I. Introduction

With the recent outbreak of high-profile financial and currency crises throughout the world and in emerging market economies in particular, the issue of proper regulation of the banking system has become critical to the policy discussion. On the one hand, it is recognized that less government interference and hence greater operation of market forces are conducive to more rapid development of the financial system which, in turn, tends to generate substantial economic benefits resulting in higher economic growth (Levinc, 1997). On the other hand, it is also increasingly apparent that as financial systems are liberalized and rely more and more on market forces, they also become more vulnerable and subject to greater instability (Demirguc-Kunt and Detragiache, 1998; Caprio and Honohan, 1999). For this reason, regulation and supervision of financial institutions must be stronger and more carefully designed in order to reduce the instability of the system and minimize the probability of crisis.

One crucial issue in banking system regulation is whether the government should provide a safety net for depositors, and if so, what form it should take. Starting with the classic work by Diamond and Dybvig (1983), there is a strong argument in favor of establishing a liquidity insurance scheme for depositors in order to prevent the disruptive and costly effects of the types of depositor runs that are endemic to a fractional reserve banking system. Of course, the cost of such a scheme is moral hazard, as insured depositors no longer have a strong incentive to monitor the behavior of banks where they hold their deposits, and thus banks may also have an incentive to assume greater risks. Therefore, the design of a deposit insurance scheme faces a tradeoff between the reduction of risk coming from bank runs and the additional risk induced by moral hazard.

Recently there has been significant interest in the literature in examining how depositors choose among different banks, and whether this choice reflects market discipline, whereby depositors would effectively monitor bank management and reward well-managed banks for good behavior while punishing poorly managed banks for bad behavior. If depositor behavior is found to be consistent with market discipline, then this suggests that there is limited moral hazard in the system, an indication that depositors do not perceive their deposits to be fully protected.

In this paper we use Colombia as a case study to explore depositor behavior and market discipline, and to address an apparent contradiction that has arisen in the empirical literature. On the one hand, cross-country analysis (Demirguc-Kunt and Detragiache, 1999) finds that countries with explicit and more extensive deposit insurance tend to be more fragile and, hence, more prone to banking crises, which is consistent with deposit insurance leading to an increase in moral hazard and a corresponding breakdown in market discipline. On the other hand, individual country studies of depositor behavior have been generally supportive of market discipline, even in cases of explicit and relatively extensive deposit insurance.

Although some explanations related to the design and/or credibility of the deposit insurance system may account partially for the results of the individual country studies, we also argue that the tests conducted to date have tended to be relatively weak and perhaps have biased the results in favor of finding market discipline. Thus, we set out to strengthen the traditional market discipline test by including a set of controls that may reflect more completely the set of criteria used by depositors to choose one bank over another. Also, as argued by Calomiris and Powell (2000), a true test for market discipline should also involve a second step in which it is determined whether banks effectively respond to the signals provided by depositors. We also build on this idea, showing how this test may be developed further.

Colombia constitutes an interesting case study for analyzing both market discipline and depositor behavior in general. Throughout our period of analysis (1985-99) Colombia had an explicit deposit insurance system, with several design features that might have limited moral hazard, but others that may have increased it. Anecdotal evidence of a deposit run on an individual bank that occurred in mid-1999 raises questions about the incentives of depositors and the quality of the information they use to assess bank performance. Thus, whether Colombian depositors discipline their banks is an open question. Furthermore, throughout the 1990s Colombia underwent rapid change starting with a liberalization program at the beginning of the decade that increased flexibility of interest rates and opened the financial system to greater competition. A significant number of new banks, both foreign and domestic, entered the market and quickly acquired market share (Barajas, Steiner, and Salazar, 2000). However, after several years of extremely high growth and profitability, the banking system began to show signs of fragility and distress toward the end of the decade, as numerous banks and other financial institutions encountered mounting non-performing loans and diminishing solvency. Therefore, it is crucial to understand depositor behavior during the initial expansion, when new banks entered the market, and whether depositors played a positive role in disciplining banks.

As we will show in detail, our empirical analysis finds strong evidence that deposit growth depends on bank performance fundamentals, even after strengthening the test by incorporating a more complete set of bank-specific controls. Furthermore, we also find that banks respond to signals from depositors in a manner consistent with market discipline. Interestingly, the response from banks seems to be asymmetric, occurring only when they perceive they are being punished by depositors, and does not appear to extend to the cases in which fundamentals are the weakest.

These results suggest the existence of market discipline, with depositors taking into account bank fundamentals and sending signals to banks, who then adjust key fundamentals accordingly. Thus, moral hazard appears to be limited in spite of having an explicit deposit insurance system. This may be due to some design features of Colombia’s insurance scheme, which is compulsory, has co-insurance and risk-weighted premiums, and with relatively small coverage which is declining continuously in real terms. We also suggest that the observed appropriate response of banks could also be the result of effective regulatory oversight rather than market discipline per se. It may be the case that banks are simply adjusting their fundamentals in order to comply with the regulations, rather than as a response to deposit outflows.

The paper is divided into six sections, including this introduction. In the second section we review the current literature regarding depositor behavior in the presence of deposit insurance. In the third section we discuss the Colombian context, describing its deposit insurance scheme and briefly discussing the recent case of a rumor-induced bank run, a good example that depositors may react to non-fundamental news. In the fourth section we present our estimation approach, showing how we build upon the standard tests for market discipline. We also discuss the idea that disciplining cannot be detected exclusively through the behavior of depositors, as it is also important to assess how banks react to the actions of depositors. In the fifth section we present the results of three sets of econometric estimations. The first one looks at the determinants of the rate of growth of deposits, the second at the determinants of interest rates, and the third at the reaction of banks to changes in depositor behavior. In the sixth section we conclude.

II. Deposit insurance and depositor behavior

When deciding on a deposit insurance system, policymakers face a tradeoff between two types of risk: (1) the risk of non-fundamental or inefficient deposit runs (Diamond and Dybvig, 1983), vs. (2) moral hazard. If policymakers believe that deposit runs are not a significant source of instability², and that moral hazard is the predominant risk, then they should opt to eliminate deposit insurance, as in the extreme case of New Zealand, where depositor protection is explicitly denied and therefore the system relies entirely on market discipline and transparency (García, 1999 and 2000). If, on the contrary, policymakers believe that deposit runs are extremely dangerous and likely, and that moral hazard is not significant, then they should choose the opposite extreme, a full explicit guarantee. In practice, both extremes are very rare, with the majority of countries in the world adopting some limited form of deposit insurance, and therefore assuming a certain level of additional moral hazard in exchange for greater protection against bank runs. They also may incorporate certain design features into the scheme that limit the propensity for moral hazard and thus risk-taking by banks (Demirguc-Kunt and Huizinga, 1999).

In the end, which type of risk dominates? A recent study (Demirguc-Kunt and Detragiache, 1999) sought to answer this question for a broad sample of countries exhibiting varying extents of coverage in their deposit insurance systems. Using the capacity to predict financial crisis as the prime criterion, the authors investigate what happens to bank fragility when the extent and coverage of deposit insurance increases. Based on a sample of 61 countries for 1980-97, they find that bank fragility increases as the deposit insurance scheme becomes becomes more explicit and extensive, an indication that moral hazard may be dominating over whatever stabilizing effects deposit insurance has on the risk of bank runs³.

However, recent empirical studies of depositor behavior in individual countries have given support for the opposite result: the presence of market discipline in different countries and with varying types of deposit insurance systems. Using panel data estimation on individual bank balance sheet information, these studies examine market discipline by testing whether a significant relationship arises between depositor behavior and bank performance and management indicators, or “fundamentals.” Park and Peristiani (1998) show evidence of market discipline in the U.S. thrift industry throughout the 1980s, as depositors were shown to demand a higher interest rate and deposit growth was shown to be lower as banks’ activities became riskier. Regarding developing countries, Martínez Pería and Schmukler (1999) consider the banking systems of Argentina, Chile, and Mexico, and find support for market discipline in all three cases, as the growth of deposits was significantly related to bank fundamentals, once controlling for systemic and macroeconomic variables also affecting demand for deposits. This result is shown to hold even in the case of small, insured depositors. Schumacher (1996, 2000) finds evidence of market discipline in Argentina during the 1994–95 “Tequila” crisis. Finally, for Argentina during the 1990s, Moore (1997) finds depositor growth to be significantly linked to bank fundamentals⁴, and Calomiris and Powell (2000) find deposit interest rates as well as deposit growth to be significantly related to bank fundamentals.

