Why Don′t They Lend? Credit Stagnation in Latin America
  • 1 0000000404811396https://isni.org/isni/0000000404811396International Monetary Fund

This study examines the recent marked slowdown in bank credit to the private sector in Latin America. Based on a study of eight countries—Argentina, Bolivia, Brazil, Chile, Colombia, Peru, Mexico, and Venezuela—the magnitude of the slowdown is documented, comparing it to historical behavior and to similar episodes in other regions of the world. Second, changes in bank balance sheets are examined to determine whether the credit slowdown is merely a reflection of a downturn in bank deposits or whether the asset side has changed. Third, following an econometric disequilibrium approach used in recent studies of bank credit in East Asia and Finland, the paper investigates the possible causes in three countries: Colombia, Mexico, and Peru. While both supply and demand factors appear to have played key roles, their relative importance has varied across the three countries.


This study examines the recent marked slowdown in bank credit to the private sector in Latin America. Based on a study of eight countries—Argentina, Bolivia, Brazil, Chile, Colombia, Peru, Mexico, and Venezuela—the magnitude of the slowdown is documented, comparing it to historical behavior and to similar episodes in other regions of the world. Second, changes in bank balance sheets are examined to determine whether the credit slowdown is merely a reflection of a downturn in bank deposits or whether the asset side has changed. Third, following an econometric disequilibrium approach used in recent studies of bank credit in East Asia and Finland, the paper investigates the possible causes in three countries: Colombia, Mexico, and Peru. While both supply and demand factors appear to have played key roles, their relative importance has varied across the three countries.

After experiencing moderate to high rates of growth during most of the 1990s, several Latin American countries witnessed a significant slowdown over the past two years. As Table 1 illustrates, Argentina, Bolivia, Brazil, Chile, Colombia, Peru, and Venezuela recently experienced declines in their growth rates, ranging from about 1 percentage point in Brazil to 7 percentage points in Argentina. One prominent exception to this behavior is Mexico, for which economic growth accelerated by about 1 percentage point on average after 1995.

Table 1.

Average Real Growth Rates of GDP and Credit to the Private Sector

Selected Latin American Countries

article image
Source:International Financial Statistics

In many cases, the evolution of bank credit to the private sector has followed a similar, or even more pronounced, cyclical pattern. When deflated by consumer prices, during the above periods credit growth decelerated by about 5 percentage points in Chile, by over 9 percentage points in Brazil, by 13 in Argentina, 18 percentage points in Bolivia, and by over 20 percentage points in Colombia and Peru. Mexico and Venezuela stand in contrast, however; in Mexico credit growth fell 34 percentage points, even as economic growth was accelerating;1 and in Venezuela economic growth declined while credit enjoyed a modest recovery.

Given that in most Latin American countries only a few large corporations can tap foreign capital markets, understanding the recent slowdown in domestic bank credit becomes a very relevant issue. In policy circles it is widely believed that the credit slowdown is an important driving force behind the recent economic slump, and that monetary policy may be, at least partly, to blame. This idea is consistent with the “credit channel” literature, which, starting with the seminal work by Bernanke and Blinder (1988), Romer and Romer (1990), and others, shows how the provision of credit by banks to the private sector constitutes a key conduit through which monetary policy affects the real economy.

However, tightness of monetary policy is not the only possible reason why credit is stagnating in Latin America. It may be the case that, due to deteriorating economic conditions, there is little demand for credit from the private sector, as agents′ balance sheets and investment prospects have worsened. Regarding supply, it may be that banks have suffered from sluggish growth in their loanable funds (that is, deposits), thus affecting their ability to lend. Finally, banks may have become less willing to lend, perhaps partly due to deteriorating economic conditions but also to their perception of risk and to the regulations regarding their risk taking. Policy implications as to how to address this credit stagnation will depend crucially on a correct interpretation of the underlying factors. Interventions, if warranted, might range from programs to alleviate corporate debt burdens, enhancing the provision of liquidity on the part of the central bank, or revising the regulatory framework regarding the level of provisions.

