Loan growth and bank risk: new evidence

Financ Mark Portf Manag (2013) 27:365–379DOI 10.1007/s11408-013-0217-6

Loan growth and bank risk: new evidence

Juan Sebastián Amador · José E. Gómez-González ·Andrés Murcia Pabón

Published online: 24 September 2013© Swiss Society for Financial Market Research 2013

Abstract This study provides new evidence on the relationship between abnormalloan growth and banks’ risk-taking behavior using data from a rich panel of Colombianfinancial institutions. We show that abnormal credit growth during a prolonged periodleads to an increase in banks’ riskiness, accompanied by a reduction in solvency andan increase in the ratio of nonperforming loans to total loans. We also show thatabnormal credit growth played a fundamental role in the bank-failure process duringthe late 1990s financial crisis in Colombia. Our results have important implicationsfor financial regulation and macro-prudential policy.

Keywords Abnormal loan growth · Hazard duration models · FGLS estimation ·Emerging market economies

JEL Classification G20 · G21

The findings, recommendations, interpretations, and conclusions expressed in this paper are those of theauthors and do not necessarily reflect the view of the Banco de la República or its Board of Directors.

J. S. AmadorDepartment of Inflation and Macroeconomic Programming,Central Bank of Colombia, Bogotá, Colombiae-mail: [email protected]

J. E. Gómez-González (B)Research Department, Central Bank of Colombia,Bogotá, Colombiae-mail: [email protected]

A. M. PabónMonetary and Reserve Affairs Office, Central Bank of Colombia,Bogotá, Colombiae-mail: [email protected]

123

366 J. S. Amador et al.

1 Introduction

An important conclusion of modern economic theory is that finance is good for growth(Cecchetti and Kharroubi 2012). However, despite the well-known advantages of creditmarket development and growth for the economy in the long run, excessive loan growthmay have negative effects on the financial system and the economy at large.

The history of financial crises and, particularly, the recent international financialcrisis, clearly illustrates what can go wrong when there is excessive credit growth.Many financial crises have been preceded by episodes of abnormal credit growth thatled to the development of asset price bubbles.

This scenario may arise if, during expansionary periods, banks, firms, and house-holds tend to underestimate risk, taking actions that increase their probability of finan-cial difficulty in the future (see, e.g., Altunbas et al. 2010, 2012). Some authors relatethis pattern to myopic behavior by private agents (García-Suaza et al. 2012; Borioet al. 2001); others highlight the presence of asymmetric information and financialfrictions in credit markets (Holmstrom and Tirole 1997; Mendoza and Bianchi 2010).

Nonetheless, there are many reasons why individual banks expand their balancesheets, and credit expansion does not necessarily imply more risk for banks. Forinstance, banks may be interested in diversifying their loan portfolios or cross-selling.Banks may try to take advantage of new profitable lending opportunities and expandgeographically, obtaining better diversification of risk. From a policymaker’s perspec-tive, it is crucial to disentangle the effects of a bank’s balance-sheet expansion onits future financial health. Hence, it is imperative to identify whether loan growth isaccompanied by adequate risk management from a microeconomic perspective.

Several recent papers study the relationship between credit growth and the poste-rior performance of banking sector financial indicators from an aggregate perspective(Dell’Ariccia et al. 2008; Demyanyk and Van Hemert 2011; Gorton 2009). However,there are relatively few works that study this important topic from an individual insti-tution’s perspective (e.g., Leaven and Majnoni 2003; Berger and Udell 2004). To date,the intertemporal relation between loan growth and banks’ risk-taking behavior hasnot been investigated (an exception is the study by Foos et al. 2010). To our knowledge,there are no studies of this kind for banks in an emerging economy.

In this paper, we study the relationship between abnormal loan growth and banks’risk-taking behavior using information on individual Colombian banks’ balance sheetsbetween June 1990 and March 2011. We conduct two empirical exercises. On the onehand, we test the relationship between abnormal loan growth and a bank’s survivalprobability using information on individual bank characteristics during the Colombianfinancial crisis of the late 1990s. On the other hand, we test the effect of abnormal loangrowth on bank financial health (solvency, nonperforming loans, and profitability)using cross-sectional time-series data on Colombian financial institutions between1990 and 2011.

Our results show that abnormal loan growth is positively and significantly associatedwith nonperforming loans, and negatively and significantly related with bank solvency.More importantly, we find a significant and positive impact of abnormal loan growthon the conditional probability of failing after a strong negative shock affecting thefinancial system at large. Thus, we provide evidence that during periods of accelerated

123

Loan growth and bank risk 367

credit expansion, banks undertake higher risks that affect negatively the soundness ofthe financial system. This result calls for policy action, such as implementing time-varying minimum capital requirements or imposing levies on abnormal loan growthof individual financial institutions (see, e.g., Weißbach and von Lieres und Wilkau2010; Hofmann 2005).