Considering the theoretical and cross-country empirical backing for moral hazard when deposit insurance is in place, this strong support for market discipline in individual country studies is somewhat puzzling. In fact, it has been argued that, even in the absence of explicit deposit insurance, depositors and bank managers may often tend to behave as if their deposits were insured, expecting a rescue if their bank were to experience serious difficulties. However, the results from country studies suggest the opposite, that even though deposits are explicitly insured, depositors behave as if they were not, refusing to trust the deposit guarantee fully. Furthermore, from the Demirguc Kunt-Detragiache (1999) study, the Mexican and Chilean deposit insurance systems exhibit some of the characteristics making them even more prone to fragility and hence, to moral hazard⁵: both systems are explicit and un-funded but callable, they cover foreign-currency deposits, and are government-managed. Additional risk elements are introduced in Mexico by having unlimited coverage, and in Chile by being funded entirely by the government. Argentina, on the other hand, appears to be a case in which market discipline would be more plausible, as there is no explicit safety net for depositors, and the currency board arrangement practically rules out any scope for the central bank to serve as a lender of last resort.

There are three possible explanations that may account for the empirical results supporting market discipline even in the presence of extensive safety nets for depositors. First, if the tests have been properly specified, then they would suggest that the moral hazard problem in depositor behavior is not very important. This may be because deposit insurance is not credible, or because it is designed in such a way to limit moral hazard. However, this conflicts with the cross-country results indicating greater fragility, and it is not clear how fragility can increase if market discipline is left intact.

A second possibility is that deposit insurance reduces but does not eliminate market discipline. A recent cross-country study by Demirguc-Kunt and Huizinga (1999) conducts market discipline tests for a number of developed and developing countries, and then pools the bank and country-specific data in order to test whether there is an impact of deposit insurance on market discipline. They find some evidence that the existence of deposit insurance lowers the responsiveness of deposit interest rates to changes in bank liquidity, an indication that market discipline is weakened. However, the study was not successful in finding significant market discipline effects in regressions explaining the growth of deposits. Therefore, if market discipline does exist, then it is most likely operating through the level of interest rates rather than through the demand for deposits.

Although these explanations may partly account for why market discipline is not rejected in countries with deposit insurance, we consider a third possibility as well: that the individual country tests have not been very strong, and therefore may be biased in favor of finding market discipline. We explore this in the present study, with particular reference to Colombia, by strengthening the test and incorporating control variables that reflect the return to depositors as well as other non-fundamental characteristics which may lead depositors to prefer a certain type of bank over another.

Finally, as argued in Calomiris and Powell (2000), even if there is evidence that depositors choose banks according to their fundamentals, it does not necessarily follow that market discipline exists. It must also be true that banks are effectively “disciplined”, in that they react appropriately by adjusting their fundamentals in response to the signals provided by depositors. Calomiris and Powell suggest an indirect test based on the behavior of deposit interest rates, and we build upon it by testing the direct response of individual fundamental variables to what we define as “fundamental” deposit withdrawals.

III. Deposit insurance in Colombia and a case of a non-fundamental bank run

Following the financial crisis of the early 1980s, FOGAFIN (Fondo de Garantiás de Instituciones Financieras, or Financial Institutions Guarantee Fund) was created in 1985 (Law 117). One of its key obligations was to develop a deposit insurance scheme, whose current main features are (see Table 1):

Table 1.

Colombia’s Deposit Insurance Scheme

Source: Superintendencia Bancaria de Colombia

Table 1.

Colombia’s Deposit Insurance Scheme

		1994	1995	1996	1997	1998	1999
Characteristics:
	Premium (% of insured deposits)	0.15	0.15	0.15	0.15	0.15	0.30
	Refund for “Investment Grade”					50%	300%
	Refund for “Good Grade”					25%	25%
	Coverage limit (col$mill)	10	10	10	10	10	10
	Deductible	25%	25%	25%	25%	25%	25%
Coverage indicators:
		(Percentage of total financial system)
Eligibility
	By size: Number of accounts smaller than the coverage limit	98.61%	98.33%	98.00%	97.79%	97.58%	97.87%
	By type: Number of eligible accounts		98.56%	92.70%	95.15%	97.95%	98.12%
Value covered
	Total insured deposits	48.16%	45.07%	41.51%	40.19%	37.72%	34.63%

Source: Superintendencia Bancaria de Colombia

View Table

Table 1.

Colombia’s Deposit Insurance Scheme

		1994	1995	1996	1997	1998	1999
Characteristics:
	Premium (% of insured deposits)	0.15	0.15	0.15	0.15	0.15	0.30
	Refund for “Investment Grade”					50%	300%
	Refund for “Good Grade”					25%	25%
	Coverage limit (col$mill)	10	10	10	10	10	10
	Deductible	25%	25%	25%	25%	25%	25%
Coverage indicators:
		(Percentage of total financial system)
Eligibility
	By size: Number of accounts smaller than the coverage limit	98.61%	98.33%	98.00%	97.79%	97.58%	97.87%
	By type: Number of eligible accounts		98.56%	92.70%	95.15%	97.95%	98.12%
Value covered
	Total insured deposits	48.16%	45.07%	41.51%	40.19%	37.72%	34.63%

Source: Superintendencia Bancaria de Colombia

(i) A guarantee that depositors will receive their funds when a financial institution intervened by the Banking Superintendency is unable honor its obligations to them. All financial institutions registered in FOGAFIN and under the tutelage of the Banking Superintendency are required to purchase deposit insurance: commercial banks, savings and loans (CAV’s), financial corporations, commercial finance companies, leasing companies and capitalization societies.

(ii) Insured liabilities are demand deposits, CD’s, savings accounts, UPAC accounts⁶, receipts payable, tax collection services and capitalization titles. FOGAFIN will cover 75% of the amount deposited up to a maximum coverage of col$10’000.000 per account, regardless of the number of persons owning the account. In each institution, and regardless of the number of accounts, one person is only insured up to col$10.000.000, while accounts in different institutions are insured separately. Insurance only covers deposits payable in Colombia, and regarding interest-earning liabilities, insurance covers principal, monetary correction and regular interests.

(iii) Yearly premiums amount to 0.3% of all liabilities (and 0.3% of reserves in the case of capitalization societies), paid quarterly, and are differential among institutions, through a system of reimbursements based on efficiency and solvency indicators. At the end of each year, financial institutions are reimbursed 50% of premiums paid if their CDs and Time Savings Deposits have been granted “Investment Grade” by a specialized grading agency, and 25% if they have been awarded a “Good Grade⁷.”

One particular issue to be highlighted is the effective coverage of the scheme. The bottom portion of Table 1 describes qualitative and quantitative aspects of deposit insurance coverage in Colombia. Almost all financial system accounts – just under 98% in 1999 – are smaller than the maximum coverage of col$10 million and are thus fully covered⁸. Also, almost all accounts –just over 98% in 1999 – were of types covered by the insurance scheme. However, in terms of value, overall coverage is much smaller – 35% in 1999 – owing to the existence of a small number of very large accounts. Furthermore, this percentage has been declining continuously throughout the nineties reflecting the fact that the coverage limit has been fixed in nominal terms while annual inflation has been between 10 and 15%.

The effective coverage can also be analyzed on an ex post basis. In Table 2 we can observe actual payments made by FOGAFIN in 1996-2000, following a period of enormous financial distress. Payments are expressed as a percentage of insured liabilities. The enormous variance among different institutions partly reflects the different composition of deposits (by size and type) across institutions. In one case, deposit insurance covered as little as 4% of an institution’s liabilities, while in another it covered almost 53%.

Table 2.

Colombia: Actual Insurance Payments to Depositors (1996-2000)

Source: Fogafin

Table 2.

Colombia: Actual Insurance Payments to Depositors (1996-2000)

Intervened Institution	Amount paid/Insured liabilities
COMMERCIAL BANKS
BANCO SELFIN S.A	9.9%
BANCO ANDINO	5.9%
BANCO PACIFICO S. A.	5.7%
FINANCIAL CORPORATIONS
CORFIPACIFICO	10.6%
COMMERCIAL FINANCE COMPANIES
LA FORTALEZA S.A.	40.6%
BERMUDEZ Y VALENZUELA S.A.	32.8%
C.F.C.PACIFICO	19.6%
FINDESARROLLO	16.7%
LEASING COMPANIES
LEASING SELFIN S.A	20.0%
LEASING FINANCIERA CAUCA S.A. (FINANCAUCA)	16.2%
LEASING PATRIMONIO	9.2%
FINANCIERA ARF1N	4.4%
CAPITALIZATION SOCIETIES
CAPITALIZA DORA GRANCOLOMBIANA	52.8%

Source: Fogafin

View Table

Table 2.

