SOFOL funding sources (debt and bank credit) led to a severe

ontraction in their lending. We collect information at the firm levelsing quarterly data from 2001 through 2011 for 68 SOFOLES. Theample period includes both a pre-crisis period (2001–2006) andhe financial crisis in Mexico (2008–2009). Using panel data esti-

ation, we test our liquidity shock hypothesis using the followingaseline regression:

log(loani,t) = ˛1time + ˛2firm + ˇ1�Bank loani,t−1

+ ˇ2�Debti,t−1 + �1�Bank loani,t−1 × Crisis

+ �2�Debti,t−1 × Crisis + Xi,t−1 + εi,t (1)

In this specification, our dependent variable, �log(loani,t), ishe quarterly growth rate of total loans of SOFOL i during quarter. Among our explanatory variables, included with lagged values,

Bank loani,t is the percentage change in bank loans, and �Debti,ts the percentage change in debt securities. These are the main vari-bles of interest and they measure the funding sources available tonance SOFOL loans. We control for firm-specific characteristics inector Xi,t−1, including size (log of total assets), capital (equity capi-al/total assets), holdings of liquid assets (liquid assets/total assets),

proxy for loan quality (nonperforming loans/total loans), profitsreturn on equity/average assets), and dummy variables to controlor foreign ownership or bank ownership.

We also include proxies for RMBS issuances associated with loanecuritizations. Although RMBS issuance is not a funding source forhe loans in our dependent variable, �log(loani,t), we include it as

control in order to capture a possible substitution between loansept on books and those securitized into RMBS.9 As discussed inection 2, at their peak in 2007 SOFOLES were using loan securiti-ations as an important off-balance-sheet funding source for theirortgage loans and were competing with other large banks for

oans into RMBS in Mexico between 2002 and 2008 from footnoteso their balance sheets in CNBV filings available at Capital IQ.10 We

9 Accounting practices in Mexico treat loan securitizations as loan sales, whichre off-balance-sheet transactions. Our dependent variable measures the growthf on-balance-sheet loans only and excludes securitized loans. Similarly, the RMBSssuance associated with those loan securitizations is an off-balance-sheet debt andhus is not included in our debt variable.10 In almost all cases, the information on the loan portfolio backing those securitiesas not disclosed. Only a couple of institutions disclosed information on securitized

40 J.M. Berrospide, R. Herrerias / Journal of Financial Stability 18 (2015) 33–54

F l banka

S

iqi

fmsstitama

lftsgm

c9

toaSWttd

ig. 6. The evolution of foreign interbank liabilities, and deposits of all commerciand 2011. Shaded areas represent the quarters of the financial crisis.

ource: Bank of Mexico.

nclude as controls both a dummy variable—Structured Debt—foruarters in which an issuance occurs and the size of the RMBS

ssuance scaled by assets.Our specification also includes time dummies, time, to account

or time effects in lending, such as seasonal effects or any otheracroeconomic changes that affect all institutions equally and

imultaneously; and firm-fixed effects, firm, to account for unob-erved heterogeneity that is constant over time and correlated withhe other independent variables in the model. In alternative spec-fications, we also consider macroeconomic aggregates, such ashe market share of commercial banks in total credit markets, to

ccount for the impact of higher competition from banks in creditarkets, and two proxies for loan demand: GDP growth to capture

ggregate economic conditions, and the interbank interest rate to

oans in the footnotes to their balance sheets, and only for a few quarters. In thoseew instances, the amount of RMBS issued closely matched the loan portfolios sold tohe securitization conduits. Carballo-Huerta and González-Ibarra (2008) argue thatecuritized mortgage loans represented a small portion of the outstanding mort-age credit portfolio (7.4 percent in June 2008), concluding that the Mexican RMBSarket remained small-scale.

iesd

piW

s in Mexico, and loans and liquid assets of banks funding SOFOLES, between 2002

ontrol for borrowing costs.11 We winsorize the data at the 1st and9th percentiles to deal with potential outliers.

According to our hypothesis of a liquidity shock leading to a con-raction in lending, we expect positive and significant coefficientsn SOFOL funding sources (ˇ1 > 0 and ˇ2 > 0). Intuitively, andccording to a simple business model, more funding allowsOFOLES to originate more loans and sustain higher loan growth.e are interested in studying the effects of the financial crisis on

he relationship between loan growth and funding growth. We cap-ure this situation by the two interaction terms between a crisisummy (which takes the value of 1 after the Lehman failure—that

s, from 2008:Q4 through 2009:Q4—and 0 otherwise) and our rel-

vant funding variables. We expect �1 and �2 to be statisticallyignificant if the sensitivity of lending to funding increases orecreases during the crisis. We also hypothesize that institutions

11 Footnotes to the balance sheets of SOFOLES (CNBV filings) show that the mainricing determinant for loans and debt contracts, and thus a good proxy for borrow-

ng costs, is the short-term interbank rate published by the central bank of Mexico.e use the 28-day and the 91-day interbank rates in our regression analysis.

al of F

wqiecciwrfi

bilattlibefhgc

3

pcottBStfcp

pts0ntm

oottNtmtPs

lc

sr

vwtsmlS

4

4

erWmatidvTspthrmicl

ocAaasrlagca

anttthe loan growth registered on the balance sheets of SOFOLES thatsecuritized their loans into RMBS was between 6 and 7 percentagepoints lower than the loan growth of their counterparts that did

14

J.M. Berrospide, R. Herrerias / Journ

ith more liquid assets and equity capital or with better credituality may be in better shape to withstand the effects of the liquid-

ty shock. Similarly, SOFOLES owned by foreign firms or banks arexpected to lend more than their national- and nonbank-ownedounterparts, as relationships with a bank or a foreign firm wouldreate close ties and internal capital arrangements. Finally, liquid-ty restrictions in the form of increased costs of funding, together

ith loose lending policies and decreased market share, may alsoesult in severe financial distress and a general contraction of theirnancial activities.

We implement additional estimations using annual dataetween 2003 and 2011. For that purpose, we obtained detailed

nformation on the share of commercial and development bankoans and the government housing funds from footnotes to thennual financial reports of about 29 SOFOLES available at Capi-al IQ and the CNBV.12 The analysis using annual data allows uso address several challenges in the estimation because the sameevel of detail is not available in the quarterly data. In particular, thempact attributed to the liquidity shock through bank loans maye understated using quarterly data, as this variable includes gov-rnment loans extended through development banks and housingunds. With this information, we are in a better position to assessow commercial bank funding pressures that resulted from thelobal financial crisis were transmitted to the SOFOL sector throughommercial bank loans.

.1. Summary statistics

Table 2 presents some descriptive statistics for the entire sam-le, divided into three subsamples: pre-crisis (2001:Q1–2008:Q3),risis (2008:Q4–2009:Q4), and post-crisis (2010:Q1–2011:Q3). Webserve that loan growth slows down for the median SOFOL fromhe pre-crisis period (8 percent) to the crisis period (0 percent) andhen declines during the post-crisis period (negative 1.5 percent).ank loan growth follows a very similar pattern, suggesting thatOFOLES did use bank loans as the main funding source to financeheir loans. Bank loan growth slows down for the median entityrom the pre-crisis period (8 percent) to the crisis period (0 per-ent) and then declines during the post-crisis period (negative 1.2ercent).

Table 2 also shows that during the financial crisis in Mexico, therofitability of the SOFOL sector dropped as the credit quality ofheir loan portfolio deteriorated dramatically. The median SOFOLaw a drop in ROA from 1.9 percent in the pre-crisis period to about.3 percent in the crisis period, while at the same time their ratio ofonperforming loans to total loans increased significantly duringhe crisis (from 1 percent to 5 percent) and post-crisis periods (the

edian ratio reached 7 percent in 2011).We also include summary statistics of the variables used in

ur regression analysis with quarterly data in Table 3. Of a totalf 68 institutions, our dataset covers about 40 SOFOLES per quar-er, on average. Note that the loan growth rate of the average andhe median SOFOL closely follows the growth rate of bank loans.ote also that debt growth is 2.5 for the mean SOFOL but is 0 for

he median SOFOL, which reflects the fact that debt from capitalarkets was an important funding source for only the largest insti-

utions (26 of 68 SOFOLES issued debt between 2001 and 2011).anel B of Table 3 also reports the correlation matrix for the regres-ion variables. The size of pairwise correlations looks relatively

12 Detailed information on the breakdown of funding is only available for theargest public SOFOLES, which have to file annual reports to the CNBV as theyontinue issuing debt or equity securities.

n

an

et

inancial Stability 18 (2015) 33–54 41

mall, which alleviates potential multicollinearity issues in ouregression specification.

Finally, Panel C of Table 3 presents summary statistics of ourariables using annual data. Given the lower frequency of the data,hich are available for 29 SOFOLES only, the number of observa-

ions ranges between 155 and 196. Consistent with the quarterlytatistics, the annual loan growth rates of the average and theedian SOFOL follow the annual growth rate of commercial bank

oans. Median annual debt growth rate is zero because severalOFOLES did not issue debt securities.

. Econometric results

.1. Results with quarterly data

Table 4 presents our basic regression results using six differ-nt specifications of Eq. (1): columns (1) and (2) are pooled OLSegressions, while columns (3) and (4) are fixed-effect regressions.

e include columns (5) and (6) for robustness purposes. Further-ore, columns (2), (4), (5) and (6) include quarterly dummies. In

ll specifications, all firm-level variables enter the regression withhe expected sign and almost all are significant. More importantly,n all specifications we find that the coefficients of bank loans andebt, our core variables of interest, are positive and significant, pro-iding empirical support for our main liquidity shock hypothesis.he coefficient on bank loans in our OLS models is about 0.17 anduggests that a 1 standard deviation reduction in bank loans (a 30ercentage point decrease) leads to a 5 percentage point contrac-ion in the loan growth rate. Also consistent with our liquidity shockypothesis, the effect of debt growth on the quarterly loan growthate is positive and statistically significant, though it is smaller inagnitude. The coefficient of 0.057 in our second OLS specification

mplies that a 1 standard deviation decline in debt growth (18 per-entage point decrease) leads to a 1 percentage point reduction inoan growth.