This paper makes three contributions to the banking and finance literature. First, toour knowledge, this is the first paper to study the relationship between credit growthand banks’ risk-taking behavior using micro data for an emerging, bank-dependenteconomy. Second, our empirical methodology allows studying the time horizon duringwhich the presence of abnormal credit growth affects the financial performance ofbanking institutions in Colombia. And third, this is the first study reporting evidence onthe effect of abnormal loan growth on the conditional probability of failing during timesof financial distress. We benefit from an especially informative dataset of financialinstitutions covering a time span during which two complete financial cycles occurred.

The remainder of the paper is organized as follows. Section 2 presents a brief reviewof related literature. In Sect. 3, we describe the data used for our empirical analysis.In Sect. 4, we present our survival analysis results and show the cross-sectional time-series results. Section 5 concludes.

2 Literature review

Banking credit is an essential source of funding for both firms and households. Ifall agents had identical and complete access to perfect capital markets, the agents’financial structure would be irrelevant for investment decisions because internal andexternal funds would be perfect substitutes. However, there is an extensive economicliterature, based on Fazzari et al. (1988) seminal paper, showing that internal andexternal capital are far from perfectly substitutable. Moreover, depending on individ-ual characteristics, there is a hierarchy for each individual’s access to external sourcesof funding. For instance, differences in cash flow and size significantly affect firms’access to capital markets (see, e.g., Gourio and Miao 2010). In fact, as shown byKashyap et al. (1993), small firms and firms with low levels of cash flow have verylimited access to capital markets and rely almost entirely on banking credit for exter-nal funding. These constraints appear to be stronger in emerging market economieswith underdeveloped capital markets, in which informational imperfections abound.Similarly, differences in wealth and income significantly affect households’ access toexternal funding. Low-income households and households with a low level of wealthaccumulation (young households and poor households) have very limited access tocredit (Hurst and Lusardi 2004).

Given that banking credit is the main source of external funding in most economies(this is especially true for bank-based economies such as Germany, France, Japan,and Colombia), and may be the only source of funding for many firms and house-holds, credit availability and sound credit flows are essential for the development ofinvestment projects and refinancing. Moreover, endogenous growth theory postulatesa positive impact of financial deepening and credit growth on economic activity indi-cators in the long run (Bencivenga and Smith 1991). This hypothesis is tested and

123


confirmed by several empirical studies, including Mishkin (2001), Levine (2001), andBekaert et al. (2001).

Nevertheless, there is a downside of excessive credit expansions. There is abundantempirical literature documenting that many financial crises were preceded by abnormalcredit growth that led to the development of asset price bubbles. In fact, Borio andLowe (2002) and Borio (2009) show that excessive credit growth is the main leadingindicator of a financial crisis in a 12-month horizon.

The 2007–2008 international financial crisis triggered a revitalized interest in under-standing the role of credit in the economy, especially the relation between credit growthand financial crises. Recent literature supports the findings of the early warning tra-dition (e.g., Kaminsky and Reinhart 1999; Goldstein et al. 2000, showing that notonly are financial crises typically preceded by credit booms gone bust (Schularickand Taylor 2012; Jorda et al. 2011), but also that excessive credit growth is the mainpredictor of financial distress over a 12-month time window (Borio and Lowe 2002;Borio 2009; Tenjo and López 2010; Allesi and Detken 2011).

Although there is striking evidence about the impact of abnormal credit growth onfinancial stability from a macroeconomic perspective, there is very limited evidenceof this relation from a microeconomic standpoint.

In an early study employing individual US bank data, Sinkey and Greenwald (1991)show that excessive past loan growth is positively associated with current loan losses.Of special interest is their finding of substantial cross-sectional variation in this link.Leaven and Majnoni (2003) study this link indirectly, using Bankscope data for 45countries. They find that bank provisions behave countercyclically and that there isa negative and significant contemporaneous relation between loan growth and loanlosses. Bikker and Metzemakers (2005) achieve similar results using banking data fora sample of OECD countries during the period 1991–2001. Other papers studying thisrelation for transition and developing economies obtain comparable results (Kraft andJankov 2005; Cotarelli et al. 2005).

All the papers discussed above abstract from the time dimension of the relationbetween loan growth and individual bank performance and several other papers alsostudy this issue. Salas and Saurina (2002) find that loan growth of Spanish savingsbanks is positively and significantly associated with loan losses 3–4 years later. Fooset al. (2010) present a similar study based on Bankscope data for banks in 16 developedeconomies and find that abnormal loan growth leads to an increase in loan losses anda decrease in bank solvency about three years later.

There are no papers studying the intertemporal relation between loan growth andindividual banking performance for emerging market economies. Moreover, there areno studies on the relationship between abnormal loan growth and bank failure for theseeconomies. This paper fills this gap by providing evidence of these links for Colombia.