Colombia: Actual Insurance Payments to Depositors (1996-2000)

Intervened Institution	Amount paid/Insured liabilities
COMMERCIAL BANKS
BANCO SELFIN S.A	9.9%
BANCO ANDINO	5.9%
BANCO PACIFICO S. A.	5.7%
FINANCIAL CORPORATIONS
CORFIPACIFICO	10.6%
COMMERCIAL FINANCE COMPANIES
LA FORTALEZA S.A.	40.6%
BERMUDEZ Y VALENZUELA S.A.	32.8%
C.F.C.PACIFICO	19.6%
FINDESARROLLO	16.7%
LEASING COMPANIES
LEASING SELFIN S.A	20.0%
LEASING FINANCIERA CAUCA S.A. (FINANCAUCA)	16.2%
LEASING PATRIMONIO	9.2%
FINANCIERA ARF1N	4.4%
CAPITALIZATION SOCIETIES
CAPITALIZA DORA GRANCOLOMBIANA	52.8%

Source: Fogafin

In Table 3 we compare the main characteristics of the Colombian deposit insurance scheme with those of several industrialized countries and a world average (Beck, 2000). According to the Demirguc-Kunt and Huizinga (1999) analysis, there appear to be several features of Colombian deposit insurance that would increase the likelihood of moral hazard: the system is explicit, funded, and managed by the government. On the other hand, other features that may work in the opposite direction: the coverage limit is relatively small (below the world average of 3 times GDP per capita) and declining in real terms, the scheme is privately funded, membership is compulsory, there is co-insurance by depositors, and premiums are risk-adjusted. Whether or not deposit insurance has prevented depositors from exerting market discipline in Colombia is therefore an open question, one that has to be addressed empirically.

Table 3.

Cross-country comparison of deposit insurance schemes

^*Minimum requirements.

^**As of July 2000, this is around USD4600, or 2 to 3 times per capita GDP.

^***Colombia’s financial system does not allow for foreign currency deposits.

Source: Beck (2000).

Table 3.

Cross-country comparison of deposit insurance schemes

	Germany	EU*	US	World average (68 countries)	Colombia
Coverage limit	30% of equity	ECU$20000	US$100000	3 times per capita GDP	col$10 million**
Coinsurance	No	10%	No	Yes, in 17	25%
Foreign currency deposits covered	Yes	Can be excluded	Yes	Yes, in 48	n.a.***
Interbank deposits Covered	No	No	Yes	Yes, in 18	No
Funding	Funded, but additional funds callable	Not regulated	Funded	Funded, in 58 cases	Funded
Sources of Funding	Banks only	Not regulated	Joint	Private: 15 Joint: 51 Public: 1	Private
Management	Private	Not regulated	Public	Private: 11 Joint: 24 Public: 33	Public
Membership	Voluntary	Compulsory	Compulsory	Compulsory in 55 cases	Compulsory
Risk adjusted Premiums	Yes	Not regulated	Yes	Yes, in 21 cases	Yes

^*Minimum requirements.

^**As of July 2000, this is around USD4600, or 2 to 3 times per capita GDP.

^***Colombia’s financial system does not allow for foreign currency deposits.

Source: Beck (2000).

View Table

Table 3.

Cross-country comparison of deposit insurance schemes

	Germany	EU*	US	World average (68 countries)	Colombia
Coverage limit	30% of equity	ECU$20000	US$100000	3 times per capita GDP	col$10 million**
Coinsurance	No	10%	No	Yes, in 17	25%
Foreign currency deposits covered	Yes	Can be excluded	Yes	Yes, in 48	n.a.***
Interbank deposits Covered	No	No	Yes	Yes, in 18	No
Funding	Funded, but additional funds callable	Not regulated	Funded	Funded, in 58 cases	Funded
Sources of Funding	Banks only	Not regulated	Joint	Private: 15 Joint: 51 Public: 1	Private
Management	Private	Not regulated	Public	Private: 11 Joint: 24 Public: 33	Public
Membership	Voluntary	Compulsory	Compulsory	Compulsory in 55 cases	Compulsory
Risk adjusted Premiums	Yes	Not regulated	Yes	Yes, in 21 cases	Yes

^*Minimum requirements.

^**As of July 2000, this is around USD4600, or 2 to 3 times per capita GDP.

^***Colombia’s financial system does not allow for foreign currency deposits.

Source: Beck (2000).

On a more anecdotal level, an incident in mid-1999 involving a deposit run on a relatively sound bank raises questions on both sides of the market discipline/moral hazard question. On May 26^th, an unfounded rumor that a large private bank was on the brink of bankruptcy was spread through the internet. This provoked a depositor panic, and in one day the bank’s deposits fell by 5% of total liabilities in net terms, with an even larger gross loss as some as some government entities that had withdrawn their deposits early in the day returned them late in the afternoon. The net deposit loss was covered, in almost equal parts, through credit from the central bank, other financial institutions, and real sector corporations. Deposits gradually came back, but were not fully restored until the end of October. The person who spread the rumor was eventually caught and criminally charged with having provoked an economic panic.

This incident illustrates on the one hand, the willingness of the public to move quickly and withdraw their deposits in response to perceived increased risk of bankruptcy, suggesting that depositor risk has not been completely eliminated by the insurance scheme, and that depositors have an incentive to monitor banks. On the other hand, the large response to unfounded and even blatantly incorrect news suggests that depositors do not have very reliable information, and this may prevent them from monitoring bank behavior effectively. In other words, it may be the case that depositors have the incentive but not the means to impose market discipline.

IV. Extending the tests for market discipline

In this section we present our estimation approach, describing first how we extend the conventional test for market discipline on the depositor side, and then how we develop a more direct and flexible test for the behavior of banks in response to depositors’ signals.

V. Estimation Results

We ran regressions using semi-annual data for 1985-1999 based on individual bank balance sheets and income statements, which yield a set of bank-specific variables which are defined below. Except for some dummy variables, all have both a time (t) and a cross-section dimension (i), which we suppress for notational simplicity. The fundamental variables that explain the probability of default are defined as follows¹¹:

NPL	=	non-performing loans/total loans
NPLASS	=	non-performing loans/assets
PROV	=	loan loss provisions/assets
KASS	=	capital/assets
COVGE	=	KASS + PROV – NPLASS
ROE	=	return on equity
LIQUIDITY	=	total reserves/assets

View Table

NPL	=	non-performing loans/total loans
NPLASS	=	non-performing loans/assets
PROV	=	loan loss provisions/assets
KASS	=	capital/assets
COVGE	=	KASS + PROV – NPLASS
ROE	=	return on equity
LIQUIDITY	=	total reserves/assets

Based on the analytical framework discussed above, we test whether depositors react to changes in fundamentals. In particular, depositors should react negatively to increases in non-performing loans (NPL and NPLASS), and positively to increases in provisions (PROV), in the capital-asset ratio (KASS), in the coverage of non-performing loans (COVGE), and in the return-to-equity (ROE).

Though most of the empirical literature considers higher liquidity as an indication of a lower probability of default, we allow for differences in depositors’ assessment of liquidity depending on the business cycle. In particular, while in “bad times” (i.e., a sharp recession) holding liquid assets might make a bank less vulnerable and depositors more confident, in “normal times”, higher liquidity implies a lower return on assets, with little offsetting positive effect. We classify each time observation as being “normal” or “bad times” according to the real GDP growth rate in relation to its trend growth rate for the 1970–1996 period¹², and produce a time-varying dummy variable, BADTIMES. The product of this variable with LIQ results in BTLIQ, which captures whether there are indeed different effects of liquidity during “bad times”. If our priors are correct, depositors should view liquidity negatively, except during “bad times.”

As we discussed in Section IV, we control for other non-fundamental variables that might affect both the rate of growth of deposits and the level of interest rates. Total real assets (ASSETS) allow us to test whether depositors believe that larger banks are “too big to fail”, and therefore that deposits held there are safer. Our two controls for the return to deposits are the number of branches (BRANCH), and (in the deposit growth equation) the deposit interest rate (r). We expect the number of branches to reflect the quality of payments services offered by the banks; all else constant, deposits should grow faster in those banks that lower transaction costs or offer more payments services by providing more branches. We measure the interest rate implicitly from the balance sheets and income statements, as the ratio of interest paid to the average stock of deposits over a given six-month period.

In addition, we control for the effect of macroeconomic shocks that affect all banks equally, such as changes in GDP growth and interest rates on government securities. Our best results were obtained using period effects, but our estimations using specific macroeconomic controls yielded similar overall results. Finally, we included dummy variables in order to distinguish between private and state-owned banks (STATE) and between domestic and foreign-owned institutions (FOR).

Our regression analysis was done using a sample encompassing virtually the entire commercial banking system (25-33 banks, depending on the period), excluding only a few smaller banks for which there was missing information. We allowed for different bank-specific intercepts by using fixed effects (FE) and random effects (RE) estimation, and chose which to report according to the Hausman test. In all cases differences in bank-specific intercepts were overwhelmingly accepted, as was the existence of common period effects.