I. Credit Stagnation in Latin America: A First Look

In this section we describe the recent performance of bank credit in eight countries mentioned in the previous section. First, we show the evolution of credit in historical context, comparing the recent behavior to previous cycles over a period of about 30 years. Secondly, we compare the slowdown with several international cases where pronounced credit contractions were studied. Thirdly, we focus on the last 20 years, using a simple balance sheet decomposition to detect where the major changes occurred in the behavior of banks from one period to the next.

The Credit Slowdown in Historical and International Context

Several differences and similarities arise when comparing the evolution of bank credit over the past 30 years in Latin America. Using International Financial Statistics (IFS) data, in Figure 1 we plot the ratio of private sector bank credit to GDP for the 1960–2000 period. We show this ratio both from Deposit Money Banks (DMB), as well as from the entire banking system, DMB plus Other Banking Institutions (OBI).2 While six out of eight countries—Brazil and Chile being the exception—exhibit a slowdown with some amount of decline in the ratio in recent years,3 the patterns for the entire period are not all alike. In Peru and Bolivia the slowdown seems to be an interruption in a process of rapid credit growth after hyperinflation had driven credit close to zero. Brazil, Chile, Mexico, and, especially, Argentina all experienced pronounced credit cycles since the 1960s. For Colombia there has been a modest but steady upward trend since 1960 and for Venezuela a sustained downward trend since the early 1980s.

Figure 1.
Figure 1.

Latin America: Credit–GDP Ratios 1960–2000

Citation: IMF Staff Papers 2002, 005; 10.5089/9781589061231.024.A006

Given this recent decline in bank credit to the private sector, it is relevant to examine whether there has been a substitution towards financing by nonbank financial institutions. For most of the countries this is clearly not the case, since the definition of OBIs is sufficiently broad to include such institutions as savings and loan corporations, credit cooperatives, investment banks, and financial funds. In two of the countries, however, IFS provides an additional category, “Nonbank Financial Institutions” (NFI)—leasing companies, brokerage houses and distributor companies in Brazil, and pension funds in Chile. Substitution toward NFIs does not appear to have taken place in Brazil, as nonbank credit amounted to less than 4 percent of bank credit in 2000. In Chile, on the other hand, the credit slowdown corresponded to a sharp rise of pension funds in the late 1980s, reaching almost one–quarter of bank credit by 1991. In the case of Mexico, information for savings and loan institutions (SLIs), investment societies, and brokerage houses shows that SLIs constituted a very small share of bank credit—less than 1 percent—while security holdings by the other nonbank institutions have been expanding rapidly since 1996, suggesting that corporate financing through the stock market has substituted to some degree for the slowdown in bank credit.4

How does the recent experience compare with historical behavior of bank credit? We undertook a second exercise using the 1960–2000 data, based on the Gourinchas, Valdes, and Landerretche (2001) study of credit booms. We used the Hodrick–Prescott (H–P) filter to calculate the trend in the credit–GDP ratio and then identified periods of credit booms as those in which the observed ratio is significantly above trend.5 As Table 2 and Figure 2 show, several countries experienced credit booms prior to their recent slowdowns; for instance Bolivia, Brazil, and Mexico all had credit booms in the early 1990s. Although the well–known end–of–period problem associated with the H–P filter does not allow us to derive similar conclusions regarding the severity of the recent slowdowns, we do observe from mid–period observations that, prior to the more recent recoveries, both Chile and Venezuela experienced severe credit contractions during the early 1990s, when credit–GDP fell well below trend.

How does the recent Latin American experience compare with those of credit slowdowns in other regions of the world? We examined the following international cases of pronounced credit contractions: Finland in 1992–97 (Pazarbasioglu, 1997), Indonesia in 1997–99, and Thailand in 1998–2000 (both in Ghosh and Ghosh, 1999); Japan in 1993–99 (Woo, 1999); Korea in 1997–98 (in Ghosh and Ghosh, 1999, and Kim, 1999); and the United States in 1990–93 (Bernanke and Lown, 1991; and Peek and Rosengren, 1995). In Table 3 we compare the episodes by showing the declines in credit–GDP with respect to a peak level for each of the slowdown episodes. For the non–Latin American countries, experiences vary widely. Finland, Indonesia, and Thailand register dramatic declines in a relatively short period of time, Korea experienced a short–lived and small drop, whereas

Table 2.

Latin American Credit Slowdowns in Historical Context

article image
Sources: International Financial Statistics; and authors′ calculations.