3 The data

We use balance-sheet data from 64 financial institutions, provided by Colombia’sFinancial Superintendence (the country’s financial oversight body), covering theperiod June 1990–March 2011. Our sample includes 42 banks and 22 financial

123


corporations. All institutions with at least 48 months of reported data and no sig-nificant missing data were included. As of March 2011, this sample covers over 90 %of Colombia’s financial system’s total assets, so it can be regarded as representativeof the Colombian financial system at large.

Our main interest is in identifying the intertemporal effects of loan growth onindividual financial institution performance. Following previous studies (particularlyFoos et al. (2010), abnormal credit growth, ALGi t , is defined as the difference betweeninstitution i’s annual loan growth rate at period t and the median1 of all institutions’annual loan growth rate at t . Defining abnormal credit growth, this way allows us tocontrol for the effect of the prevailing macroeconomic conditions on banks’ willingnessto extend new loans and focus on the cross-sectional differences at each point in time.In robustness tests, we used alternative percentiles of the distribution of loan growth indefining ALGi t . However, as results were qualitatively identical under all definitions,in the remainder of the paper, we report results using the median only.

We conduct two different empirical exercises to answer two different (but related)questions. On the one hand, we test the impact of abnormal loan growth on banks’survival probability using information on individual bank characteristics during thefinancial crisis of the late 1990s. On the other hand, we test the effect of abnormal loangrowth on banks’ financial health (solvency, nonperforming loans, and profitability)using cross-sectional time-series data on Colombian financial institutions between1990 and 2011.

Our two empirical exercises use data with different frequency. The first one usesmonthly data for identifying the time of failure. However, the covariates are measuredfor just one period of time, as our empirical model (the Cox proportional hazardsspecification) uses fixed-time covariates as regressors. The second exercise (cross-sectional time-series models) uses quarterly data.

First, we use a duration or hazard function model to study the time to failure offinancial institutions during the Colombian financial crisis. This methodology enablesus to answer questions relevant to both financial supervisors and financial institutions,such as: Does abnormal credit growth significantly affect the probability of failingafter a financial shock? and Does the amount of time during which an institution’sloans portfolio expanded affect the probability of failing?

Because we are interested in time to failure during the financial crisis, the period ofobservation for the first empirical exercise is the 42 months starting with June 1998,the moment at which the crisis began (see Gómez-González and Kiefer 2009), andDecember 2001, when the system began to recover. Financial data as of June 1998were collected for each of the 54 institutions considered in this empirical analysis. Wehave in total 2,120 bank-time observations.

Following previous studies and theoretical expectations, the following financialratios were considered in investigating time to failure: average abnormal loan growth(ALG j

i ), calculated as the j-months’ (prior to June 1998) average of ALGi t , where jtakes on different values, such as 1 month, 12 months, 24 months, 36 months, and 48

1 We use the median rather than the mean of the growth rate of all institutions at a given point in timebecause our data are highly dispersed and for every given point we find several extreme values.

123


Table 1 Summary statisticsused in the duration analysis

All data correspond topercentages

Variable Mean SD Min Max

ALG48i 0.155 0.517 −0.429 2.713

CAPi 0.150 0.800 −0.568 0.436

SIZEi 0.019 0.020 0.003 0.096

PROFi 0.029 0.021 −0.004 0.086

COMPi 0.652 0.124 0.082 0.832

BANKi 0.693 0.465 0 1

months; solvency (CAPi ), defined as the ratio of tier 1 and tier 2 bank capital to risk-weighted assets; bank size (SIZEi ), defined as the assets of institution i divided by totalsystem assets; gross profit margin (PROFi ); asset composition (COMPi ), defined asthe percentage of assets represented by loans; and a dummy variable (BANKi ) takingthe value 1 when the financial institution is a bank and 0 otherwise. The variableCOMP can be interpreted as controlling for portfolio characteristics potentially relatedto volatility. This set of variables is similar to that used in traditional CAMEL models.Table 1 presents summary statistics of the included financial variables.

Twelve failures were observed between June 1998 and December 2001, a failurerate of 24 %, and an average failure rate of 0.9 % per period.

Second, we use cross-sectional time-series models to study the relation betweenabnormal credit growth and financial institution health using quarterly data betweenJune 1990 and March 2011. We estimate three groups of models: one for solvency, onefor nonperforming loans, and one for profitability. Solvency and nonperforming loanscorrespond to CAPi t and NPLi t , respectively, as defined above. Profitability, ROAi t ,corresponds to the traditional indicator of asset profitability. These sets of estimationsare performed to identify the channels through which abnormal loan growth affectsbanks’ financial health.

Loan portfolio expansions can be financed either by issuing new debt or by issuingnew capital. If these expansions are financed by issuing new capital, the effect ofloan growth on solvency should be negligible or even null. If, on the contrary, creditexpansion is financed with additional debt, the impact of abnormal loan growth onsolvency should be negative and would imply riskier behavior by banks.