A. Estimations for depositor behavior

We focus our analysis on the deposit growth equation¹³, the results of which are shown in Tables 4A and 4B. As expected, deposit growth depends positively and significantly on the interest rate and on the number of branches, thus showing that return is an important factor in choosing among banks. In all estimations bank size appears with a positive coefficient, in line with the “too big to fail” argument. However, this coefficient is significant only in about half of the estimations.

Table 4a.

Determinants of the Rate of Growth of Real Deposits (DRD)

(Period and bank-specific cffccts, semi-annual data, 1985(1)-1999(2))

t-ratios in parentheses; (*) significant a! 90%; (**) significant at 95%

Table 4a.

Determinants of the Rate of Growth of Real Deposits (DRD)

(Period and bank-specific cffccts, semi-annual data, 1985(1)-1999(2))

		(1)	(2)	(3)	(4)	(5)
Estimation method:		FE	FE	FE	FE	FE
Constant		-0.126 (2.28)**	-0.436 (8.541)**	-0.139 (2.571)**	-0.332 (6.675)**	-0.100 (1.577)
Return on deposits
r		0.694 (2.774)**	1.184 (5.421)**	0.763 (3.025)**	0.996 (4.510)**	0.687 (2.753)**
BRANCH		0.677E-03 (1.584)	0.609E-03 (1.646)*	0.646E-03 (1.512)	0.673E-03 (1.788)*	0.712E-03 (1.670)*
Probability of default: bank size and lagged fundamentals
ASSETS		3.52E-08 (1.802)*	2.93E-08 (1.738)*	0.340E-07 (1.748)*	0.346E-07 (2.013)**	0.313E-07 (1.597)
NPL(-1)		-0.138 (1.020)
KASS (-1)			2.332 (14.742)**
COVGE(-1)					1.963 (13.666)**
LIQ(-1)						-0.503 (-1.962)**
BTLIQ(-1)						0.897 (2.308)**
ROE(-1)				-0.017 (1.485)
Sector dummies
STATE		.026 (0.461)	.099 (2.116)**	.009 (0.179)	0.117 (2.437)**	.017 (0.319)
FOR		.007 (0.138)	-0.025 (0.561)	0.007 (0.139)	-0.048 (-1.049)	.016 (0.313)
Hypothesis tests
Bank effects		105.87	99.19	106.02	97.41	110.37
	p-value	0.00	0.00	0.00	0.00	0.00
Period effects		119.34	183.27	121.78	186.11	116.12
	p-value	0.00	0.00	0.00	0.00	0.00
Hausmann test		27.02	23.24	23.55	29.31	31.67
	p-value	0.00	0.00	0.00	0.00	0.00
No. of observations		709	709	709	709	709
	R2	0.293	0.470	0.294	0.451	0.299

t-ratios in parentheses; (*) significant a! 90%; (**) significant at 95%

View Table

Table 4a.

Determinants of the Rate of Growth of Real Deposits (DRD)

(Period and bank-specific cffccts, semi-annual data, 1985(1)-1999(2))

		(1)	(2)	(3)	(4)	(5)
Estimation method:		FE	FE	FE	FE	FE
Constant		-0.126 (2.28)**	-0.436 (8.541)**	-0.139 (2.571)**	-0.332 (6.675)**	-0.100 (1.577)
Return on deposits
r		0.694 (2.774)**	1.184 (5.421)**	0.763 (3.025)**	0.996 (4.510)**	0.687 (2.753)**
BRANCH		0.677E-03 (1.584)	0.609E-03 (1.646)*	0.646E-03 (1.512)	0.673E-03 (1.788)*	0.712E-03 (1.670)*
Probability of default: bank size and lagged fundamentals
ASSETS		3.52E-08 (1.802)*	2.93E-08 (1.738)*	0.340E-07 (1.748)*	0.346E-07 (2.013)**	0.313E-07 (1.597)
NPL(-1)		-0.138 (1.020)
KASS (-1)			2.332 (14.742)**
COVGE(-1)					1.963 (13.666)**
LIQ(-1)						-0.503 (-1.962)**
BTLIQ(-1)						0.897 (2.308)**
ROE(-1)				-0.017 (1.485)
Sector dummies
STATE		.026 (0.461)	.099 (2.116)**	.009 (0.179)	0.117 (2.437)**	.017 (0.319)
FOR		.007 (0.138)	-0.025 (0.561)	0.007 (0.139)	-0.048 (-1.049)	.016 (0.313)
Hypothesis tests
Bank effects		105.87	99.19	106.02	97.41	110.37
	p-value	0.00	0.00	0.00	0.00	0.00
Period effects		119.34	183.27	121.78	186.11	116.12
	p-value	0.00	0.00	0.00	0.00	0.00
Hausmann test		27.02	23.24	23.55	29.31	31.67
	p-value	0.00	0.00	0.00	0.00	0.00
No. of observations		709	709	709	709	709
	R2	0.293	0.470	0.294	0.451	0.299

t-ratios in parentheses; (*) significant a! 90%; (**) significant at 95%

Table 4b.

Determinants of the Rate of Growth of Real Deposits (DRD)

(Period and bank-specific effects, semi-annual data, 1985(1)-1999(2))

t-ratios in parentheses; (*) significant at 90%, (**)significant at 95%

¹Also ran this specification restricting the coefficients for LIQ and BTLIQ to be equal in absolute value. This restriction yields an F-statistic of 0.4408 and a p-value of 0.503.

²Two-stage least squares estimation using the lending rate, the required reserve ratio, and the ratio of noninterest expenses to assets as instruments for the deposit interest rate.

Table 4b.

Determinants of the Rate of Growth of Real Deposits (DRD)

(Period and bank-specific effects, semi-annual data, 1985(1)-1999(2))

		(6)	(7)	(8)1	(9)2
Estimation method:		FE	FE	FE	FE-2SLS
Constant		-0.273 (4.745)**	-0.419 (6.891)**	-0.400 (6.549)**	-0.133 (1.222)
Return on deposits
r		0.947 (4.253)**	1.154 (5.228)**	1.126 (5.136)**	1.019 (4.336)**
BRANCH		0.732E-03 (1.947)*	0.645E-03 (1.746)*	6.83E-04 (1.858)*	0.001 (4.514)**
ProbabiIity of default: bank size and lagged fundamentals
ASSETS		0.302E-07 (1.753)*	0.249E-07 (1.467)	0.240E-07 (1.420)	-0.836 (1.222)
NPL(-1)			0.166 (1.391)
NPLASS(-1)				-0.547 (1.402)
PROV(-I)				1.427 (2.587)**
KASS (-1)			2,404 (14.657)**	2.396 (14.882)**
COVGE(-1)		1.980 (13.609)**			1.919 (12.871)**
LIQ(-1)		-0.605 (2.672)**	-0.430 (1.926)*	-0.416 (1.852)*	-0.852 (3.079)**
BTLIQ(-1)		0.702 (2.049)**	0.614 (1.822)*	0.617 (1.839)*	0.740 (2.156)**
ROE(-1)		0.005 (0.500)	0.012 (1.197)
Sector dummies
STATE		0.133 (2.747)**	0.086 (1.727)*	0.076 (1.540)	0.092 (1.740)*
FOR		-0.044 (0.961)	-0.014 (0.307)	-0.023 (0.513)	-0.042 (0.922)
Hypothesis tests
Bank effects		102.15	103.54	103.41
	p-value	0.00	0.00	0.00
Period effects		179.29	172.36	179.39
	p-value	0.00	0.00	0.00
Hausmann test		36.67	24.93	29.08
	p-value	0.00	0.01	0.00
Joint significance of fundamentals
	F-statistic	49.15	0.00	45.66
	p-value	0.00	0.00	0.00
No. of observations		709	709	709	709
	R2	0.458	0.478	0.482	0.456

t-ratios in parentheses; (*) significant at 90%, (**)significant at 95%

¹Also ran this specification restricting the coefficients for LIQ and BTLIQ to be equal in absolute value. This restriction yields an F-statistic of 0.4408 and a p-value of 0.503.

²Two-stage least squares estimation using the lending rate, the required reserve ratio, and the ratio of noninterest expenses to assets as instruments for the deposit interest rate.

View Table

Table 4b.