It is worth noting that the coefficient on the crisis dummy inur OLS regression, column (1), is negative and significant, indi-ating that loan growth was negative during the crisis period.lthough the interaction terms of crisis and the funding variablesre negative—suggesting that the correlation between loan growthnd funding was reduced during the crisis—they are not statisticallyignificant.13 This result seems to indicate that the strong positiveelationship between loan growth and funding (particularly bankoans) prevailed during the crisis. The bank-owned dummy vari-ble is positive and statistically significant, suggesting that the loanrowth of SOFOLES with a bank parent was between 3 and 4 per-entage points greater than the loan growth of SOFOLES with noffiliation to a commercial bank.

RMBS issuance measured by our Structured Debt dummy is neg-tive and significant in all specifications. We interpret the stronglyegative coefficient as reflecting the substitution between loanshat SOFOLES originated and kept on their books and loans thathey sold to securitization conduits. Thus, this result suggests that

ot securitize their loans.

13 An F-test for joint significance of the change in funding (debt and bank loan)nd their interaction with our Crisis dummy shows that the total combined effect isot statistically significant either (p-values are 0.34 and 0.82, respectively).14 A potential omitted-variable problem in our estimation could result from thexclusion of RMBS issuances from our baseline regression if the issuance of struc-ured debt is correlated with unobservable firm characteristics such as lending

42 J.M. Berrospide, R. Herrerias / Journal of Financial Stability 18 (2015) 33–54

Table 2SOFOLES sample summary statistics, 2001–2011 (percentages).

Mean Median Min Max Std. dev.

Pre-crisis (2001:Q1–2008:Q3)Loans/assets 82.17 90.61 0.00 200.63 25.35Liquid assets/assets 13.67 5.49 −2.39 99.70 21.91Net income/assets (ROA) 1.44 1.88 −52.25 33.44 11.20Non-performing loans/loans 4.22 1.39 0.00 59.58 8.15Equity capital/assets 27.42 15.47 0.41 103.05 27.67Loan growth 14.22 7.72 −46.93 150.61 28.97Bank loan growth 13.10 7.67 −63.62 170.39 31.73Debt growth 3.56 0.00 −81.35 126.82 18.02RMBS issuance/assets 0.40 0.00 0.00 29.70 22.34

Crisis (2008:Q4–2009:Q4)Loans/assets 80.68 85.63 5.00 106.73 15.86Liquid assets/assets 9.13 5.93 0.09 51.82 9.27Net income/assets (ROA) −1.02 0.33 −52.25 26.68 12.15Non-performing loans/loans 9.57 4.69 0.00 65.09 13.52Equity capital/assets 17.82 14.54 −3.42 68.31 13.33Loan growth 0.67 −0.01 −46.93 82.81 16.41Bank loan growth 0.24 0.32 −63.62 111.46 22.46Debt growth −3.46 0.00 −81.35 44.44 15.83RMBS issuance/assets 0.00 0.00 0.00 4.00 0.30

Post-crisis (2010:Q1–2011:Q3)Loans/assets 78.83 84.69 28.90 147.39 18.89Liquid assets/assets 8.97 4.99 0.14 63.58 11.63Net income/assets (ROA) −1.55 0.22 −52.25 28.16 10.14Non-performing loans/loans 15.45 7.33 0.00 95.05 18.75Equity capital/assets 19.67 14.80 −1.48 75.79 14.71Loan growth −1.72 −1.47 −46.93 36.45 11.92Bank loan growth −2.07 −1.23 −63.62 77.52 18.33Debt growth −2.24 0.00 −81.35 126.82 19.80RMBS issuance/assets 0.00 0.00 0.00 0.00 0.00

Full sample (2001:Q1–2011:Q3)Loans/assets 81.64 89.53 0.00 200.63 23.94Liquid assets/assets 12.69 5.55 −2.39 99.70 20.15Net income/assets (ROA) 0.84 1.57 −52.25 33.44 11.22Non-performing loans/loans 5.96 2.10 0.00 95.05 11.06Equity capital/assets 25.61 15.30 −3.42 103.05 25.66Loan growth 10.91 5.68 −46.93 150.61 27.12

uoatgtaiOnbc

ncbsoT

swRa

WdstfiIbbbfcrp

So far, the impact of the liquidity shock on loan growth isstatistically significant but seems to be economically modest.An explanation for this result is that our measure of the funding

Bank loan growth 9.77 4.97Debt growth 2.17 0.00RMBS issuance/assets 0.30 0.00

To account for the possibility of omitted variables resulting fromnobserved firm characteristics, we exploit the panel structure ofur data using a firm-fixed-effect model, as shown in columns (3)nd (4) of Table 4. The coefficient of 0.09 on bank loans is about halfhe coefficient in the OLS regression, suggesting that firm hetero-eneity in the form of unobserved factors contribute significantlyo our OLS estimates. The coefficient on Debt Growth is also smallernd indicates that the 1 standard deviation decline in debt growthmplies a reduction in loan growth of less than 1 percentage point.ther firm characteristics, such as the liquid assets ratio and theonperforming loans ratio, are key determinants of the loan growthehavior. We do not find any significant role for the ratio of equityapital to total assets.

Our estimates in Table 4 control for firm heterogeneity but doot necessarily capture the role of loan demand. To alleviate thisoncern we include the quarterly growth of GDP and the inter-ank rate in our regression analysis. We also include the market

hare of banks in credit markets to control for their competition. Allf these macro-variables enter the regression with a lagged value.able 5 presents these results for the same specifications as before.

tandards, which may in turn be correlated with our funding variables. To dealith this issue, we run our baseline regression excluding the 7 SOFOLES that issuedMBS. Our results are unchanged; our two funding variables, �Bankloanand �Debt,re positive and statistically significant, though their coefficients are slightly lower.

tmuabGab

−63.62 170.39 30.03−81.35 126.82 18.23

0.00 29.70 1.99

e drop the time (quarterly) dummies and include four quarterummies to control for seasonal effects instead. Most variables areignificant and enter the regression with the expected sign, andhe coefficients on our proxies for the liquidity shock and the otherrm controls are relatively similar in magnitude to those in Table 4.

n our OLS specifications (columns (1) and (2)), the interactionetween the crisis dummy and our funding variables is negativeut insignificant, as in Table 4. GDP growth and the commercialank market share seem relevant and suggest an important roleor loan demand and market competition.15 Our estimates indi-ate that a 1 standard deviation reduction in the quarterly growthate of GDP (a 3 percentage point decrease) leads to a 2 percentageoint decline in loan growth.

15 In unreported regressions, we also account for changes in loan demand usinghe changes in the unemployment rate, changes in sectoral GDP (construction and

anufacturing), and a consumer confidence index in Mexico. None of these meas-res are statistically significant. Other proxies for loan demand such as changes inverage real income among middle- and low-income groups in Mexico, which maye more appropriate for SOFOLES given their customer profiles, are unavailable. OurDP growth measure is a broad and crude proxy for demand, and thus our estimatesimed at controlling for changes in loan demand should be interpreted as an upperound.

J.M.

Berrospide, R

. H

errerias /

Journal of

Financial Stability

18 (2015)

33–54

43Table 3Summary statistics for the regressions analysis, quarterly (2001–2011) and annual data (2003–2011).

Variable Obs. Mean Median Std. dev. Min Max

Panel A. Quarterly dataLoan growth 1531 10.91 5.68 27.12 −46.93 150.61Log (total assets) 1563 13.79 13.73 1.94 10.11 17.52Non-performing loans/loans 1508 5.70 2.01 10.55 0.00 95.05Net income/assets 1457 0.92 1.57 11.04 −52.25 33.44Liquid assets/assets 1563 12.71 5.58 20.20 0.00 99.00Equity capital/assets 1562 25.56 15.24 25.65 0.00 99.00Foreign dummy 1638 0.29 0.00 0.45 0.00 1.00Owned bank dummy 1638 0.25 0.00 0.44 0.00 1.00Bank loan growth 1372 10.35 5.29 30.26 −63.62 170.39Debt growth 1451 2.50 0.00 18.30 −81.35 126.82Structured debt dummy 1638 0.03 0.00 0.17 0.00 1.00Crisis dummy 1638 0.09 0.00 0.29 0.00 1.00GDP growth 1508 0.63 1.38 3.30 −10.41 5.81Banks market share 1605 81.23 80.18 6.03 72.37 91.35Interbank rate 1605 7.91 7.71 2.31 4.85 18.01

Variable Loan Grw. Log (tot.assets)

Non-perf.loans/loans

Netincome/assets

Liquidassets/assets

Equitycapital/assets

For.dummy

Ownedbank Dum.

Bank loangrowth

Debtgrowth

Struct. debtdummy

CrisisDum.