As mentioned above, credit expansions do not always imply future loan portfoliodeterioration. If new loans are provided to solvent borrowers with profitable projects,there should be no significant impact of loan growth on the future ratio of nonper-forming loans to total loans. However, if new loans are extended to riskier customersor projects, an increase in this ratio is expected several months in the future.

Finally, we estimate the effect of cumulative abnormal loan growth on bank prof-itability. Banks may be seeking higher returns by expanding their balance sheets andincurring higher risks. We formally test this hypothesis by studying the impact ofabnormal loan growth on asset profitability.

In this second empirical exercise, we exclude solvency as an explanatory variableand include instead an inverse leverage ratio (LEVi t ). We also replace SIZEi t with(SIZE×ALG)i t , the interaction between size and abnormal credit growth. This changeof regressor is made for stationarity purposes (while panel-data unit-root tests do

123


Table 2 Summary statistics used in the cross-sectional time-series analysis

Variable No. observations Mean SD Min Max

NPLi t 3,073 0.081 0.151 0.000 4.313

ROAi t 2,106 0.001 0.004 −0.064 0.025

CAPi t 2,050 0.142 0.132 −2.116 2.656

ALGi t 2,890 0.090 1.177 −1.346 43.175

SIZEi t 3,172 0.026 0.029 0.002 0.207

(SIZE × ALG)i t 2,890 0.166 0.112 −0.332 0.188

BANKi t 5,376 0.6563 0.475 0 1

LEVi t−1 3,172 0.141 0.098 −0.382 0.982

All data correspond to percentages, except for number of observations and the interaction between size andabnormal loan growth

not reject the null hypothesis of unit root for SIZEi t , they do so for the interactionvariable). Additionally, we excluded PROFi and LOANi, and included PROVit, theratio of provisions to total loans.

Table 2 presents summary statistics of the financial variables included in the secondempirical exercise.

4 Empirical results

In this section, we present a quick overview of the two empirical modeling strategiesused in this study, together with the respective estimation results.

4.1 Duration models for studying bank failure

We use a duration or hazard function model to study the time to failure of financialinstitutions. This approach generalizes the more common binary response (logit orprobit) approach by modeling not only the occurrence of failure but also the time tofailure—allowing finer measurement of the effect of different variables on failure.

Most of the papers that apply these models to explain time to bank failure useCox (1972) semi-parametric proportional hazards model; an exception is the work ofCarree (2003), who uses several parametric models to explain bank failure in Russia.The proportional hazards model is the most frequently used because it makes noassumptions about the particular functional form of the baseline hazard, and becauseestimated hazard functions of bank failure in many cases are non-monotonic, thusreducing the number of parametric models that can be used.

We begin by estimating the unconditional (raw: no covariates) survivor functionusing the Kaplan–Meier nonparametric estimator, which takes into account censoreddata. Suppose that bank failure is observed at different moments in time, t1, t2, . . . , tm ,and that di banks fail at time ti . For t ≥ ti ,

S(t) =∏

ti ≤t

[1 − di

Ni

], (1)

123


Fig. 1 Kaplan–Meier survival estimate

where Ni represents the total number of banks still operating at time ti . Figure 1 showsthe estimated survival function for the sample of financial intermediaries included inthis study.

The failure pattern of banks and of other financial institutions during the Colombianfinancial crisis was similar in terms of percentage of entities failing. This suggests thatthe survival functions of both groups might be similar. To corroborate that intuition,tests of equality of the survival functions were done. The results obtained from thelog-rank, Cox, and Wilcoxon tests give us confidence that pooling is appropriate, asthere is no evidence to reject the null hypothesis of equality of the survival functionsof both groups. Therefore, in the remainder of this paper, we treat all the institutionsas one group.

Figure 2 shows the estimated smoothed hazard function for the group of financialinstitutions. Note how the hazard rate of failure is clearly non-monotonic. This behaviorof the baseline hazard reflects events that affected all institutions, such as changes inmacroeconomic conditions during the period of study. Of particular importance, therewas a change in the exchange rate regime in September 1999 from a crawling-pegsystem to a free-floating system. The form of the estimated hazard function showsthat the most commonly used parametric models for the distribution of duration donot seem to be appropriate for modeling the baseline hazard of bank failure in Colombiaduring the period of financial distress.

Our objective is to understand how bank-specific variables affected the conditionalprobability of failure and time to failure after the shocks that initiated the financialcrisis. In ordinary regression models, explanatory variables affect the dependent vari-able by moving its mean. However, in duration models, it is not straightforward to

123


Fig. 2 Smoothed hazard function estimate

see how explanatory variables affect duration and interpretation of the coefficients inthese types of models depends on the particular specification of the model. However,there are two widely used special cases in which the coefficients can be given a partialderivative interpretation: the proportional hazards model and the accelerated lifetimemodel (Kiefer 1988).

Following previous literature on the application of duration models to bank failureand building on the above analysis indicating that conventional candidates for para-metric models are inappropriate, this paper estimates a proportional hazards modelin which no parametric form is assumed for the baseline hazard function. As shownbelow using a specification test, this assumption appears to be appropriate for theproblem of interest.