Determinants of the Rate of Growth of Real Deposits (DRD)

(Period and bank-specific effects, semi-annual data, 1985(1)-1999(2))

		(6)	(7)	(8)1	(9)2
Estimation method:		FE	FE	FE	FE-2SLS
Constant		-0.273 (4.745)**	-0.419 (6.891)**	-0.400 (6.549)**	-0.133 (1.222)
Return on deposits
r		0.947 (4.253)**	1.154 (5.228)**	1.126 (5.136)**	1.019 (4.336)**
BRANCH		0.732E-03 (1.947)*	0.645E-03 (1.746)*	6.83E-04 (1.858)*	0.001 (4.514)**
ProbabiIity of default: bank size and lagged fundamentals
ASSETS		0.302E-07 (1.753)*	0.249E-07 (1.467)	0.240E-07 (1.420)	-0.836 (1.222)
NPL(-1)			0.166 (1.391)
NPLASS(-1)				-0.547 (1.402)
PROV(-I)				1.427 (2.587)**
KASS (-1)			2,404 (14.657)**	2.396 (14.882)**
COVGE(-1)		1.980 (13.609)**			1.919 (12.871)**
LIQ(-1)		-0.605 (2.672)**	-0.430 (1.926)*	-0.416 (1.852)*	-0.852 (3.079)**
BTLIQ(-1)		0.702 (2.049)**	0.614 (1.822)*	0.617 (1.839)*	0.740 (2.156)**
ROE(-1)		0.005 (0.500)	0.012 (1.197)
Sector dummies
STATE		0.133 (2.747)**	0.086 (1.727)*	0.076 (1.540)	0.092 (1.740)*
FOR		-0.044 (0.961)	-0.014 (0.307)	-0.023 (0.513)	-0.042 (0.922)
Hypothesis tests
Bank effects		102.15	103.54	103.41
	p-value	0.00	0.00	0.00
Period effects		179.29	172.36	179.39
	p-value	0.00	0.00	0.00
Hausmann test		36.67	24.93	29.08
	p-value	0.00	0.01	0.00
Joint significance of fundamentals
	F-statistic	49.15	0.00	45.66
	p-value	0.00	0.00	0.00
No. of observations		709	709	709	709
	R2	0.458	0.478	0.482	0.456

t-ratios in parentheses; (*) significant at 90%, (**)significant at 95%

¹Also ran this specification restricting the coefficients for LIQ and BTLIQ to be equal in absolute value. This restriction yields an F-statistic of 0.4408 and a p-value of 0.503.

²Two-stage least squares estimation using the lending rate, the required reserve ratio, and the ratio of noninterest expenses to assets as instruments for the deposit interest rate.

Regarding the fundamental variables, Table 4A shows estimations in which they are introduced one by one, and Table 4B shows the estimations when more than one are included at a time. The results reported in Table 4A give evidence of market discipline. Though deposit growth does not appear to depend on the level of non-performing loans (NPL) nor on the return to equity (ROE)¹⁴, it does depend, and with the expected sign, on the other three fundamental variables. In particular, deposits grow faster in banks with a higher capital base (KASS) and in banks where non-performing loans are better covered by both capital and provisions (COVGE). As for liquidity, our results give support to our suspicion that depositors’ attitudes toward liquidity differ according to the macroeconomic environment. In normal times depositors tend to have a negative view of liquidity, as reflected in the negative significant coefficient of LIQ. During an economic downturn, on the other hand, depositors’ view of liquidity changes, presumably as the bank becomes better equipped to withstand a draining of resources. This is shown by the positive coefficient of BTLIQ.

Since cross-sectional differences across banks are captured by the bank-specific intercepts in the FE estimations, the dummy variables for bank ownership (STATE and FOR) reflect only changes in ownership over time for individual banks. Two major types of changes in ownership occurred during our sample period, particularly during the early 1990s: privatization of previously state-owned banks, and foreign acquisition of domestically-owned banks¹⁵. The estimations in Table 4A show some evidence of a deposit loss stemming from privatization (the dummy variable STATE has a positive and significant coefficient in most regressions), but no discernible change in deposits following acquisition by foreign investors.

Since we are already controlling for fundamental variables capturing quality of bank management, we interpret the positive coefficient on STATE as an indication that, other things constant, depositors perceived that a privatized bank was more likely to default than its state-owned predecessor, presumably because the government would be more likely to bail out the latter in case of distress.

Most of the above results carry on to Table 4B, where now we include more than one fundamental variable at a time. Regarding the “too big to fail argument”, we always obtain a positive coefficient for the level of assets, although the coefficient is only significant, and at the 90% level, in one of the regressions. The results on a bank’s interest rates and the number of branches continue to hold and have the expected signs. The results on changes in ownership are maintained as well, with privatization lowering deposit growth but with no significant effect from foreign acquisition of domestic banks.

Turning to the fundamentals, there is evidence that depositors arc sensitive to these variables, as shown by their overwhelming joint significance. Individually, however, not all fundamentals appear to matter to depositors. Once again the percentage of non-performing loans and the return-on-equity (ROE) do not appear to have any effect on depositor behavior. On the other hand, the coverage ratio (COVGE), which a previous study (González-Hermosillo, 1999) found to be a good leading indicator for bank failure during the 1980s banking crisis, does have a positive and significant effect on deposit growth. When we decompose this variable into its three components: provisions (PROV), KASS, and the ratio of non-performing loans to assets (NPLASS), we find that the first two have the expected positive sign and are significant at the 95% level, whereas NPLASS has the expected negative sign but is not statistically significant.

In all four regressions a higher level of liquidity (LIQ) is associated in a statistically significant manner with a lower rate of growth of deposits during normal times, while BTLIQ is associated with a higher rate of growth of deposits. In the final column of Table 4B we test for the linear restriction that the sum of the coefficients on LIQ and on BTLIQ is equal to zero, which is not rejected. Thus, our reading on this is as follows. In “normal times” (i.e. when the rate of growth of GDP is not considerably below trend), the bad news coming out of a high level of liquidity (i.e. a low return on assets) dominates the good news (i.e. the bank will more easily accommodate a deposit withdrawal). On the other hand, in “bad times” depositors perceive that the level of liquidity is irrelevant – the negative effect associated with a low return on assets appears to be offset by the positive effect stemming from a better ability to cope with a deposit withdrawal.

In order to capture cross-sectional differences according to bank ownership, we also ran OLS regressions where we eliminated bank-specific intercepts. Since all other results were very similar to those obtained in the FE estimations, for simplicity we report only the results for STATE and FOR in Table 5. Our results give some support to the hypothesis that depositors prefer state-owned banks, possibly due to a greater perceived probability of bailout and thus lower risk of default. The strongest evidence comes from specifications where we exclude non-performing loans and incorporate macro controls¹⁶. On the other hand, there does not appear to be a preference for foreign-owned banks; their ability to attract deposits and gain market share are solely linked to their fundamentals and the return they offered to depositors.

Table 5.

OLS Regressions for Differences Across Groups of Banks

Dependent variable: Real deposit growth (DRD)

t-ratios in parentheses; (*) significant at 90%, (**) significant at 95%.

¹Macro controls included: overall deposit growth (AGGDRD), the real interest rate on government paper (RGINT), and a dummy variable for the 1990 financial liberalization (LIB).

Table 5.

OLS Regressions for Differences Across Groups of Banks

Dependent variable: Real deposit growth (DRD)

Specification:	(6)	(7)	(8)	(9)
Estimations without period effects or macro controls
STATE	0.038 (1.544)	-0.001 (0.050)	0.001 (0.031)	0.038 (1.545)
FOR	-0.007 (0.378)	-0.007 (0.375)	-0.007 (0.358)	-0.007 (0.375)
Estimations with macro controls1
STATE	0.049 (2.031)**	-0.002 (0.082)	0.003 (0.130)	0.050 (2.052)**
FOR	-0.009 (0.444)	-0.008 (0.425)	-0.008 (0.431)	-0.009 (0.482)

t-ratios in parentheses; (*) significant at 90%, (**) significant at 95%.

¹Macro controls included: overall deposit growth (AGGDRD), the real interest rate on government paper (RGINT), and a dummy variable for the 1990 financial liberalization (LIB).

View Table

Table 5.

OLS Regressions for Differences Across Groups of Banks

Dependent variable: Real deposit growth (DRD)

Specification:	(6)	(7)	(8)	(9)
Estimations without period effects or macro controls
STATE	0.038 (1.544)	-0.001 (0.050)	0.001 (0.031)	0.038 (1.545)
FOR	-0.007 (0.378)	-0.007 (0.375)	-0.007 (0.358)	-0.007 (0.375)
Estimations with macro controls1
STATE	0.049 (2.031)**	-0.002 (0.082)	0.003 (0.130)	0.050 (2.052)**
FOR	-0.009 (0.444)	-0.008 (0.425)	-0.008 (0.431)	-0.009 (0.482)

t-ratios in parentheses; (*) significant at 90%, (**) significant at 95%.

¹Macro controls included: overall deposit growth (AGGDRD), the real interest rate on government paper (RGINT), and a dummy variable for the 1990 financial liberalization (LIB).