GDP Grw. BanksMkt.share

Interb. rate

Panel B. Correlation matrix, quarterly dataLoan growth 1Log (total assets) −0.29 1Non-perf. loans/loans −0.27 −0.03 1Net income/assets −0.10 0.20 −0.27 1Liquid assets/assets 0.24 −0.32 0.04 −0.07 1Equity capital/assets 0.21 −0.55 −0.06 0.05 0.29 1Foreign dummy 0.00 0.38 −0.11 −0.08 −0.12 −0.29 1Owned bank dummy 0.03 0.32 −0.10 0.05 −0.19 −0.14 0.26 1Bank loan growth 0.38 −0.27 −0.23 −0.10 0.08 0.11 −0.02 0.01 1Debt growth 0.02 0.12 −0.08 0.05 −0.05 −0.11 −0.01 0.01 −0.10 1Structured debt dum. −0.07 0.25 −0.05 0.02 −0.03 −0.16 −0.03 0.02 −0.05 0.15 1Crisis dummy −0.12 0.05 0.10 −0.05 0.02 −0.02 −0.04 −0.04 −0.09 −0.10 −0.06 1GDP growth 0.06 −0.01 −0.01 −0.01 0.01 0.02 0.00 0.01 0.07 0.04 −0.01 −0.19 1Banks market share −0.23 0.05 0.24 −0.15 0.07 0.05 −0.09 −0.10 −0.16 −0.11 −0.06 0.43 −0.03 1Interbank rate 0.15 −0.09 −0.20 0.07 −0.02 0.03 0.06 0.05 0.16 0.11 0.01 −0.06 −0.08 −0.35 1

Variable Obs. Mean Median Std. dev. Min Max

Panel C. Annual dataLoan growth 187 20.25 13.09 46.09 −65.53 172.61Log (total assets) 187 14.79 15.00 1.72 10.44 17.51Non-performing loans/loans 187 5.46 2.50 9.34 0.00 56.78Net income/assets 187 0.70 1.21 6.39 −41.36 15.32Liquid assets/assets 186 6.78 4.90 9.96 0.10 88.03Equity capital/assets 187 17.27 11.57 15.16 0.00 99.49Foreign dummy 196 0.32 0.00 0.47 0.00 1.00Owned bank dummy 196 0.33 0.00 0.47 0.00 1.00Commercial bank loan growth 174 24.82 15.75 45.49 −66.94 165.48Debt growth 155 16.29 0.00 44.30 −64.88 138.63Structured debt dummy 196 0.12 0.00 0.33 0.00 1.00Crisis dummy 196 0.19 0.00 0.39 0.00 1.00GDP growth 196 2.39 3.56 2.15 −1.94 4.53Banks market share 196 80.66 78.72 6.41 73.00 90.42Interbank rate 196 7.42 7.51 1.27 4.91 9.61

44 J.M. Berrospide, R. Herrerias / Journal of Financial Stability 18 (2015) 33–54

Table 4SOFOLES, loan growth regression. This table reports estimates from regressions using our benchmark specification described by Eq. (1) and using quarterly data from 2001:Q1to 2011:Q3. The dependent variable is the quarterly loan growth for each SOFOL in the sample. All the non-dummy explanatory variables enter the regression with a laggedvalue. Nonperforming loans is the ratio of loans past due 90 days and non-accrual to total loans. Foreign and Bank-owned dummies are time-invariant dummy variables thattake the value of 1 for foreign SOFOLES and those SOFOLES affiliated to a bank or a financial group in Table 1, respectively. The crisis dummy takes the value of 1 between2008:Q4 and 2009:Q4. Structured debt dummy takes the value of 1 for SOFOLES that securitized credit portfolios in the debt market (RMBS). Robust standard errors arereported in brackets.

OLS Fixed effect Fixed effects and post-crisisstructural break dummy

(1) (2) (3) (4) (5) (6)

Size (log of total assets) −1.740*** −1.288** −8.587*** −8.322*** −10.390*** −10.315***

[0.575] [0.575] [0.888] [1.642] [2.132] [2.128]Non-performing loans −41.604*** −37.743*** −42.039*** −38.256*** −40.310*** −40.712***

[5.953] [6.037] [6.343] [6.704] [7.452] [7.503]Net income/total assets −0.197** −0.242*** −0.180** −0.176** −0.186** −0.183**

[0.079] [0.084] [0.088] [0.088] [0.091] [0.091]Liquid assets/total assets 32.018*** 33.559*** 26.070*** 28.238*** 26.896*** 27.184***

[8.098] [8.195] [8.920] [8.903] [9.828] [9.832]Equity/total assets 6.596 9.01 14.261 13.559 10.193 9.94

[6.929] [6.642] [9.744] [10.575] [11.869] [11.865]Foreign dummy 1.76 0.982

[1.488] [1.612]Bank-owned dummy 3.904** 3.230*

[1.703] [1.754]Growth in bank loans 0.177*** 0.171*** 0.089*** 0.091*** 0.070** 0.075**

[0.035] [0.037] [0.030] [0.031] [0.031] [0.032]Growth in debt 0.066*** 0.057*** 0.037* 0.042** 0.041* 0.045**

[0.016] [0.020] [0.021] [0.021] [0.021] [0.022]Structured debt dummy −4.394** −5.832*** −5.389*** −6.892*** −7.093*** −7.068***

[2.010] [2.106] [1.632] [1.768] [1.834] [1.843]Crisis −4.615*** −1.547

[1.664] [1.552]Growth in bank loans × crisis −0.098 −0.1 0.009 −0.006 −0.074

[0.071] [0.072] [0.103] [0.100] [0.108]Growth in debt × crisis −0.049 −0.046 −0.01 −0.033 −0.062

[0.051] [0.051] [0.046] [0.051] [0.058]Constant 31.199*** 15.700* 128.639*** 124.837*** 165.049*** 164.016***

[8.656] [9.302] [13.007] [25.672] [34.096] [34.026]

Observations 1303 1303 1303 1303 1303 1303R-squared 0.3 0.34 0.42 0.44 0.46 0.46Quarterly time effects No Yes No Yes Yes Yes

* Significance at the 10%.

cbtt

ffitoacautTS

SAorsd

lisTic

epcltc

** Significance at the 5%.*** Significance at the 1%.

ontraction nets out the effect of the severe decline in commercialank loans because it includes government funding to SOFOLEShrough loans from development banks. We isolate the effect onhe government funding using annual data in Section 4.2.3.

SOFOLES specialized in different market segments. To controlor different types of loans, Table 6 presents the results usingxed-effect regressions and includes macroeconomic controls forhree different types of SOFOLES (columns 1 through 3) dependingn whether they specialize in commercial loans (e.g., equipmentnd working capital), mortgage loans, or consumer loans (e.g.,redit card and auto loans).16 The time-invariant dummy vari-bles for Foreign and Bank Ownership are excluded because wese fixed-effect estimations. The coefficients on our measures for

he liquidity shock are significant only for the mortgage SOFOLES.he coefficient on Bank loans for commercial and consumerOFOLES is positive but insignificant, suggesting weak evidence of a

16 We grouped SOFOLES by type based on their loan share. For example, mortgageOFOLES are those with more than 50 percent of their loan share in mortgage loans.s a robustness exercise, we also use loan growth regressions for different typesf loans (commercial, real estate, and consumer) and obtained qualitatively similaresults, though in those cases the number of observations becomes very small forome loan categories. For those reasons we report the results using SOFOLES typeefined by loan shares.

p

stsifSaet

iquidity shock for those entities. Similarly, the coefficient on Debts significant only for mortgage SOFOLES. The interbank rate is alsoignificant for these institutions only, though its effect seems small.ogether, these results suggest that institutions that specializedn mortgage lending were the ones most affected by the severeutback in bank funding.

As before, size and the ratio of nonperforming loans are keyxplanatory variables of loan growth across all specifications. Non-erforming loans are strongly negative for both mortgage andonsumer SOFOLES, though they have the largest impact for theatter type. Both commercial and mortgage SOFOLES that securi-ized their loans into RMBS had a lower loan growth than theirounterparts that did not (11 percentage points and 4 percentageoints, respectively).

Overall, our findings provide support for our hypothesis anduggest a role for the liquidity shock in explaining the sharp con-raction of SOFOL lending during the financial crisis in Mexico. Thiseems to be the case mainly for those institutions that specializedn mortgage loans. The ratio of liquid assets to total assets is positiveor all three segments; however it is only significant for commercial

OFOLES indicating that liquidity buffers left them in better shapes compared to the other SOFOLES. Finally, demand factors are rel-vant only for commercial SOFOLES, for which the coefficient onhe GDP growth rate is positive and significant.

J.M. Berrospide, R. Herrerias / Journal of Financial Stability 18 (2015) 33–54 45

Table 5SOFOLES, loan growth regression with macro-variables. This table reports estimates from regressions using our benchmark specification described by Eq. (1) augmented bymacroeconomic variables (real GDP growth, the share of commercial banks in credit markets, and the interbank rate). The dependent variable is the quarterly loan growthfor each SOFOL in the sample. All the non-dummy explanatory variables enter the regression with a lagged value. Nonperforming loans is the ratio of loans past due 90days and non-accrual to total loans. Foreign and bank-owned dummies are time-invariant dummy variables that take the value of 1 for foreign SOFOLES and those SOFOLESaffiliated to a bank or a financial group in Table 1, respectively. The crisis dummy takes the value of 1 between 2008:Q4 and 2009:Q4. Structured debt dummy takes the valueof 1 for SOFOLES that securitized credit portfolios in the debt market (RMBS). Spread is the TED spread measured as the difference in yields between the 3-month LIBOR andthe 3-month Treasury bill. Robust standard errors are reported in brackets.

TED Spread and allowance for loan losses

OLS Fixed effect FE and Post-crisisstruct. breakdummy

OLS Fixed effect

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Size (log of total assets) −1.424** −1.409** −7.795*** −7.827*** −9.28*** −1.203** −1.755*** −1.333** −8.24***

[0.581] [0.585] [1.299] [1.296] [1.533] [0.591] [0.575] [0.591] [1.304]Non-performing loans −37.98*** −37.93*** −38.85*** −38.88*** −39.80***

[6.098] [6.038] [6.586] [6.598] [7.020]Net Income/total assets −0.238*** −0.234*** −0.179** −0.178** −0.192** −0.190** −0.132 −0.181** −0.152*

[0.079] [0.079] [0.088] [0.088] [0.091] [0.087] [0.085] [0.083] [0.087]Liquid assets/total assets 33.71*** 33.65*** 27.66*** 27.65*** 26.20*** 32.10*** 30.037*** 32.16*** 28.31***

[8.107] [8.096] [8.894] [8.906] [9.859] [8.653] [8.555] [8.585] [8.832]Equity/total assets 8.735 8.56 15.476 15.505 13.581 13.165* 11.301 12.859* 15.431

[6.459] [6.475] [9.880] [9.923] [11.018] [6.840] [7.130] [6.720] [9.961]Foreign dummy 1.183 1.18 2.137 3.317** 2.304

[1.586] [1.587] [1.680] [1.578] [1.649]Bank-owned dummy 3.427* 3.414* 3.421* 4.146** 3.617**

[1.721] [1.725] [1.821] [1.790] [1.796]Growth in bank loans 0.162*** 0.168*** 0.089*** 0.088*** 0.074** 0.201*** 0.220*** 0.198*** 0.092**