The model was estimated using the partial likelihood method. Results are presentedin Table 3, which shows the values of the estimated coefficients and their standarderrors. We present four different specifications of the empirical model. Each specifi-cation includes a different subset of the covariates mentioned in the data section ofthis paper. One first important conclusion from Table 1 is that the null hypothesis thatnone of the indicators included in the model is important in explaining the behavior ofduration is clearly rejected in all specifications. This provides evidence for the idea thatfailure of financial institutions during the period of financial distress can be explainedby differences in the financial health and prudence of the institutions.

In all specifications, cumulative abnormal loan growth positively and significantlyaffects the probability of failing. Results shown in Table 1 use an average of 4 years forthe abnormal loan growth variable. Results are qualitatively identical when we usedalternative averages of 1 year, 2 years, or 3 years. Significance is lost when consideringonly point abnormal loan data as of June 1998. This result indicates that it is sustainedabnormal loan growth that matters for the probability of bank failure. The value of thecoefficient is near 0.01 in the different specifications, indicating that a 1 % increase in

123


Table 3 Cox proportional hazards model

Variable Model

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

ALG48i 0.008** 0.009** 0.008** 0.008**

(0.004) (0.004) (0.004) (0.004)

CAPi −0.079* −0.072* −0.125*** −0.122**

(0.042) (0.042) (0.048) (0.052)

SIZEi −0.005* −0.005 −0.005 −0.005

(0.003) (0.003) (0.003) (0.003)

PROFi −0.303 −0.040

(0.189) (0.248)

COMPi −0.054** −0.050

(0.023) (0.033)

BANKi −1.340* −0.587 −1.239 −1.142

(0.739) (0.858) (0.792) (0.992)

Log-likelihood −34.954 −33.801 −32.642 −32.629

Prob > χ2 0.003 0.002 0.001 0.002

Four different empirical specifications are reported (Models 1–4). From the set of regressors, abnormal loangrowth, solvency, size, profitability, asset composition, and the dummy variable distinguishing betweenbanks and other financial institutions, Model 1 includes all of them except for profitability and asset com-position; Model 2 excludes asset composition only; Model 3 excludes profitability. Model 4 includes allregressors. Note that parameter estimates and standard errors (in parentheses) are reported for all variablesincluded in each model. Following regular conventions, ***, **, and * denote significance at the 1, 5, and10 % level, respectively

the average 4-year abnormal credit growth leads to a 1 % increase in the probabilityof failing.

Not surprisingly, in all cases, increases in the capitalization ratio lead to significantreductions in the hazard of failing. Poorly capitalized banks are more likely to fail thanotherwise identical financial institutions. The other included variables have intuitivesigns for their estimated coefficients, but are not always statistically different fromzero at standard significance levels.

As a robustness check of our results, and to confirm that excessive loan growth priorto the crisis exposed banks to higher solvency risk once the crisis hit, we estimateda cross-sectional probit model relating survival during the crisis to pre-crisis loangrowth. The results obtained under the probit model confirm those obtained under theCox hazard function specification. Specifically, they show that abnormal loan growthin the pre-crisis period significantly affected a bank’s failure probability during thecrisis. Table 4 shows results for the probit model including all five covariates.

4.2 Cross-sectional time-series analysis

In this section, we test the effect of abnormal loan growth on bank financial health (sol-vency, nonperforming loans, and profitability), using quarterly cross-sectional time-

123


Table 4 Cross-sectional probitmodel

Note that parameter estimatesand standard errors (inparentheses) are reported for allvariables included in eachmodel. Following regularconventions, ***, **, and *denote significance at the 1, 5,and 10 % level, respectively

ALG48i 0.0231782**

(0.0116581)

CAPi 0.0203378

(0.0541325)

SIZEi −0.0032339

(0.0024216)

BANKi −0.1932816

(0.7764351)

PROFi −0.2882576*

(0.1750316)

CONSTANT −0.0054868

(1,137646)

Pseudo − R2 0.4139

series data between 1990 and 2011. The number of cross-sectional units is relativelysmall, whereas the number of time periods is relatively large. More importantly, it isexpected that the panel’s time dimension grows faster than the cross-sectional dimen-sion. In this context, and contrary to traditional panel data settings, it appears rea-sonable to specify a common conditional mean function across the units, with het-erogeneity taking the form of different variances rather than shifts in the means. Theasymptotic theory here is with respect to time going to infinity, while the number ofcross-sectional units is fixed.

Correlations across financial institutions are also very relevant in our dataset, asthese institutions have established relations in different financial markets (e.g., moneymarkets). To summarize, rather than using traditional panel data techniques, such asfixed effects, random effects, or dynamic panel data models, for our quarterly datasample we estimate a cross-sectional time-series model by feasible generalized leastsquares (FGLS) and estimate the structure of the variance-covariance matrix of theerror terms.