One comment is in order regarding the interpertation of the positive coefficient on the desposit interest rate. It may be argued that this variable reflects the riskiness of a particular bank rather than the return to the depositor. Indeed, as we will show in Section V.C., riskier banks do in fact tend to set higher interest rates on deposits. However, if the risk information contained in the interest rate is fully related to the fundamental variables, then the 2SLS procedure should correct for this, and one would expect the coefficient on the interest rate to be non-significant. On the other hand, if the interest rate also contains substantial risk information not related to the fundamentals, then the 2SLS procedure should yield a significant negative coefficient under market discipline, since depositors would move out of riskier banks. Of course, if there is no market discipline and moral hazard is rampant, one would expect the interest rate - as well as the fundamental variables - to be non-significant. Since our estimations overwhelmingly show a positive significant coefficient, our conclusion is that the return element is crucial and is dominating whatever risk information may be contained in the interest rate¹⁷.

B. Extensions to the estimations for depositor behavior

In this section we examine three extensions to our empirical analysis of depositor behavior: addressing possible endogeneity of the interest rate, testing whether there were changes in market discipline over time, and re-estimating the deposit growth equation using explicit macroeconomic controls instead of period effects.

First, we address the possible endogeneity between deposit interest rates and fundamentals (in fact, in the following section deposit interest rates are shown to respond significantly to prior changes in fundamentals), employing a 2SLS procedure. We include as instruments for the deposit interest rate the lending interest rate, the required reserve ratio (which may be viewed as an exogenous policy variable), and the ratio of noninterest expenses to assets (which reflects the individual product mix chosen by the bank as well as its level of efficiency), and then enter the predicted value for the deposit interest rate as a regressor in the deposit growth equation. We report the 2SLS results for equation (9), the most parsimonious specification in that it summarizes the fundamental variables in three: COVGE, LIQ, and BTLIQ. The previous results hold; the interest rate and number of branches continue to be positively and significantly related to deposit growth, the respective fundamentals continue to be significant and in the expected direction, and there is a relative drop in deposit growth as a result of privatization.

Secondly, we examine whether market discipline, as measured by the sensitivity of deposit growth to bank fundamentals, has changed over time. Given that coverage has been declining in real terms, and that depositors may have undergone a learning process in which their assessment of banks performance has improved, we might expect to observe some increase in market discipline. Table 6 presents results of estimations in which we include interaction terms between a time variable and each of the fundamentals, and find that depositor sensitivity has increased for those fundamentals which performed best in the previous regressions, COVGE, KASS, and PROV. For NPLASS however, which did not perform well in the previous estimations, we find that depositor sensitivity has declined over time. Finally, for LIQ and ROE we do not detect a significant change in sensitivity. Thus, we find some evidence of increasing market discipline throughout the sample period.

Table 6.

Changes Over Time in the Responsiveness to Bank Fundamentals: Rate of Growth of Real Deposits (DRD)¹

(Period and bank-specific effects, semi-annual data, 1985(1)-1999(2))

t-ratios in parentheses; (*) significant at 90%, (**) significant at 95%

¹For simplicity we present only the coefficients for the fundamentals.

²Significant at 89%.

Table 6.

Changes Over Time in the Responsiveness to Bank Fundamentals: Rate of Growth of Real Deposits (DRD)¹

(Period and bank-specific effects, semi-annual data, 1985(1)-1999(2))

Specification		(6)	(7)	(8)	(9)
Estimation method:		FE	FE	FE	FE
Probability of default: bank lagged fundamentals
	NPL(-1)		0.428 (2.237)**
	NPL(-1)*Time		-0.042 (2.171)**
	NPLASS(-1)			0.935 (1.528)
	NPLASS(-1)*Time			-0.132 (3.031)**
	PROV(-1)			-0.357 (0.364)
	PROV(-1)*Time			0.128 (2.109)**
	KASS (-1)		1.775 (4.084)**	1.558 (3.310)**
	KASS (-1)*Time		0.036 (1.608)2	0.047 (1.893)*
	COVGE(-1)	0.560 (1.822)**			-0.934 (1.979)**
	COVGE(-1)*Time	0.088 (5.275)**			1.124 (6.455)**
	LIQ(-1)	-0.214 (0.530)	-0.298 (0.741)	-0.245 (0.613)	0.374 (0.571)
	LIQ(-1)*Time	-0.003 (0.143)	0.009 (0.403)	0.003 (0.162)	-0.223 (0.929)
	ROE(-1)	0.008 (0.458)	0.013 (0.688)
ROE(-1)*Time		-0.001 (0.086)	0.000 (0.212)
Bank effects		109.451	112.595	116.596	109.883
	p-value	0.000	0.00	0.00	0.00
Period effects		185.999	171.513	178.862	187.422
	p-vaiue	0.000	0.00	0.00	0.00
Hausmann test		26.100	19.3	20.69	24.18
	p-value	0.006	0.114	0.079	0.004
No. of obs.		709	709	709	709
R2		0.478	0.487	0.493	0.488

t-ratios in parentheses; (*) significant at 90%, (**) significant at 95%

¹For simplicity we present only the coefficients for the fundamentals.

²Significant at 89%.

View Table

Table 6.

Changes Over Time in the Responsiveness to Bank Fundamentals: Rate of Growth of Real Deposits (DRD)¹

(Period and bank-specific effects, semi-annual data, 1985(1)-1999(2))

Specification		(6)	(7)	(8)	(9)
Estimation method:		FE	FE	FE	FE
Probability of default: bank lagged fundamentals
	NPL(-1)		0.428 (2.237)**
	NPL(-1)*Time		-0.042 (2.171)**
	NPLASS(-1)			0.935 (1.528)
	NPLASS(-1)*Time			-0.132 (3.031)**
	PROV(-1)			-0.357 (0.364)
	PROV(-1)*Time			0.128 (2.109)**
	KASS (-1)		1.775 (4.084)**	1.558 (3.310)**
	KASS (-1)*Time		0.036 (1.608)2	0.047 (1.893)*
	COVGE(-1)	0.560 (1.822)**			-0.934 (1.979)**
	COVGE(-1)*Time	0.088 (5.275)**			1.124 (6.455)**
	LIQ(-1)	-0.214 (0.530)	-0.298 (0.741)	-0.245 (0.613)	0.374 (0.571)
	LIQ(-1)*Time	-0.003 (0.143)	0.009 (0.403)	0.003 (0.162)	-0.223 (0.929)
	ROE(-1)	0.008 (0.458)	0.013 (0.688)
ROE(-1)*Time		-0.001 (0.086)	0.000 (0.212)
Bank effects		109.451	112.595	116.596	109.883
	p-value	0.000	0.00	0.00	0.00
Period effects		185.999	171.513	178.862	187.422
	p-vaiue	0.000	0.00	0.00	0.00
Hausmann test		26.100	19.3	20.69	24.18
	p-value	0.006	0.114	0.079	0.004
No. of obs.		709	709	709	709
R2		0.478	0.487	0.493	0.488

t-ratios in parentheses; (*) significant at 90%, (**) significant at 95%

¹For simplicity we present only the coefficients for the fundamentals.

²Significant at 89%.

Thirdly, we re-ran the regressions using explicit macroeconomic controls rather than common period effects, the detailed results of which are available upon request. While the principal results reported in Tables 4A and 4B continue hold, this is not the case for the liquidity variables LIQ and BTLIQ, which continue to be significant, but change sign. Two macro controls performed well in all regressions: the growth rate of aggregate bank deposits, and the real interest rate on government paper. A third variable, a liberalization dummy for 1991, was positive and significant in some of the deposit growth equations, thus suggesting that the liberalization program may have increased the attractiveness of bank deposits above and beyond the return or risk characteristics of banks.

C. Response of banks to changes in deposit growth

In the previous sections we showed that depositor behavior appears to be broadly consistent with market discipline, particularly as seen in the deposit growth regressions; depositors tend to prefer banks with stronger fundamentals. Now we turn to the question of whether banks respond once depositors have revealed their preferences. Thus, we measure to what extent bank fundamentals change in response to past changes in deposits, and we zero in on the component of deposit growth that is directly attributable to an individual bank’s fundamentals, which we term the “fundamental” deposit growth, (DRDFUND). We define DRDFUND as the real growth of deposits explained by fundamental variables from regression (8) in Table 4B. Note that this variable is bank-specific, but below we drop the i subscript for simplicity.