Defined as a period containing at least one year in which the credit–GDP ratio is at least 5 percentage points (absolute) or 25 percent (relative) above its trend value.

Trend calculated using a Hodrick–Prescott filter on the original series, with smoothing factor =100.

Twelve–month growth rate based on the latest monthly observation available: August 2001 (Brazil, Venezuela, Mexico, and Peru) and September 2001 (Chile, Argentina, Bolivia, Colombia).

Average taken over the longest period within 1960–2000 for which data are available.

Figure 2.
Figure 2.

Latin America: Absolute Deviations in the Credit–GDP Ratio with Respect to Trend

Citation: IMF Staff Papers 2002, 005; 10.5089/9781589061231.024.A006

Table 3.

Latin American Credit Slowdowns in Comparison to Selected International Cases

Credit to the Private Sector by the Banking System (except where otherwise indicated

article image
Sources: International Financial Statistics; Bank of Korea; and authors' calculations.

Japan and the United States exhibit much more modest but extended downturns in credit. Of all the countries in the table, the largest total decline in credit–GDP was 56 percentage points, experienced by Venezuela (1983–95), followed closely by Finland′s 44 percentage point contraction (1992–97). On a per year basis, the largest declines were those of post–crisis Thailand and Indonesia, where over 10 percentage points were lost per year. Credit–GDP fell by almost 30 percentage points in the case of all banks in Chile (1985–91) and by about 19 percentage points for DMBs in Mexico, or over 3 percentage points per year, and over 4 percentage points per year for all banks in Colombia.

On the surface it may appear that the recent credit slowdowns in Latin America have been less pronounced than in other parts of the world. However, the experience of earlier episodes—including Chile, Finland, and Venezuela—show that they did become severe over time. So far the recent Latin American slowdowns have been relatively short, lasting about 3 years as opposed to 13 years in Venezuela, 7 in Chile, and 6 in Finland. In Latin America we might be witnessing the first stages of a more protracted process. Table 3 also shows that the credit contractions continued well into 2001 in all countries, and even intensified in several cases.

Table 4.

The Recent Credit Slowdown in Latin America—A Summary

article image
Sources: International Financial Statistics;and authors′ calculations.

The Latin American Credit Slowdown: Some Stylized Facts

We now turn to a closer look at the evolution of credit and bank behavior in our eight countries, focusing on a shorter sample period, 1980–2000. In Tables 4 and 5 we show average annual real growth rates of credit and deposits for different subperiods, as well as the ratios of credit to deposits and credit to total assets. For the six countries in which the slowdown is recent (Argentina, Bolivia, Brazil, Colombia, Mexico, and Peru; see Table 4), we divide the period into three portions: (1) the preliberalization period of the 1980s, characterized by relatively repressed financial markets and thereby low credit growth; (2) the expansion period of the early 1990s, spurred in part by financial liberalizations undertaken at the beginning of the decade; and (3) the recent slowdown period. The expansion is defined as ending in the year when credit–GDP reached its peak of the 1990s. In Brazil and Mexico, credit peaked in 1995; in Argentina, Bolivia, Colombia, and Peru, in 1998.

For the other two countries, Chile and Venezuela (Table 5), the slowdown occurred earlier, and the more recent period is characterized by a recovery in credit.

Table 5.

Earlier Credit Slowdowns in Latin America—A Summary

article image
Sources:International Financial Statistics;and authors′ calculations.

In Chile, financial markets were liberalized much sooner than in the rest of the region, while Venezuela observed a positive terms of trade shock after 1973, and faced severe macroeconomic distress in the early 1990s. Thus, we divide the sample period for these two countries into the following subperiods: (1) credit expansion in the 1970s and early 1980s; (2) credit slowdown in the late 1980s and early 1990s; and (3) credit recovery in the late 1990s.

There are several consistent patterns across the first group of countries. The acceleration and subsequent slowdown observed in credit growth also occurred on the deposit side. The latter may have been the result of deregulation and financial reform programs that reduced taxes on financial intermediation, liberalized interest rates, and encouraged savings through the banking system. In the more recent period there has been a disintermediation process following a period of financial turmoil, in which capital outflows took place and bank deposits fell. Thus, it may be that the slowdown in credit was driven primarily by a slowdown in deposits, and banks merely reacted passively in response to a squeeze on their loanable funds. However, it appears that although deposit growth was in fact a key factor, it was not the entire story. Indeed, except in the case of Brazil, the slowdown in credit is more pronounced than that in deposits. In fact, in some cases deposits continued to grow in real terms while real credit fell.