Our empirical model is specified as follows:

yit = α +8∑

s=1

βsALGi,t−s + γ LEVi,t−1 + φPROVi,t−1

+υ(SIZE × ALG)i t + δBANKi t + εi t , (2)

where yit represents the dependent variable, which, depending on the specification,may be the capital ratio of institution i at time t , the ratio of nonperforming loans tototal loans of institution i at time t , or profitability of bank i at time t .

The variance–covariance matrix of the error terms is specified to account for het-eroskedasticity across panels (the variance of each panel differs) and to account forautocorrelation of order one specific to each panel.

123


Table 5 FGLS estimation usingquarterly data 1990:2–2011:1

Dependent variables are NPLi t(A), CAPi t (Panel B), andROAi t (Panel C). Note thatparameter estimates andstandard errors (in parentheses)are reported for all variablesincluded in each model.Following regular conventions,***, **, and * denotesignificance at the 1, 5, and 10 %level, respectively

Panel ANPLi t

Panel BCAPi t

Panel CROAi t

LONGRUNALGi t 0.0001555*** −0.00773*** −0.00002

(0.0000306) (0.0029265) (0.0000359)

(SIZE × ALG)i t −0.0000002*** −0.00001* 0.00000

(0.0000001) (0.0000053) (0.0000000)

BANKi t −0.0071296*** 1,77713*** 0.01077*

(0.0015757) (0.1490219) (0.0056214)

LEVi t−1 −0.0005747*** 0.38360*** 0.00201***

(0.0001233) (0.0117778) (0.0001131)

PROVi t−1 0.0144878*** 0.09790*** −0.00131***

(0.0001863) (0.0160990) (0.0001594)

CONSTANT 0.0315142*** 5,93266*** −0.01576**

(0.0026214) (0.2255836) (0.0060882)

Waldχ2(12)

6172.840 2621.080 380.290

Prob > χ2 0.000 0.000 0.000

Our main interest is in estimating the sign and magnitude of β, which stands for thelong-run effect of abnormal credit growth on the dependent variable. This coefficientis estimated as the sum of the βs’s and its variance is estimated using the delta method.

Table 5, Panel A presents a summary of our main findings and results when usingnonperforming loans as the dependent variable. The variables included as regres-sors are jointly significant in explaining deviations of NPLi t around its mean, asindicated by the Wald statistic. Every individual financial variable included in theregression is significant at the 1% level. Our main finding is that sustained increasesin abnormal loan growth lead to significant increases in the ratio of nonperform-ing loans to total loans. Specifically, a 1 % point upsurge in the 2-year abnormalloan growth leads to a 1.6 % percentage point increase in the ratio of nonperformingloans.

This result supports the hypothesis that when banks’ lending increases over a longperiod of time, an important portion of new loans are extended to clients withoutcredit history or that under normal circumstances would have been rejected. Therefore,banks with sustained periods of abnormal loan growth frequently take higher risks,and eventually experience loan-portfolio deterioration. It is noteworthy that the effectof abnormal loan growth on nonperforming loans differs according to bank size: largerbanks tend to experience less loan-portfolio weakening than do smaller banks. Thesigns of all other included variables are as expected.

Table 5, Panel B shows results of estimating Eq. (2) using solvency as the dependentvariable. As expected, abnormal credit growth exerts a significantly negative influenceon bank solvency in the long run. This result shows that, on average, banks do notincrease their capital buffers in accordance with the additional risk they are undertakingwhen significant credit expansions occur. In line with García-Suaza et al. (2012), largerbanks further reduce capital buffers when experiencing abnormal loan growth for along period of time.

123


Table 5, Panel C shows estimation results when using ROAi t as the dependentvariable. A very interesting result is observed. Note that although abnormal creditgrowth may have a positive impact on short-run profitability, in the long run the effectis null. A very similar result is obtained when interest rate spreads, measured as thedifference between interest income and interest expense, are used as the dependentvariable. Hence, this result, together with those obtained when using solvency andthe ratio of nonperforming loans as the dependent variables, supports the hypothesisof bank short-sightedness. In other words, when banks expand their balance sheetssignificantly, seeking either higher immediate returns or a larger market share, theydo not appropriately hedge against the higher risks they are incurring.

Prudential policies may need to be implemented. One possibility is to impose indi-vidual additional capital charges to assure that banks internalize the potential costs oftheir riskier behavior when expanding their balance sheets. Another alternative wouldbe to impose levies on the origination of new credit in the presence of abnormal loangrowth. For instance, in March 2013 Peru implemented an individual marginal reserverequirement based on credit growth.

5 Conclusion

This study provides new evidence on the relationship between abnormal loan growthand banks’ risk-taking behavior using data from a rich panel of Colombian financialinstitutions.