[0.036] [0.036] [0.030] [0.031] [0.032] [0.049] [0.047] [0.049] [0.045]Growth in debt 0.051*** 0.053*** 0.034* 0.036* 0.038* 0.061* 0.078** 0.057* 0.024

[0.017] [0.018] [0.020] [0.021] [0.021] [0.035] [0.030] [0.030] [0.028]Structured debt dummy −5.006** −5.031** −5.73*** −5.67*** −6.00*** −6.059** −3.677* −5.441*** −6.91***

[1.911] [1.941] [1.626] [1.646] [1.681] [2.463] [2.161] [2.003] [2.079]Real GDP growth 0.609* 0.612* 0.531* 0.573* 0.397 0.681** 0.626**

[0.333] [0.340] [0.297] [0.320] [0.314] [0.317] [0.294]Comm. bank Mkt. share −0.441*** −0.434*** −0.228 −0.237* −0.096 −0.483*** −0.301**

[0.081] [0.083] [0.139] [0.140] [0.186] [0.083] [0.148]Interbank rate 0.244 0.247 −0.062 −0.07 −0.375 0.149 −0.262

[0.278] [0.278] [0.288] [0.289] [0.369] [0.299] [0.318]Crisis 0.062 0.502 2.065

[1.706] [1.665] [1.736]Grw. in bank loans × crisis −0.088 0.007 −0.053

[0.069] [0.101] [0.105]Grw. in debt × crisis −0.031 −0.02 −0.063

[0.046] [0.044] [0.054]Allowance/loans −44.93*** −52.36*** −44.06*** −47.42***

[8.793] [8.790] [8.128] [10.091]Grw. in bank loans × spread −0.045 −0.058 −0.043 0.009

[0.053] [0.054] [0.054] [0.056]Grw. in debt × spread −0.002 0.001 0.008 0.042

[0.054] [0.047] [0.048] [0.045]Struct. debt dum × spread 0.782 0.916 1.23 2.281

[2.597] [2.059] [2.084] [2.302]Spread −1.748 0.884 1.703

[1.544] [1.572] [1.338]Constant 56.53*** 55.66*** 135.32*** 136.47*** 13.428 29.697*** 59.99*** 147.16***

[12.488] [12.757] [13.688] [14.018] [9.376] [8.963] [11.723] [14.439]

Observations 1303 1303 1303 1303 1303 1319 1319 1319 1319R-squared 0.32 0.32 0.42 0.42 0.45 0.33 0.29 0.31 0.42Time effects No No No No No Yes No No No

4

Stl

fiH

cmation, we were able to obtain their balance sheet data from Capital

* Significance at the 10%.** Significance at the 5%.

*** Significance at the 1%.

.1.1. The impact of financial deregulationTo further investigate the role of deregulation, we use the

OFOLES that transformed into SOFOMES but continued filing tohe CNBV. This information was available only for 10 (relativelyarge) deregulated SOFOLES out of the 35 in our sample.17 As they

17 The 10 SOFOLES used in this analysis are: Financiera Independiente, Metro-nanciera, Crédito Inmobiliario, Hipotecaria Su Casita, Hipotecaria Vértice, Fincasaipotecaria, Ford Credit de México, Servicios Financieros Navistar, GMAC Financiera,

Io

aSae

ontinued issuing debt in capital markets and thus filing their infor-

Q. Although the incorporation of additional information for 10ut of the 35 deregulated SOFOLES helps ameliorate the potential

nd Hipotecaria Casa Mexicana. These entities represent some of the largestOFOLES that converted into SOFOMES and had total assets of 102 million pesost the time of conversion (about 50 percent of the total assets of the 35 deregulatedntities in our sample).

46 J.M. Berrospide, R. Herrerias / Journal of F

Table 6SOFOLES, loan growth regression by type of SOFOL (fixed-effect regressions). Thistable reports coefficients of fixed effects regressions using the sample of SOFOLESsplit by sectors and using quarterly data from 2001:Q1 to 2011:Q3. The dependentvariable is the quarterly loan growth in percentage points for each SOFOL in the sam-ple. All the non-dummy explanatory variables enter the regression with a laggedvalue. The crisis dummy takes the value of 1 between 2008:Q4 and 2009:Q4. Struc-tured debt dummy takes the value of 1 for SOFOLES that securitized credit portfoliosin the debt market (RMBS). Robust standard errors are reported in brackets.

Commercial(1)

Mortgage(2)

Consumer(3)

Size (log of total assets) −11.453*** −7.143*** −6.581***

[3.243] [1.737] [1.592]Non-performing loans −36.217*** −48.549*** −73.255***

[8.884] [16.016] [25.273]Net income/total assets −0.188 −0.057 −0.02

[0.137] [0.269] [0.110]Liquid assets/total assets 27.255** 31.467 14.082

[12.230] [19.826] [18.971]Equity/total assets 12.88 20.064 4.951

[15.236] [24.052] [15.342]Growth in bank loans 0.017 0.233*** 0.059

[0.042] [0.061] [0.039]Growth in debt −0.014 0.059** 0.018

[0.052] [0.026] [0.026]Structured debt dummy −11.095** −4.469***

[5.399] [1.469]Real GDP growth 0.883* −0.134 0.765

[0.536] [0.461] [0.484]Commercial bank market

share−0.199 0.172 0.06

[0.296] [0.177] [0.203]Interbank rate 0.164 −0.893** 0.244

[0.583] [0.353] [0.480]Crisis −0.276 −1.054 −0.97

[2.680] [4.096] [2.353]Growth in bank

loans × crisis0.088 −0.229 0.162

[0.115] [0.562] [0.118]Growth in debt × crisis 0.062 −0.184 0.062

[0.088] [0.131] [0.053]Constant 165.650*** 105.075*** 89.742***

[33.230] [22.248] [22.405]

Observations 629 419 240R-squared 0.45 0.58 0.54

* Significance at the 10%.

sb

gattipSlg

ividSao(v

tstto

rnpactgt

SrgiveCwgcl(

eislipigTcgktt

4

tswieod

4.2.1. Potential omitted variablesTo mitigate the concern of lending standards possibly being an

omitted variable, we include in our specifications all the relevant

18 In unreported regressions, we use alternative measures for the impact of deregu-lation: (1) a dummy variable for the 10 entities that were deregulated that changesvalue from 0 to 1 at the time of their conversion into SOFOMES, (2) the number of

** Significance at the 5%.*** Significance at the 1%.

urvivorship bias in our analysis, our results may still be affectedy these data limitations.

Table 7 uses univariate analysis and addresses the role of dere-ulation by comparing the mean values of the variables in ournalysis for two groups of SOFOLES. The first group correspondso 19 SOFOLES that remained in operation and were regulatedhroughout our sample period. We compare the behavior of thesenstitutions before and after 2006, when the financial deregulationolicy was implemented. The second group corresponds to the 10OFOLES that converted into SOFOMES. Notice that the 10 deregu-ated SOFOLES are, on average, larger (by total assets) than the firstroup of SOFOLES.

Our univariate analysis for the first group, the leftmost column,ndicates that indeed there are significant differences in all theariables (liquidity, credit quality, capital, loan growth, and fund-ng sources) except profitability before and after 2006. Althougheregulation could explain the severe loan growth contraction forOFOLES in the control group, the impact of the financial crisis may

lso explain some of the differences. To further separate the impactf the financial crisis, we turn to the second group of SOFOLESthe treatment group) and examine the differences across the sameariables before deregulation (since we know the exact date when

Stfit

inancial Stability 18 (2015) 33–54

hey deregulated), after deregulation but before the financial cri-is (deregulation before 2008:Q4), and after deregulation and afterhe financial crisis (deregulation after 2008:Q4). As seen in the lastwo columns, the effect of deregulation on loan growth and fundingccurs only after the financial crisis.

Before the crisis, deregulation seems to have caused moreisk-taking, but the differences in loan growth and funding areot significant. Credit quality measures deteriorated (e.g., non-erforming loans and allowance for loan losses both doubled)nd profitability decreased but remained positive. After the finan-ial crisis, however, credit quality worsened (nonperforming loansripled) and profits turned negative. Loan growth plunged as therowth in both debt and bank loans dramatically dropped. Thus,his first evidence suggests a relatively minor role for deregulation.

We formally test for the impact of financial deregulation onOFOL lending by using regression analysis. We augment ouregression specification with a time dummy variable for dere-ulation that takes the value of 1 after 2006.18 We also includenteractions between the deregulation dummy and our fundingariables (debt growth and bank loan growth) in OLS and fixed-ffect regressions of loan growth. Our results are shown in Table 8.onsistent with the results of our univariate analysis, there is onlyeak evidence that deregulation has a role in explaining the loan

rowth of SOFOLES. The coefficients on size, liquidity, profits, andapital, and the growth rate of funding sources are all very simi-ar in magnitude and significance to the ones in previous resultsTables 4 and 5).

The regulation dummy is never statistically significant andnters the regression with the expected negative sign only when wenclude the macroeconomic controls for loan demand, bank markethare, and interest rate spreads. The interaction between deregu-ation and the growth in bank loans is negative and significant onlyn column 2 of Table 8. This result indicates that the deregulationolicy caused a contraction in loan growth only to the extent that

t reduced the funding through bank loans. The coefficient on therowth in bank loans shrinks by 0.12 after the 2006 deregulation.his finding suggests that loan growth slowed down as fundingontracted during the deregulation process. In sum, our results sug-est that there is weak evidence that financial deregulation was aey factor in the collapse of the SOFOL sector. If it had any role,he effect seems to have occurred during the financial crisis and isherefore hard to disentangle from other factors.

.2. Robustness tests

We perform several robustness checks to account for the poten-ial biases in our results due to omitted variables such as lendingtandards, and the possible endogeneity of the SOFOLES fundinghich may arise if they adjusted their demand for funding in antic-

pation of further cuts in lending. Finally, we repeat our panelstimations using annual data, which allow us to isolate the effectf the government funding and thus to identify the effect of therop in commercial bank loans.

OFOLES that became regulated in every quarter after 2006, and (3) interactions ofhe previous variables with the funding variables in fixed-effect regression speci-cations. Results are qualitatively similar and show, in general, weak evidence forhe importance of financial deregulation on loan growth.

J.M.