This behavior was not always symmetrical across the expansion and slowdown. In Argentina and Colombia, the expansion phase was more pronounced for deposits, while the slowdown was more pronounced for credit. For the other three countries, the entire cycle of the 1990s was more pronounced for credit; thus the loan–deposit and loan–asset ratios increased during the expansion and fell during the slowdown. This is also the case in the earlier slowdowns, Chile and Venezuela. In fact, in Venezuela the recent credit recovery period is characterized by a slowdown in deposits.

A Decomposition of Credit Growth

In this section we present a summarized balance sheet for commercial banks, and observe the major changes that took place in bank activities in the subperiods described above. We decompose credit growth to the private sector into banks’ sources of funds or alternative uses of funds. Based on IFS data, we define the major balance sheet categories as:

article image

The distinction between “sources” and “uses” is not clear cut. For example, accumulating foreign assets may be viewed as an alternative use, while contracting a foreign debt could be considered a source of funds. The purpose was to keep a simple classification that would yield a small number of groups that could be easily identified. We proceeded to decompose credit growth to the private sector in the three periods by using the following balance sheet identity:


This decomposition yields an average real growth rate for credit in each subperiod, which is then decomposed into each source and alternative use of funds. We then calculate the differences in growth rates between subperiods and also decompose these into sources and alternative uses. We present a summary of this calculation in Tables 6 and 7, where we indicate the four balance sheet factors that had the greatest impact on changes in credit growth in each phase and for each country.6 Also, we define “contributing” factors as those that moved in a direction consistent with the change in credit growth rate, and “offsetting” factors as those that moved in the opposite direction. For example, in Argentina real credit accelerated from an average growth rate of –6.7 percent in the preliberalization period to 7.7 percent in the expansion period, thus a turnaround of 14.4 percentage points. Table 6 shows that two major factors contributed to this, namely deposits and capital, whose growth rate accelerated by 20.7 and 2.8 percentage points, respectively. Their effect was offset to a certain extent, however, by a deceleration in the banks’ position with respect to the central bank (4.8 percentage points) and with respect to the nonfinancial public sector (2.1 percentage points).

A number of characteristics of the credit cycle become apparent. First, the single most important factor contributing to the changes in credit growth was deposit growth. For five of the eight countries it had the largest impact during the expansion, and for another five it had the largest impact during the slowdown. There are no cases in which deposits were not the dominant factor in at least one of the two phases, and in four countries—Argentina, Bolivia, Colombia, and Peru—they were the dominant factor in both phases.

Second, the comovement of deposit and credit growth dropped during the slowdown. Taking a simple average of the relative contribution of deposits to credit growth changes across the eight countries, we find that deposits accounted for 63 percent of the acceleration in credit growth during the expansion, and for 53 percent of the deceleration during the slowdown. Excluding Venezuela, where the recent credit upswing came together with a decline in deposits, the comovement drops from 77 to 50 percent, when comparing expansion and contraction phases.

Third, the net position with the central bank generally played an important role, albeit for different reasons. In three countries net central bank credit has been highly procyclical. In the case of Brazil, it was the dominant accelerating factor during the expansion and the dominant decelerating factor during the contraction. In Chile, it was the dominant accelerating element during the expansion and a major decelerating factor during the contraction. In Mexico, net central bank credit was the dominant factor decelerating credit in the recent slowdown. On the other hand, in Argentina and Peru, and to a lesser extent in Bolivia and Colombia, it played an offsetting role, contracting during the expansionary period and expanding during the contraction. In the case of Peru it was the major offsetting factor in both phases, as reserve requirements were kept high during the expansion and relaxed during the decline.

Table 6.

The Recent Credit Slowdown In Latin America: Major Factors Affecting Changes in Credit Growth

article image
Sources: International Financial Statistics; and authors′ calculations.
Table 7.

Earlier Credit Slowdowns in Latin America: Major Factors Affecting Changes in Credit Growth

article image
Sources: International Financial Statistics; and authors′ calculations.