We conduct two empirical exercises. On the one hand, we investigate the impactof abnormal loan growth on bank survival probability using information on individualbank characteristics during the Colombian financial crisis of the late 1990s. On theother hand, we test the effect of abnormal loan growth on bank financial health usingcross-sectional time-series data on Colombian financial institutions between 1990 and2011.

Our main findings support the hypothesis of banks’ intertemporal short-sightedness.We show that abnormal loan growth during a sustained period led to reductions inbanks’ capital ratios and to increases in the ratio of nonperforming loans to totalloans. Although abnormal credit growth may have a positive impact on short-runprofitability, in the long run the effect is null. Significant credit expansions do notgenerate corresponding increases in bank safety margins or in long-run profitability.

Concordant with these results, we also show that during the Colombian financialcrisis of the late 1990s, sustained abnormal loan growth was one of the most significantvariables in explaining observed differences in the process of bank failure.

These findings suggest that additional regulatory measures should be undertaken toassure financial soundness when abnormal loan growth is observed. One possibility isto impose individual additional capital requirements to assure that banks internalize thepotential costs of their riskier behavior when expanding their balance sheets. Anotheralternative is to impose levies on the origination of new credit in the presence ofindividual abnormal loan growth.

Acknowledgments We gratefully acknowledge comments received from Fernando Tenjo, Leonardo Vil-lar, Juan M. Ramírez, Cristina Ferández, Luis A. Zuleta, Luis Melo, Hernán Rincón, Jair Ojeda, and partici-

123


pants at the Banco de la Republica’s Economics Workshop and Fedesarrollo’s workshop. We are especiallygrateful to Professor Markus Schmid and anonymous referees for superb comments and suggestions madeto a previous version of this article.

References

Allesi, L., Detken, C.: Quasi real time early warning indicators for costly asset price boom/bust cycles: arole for global liquidity. Eur. J. Polit. Econ. 27, 520–533 (2011)

Altunbas, Y., Gambacorta, L., Marques-Ibanez, D.: Bank risk and monetary policy. J. Financ. Stab. 6,121–129 (2010)

Altunbas, Y., Gambacorta, L., Marques-Ibanez, D.: Do bank characteristics influence the effect of monetarypolicy on bank risk? Econ. Lett. 117, 220–222 (2012)

Bekaert, G., Harvey, C., Lundblad, C.: Does financial liberalization spur growth? NBER Working Paper8245 (2001)

Bencivenga, V., Smith, B.: Financial intermediation and endogenous growth. Rev. Econ. Stud. 58, 195–209(1991)

Berger, A.N., Udell, G.F.: The institutional memory hypothesis and the procyclicality of bank lendingbehavior. J. Financ. Intermed. 13, 458–495 (2004)

Bikker, J., Metzemakers, P.: Bank provisioning behaviour and procyclicality. J. Int. Financ. Mark. Inst.Money 15, 141–157 (2005)

Borio, C.: Ten propositions about liquidity crises. Bank for International Settlements Working Paper 293(2009)

Borio, C., Furfine, C., Lowe, P.: Procyclicality of the financial system and financial stability: issues andpolicy options”, Bank for International Settlements Working Paper 1 (2001)

Borio, C., Lowe, P.: Asset prices, financial and monetary stability: exploring the nexus. Bank forInternational Settlements Working Paper 114 (2002)

Carree, M.A.: A hazard rate analysis of Russian commercial banks in the period 1994–1997. Econ. Syst.27, 255–269 (2003)

Cecchetti, S.G., Kharroubi, E.: Reassessing the impact of finance on growth. Bank for International Settle-ments Working Paper 381 (2012)

Cotarelli, C., Dell’Ariccia, G., Vladkova-Hollar, I.: Early birds, late risers, and sleeping beauties: Bankcredit growth to the private sector in Central and Eastern Europe and in the Balkans. J. Bank. Finance29, 83–104 (2005)

Cox, D.R.: Regression models and life-tables. J. R. Stat. Soc. B 34, 187–220 (1972)Dell’Ariccia, G., Igan, D., Laevem, L.: Credit booms and lending standards: evidence from the subprime

mortgage market. IMF Working Paper 08/206 (2008)Demyanyk, Y., Van Hemert, O.: Understanding the subprime mortgage crisis. Rev. Financ. Stud. 24, 1848–

1880 (2011)Fazzari, S., Hubbard, G., Petersen, B.: Finance constraints and corporate investment. Brookings Papers

Econ. Activity 1, 141–195 (1988)Foos, D., Norden, L., Weber, M.: Loan growth and riskiness of banks. J. Bank. Finance 34, 2929–2940

(2010)García-Suaza, A., Gómez-González, J.E.: The competing risks of acquiring and being acquired: evidence

from Colombia’s financial sector. Econ. Syst. 34, 437–449 (2010)García-Suaza, A., Gómez-González, J.E., Murcia Pabón, A., Tenjo-Galarza, F.: The cyclical behavior of

bank capital buffers in an emerging economy: size does matter. Econ. Model. 29, 1612–1618 (2012)Goldstein, M., Kaminsky, G., Reinhart, C.: Early Warning System: An Assessment of Vulnerability. MPRA

Paper 24579. University Library of Munich, Germany (2000)Gómez-González, J.E., Kiefer, N.M.: Bank failure: evidence from the Colombian financial crisis. Int. J.