Berrospide, R

. H

errerias /

Journal of

Financial Stability

18 (2015)

33–54

47

Table 7Impact of the 2006 financial deregulation of SOFOLES, Univariate analysis. This table reports mean values for each variable used in our regression analysis and t-tests for the null hypothesis that means are not equal acrossgroups. SOFOLES that remained regulated (Panel A) are those that remain in operation throughout our sample period (19 institutions). SOFOLES that converted into SOFOMES (Panel B) are those that became unregulatedbut remained filing their financial information to the CNBV (10 institutions). We compare mean values across groups before and after the 2006 financial deregulation and after the 2006 deregulation but before and after thefinancial crisis. The crisis period starts in 2008:Q4.

Panel A

SOFOLES that remained regulated

Before 2006 After 2006 Difference(1) (2) (2)–(1)

Total Assets (Millions of pesos) 3408.06 3260.47 −147.59Liquid Assets/Assets 18.208 10.819 −0.07***

Non-Performing Loans/Loans 4.339 8.439 0.04***

Allowance for Losses/Loans −1.526 −5.147 −0.04***

Equity Capital/Assets 33.077 22.588 −0.10***

Net Income/Assets −1.959 −0.344 1.62Loan Growth 20.794 8.080 −12.71***

Debt growth 2.816 −0.250 −3.07**

Bank Loan Growth 15.028 7.753 −7.28**

Panel B

SOFOLES that converted into SOFOMES (unregulated but filing to CNBV)

Before Deregulation After Deregulation Difference After Deregulationbut Before Crisis

Difference After Deregulationand After Crisis

Difference

(3) (4) (4)–(3) (5) (5)–(3) (6) (6)–(3)

Total Assets (Millions of pesos) 7697.05 9719.27 2022.22* 8256.38 559.33 10180.4 2483.35*

Liquid Assets/Assets 6.013 7.260 0.01 6.227 0.00 7.585 1.57Non-Performing Loans/Loans 2.690 12.635 0.10*** 5.036 0.02*** 15.168 12.48***

Allowance for Losses/Loans −2.447 −9.215 −0.07*** −4.227 −0.02*** −10.695 −8.25***

Equity Capital/Assets 14.112 15.208 0.01 17.927 0.04 14.380 0.27Net Income/Assets 4.323 −0.949 −5.27 8.176 3.85*** −3.801 −8.12***

Loan Growth 11.489 −1.227 −12.72*** 6.099 −5.39 −3.417 −14.91***

Debt Growth 7.313 −2.465 −9.78*** 8.789 1.48 −5.113 −12.43***

Bank Loan Growth 10.223 −2.572 −12.79*** 3.330 −6.89 −4.200 −14.42***

* Mean differences: significance at the 10%.** Mean differences: significance at the 5%.

*** Mean differences: significance at the 1%.

48 J.M. Berrospide, R. Herrerias / Journal of Financial Stability 18 (2015) 33–54

Table 8SOFOLES, role of financial deregulation on loan growth, regression analysis. This table reports results for the regression specification that tests the impact of financialderegulation on SOFOLES loan growth. Dummy dereg. is a deregulation time dummy and takes the value of 1 for each quarter after 2006. All the non-dummy explanatoryvariables enter the regression with a lagged value. The specification includes interactions between the deregulation dummy and funding variables (debt growth and bankloan growth) in OLS and fixed-effect regressions of loan growth. Robust standard errors are reported in brackets.

OLS Fixed effects

(1) (2) (3) (4)

Size (log of total assets) −1.587*** −1.615*** −8.811*** −8.303***

[0.538] [0.541] [1.580] [1.572]Non-performing loans −38.127*** −37.578*** −38.678*** −37.514***

[6.025] [5.756] [6.397] [6.289]Net income/total assets −0.233*** −0.223*** −0.173** −0.172**

[0.081] [0.081] [0.088] [0.087]Liquid assets/total assets 32.969*** 32.491*** 27.417*** 28.470***

[8.241] [8.251] [8.958] [8.855]Equity/total assets 9.239 8.48 12.771 11.575

[6.630] [6.613] [10.608] [10.475]Foreign dummy 1.549 1.616

[1.528] [1.545]Bank owned dummy 3.483** 3.485*

[1.696] [1.748]Growth in bank loans 0.162*** 0.231*** 0.088*** 0.142***

[0.035] [0.040] [0.030] [0.047]Growth in debt 0.047*** 0.060** 0.033* 0.035

[0.016] [0.026] [0.020] [0.029]Dummy deregulation −0.05 1.997 −1.757 −0.484

[1.561] [1.907] [1.594] [1.780]Growth in bank loans × dummy dereg. −0.121** −0.091

[0.053] [0.058]Growth in debt × dummy dereg. −0.023 −0.002

[0.041] [0.039]Real GDP growth (annualized) 0.646* 0.628* 0.399 0.418

[0.329] [0.335] [0.293] [0.293]Commercial bank market share −0.429*** −0.497*** −0.064 −0.167

[0.117] [0.127] [0.168] [0.169](Interbank − treasury) spread 1.431 1.051 −2.601 −2.591

[1.584] [1.382] [1.827] [1.793]Constant 54.173*** 59.186*** 132.760*** 133.192***

[11.889] [12.465] [19.258] [19.216]

Observations 1303 1303 1303 1303R-squared/pseudo R-squared 0.31 0.32 0.42 0.43Firm fixed effects No No Yes Yes

*

ctfsccFfiaeT

sturpcatrkCa

saosdc

i

Significance at the 10%.** Significance at the 5%.

*** Significance at the 1%.

ontrols, both macro-variables and firm-specific characteristicshat have been proposed in the literature as the determinantactors.19 Unobserved factors related to lending standards deci-ions are captured by firm fixed effects. However, a potentialoncern about omitted variable bias may still arise if our specifi-ation fails to capture changes in unobserved lending standards.or example, if changes in lending standards are omitted then ournding proxies may be correlated with the error term and thus

n increase in lending standards would give an observationallyquivalent prediction for the contraction in bank loans and debt.o further account for this possibility, we extend our regression

19 In Bassett et al. (2014), bank-specific factors that play a role in setting lendingtandards include proxies for profitability, asset quality, and balance sheet composi-ion. Macroeconomic variables that influence banks’ lending standards include thenemployment rate, GDP growth, and changes in short term rates (federal fundsate). Consistent with this view, our baseline regression specification includes bankrofitability (ROA), asset quality (nonperforming loans/loans), and balance sheetomposition (share of liquid assets in total assets to capture the mix between loansnd liquid assets). We also include macroeconomic variables such as GDP growth,he market share of banks to capture bank competition, and the short term interestate spread. Kara et al. (2011) show that European banks active in securitization mar-ets loosened more aggressively their lending standards in the run-up to the crisis.onsistent with this view, we also include controls for the use of loan securitizationss additional proxies for lending standards.

lcfictctttssttNsS

pecification and include both firm’s fixed effects and their inter-ction with an after crisis dummy variable (which takes the valuef 1 after the Lehman failure in 2008:Q4). The main purpose of thispecification is to capture the change in unobserved lending stan-ards (fixed effects) in the post crisis period. Results are shown inolumns (5) and (6) of Table 4, and in column (5) of Table 5.

Compared to previous specifications (columns (1) through (4)n Tables 4 and 5), the coefficients on the funding variables (bankoans and debt) remain almost unchanged and they are statisti-ally significant. Most of the unreported interaction terms of thexed effects and the after crisis dummy are negative and signifi-ant, indicating that indeed part of the SOFOL loan contraction inhe post crisis period was driven by changes in unobserved firmharacteristics. These changes should be capturing the portion ofhe increase in lending standards not accounted for by the firm con-rols and macro-variables in our baseline specifications. However,he interaction terms may also absorb any other changes in unob-erved characteristics which may reduce the explanatory power ofome of the specific controls in our specification. This seems to behe case of SOFOL liquid assets and of the macroeconomic variables

hat control for demand conditions and competition from banks.evertheless, our funding variables remain significant across all

pecifications and imply that our results are robust to changes inOFOL unobserved characteristics after the crisis.

al of F

aittoclca

w(isr

4

ofipdaetfmflwotevTs

stf(fta

�

tHvtiistat“2p

aa2tcbeivatc

sBfidtbtiWtl

4

tSdttatgkAfbt

milw

J.M. Berrospide, R. Herrerias / Journ

We also replace nonperforming loans ratio with the ratio ofllowances for loan losses to total loans (allowance ratio), whichs a forward looking measure of loan quality as financial institu-ions create reserves based on their expected losses. We reporthe results in columns (6) through (9) in Table 5. The coefficientn the allowance ratio in all 4 estimations is negative and signifi-ant, and slightly higher than the coefficients on the nonperformingoan ratio. More importantly, after including these variables, theoefficients on bank loans and debt remain positive and significantnd slightly larger compared to estimations in columns (1) to (4).

Finally, to address a potential bias introduced by the way inhich we captured the effects of the financial crisis, in columns

6) through (9) of Table 5 we substitute the crisis dummy and itsnteraction with our funding proxies with a continuous variableuch as the TED spread, which is a more precise measure of creditisk perception. All our results remain qualitatively unchanged.

.2.2. Endogeneity of SOFOLES fundingThe SOFOL credit contraction also could have been the result

f a generalized decline in loan demand from households andrms associated with the economic recession in Mexico during oureriod of study. Furthermore, SOFOLES may have adjusted theiremand for funding from banks and investors in debt markets innticipation of further cuts in lending, perhaps as a result of anxpected deterioration in the quality of their loan portfolios afterhe end of the credit boom.20 For this reason, the change in SOFOLunding sources may be endogenous, which would render our esti-

ates inconsistent. We deal with the potential endogeneity ofunding by analyzing the historic dependence of SOFOLES on bankoans and market debt in a cross-sectional analysis. For this purpose

e employ quarterly data on the 31 institutions that remained inur sample between 2006 and 2009. We hypothesize that SOFOLEShat historically depended more on those funding sources shouldxhibit a larger contraction in their lending during the most rele-ant quarters of the financial crisis (that is, at the peak of the crisis).he rationale is that, given their business model, a SOFOL’s fundingources are sticky and hard to substitute during a crisis.