${\mathit{\text{DRDFUND}}}_t=-0.547{\mathit{\text{NPLASS}}}_{t-1}+1.427{\mathit{\text{PROV}}}_{t-1}+2.396{\mathit{\text{KASS}}}_{t-1}-0.416{\mathit{\text{LIQ}}}_{t-1}+0.617{\mathit{\text{BTLIQ}}}_{t-1}$

The first test therefore consists of regressing each fundamental variable on lagged DRDFUND, to determine whether this period’s fundamentals are sensitive to depositors’ preference for strong fundamentals in the previous period. If market discipline holds, we should expect negative coefficients on DRDFUND in the equations for KASS, COVGE, PROV, and LIQ, and a positive coefficient in the case of NPLA. For instance, if depositors punished a bank last period for weak fundamentals (DRDFUND falls), then the bank should react today by improving its fundamentals, increasing the capital base or reducing the level of non-performing loans. The results of this test are not very encouraging, however, as shown in the top portion of Table 7. Only PROV clearly appears to behave in a manner consistent with market discipline, where a previous deterioration leads to an increase today in provisions. Liquidity responds with the expected sign but the coefficient is not statistically significant. All other variables appear to exhibit the opposite behavior: if fundamentals deteriorate (improve) in the previous period, and therefore depositors withdraw (increase) deposits, then today fundamentals tend to deteriorate (improve) even more.

Table 7.

Response of Banks to Fundamental Growth of Deposits

(Period and bank-specific effects, semi-annual data, 1985(2)-1999(2))

t-ratios in parentheses; (“) significant at 90%; (**) significant at 95%

¹Regressions included non-fundamental controls: ASS, BRANCH, STATE, FOR.

²In this case regressors are DLOSS1 and DLOSS2 rather than their products with DRDFUND.

Table 7.

Response of Banks to Fundamental Growth of Deposits

(Period and bank-specific effects, semi-annual data, 1985(2)-1999(2))

	Response in fundamental variables					Response in deposit rate 1		Response in lending rate1
	COVGE	KASS	PROV	NPLA	LIQ	r	r 2	r L	r L 2
No. of observations:	678	678	678	678	678	678	678	678	676
Symmetric response
Estimation method:	RE	RE	RE	RE	FE	FE		FE
DRDFUND(-1)	0.215 (14.592)**	0.162 (13.228)**	-0.038 (4.135)**	-0.143 (6.145)**	-0.114 (0.829)	-0.041 (3.400)**		-0.005 (0.317)
Constant	0.0274 (3.880)**	0.061 (11.932)**	0.028 (6.920)**	0.124 (9.225)**	0.139 (45.42)**	0.161 (24.903)**		0.365 (40.227)**
Hausman test p-value	0.623	0.672	0.792	0.514	0.006	0.001		0.000
R2	0.003	0.003	0.025	0.090	0.691	0.702		0.638
Asymmetric response: Banks only respond to deposit “losses”
Case l: Banks only respond when their fundamental deposit growth is negative
Estimation method:	RE	RE	FE	FE	RE	FE	FE	FE	FE
DRDFUND(-1)’DLOSS1(-1)	0.404 (5.167)**	0.078 (1.19)	-0.407 (9.505)**	-1.160 (11.035)**	-0.133 (1.566)	0.048 (0.872)	-0.001 (0.005)	0.005 (0.069)	-0.019 (1.319)
Constant	0.074 (5.167)**	0.096 (14.092)**	0.0185 (18.503)**	0.0862 (35.12)**	0.125 (15.206)**	0.153 (25.176)**	0.153 (25.123)**	0.364 (43.057)**	0.364 (43.119)**
Hausman test ρ-value	0.110	0.576	0.000	0.000	0.272	0.001	0.001	0.000	0.000
R2	0.049	0.006	0.463	0.632	0.005	0.697	0.696	0.638	0.639
Case 2: Banks only respond when their fundamental deposit growth is below the banking sector average
Estimation method:	RE	RE	FE	FE	FE	FE	FE	FE	FE
DRDFUND(-1)’DL OSS2-(1)	-0.091 (3.707)**	-0.101 (5.093)**	-0.032 (2.130)**	-0.045 (1.211)	0.017 (0.831)	0.079 (0.448)	0.007 (2.253)**	-0.074 (3.039)**	-0.01 (2.382)**’
Constant	0.080 (10.392)**	0.103 (17.632)**	0.022 (13.851)**	0.093 (23.176)**	0.135 (63.312)**	0.153 (24.356)**	0.149 (23.253)**	0.370 (42.871)**	0.37 (41.777)**
Hausman test ρ-value	0.180	0.582	0.077	0.000	0.000	0.001	0.004	0.000	0.000
R2	0.022	0.051	0.389	0.560	0.691	0.696	0.699	0.643	0641

t-ratios in parentheses; (“) significant at 90%; (**) significant at 95%

¹Regressions included non-fundamental controls: ASS, BRANCH, STATE, FOR.

²In this case regressors are DLOSS1 and DLOSS2 rather than their products with DRDFUND.

View Table

Table 7.

Response of Banks to Fundamental Growth of Deposits

(Period and bank-specific effects, semi-annual data, 1985(2)-1999(2))

	Response in fundamental variables					Response in deposit rate 1		Response in lending rate1
	COVGE	KASS	PROV	NPLA	LIQ	r	r 2	r L	r L 2
No. of observations:	678	678	678	678	678	678	678	678	676
Symmetric response
Estimation method:	RE	RE	RE	RE	FE	FE		FE
DRDFUND(-1)	0.215 (14.592)**	0.162 (13.228)**	-0.038 (4.135)**	-0.143 (6.145)**	-0.114 (0.829)	-0.041 (3.400)**		-0.005 (0.317)
Constant	0.0274 (3.880)**	0.061 (11.932)**	0.028 (6.920)**	0.124 (9.225)**	0.139 (45.42)**	0.161 (24.903)**		0.365 (40.227)**
Hausman test p-value	0.623	0.672	0.792	0.514	0.006	0.001		0.000
R2	0.003	0.003	0.025	0.090	0.691	0.702		0.638
Asymmetric response: Banks only respond to deposit “losses”
Case l: Banks only respond when their fundamental deposit growth is negative
Estimation method:	RE	RE	FE	FE	RE	FE	FE	FE	FE
DRDFUND(-1)’DLOSS1(-1)	0.404 (5.167)**	0.078 (1.19)	-0.407 (9.505)**	-1.160 (11.035)**	-0.133 (1.566)	0.048 (0.872)	-0.001 (0.005)	0.005 (0.069)	-0.019 (1.319)
Constant	0.074 (5.167)**	0.096 (14.092)**	0.0185 (18.503)**	0.0862 (35.12)**	0.125 (15.206)**	0.153 (25.176)**	0.153 (25.123)**	0.364 (43.057)**	0.364 (43.119)**
Hausman test ρ-value	0.110	0.576	0.000	0.000	0.272	0.001	0.001	0.000	0.000
R2	0.049	0.006	0.463	0.632	0.005	0.697	0.696	0.638	0.639
Case 2: Banks only respond when their fundamental deposit growth is below the banking sector average
Estimation method:	RE	RE	FE	FE	FE	FE	FE	FE	FE
DRDFUND(-1)’DL OSS2-(1)	-0.091 (3.707)**	-0.101 (5.093)**	-0.032 (2.130)**	-0.045 (1.211)	0.017 (0.831)	0.079 (0.448)	0.007 (2.253)**	-0.074 (3.039)**	-0.01 (2.382)**’
Constant	0.080 (10.392)**	0.103 (17.632)**	0.022 (13.851)**	0.093 (23.176)**	0.135 (63.312)**	0.153 (24.356)**	0.149 (23.253)**	0.370 (42.871)**	0.37 (41.777)**
Hausman test ρ-value	0.180	0.582	0.077	0.000	0.000	0.001	0.004	0.000	0.000
R2	0.022	0.051	0.389	0.560	0.691	0.696	0.699	0.643	0641

t-ratios in parentheses; (“) significant at 90%; (**) significant at 95%

¹Regressions included non-fundamental controls: ASS, BRANCH, STATE, FOR.

²In this case regressors are DLOSS1 and DLOSS2 rather than their products with DRDFUND.

One possible explanation for this behavior is that banks may respond asymmetrically to signals by depositors. On the upside it may be that a virtuous cycle is encountered whereby an initial improvement in fundamentals leads to further improvements, whereas on the downside bank managers would not allow a vicious cycle to ultimately doom the bank, so they would be likely to react by improving the fundamentals. Thus, in a second group of tests we allow for this type of asymmetry, whereby the bank only reacts when it perceives a deposit loss stemming from weak fundamentals.