Bus. Finance Res. 3, 15–32 (2009)Gorton, G.: Information, liquidity and the (ongoing) panic of 2007. Am. Econ. Rev. 99, 567–572 (2009)Gourio, F., Miao, J.: Firm heterogeneity and the long-run effects of dividend tax reform. Am. Econ. J.

Macroecon. 2, 131–168 (2010)Hofmann, B.: Procyclicality: the macroeconomic impact of risk-based capital requirements. Financ. Mark.

Portfolio Manag. 19, 179–200 (2005)

123


Holmstrom, B., Tirole, J.: Financial intermediation, loanable funds, and the real sector. Q. J. Econ. 112,663–692 (1997)

Hurst, E., Lusardi, A.: Household wealth, and entrepreneurship. J. Polit. Econ. 112, 319–347 (2004)Jorda, O., Schularick, M., Taylor, A.: Financial crises, credit booms, and external imbalances: 140 years of

lessons. IMF Econ. Rev. 59, 340–378 (2011)Kaminsky, G., Reinhart, C.: The twin crises: the causes of banking and balance-of-payments problems.

Am. Econ. Rev. 89, 473–500 (1999)Kashyap, A., Stein, J., Wilcox, D.: Monetary policy and credit conditions: evidence from the composition

of external finance. Am. Econ. Rev. 83, 78–98 (1993)Kiefer, N.M.: Economic duration data and hazard functions. J. Econ. Lit. XXVI, 646–679 (1988)Klein, J.P., Moeschberger, M.L.: Survival Analysis. Techniques for Censored and Truncated Data, 2nd edn.

Springer, Berlin (2003)Kraft, E., Jankov, L.: Does speed kill? Lending booms and their consequences in Croatia. J. Bank. Finance

29, 105–121 (2005)Leaven, L., Majnoni, G.: Loan loss provisioning and economic slowdowns: too much, too late. J. Financ.

Intermed. 12, 178–197 (2003)Levine, R.: International financial liberalization and economic growth. Rev. Int. Econ. 9, 688–702 (2001)Mendoza, E., Bianchi, J.: Overborrowing, financial crises and ‘macro-prudential’ taxes. In: Proceedings,

Federal Reserve Bank of San Francisco (2010)Mishkin, F.: Financial policies and the prevention of financial crises in emerging market economies. World

Bank Working Paper 2683 (2001)Salas, V., Saurina, J.: Credit risk in two institutional regimes: spanish commercial and saving banks. J.

Financ. Serv. Res. 22, 203–224 (2002)Schularick, M., Taylor, A.: Credit booms gone bust: monetary policy, leverage cycles, and financial crises,

1870–2008. Am. Econ. Rev. 102, 1029–1061 (2012)Sinkey, J., Greenwald, M.: Loan loss experience and risk-taking behavior at large commercial banks. J.

Financ. Serv. Res. 5, 43–59 (1991)Tenjo, F., López, M.: Early warning indicators for Latin America. Ensayos Sobre Política Económica 63,

232–259 (2010)Weißbach, R., von Lieres, C., Wilkau, : Economic capital for nonperforming loans. Financ. Mark. Portfolio

Manag. 24, 67–85 (2010)

Author Biographies

Juan Sebastián Amador is a graduate student at the Universidad Javeriana, Bogotá, Colombia. He holdsa B.A. degree in economics from the same university, and works at the Department of MacroeconomicProgramming and Inflation of the Central Bank of Colombia. He obtained the seventh place in the eco-nomics ECAES exam (national examination for economics graduates) in 2010.

José E. Gómez-González holds a Ph.D. degree in economics from Cornell University. Currently heworks as a Senior Economist of the Colombian Central Bank’s Research Department. He has publishedextensively in various economics journals, including Economic Modelling, Economic Systems, Emerg-ing Markets Finance and Trade, International Economic Journal, Financial Markets and Portfolio Man-agement, and Macroeconomics and Finance in Emerging Market Economies. He has lectured financialeconomics and econometrics at different Colombian universities, such as Universidad de los Andes, Uni-versidad Javeriana, Universidad del Rosario, and Universidad Nacional de Colombia.

Andrés Murcia Pabón holds an M.Phil. in economics from Université de Toulouse. Currently he worksas Senior Economist of the Colombian Central Bank’s Monetary and International Reserves Division. Hehas published in various economics journals including Economic Modelling, Financial Markets and Port-folio Management, and Ensayos Sobre Política Económica. He has lectured financial economics at Uni-versidad de los Andes.

123

Documents

Loan growth and bank risk: new evidence