We use our quarterly database and run cross-section regres-ions of loan growth in the quarters surrounding the crisis, 2008:Q3hrough 2009:Q2, on the dependence on bank funding and debtunding before the crisis. Taking into account the restricted number31) of observations in each quarter, we include only two controlsor SOFOL characteristics (size and nonperforming loans) in addi-ion to our funding variables, measured as a percentage of totalssets. We use the following regression specification:

log (Loani)crisis = + ˇ1

(Bank Loani

Total Assetsi

)Pre-crisis

+ ˇ2

(Debti

Total Assetsi

)Pre-crisis

+ �1sizei,Pre-crisis

+ �2NPLi,Pre-crisis + εi (2)

20 Another potential omitted variable concern may arise if SOFOLES securitizedheir relatively risky loans and retained their safer loans on their balance sheet.ad this been the case, then the performance of their RMBS would have containedaluable information that lenders and investors funding SOFOLES could have usedo infer lending standards at SOFOLES. This situation would have made the fund-ng contraction endogenous. This, however, was not the case as the deteriorationn the performance of the SOFOLES-originated RMBS (downgrades and defaults onecuritized loans) occurred late in 2008 and in 2009, that is, after the decline inhe funding sources. Thus, we interpret the collapse in structured debt fundings another manifestation, and not the cause, of the liquidity shock. In our view,his provides additional support in favor of our liquidity shock hypothesis. See,Default Study: Mexican Structured Finance Default and Rating Transition Study,003–2012,” by Standard and Poor’s Rating Services, May 9, 2013, for details on theerformance of RMBS in Mexico.

igai

tmIcOTkvmo(p1

inancial Stability 18 (2015) 33–54 49

Note that in this specification our funding variables, Bank loannd Debt, are expressed not as growth rates but as shares of totalssets and are measured in the pre-crisis period (either yearend006 or yearend 2007). Thus, this regression specification captureshe extent to which the loan growth rate of SOFOLES during therisis was explained by their dependence on bank loans and debtefore the crisis. Unlike in our baseline regression (1), in Eq. (2) wexpect negative and significant coefficients on our pre-crisis fund-ng variables (ˇ1 < 0 and ˇ2 < 0), which means that, if our fundingariables are exogenous, then higher dependence on bank loansnd debt during the pre-crisis period (say, one or two years beforehe crisis) would imply a larger contraction in lending during therisis.

Table 9 presents the results with explanatory variables mea-ured at yearend 2006 in Panel A and at yearend 2007 in Panel. Our findings indicate that during the two peak quarters of thenancial crisis, 2008:Q4 and 2009:Q1, dependence on pre-crisisebt and bank loans is indeed associated with a significant contrac-ion of lending. SOFOLES that before the crisis had larger shares ofank loans and debt contracted their lending during the crisis morehan SOFOLES that relied less on these funding sources. The fund-ng variables are statistically insignificant in the non-crisis quarters.

e interpret these findings as additional supporting evidence thathe liquidity shock that contributed to the large decrease in SOFOLending was exogenous.

.2.3. Results with annual dataWe repeat our estimations using annual data instead of quar-

erly data. Results are qualitatively similar, and our proxies forOFOLES funding are positive and statistically significant. As weiscussed above, one of our estimation’s challenges in using quar-erly data is that the impact attributed to the liquidity shockhrough bank loans may be understated, as we could not separate,t the firm-level, the magnitude of the government loans extendedhrough development banks and the housing funds. The Mexicanovernment stepped in to compensate for the collapse of debt mar-ets and the restricted access of SOFOLES to commercial bank loans.lthough this issue in the quarterly data works somewhat in our

avor—insofar as it biases our estimates of a sizable effect of the cut-ack in funding from commercial banks—we nevertheless mitigatehe effect by using the annual data.

The advantage of using the annual data is that we isolate theagnitude of government loans and thus cleanly identify the drop

n commercial bank loans. Fig. 7 illustrates this situation. The bankoan growth rate (commercial bank and government loans) that

e identify in the quarterly data (solid line) drops from 32 percentn 2008 to negative 7 percent in 2010. The commercial bank loanrowth rate that we identify in the annual data (dashed line) shows

more severe decline from 9 percent in 2008 to negative 63 percentn 2010.

We utilize the annual data on 29 SOFOLES and repeat estima-ions using our benchmark specification in Table 5, which includes

acroeconomic controls for loan demand and bank competition.n these specifications the variable of interest is the growth inommercial bank loans rather than the growth in total bank loans.ur estimates using OLS specifications in the first two columns ofable 10 confirm that both commercial bank loans and debt areey determinants of SOFOLES’ annual loan growth. Both fundingariables have positive and significant coefficients. Using the sum-ary statistics in panel C of Table 3, our estimates indicate that a

ne 1 standard deviation decline in commercial bank loan growthabout a 45 percentage point decrease) leads to an 8 percentageoint drop in the annual loan growth of SOFOLES. Similarly, a one

standard deviation decline (44 percentage point decrease) in the

50 J.M. Berrospide, R. Herrerias / Journal of Financial Stability 18 (2015) 33–54

Table 9Loan growth analysis, cross-section. The table reports estimation results using our cross-section analysis described by Eq. (2). Our cross section analysis is limited to 31SOFOLES that remain in the quarterly sample between 2006 and 2009. Explanatory variables are measured at yearend 2006 in Panel A and at yearend 2007 in Panel B. Thedependent variable is loan growth in the quarters surrounding the crisis, 2008:Q3 through 2009:Q2. The specification includes size and nonperforming loans to control forSOFOL characteristics in addition to funding variables, measured as a percent of total assets. Bank loan and Debt are expressed as shares of total assets and are measured inthe pre-crisis period (either yearend 2006 or yearend 2007). Robust standard errors are reported in brackets.

Firm controls as of 2006:Q4 2008:Q3 2008:Q4 2009:Q1 2009:Q2

Panel A. Firm controls measured as of 2006:Q4Size (log of total assets) −0.908 5.119** 2.657 0.749

[2.705] [1.868] [1.914] [2.336]Non-performing loans 35.99 −204.262* −204.377** −112.4

[152.879] [104.935] [95.821] [111.587]Bank Loans 7.951 −96.305*** −9.398 24.586

[25.051] [17.085] [16.526] [27.906]Debt 8.214 −140.767*** −67.271** 0.092

[38.599] [26.312] [26.025] [35.586]Constant 7.957 14.911 −23.94 −30.233

[31.046] [21.162] [20.053] [23.993]

Observations 31 31 31 29F test 0.05 11.88 3.0 0.68R-squared 0.01 0.65 0.32 0.10

Firm controls as of 2007:Q4 2008:Q3 2008:Q4 2009:Q1 2009:Q2

Panel B. Firm controls measured as of 2007:Q4Size (log of total assets) −1.156 0.286 1.904 1.642

[2.571] [2.386] [2.090] [2.305]Non-performing loans −51.757 −192.805** −85.184 −94.239

[93.281] [86.794] [62.923] [69.537]Bank loans 39.835 −55.083** −25.838 −0.547

[26.495] [24.560] [25.055] [32.383]Debt 33.055 −84.975** −64.956* −15.847

[39.517] [36.637] [34.417] [40.602]Constant −10.091 54.876 −3.589 −24.269

[35.042] [32.523] [23.664] [28.692]

Observations 30 29 28 27F-test 0.58 4.23 1.6 0.66R-squared 0.08 0.41 0.22 0.10

* Significance at the 10%.** Significance at the 5%.

*** Significance at the 1%.

Fig. 7. The annual growth rate of SOFOLES total bank loans and commercial bankloans using annual data. Shaded areas represent the years of the financial crisis.

gl

usdi(tbctfillct

rl

faihb

rowth rate of Debt also implies an 8 percent drop in their annualoan growth.

Despite the fact that we refine our measure of the funding shocksing the growth rate of commercial bank loans, this variable cantill be endogenous if, for example, it reflects changes in SOFOLES’emand for funding. To further control for the potential endogene-

ty of this funding source we implement an instrumental variableIV) approach. Since we know the identity of the commercial bankshat lend to each SOFOL, we match the SOFOLES annual data withalance sheet data for each individual lender.21 We instrument ourommercial bank loan variable with bank-specific data to capturehe financial conditions of banks that lend to SOFOLES. Thus, in therst stage regression we model the growth in commercial bank

oans of each SOFOL as a function of weighted average measures ofiquidity, risk, and funding across its lenders as well as other SOFOLontrols. The predicted value of commercial bank loan growth ishen the variable of interest in the second stage regressions.

The last two columns of Table 10 present the second stageegression results using the IV approach. Consistent with ouriquidity shock hypothesis, the coefficient on the predicted values

21 We obtain data on loan amount, maturity, and the identity of the lending bankor 19 SOFOLES from Capital IQ. For these, we employ balance sheet data (weightedverage ratios) on the liquidity, risk, and funding of each of the lending banks asnstruments for bank funding. For the remaining 10 institutions, for which we do notave the identity of individual lenders, we use the data on the average commercialank lending to SOFOLES.

J.M. Berrospide, R. Herrerias / Journal of Financial Stability 18 (2015) 33–54 51

Table 10SOFOLES, loan growth regression, 2003–2011 (annual data). This table reports estimates from regressions using annual data for a sample of 29 SOFOLES for which we obtaineddisaggregated information on commercial bank loans and government loans. The regression specification includes macroeconomic variables (real GDP growth, the share ofcommercial banks in credit markets, and the interbank rate). The dependent variable is the annual loan growth for each SOFOL in the sample. All the non-dummy explanatoryvariables enter the regression with a lagged value. The main funding variable of interest is the growth in commercial bank loans. Nonperforming loans is the ratio of loanspast due 90 days and non-accrual to total loans. Foreign and Bank-owned dummies are time-invariant dummy variables that take the value of 1 for foreign SOFOLES andthose SOFOLES affiliated to a bank or a financial group in Table 1, respectively. Structured debt dummy takes the value of 1 for SOFOLES that securitized credit portfolios inthe debt market (RMBS). The last two columns report results of the instrumental variable approach where we use bank-specific data to model the growth in commercialbank loans of each SOFOL as a function of weighted average measures of liquidity, risk, and funding across its lenders as well as other SOFOL controls. The predicted valueof commercial bank loan growth is then the variable of interest in the second stage regressions.