We define two types of deposit losses, one in absolute and one in relative terms. The first occurs when the fundamental deposit growth rate of a given bank is negative, so it defines an absolute loss of deposits owing to weak fundamentals. The second defines a deposit loss as any situation when a bank exhibits a fundamental growth rate below the banking sector average, thus indicating that this bank could increase deposits by moving its fundamentals closer to the sector average. Formally we define two dummy variables for each bank i and period t as shown below:

$\mathit{\text{DLOSS}}{1}_{it}=\{\begin{array}{c}1,\mathit{\text{DRDFUN}}{D}_{it}<0\\ 0,\mathit{\text{otherwise}}\end{array}\mathit{\text{DLOSS}}{2}_{it}=\{\begin{array}{c}1,{\mathit{\text{DRDFUND}}}_{it}<{\overline{\mathit{\text{DRDFUND}}}}_t\\ 0,\mathit{\text{otherwise}}\end{array}$

We must note that these variables constitute two extremes in deposit losses¹⁸. In the total sample of 709 observations, DLOSS1 defines only 15 observations as having been of banks being out of line in their fundamentals. On the other hand, DLOSS2 encompasses a much greater number of observations (420) in which individual banks were simply exhibiting sub-par fundamentals in relation to the rest of the banking sector.

We now test the response of the fundamental variables to deposit losses only, using the two definitions. In the case of DLOSSI, the more extreme observations of weak fundamentals, two minor changes in results occur with the respect to the symmetric estimation, as shown in the second portion of Table 6. First, the capital-asset ratio is now simply unresponsive to deposit losses, rather than being responsive in the wrong direction. Second, liquidity has the correct sign and now approaches statistical significance, with a p-value of just over 10 percent. Therefore, in these most extreme cases, the banks that suffer the strongest deposit withdrawals appear to be unlikely to respond in any desirable way other than by increasing their loan loss provisions. Their level of nonperforming loans deteriorates even more and coverage falls even further, reflecting a type of vicious cycle of deteriorating fundamentals.

For DLOSS2, a wider sample including less extreme cases of deposit losses, bank behavior appears to be more consistent with market discipline. Banks now tend to improve their coverage and capital-asset ratios when depositors have discriminated against them in the previous period, they continue to adjust their provisions upward, but do not appear to adjust their nonperforming loans or their liquidity levels.

In another set of tests, we investigated whether, in addition to changes in fundamental variables, banks adjusted their interest rates following fundamental-driven changes in deposit demand. We might expect a negative relationship for deposit interest rates, as banks could counteract the (absolute or relative) deposit outflow without having to resort to improving their fundamentals. In other words, they could attract deposits either by lowering risk (i.e., improving fundamentals) or by increasing return (the interest rate). We estimate the interest rate equation described in Table 5 but replace all fundamental variables with DRDFUND as well as its cross-product with each of the deposit loss dummy variables. The results show that deposit rates do exhibit the expected behavior symmetrically, thus they not only increase in response to a deposit loss, but also tend to fall when the fundamentals are above-par and deposits are growing rapidly. Furthermore, the asymmetric estimation does not seem to hold, thus indicating that the adjustment seems to be occurring primarily on the upside. However, a final exercise estimated the interest equation using DLOSS2 itself rather than its cross-product with DRDFUND and found that, on average, banks that suffered from sub-par fundamentals tended to pay higher interest rates, although these rates did not depend on the magnitude of the deposit loss.

Finally, in the last two columns of Table 7 we analyze the response of bank lending rates to changes in fundamental deposit growth. The behavior of lending rates may signal a bank’s level of risk-taking, which may eventually lead to changes in asset quality. For example, if a bank’s fundamentals deteriorate and it suffers a subsequent deposit loss, it may attempt to increase risk to recover profits (i.e., “gamble for resurrection”) or it may reduce risk in order to improve its fundamentals. Thus, an upward adjustment in lending rates may reflect increased risk-taking while a downward adjustment would reflect the contrary, a lower level of risk.

We find that lending rates do not appear to adjust symmetrically nor on the extreme downside in response to changes in fundamental deposit growth. However, in the case of moderate deposit losses, lending rates tend to be adjusted downward, thus suggesting that banks with sub par fundamentals tend to react by lowering risk. We also find that on average this group of banks charges lower lending rates. This prudent behavior comes at the cost of lower profits, since we found earlier that the deposit interest rates of these banks tended to be higher on average.

VI. Conclusions

In this study we have asked two types of questions: (1) What determines depositor behavior? and (2) do banks respond correctly to depositor’s signals regarding their performance? With regard to the first question, our strongest answers come from the estimations for real deposit growth, which showed that Colombian depositors respond to both risk and return factors. Banks attract deposits either by offering high interest rates (return), providing a large number of branches (return), or by exhibiting strong fundamentals, which presumably lowers the risk of default. Furthermore, state banks appear to have an inherent advantage in attracting deposits. This was seen both in a time dimension, as privatization led to a deposit loss, and in a cross-section dimension in the OLS regressions, as state banks on average exhibited higher deposit growth. On the other hand, although foreign banks have been highly successful in attracting deposits in recent years, they do not appear to possess a reputational advantage, as their deposit growth was explained solely by the return they offer and their fundamentals. Finally, there appears to be some limited evidence of a “too big to fail” perception by depositors, where in some regressions larger banks exhibited higher deposit growth.

Fundamentals appear to matter to depositors. Coverage and two of its components, the capital-asset ratio and loan-loss provisions, tend to exert a positive influence on deposit growth, and the joint significance of fundamental variables was overwhelmingly accepted in all regressions. However, neither profitability (as measured by the return to equity), nor nonperforming loans, nor liquidity ratios appear to be associated with declines in depositors’ perceived risk of losses. As we discussed earlier, the poor performance of the profitability variable is a common finding in the literature, but it remains puzzling that a bank’s nonperforming loans do not seem to affect deposit demand. Finally, we showed that depositor’s attitudes toward liquidity varied depending on whether the macroeconomic environment was normal or one of particularly low growth. During normal times, greater liquidity was associated with less active intermediation, but with very little benefit in terms of lower risk of default. Only in difficult or “bad times” did the benefits of greater liquidity become apparent.

As for the second question, we obtained evidence that banks responded to depositors’ signals in a manner consistent with market discipline, as evidenced by the responsiveness of some of the fundamental variables that were relevant to depositors: coverage, capitalization, and provisions. Our results also gave support to the hypothesis that this response was asymmetric, occurring only when the banks perceived that they were being punished by depositors. However, this disciplining behavior did not appear to extend to the cases in which fundamentals were the weakest, as these banks tended to perpetuate their problems rather than try to correct them. Finally, we found that positive signals from depositors tended to allow banks to lower their interest costs subsequently and that, on average, banks that received negative signals from depositors tended to have higher interest costs and lower interest receipts than the rest of the banking system, thus suffering a profit squeeze as a result of attempting to limit the deposit outflow and the level of risk.

These results point to the existence of market discipline in Colombia, as depositors take into account bank fundamentals, send signals to the banks, who then adjust their behavior accordingly. In other words, moral hazard appears to be limited, even though there is an explicit deposit insurance system throughout our sample period. This may be due to one or several design features, such as its compulsory nature, or the co-insurance, risk-weighted premiums, and the existence of a coverage limit that is continuously declining in real terms. With regard to the latter variable, we note that the recent cross-country study by Demirguc-Kunt and Detragiache (1999) finds evidence that the less extensive the coverage, the less moral hazard and greater market discipline is observed. Furthermore, the steady decline in coverage is consistent with the observed increase over time in depositors’ sensitivity to fundamentals, an indication that market discipline may be increasing.

Our results also have implications regarding the nature of market discipline in Colombia, in particular with regard to the behavior of loan-loss provisions. Indeed, we found this to be the only fundamental variable that responded symmetrically (and in the expected direction) to depositors’ signals. This finding is consistent with the level of provisions being largely an automatic or “endogenous” response of banks in order to comply with regulation. While depositor behavior may play an important role in disciplining banks in Colombia, there may also be an element of “regulatory discipline” which imposes appropriate adjustments by banks when their fundamentals deteriorate, most notably observed in the level of provisions.

Finally, although depositor behavior appears to exhibit market discipline and moral hazard seems to be limited, there still are two principal causes for concern. First, it is unclear why nonperforming loans do not play an important part either in depositors’ decisions nor in banks’ responses, especially when this variable has a proven empirical track record in predicting bank failures, both in Colombia’s mid-1980s crisis as well as in other countries (González-Hermosillo, 1999). A possible explanation for this is related again to regulatory discipline; since loan-loss provisioning was strengthened in the aftermath of the 1980s crisis, increases in non-performing loans throughout our 1985-99 sample period were more likely to be adequately covered, and therefore provided less of a signal of bank insolvency than was the case before the crisis. To the extent that regulation regarding provisions was strong, depositors may have needed to worry less about the level of non-performing loans per se. Second, our findings of a preference for state-owned banks and a possible “too big to fail” mentality on the part of depositors are a sign that market discipline is still imperfect in Colombia, and that some degree of moral hazard remains.