OLS IV regression (second stage)

(1) (2) (1) (2)

Size (log of total assets) −2.281 −2.743 −0.942 −1.095[3.688] [3.817] [4.010] [4.024]

Non-performing loans −92.228*** −101.219*** −90.462*** −98.849***

[32.200] [35.561] [25.871] [29.725]Net income/total assets −1.705* −1.657* −1.904** −1.903**

[0.886] [0.850] [0.877] [0.864]Liquid assets/total assets 18.806 22.816 8.634 9.198

[84.051] [84.114] [61.817] [60.443]Equity/total assets 95.254** 98.766*** 104.447*** 109.767***

[36.640] [35.010] [35.942] [35.603]Foreign dummy 16.213 18.603* 13.401 15.032

[9.971] [10.575] [10.995] [11.050]Bank owned dummy −3.434 −3.543 −4.93 −5.39

[7.196] [6.940] [6.206] [5.961]Growth in commercial bank loans 0.157** 0.156* 0.248* 0.263*

[0.074] [0.076] [0.143] [0.153]Growth in debt 0.133* 0.163** 0.179* 0.219**

[0.069] [0.060] [0.100] [0.104]Structured debt/total assets 9.3 27.507 −18.459 −7.75

[61.067] [57.094] [85.485] [83.701]Real GDP growth 2.105* 2.095**

[1.096] [1.053]Commercial bank market share −0.35 0.312

[1.032] [1.076]Interbank rate 51.874 39.302

[60.902] [57.535]Constant 23.155 31.113 5.84 −43.972

[59.884] [73.451] [61.325] [100.485]

Observations 99 99 99 99R-squared 0.37 0.36 0.35 0.32

*

oIlotrtpt

4

iibmtlr2si

ia

rom(mitrAtgicomdf

Significance at the 10%.** Significance at the 5%.

*** Significance at the 1%.

f commercial bank lending is always positive and significant.nterestingly, the magnitude of the coefficient on commercial bankoans more than doubles in magnitude relative to the coefficientn commercial bank lending in our OLS specification (found inhe first two columns, which do not correct for endogeneity). Thisesult implies that when we control for the factors associated withhe supply of bank funding—liquid assets, interbank loans, androfitability of commercial bank lending to SOFOLES—the effect ofhe liquidity shock on SOFOL credit activity is bigger.

.3. Loan growth decomposition

To have a better sense of the economic significance of the liquid-ty shock, and in an effort to disentangle supply and demand factorsn loan growth, we decompose the annual loan growth of SOFOLESy the different key explanatory factors using our regression esti-ates that correct for the endogeneity problem. Table 11 shows

hese results. Columns (1) and (2) of the table show the averageoan growth rate as well as each of the explanatory variables in our

egression model between 2005 (the peak in the credit cycle) and010 (the trough of the credit cycle). During this period, SOFOLESaw a large contraction of their loan growth (52 percentage points)n annual terms (Column (3)). Commercial bank loans and debt

tlid

ssuance also experienced a significant decline (87 and 24 percent-ge points, respectively).

We use our estimates from the second specification of our IVegression (last column in Table 10) to obtain the implied impactf individual factors. This specification considers lending from com-ercial banks and adds macroeconomic controls for loan demand

GDP growth and interest rate spread) and competition from com-ercial banks to our baseline specification. Column (4) shows the

mplied impact for each explanatory variable obtained by mul-iplying its coefficient in the IV regression by the change in theespective variable from peak to trough in Column (3) of Table 11.s seen in column (4), the total model implied impact is nega-

ive 43.98—that is, about 85 percent of the actual change in loanrowth from peak to trough. Column (5) shows the contribut-ng percentage of each explanatory variable to the total impliedhange in the annual loan growth rate. The combined change inur funding variables—commercial bank loans and debt—is theost important factor and explains 64 percent of the lending

ecline of SOFOLES between 2005 and 2010. Two other importantactors are nonperforming loans, which proxy for the deteriora-

ion in their loan portfolios and account for 21 percent of theoan decline, and our loan demand controls (GDP growth and thenterest rate spread), which account for 16 percent of the lendingecline.

52 J.M. Berrospide, R. Herrerias / Journal of Financial Stability 18 (2015) 33–54

Table 11SOFOLES, loan growth decomposition (annual regression). This table reports the SOFOL loan growth decomposition based on our estimates using the annual instrumentalvariable (IV) regression in Table 10, which corrects for the potential endogeneity of SOFOL commercial bank loans. Contributing factors to the SOFOL loan growth are theexplanatory variables in column (4) of Table 10. Model Implied impact (column 4) is obtained by multiplying coefficients of the IV model by the change in each explanatoryvariable in column (3). Column (5) shows the model implied impact for each contributing factor expressed as percent of the total estimated impact on loan growth in column(4).

Mean value at peak(2005)(1)

Mean value attrough (2010)(2)

Change(3) = (2) − (1)

Model impliedimpact(4)

Model impliedimpact (%)(5)

Loan growth (annual) 38.25 −13.53 −51.78 −43.98 100Size (log of total assets) 14.49 15.15 0.66 −0.72 2Non-performing loans 0.02 0.12 0.09 −9.30 21Net income/total assets 0.47 −1.33 −1.80 3.42 −8Liquid assets/total assets 0.09 0.05 −0.04 −0.36 1Equity/total assets 0.23 0.17 −0.06 −6.87 16Foreign dummy 0.32 0.33 0.01 0.18 0Bank owned dummy 0.32 0.28 −0.04 0.24 −1Growth in commercial bank loans 59.53 −27.61 −87.14 −22.92 52Growth in debt 14.53 −9.19 −23.72 −5.20 12

−0.03 0.23 −1 14.51 4.53 −10

−7.22 16

mcst6tic

5

ttmdiclidirskattat2out

siptntl

Fig. 8. The number of mortgage loans originated by Commercial banks and SOFOLESb

S

2d

sitIsoMrtments of the population. In contrast, the number and size of loansfor social interest houses, available for low-income segments of thepopulation, show a significant reduction after 2010 (for example,

Structured debt/total assets 0.03 0.00Commercial bank market share 75.49 90.00Demand (GDP grw., interest rate spread)

To summarize, after controlling for the endogeneity of com-ercial bank funding, our results using annual information are

onsistent with our liquidity shock hypothesis in explaining theevere credit crunch of Mexican SOFOLES, in accordance withhe analysis in Section 2. The funding contraction explains about4 percent of the decline in annual loan growth. In particular,he severe cutback in commercial bank loans is one of the mostmportant determinant and accounts for 52 percent of the loanontraction.

. The economic and social impact of the exit of SOFOLES

Finally, we illustrate the economic and social consequences ofhe severe contraction in the SOFOL sector by examining whetherhe return of the commercial banking sector to the housing finance

arket has substituted for the SOFOLES’ loss in market share. Asiscussed above, SOFOLES played a major role in mortgage lend-

ng during the last two decades. SOFOL participation in housingredit markets quadrupled between 1997 and 2005, moving fromess than 2 percent to 17 percent market share by mid-2005. Dur-ng the same time period, the market share of commercial bankseclined from 43 percent to 22 percent. In contrast, SOFOL partic-

pation in housing financing contracted dramatically after 2005 toepresent only 2 percent as of yearend 2010, whereas the markethare of commercial banks rose to 31 percent (the rest of the mar-et is served by government housing programs). As our regressionnalysis indicates, using this aggregate measure of market shares inotal lending, increased competition from commercial banks seemso be a key driver of their loan contraction. Interestingly, 9 percent-ge points of banks’ gain in market share does not compensate forhe 15 percentage point loss in SOFOLES’ market share between005 and 2010. This simple observation implies that the collapsef the SOFOL sector may have left an important fraction of the pop-lation unattended and without much access to the financial sectoro finance the purchase of a house.

In terms of the number of mortgage loan originations, Fig. 8hows that at the peak of their business in 2005, SOFOLES orig-nated more loans than commercial banks (96,582 versus 55,537er year, respectively). However, since 2005 (and mostly due to

he factors we studied in this paper), the number of loans origi-ated by SOFOLES (solid line) dropped dramatically to 38,230 byhe end of 2010. Loan originations by commercial banks (dashedine) increased to 120,211 (the peak before the financial crisis in

l9

etween 1995 and 2010.

ource: Mexican Housing Overview, 2012. Softec, S.C. México.

008:Q4) and then fell to 69,700. The gap between the solid andashed lines after 2006 shows the loss in originations.

Furthermore, to go beyond our aggregate measure in the regres-ion analysis, panels A and B in Fig. 9 illustrates the significant gapn the market for housing finance during recent years using datahat considers the type of financing and targeted social sectors.nformation on mortgage loans by commercial banks divided byegment shows that between 2010 and 2013, the number and sizef loans used to finance the acquisition of residential houses inexico increased significantly (for example, the number of loans

ose from 8 percent to 30 percent).22 Residential houses are theype of housing affordable to the medium- and high-income seg-

22 According to a recent BBVA mortgage report (Torres and Balbuena, 2013), theoan size of bank mortgage loans has increased from 726,000 pesos in 2009 to96,000 pesos in 2013.

J.M. Berrospide, R. Herrerias / Journal of F

Fig. 9. The percentage participation by housing segment by number of loans and bythe financing amount of commercial banks mortgage loans. Social interest housesinclude three categories: economic (up to 232 thousands of pesos), popular (up to394 thousands of pesos) and traditional (up to 689 thousands of pesos). The value ofmi

S

tta

ms(

gsttpcgsa

dptetsrrolcgst

6

esaabMmdidivap

iatqteftttadscht

id-level houses is up to 1.47 million of pesos, and the value of residential housess 2.95 million of pesos or more.

ource: BBVA Research, “Flash Inmobiliario Mexico”, April 4, 2013.

he number of loans fell from 65 percent to 32 percent). Amonghose affected by this severe contraction were workers not affili-ted with the social security system.23

Beyond the implications for credit markets, the contraction in

ortgage lending may have had an important impact on the con-

truction industry. According to the National Institute of StatisticsINEGI), the index for construction activity decreased 4.5 percent

23 Historically, self-employed individuals have not been part of the coverage ofovernment housing programs because these programs function through the socialecurity system. Other housing programs such as FOVI (a housing fund channeledhrough the banking sector), administered by SHF, have served informal sectors ofhe population through financial intermediaries. These programs became the majorroviders of housing finance channeling funds through SOFOLES at the time whenommercial banks withdrew from mortgage lending. Lower capital requirementsave them operational advantages over commercial banks in serving less affluentectors by financing the purchase of social interest houses in remote geographicalreas.

oalstwfirAcirbr

inancial Stability 18 (2015) 33–54 53

uring the first 9 months of 2013, compared to increases of 4.0ercent and 1.5 percent during 2011 and 2012, respectively. Fur-hermore, as noted in Section 1, the three largest homebuildersntered into bankruptcy at the end of 2012. Certainly the reduc-ion in mortgage lending is not the only factor contributing to thelowdown in the construction industry, but there is no doubt thateducing funding opportunities for a large sector of the populationesults in lower home sales which, in turn, has a negative impactn construction activities. In our view, this situation illustrates theong-lasting and severe social and economic consequences of theollapse of the SOFOL sector—triggered, as we argue here, by thelobal liquidity crisis—on the financing of houses for an importantegment of the population, as well as the detrimental effects forhe construction industry in Mexico.

. Conclusions

In the aftermath of the 1994 financial crisis in Mexico, SOFOLESmerged as the most viable alternative to cover the demand in con-umer and mortgage lending in Mexico. As niche lenders, SOFOLESllowed a number of households and small businesses to gainccess to formal credit that commercial banks did not provide. Theyecame one of the most important providers of mortgage loans inexico. However, as the housing market expanded, their businessodel became increasingly reliant on bank credit and short-term

ebt from capital markets. In mid-2005, SOFOLES were originat-ng long-term loans funded with significant amounts of short-termebt. Excessive risk-taking via maturity transformation and a frag-

le funding model left them highly vulnerable to capital marketolatility and the collapse of short-term debt markets broughtbout by the 2008 global liquidity crunch. As a result, the sectorassed from an impressive boom to a near collapse.

We attempt to disentangle the connection between the crunchn SOFOL lending and the U.S. subprime crisis using quarterlynd annual firm-level data. We also control for loan demand,he deregulation process, and other supply factors such as loanuality deterioration. Our results provide supporting evidence forhe hypothesis that the large contraction in SOFOL lending wasxplained by a liquidity shock—that is, the collapse in their mainunding sources (bank loans and market debt). After accounting forhe potential endogeneity of commercial bank loans, we find thathe funding reduction explains 64 percent of the lending contrac-ion. Demand factors, particularly the large contraction in economicctivity during the financial crisis also played a significant role, asid nonperforming loans and lower liquidity buffers. Our analy-is finds only a minor role for deregulation in explaining SOFOLES’redit contraction. The role of deregulation, if there was any, mayave been entangled with the impact of the financial crisis throughhe severe cut in funding.

Our results have important policy implications. The collapsef SOFOLES has long-lasting economic and social consequences,s the return of the banking sector, particularly in the mortgageending business for less affluent households, has not compen-ated for the loss in market share that resulted from the exit ofhe SOFOLES. First, as our results show, mortgage credit in Mexicoas severely affected by the contagion from the internationalnancial crisis through debt and credit markets. This somewhatesembles previous international contagion episodes, such as thesian Financial crisis in the late 1990s. Second, near their financialollapse, SOFOLES had a business model similar to shadow banking

nstitutions in the U.S. and in other countries. Recent financialeforms in those countries are aimed at overseeing the shadowanking sector more closely and making them subject to moreegulation for financial stability purposes. Thus, policymakers

5 al of F

idttst

rtttidm

A

aasSSctrprnR

R

A

AA

B

B

B

B

C

C

C

C

D

E

G

H

I

K

K

L

M

O

P

R

S

4 J.M. Berrospide, R. Herrerias / Journ

n emerging markets may have to adopt similar policies and, inoing so, they may need to deal with the impact of contagionhrough capital markets and, perhaps more importantly, throughhe banking sector. This is especially important when the bankingector is dominated by international banks, as they may exacerbatehe impact of global liquidity shocks through capital outflows.

Finally, as we learned from the Mexican SOFOLES, althoughelieving non-depository financial intermediaries of strict regula-ion may have been a reasonable policy response during difficultimes, deregulation and a lack of supervision may still bring unin-ended consequences. More regulation may be needed for financialnstitutions that rely on short-term wholesale funding to interme-iate funds not only through interest rate gaps but also throughaturity transformation.

cknowledgements

We would like to thank Iftekhar Hasan the editor of the JFSnd two anonymous reviewers for their very helpful commentsnd suggestions. We thank Eric Hardy and Luis Alberto López foruperb research assistance. We are grateful to Ralf Meisenzahl,eung Lee, Seon Tae Kim, Manuel Campos, Eugene Towle, Rafaelchiozer, Skander Van den Heuvel, and Aurelio Vasquez for valuableomments. Renata Herrerias thanks the hospitality and support ofhe Department of Finance at Boston University, where part of thisesearch was done and gratefully acknowledges the financial sup-ort provided by the Asociación Mexicana de Cultura, A.C. All errorsemain our own. The views expressed here are entirely ours and doot necessarily reflect those of the Board of Governors of the Federaleserve System or its staff.

eferences

iyar, Sh., 2011. How did the crisis in international funding markets affect banklending? Balance sheet evidence from the United Kingdom. Bank of England,Working Paper No. 424.

llen, F., Gale, D., 2000. Financial contagion. J. Polit. Econ. 108 (1), 2–33.MFE (Mexican Association of Finance Companies), 2012, October. Plan Estratégico

2020.assett, W.F., Chosak, M.B., Driscoll, J., Zakrajsek, E., 2014. Changes in lending stan-

dards and the macroeconomy. J. Monet. Econ. 62 (1), 23–40.eltratti, A., Stulz, R.M., 2012. The credit crisis around the globe: why did some banks

perform better? J. Financ. Econ. 105 (1), 1–17.

T

T

inancial Stability 18 (2015) 33–54

runnermeier, M.K., 2009. Deciphering the liquidity and credit crunch 2001–2008.J. Econ. Perspect. 23 (1), 77–100.

runnermeier, M.K., Pedersen, L.H., 2009. Market liquidity and funding liquidity.Rev. Financ. Stud. 22 (6), 2201–2238.

arballo-Huerta, J., González-Ibarra, J.P., 2008. Financial Innovations and Develop-ments in Housing Finance in Mexico. Bank for International Settlements, IFCBulletin No. 31.

etorelli, N., Goldberg, L., 2011. Global banks and international shock transmission:evidence from the crisis. IMF Econ. Rev. 59 (1), 41–76.

ornett, M.M., McNutt, M.J.J., Strahan, P., Tehranian, H., 2011. Liquidity risk man-agement and credit supply in the financial crisis. J. Financ. Econ. 10 (12),297–312.

ovitz, D., Liang, N., Suarez, G.A., 2013. The evolution of a financial crisis:collapse of the asset-backed commercial paper market. J. Financ. 68 (3),815–848.

iamond, D., Rajan, R., 2005. Liquidity shortages and banking crises. J. Financ. 60(2), 615–647.

ichengreen, B., Modyc, A., Nedeljkovicd, M., Sarnoe, L., 2012. How the subprimecrisis went global: evidence from bank credit default swap spreads. J. Int. MoneyFinanc. 31 (5), 1299–1318.

orton, G., 2009. Information, liquidity, and the (ongoing) panic of 2007. Am. Econ.Rev. Pap. Proc. 99 (2), 567–572.

errerias, R., 2011, March. SOFOLES: Risk and Liquidity. Dirección Estratégica: larevista de negocios del ITAM. No. 36.

xe Grupo Financiero, 2009, December. SOFOLES/SOFOMES: Enfrentando la Tor-menta. Análisis de Deuda Privada.

ara, A., Marquez-Ibanez, D., Ongena, S., 2011. Securitization and Lending Standards:Evidence from the Wholesale Loan Market. European Banking Center, DiscussionPaper No. 2011-020.

hwaja, A.I., Mian, A., 2008. Tracing the impact of bank liquidity shocks: evidencefrom an emerging market. Am. Econ. Rev. 98 (4), 1413–1442.

ongstaff, F.A., 2010. The subprime credit crisis and contagion in financial markets.J. Financ. Econ. 97 (3), 319–338.

endoza, E., 2012. Un análisis de los factores de riesgo que incrementaron elriesgo de liquidez en las SOFOLES hipotecarias de México. Estudios Económicos,National Banking and Securities Commission (CNBV), pp. 1.

ssa, F., 2008, December 8. Housing Pessimists Chew on Metrofinanciera;Company puts Figure on Money Owed to the Trusts. Asset Securitizationreport, Downloaded from Factiva on the 31/08/2012 http://global.factiva.com.ezproxy.bu.edu/hp/printsavews.aspx?pp=Print&hc=All

ickering, N., 2000, May. The SOFOLES: Niche Lending or New Leaders in the MexicanMortgage Market? Joint Center for Housing Studies, Harvard University, WP-002.

amcharan, R., Van den Heuvel, S.J., Verani, S., 2015. From wall street to mainstreet: the impact of the financial crisis on consumer credit supply. J. Financ.,http://dx.doi.org/10.1111/jofi.12209 (in press).

chnabl, P., 2012. The international transmission of bank liquidity shocks: evidencefrom an emerging market. J. Financ. 63 (3), 897–931.

ello, M., Gurwitz, Z.R., De Mollein, J.P., 2013, May 9. Default Study: Mexican Struc-tured Finance Default and Rating Transition Study, 2003–2012. Standard andPoor’s Rating Services.

orres, E., Balbuena, F., 2013, April 4. Flash Inmobiliario México, La mano que mecela cuna en el mercado hipotecario. Análisis Económico, BBVA Research.