Fourth, except in Colombia and Mexico, net foreign liabilities (NFL) played an important role in credit growth. They were a major contributing factor during both phases in Bolivia, Peru, and Chile and played a major role in Venezuela′s contraction. On the other hand, they were a major offsetting factor in the recent contractions of Argentina and Brazil.

Fifth, a fiscal factor (NFPS in Table 6) played a major role in accelerating Mexico′s credit in the early 1990s and in Venezuela′s recent credit recovery. In Bolivia and Chile, and to a lesser extent in Argentina and Colombia, credit to the public sector on occasions acted as a major offsetting factor. Regarding the slowdown, net credit to the NFPS was a significant contributing factor for all countries except Bolivia and Chile. This may reflect a fiscal expansion that squeezed out private sector credit, and/or a reduction in risk taking by banks, who may have decided to increase their holdings of relatively safe government securities rather than increase loans.

Finally, the private sector credit slowdown appears to be more complex than the expansion, as it involves changes in a larger number of balance sheet items. We generally find that the aggregate relative contribution of a small number of factors—two contributing, two offsetting—falls from the expansion to the slowdown. For example, in Argentina the major factors accounted for 115 percent of the expansion, but for only 77 percent of the slowdown.

II. Econometric Analysis

A vast empirical literature has recently emerged, which focuses on the possible causes of credit contraction in different countries.7 In many cases the approach followed is to use data at the macro level to identify whether or not specific situations of credit contraction can be identified as being a “credit crunch,” defined as a situation in which, for a given level of deposits, banks refuse to increase interest rates on their loans to market–clearing levels, and thus, excess demand for credit remains unsatisfied. Such a situation can emerge either because banks′ perceptions of corporate risk are too high, or because they simply do not have enough capital to accommodate riskier loans. In our econometric estimation, we follow this type of approach (as in Pazarbasioglu, 1997; Ghosh and Ghosh, 1999; Kim, 1999; and Barajas, Lopez, and Oliveros, 2001), which is sufficiently general to permit us to separate demand from supply–side determinants of credit growth, and allows us to capture any possible “credit crunch” phenomenon.

Methodological Issues

We estimated aggregate demand and supply functions for credit in three of the countries analyzed: Colombia, Mexico, and Peru. The econometric approach used is based on pioneering work by Laffont and Garcia (1977) and Sealey (1979), and consists of estimating the system of supply and demand functions under the assumption that, at a given point in time, the credit market may either exhibit equilibrium, or temporary excess demand or supply owing to imperfect flexibility in the interest rate in the short run. Thus, in addition to capturing the main determinants of credit growth, the approach also allows one to assess whether a situation of excess demand, or credit crunch, occurred during an episode of declining credit growth.

Actual credit observed at time t, Ch is defined as lying either on the supply curve (excess demand), on the demand curve (excess supply), or on both (equilibrium):


where Cts and Ctd are the supply and demand functions, respectively, defined as a function of the vectors of explanatory variables X1t and X2f and error terms:


Without adequate information on the price adjustment process, and assuming that the errors u1 and u2 are normally distributed, a likelihood function may be determined for the above model. Defining λt as the probability that a given observation t lies on the supply function (that is, demand is greater than supply), and given that g (·) is the joint density function for supply and demand derived from the joint density function for u1 and u2, then the density function for Ct under the assumption of excess demand is:


Similarly, the density function under the assumption of excess supply is:


The unconditional density function is then equal to:


The likelihood function is L = Πth(Ct), and the corresponding log likelihood to be maximized subject to the parameter values is:


Maximization of equation (6) permits the estimation of both equations as well as the estimation of the probability of observed credit lying on either of the curves, λ.8 As in previous studies, we used OLS estimates of the supply and demand functions to provide initial values for the coefficients and standard errors. Although Maddala (1983) warns that the likelihood function is unbounded for certain parameter values, we found the OLS estimates to perform well as starting values, and convergence tended to occur relatively quickly.

Specification Issues

In specifying the supply and demand functions, several issues emerge. First, identification of the model requires that one or more variables included in one function be excluded from the other. Past studies using this approach have used one key variable, lending capacity (LC), to distinguish the supply from the demand function. Lending capacity, defined as the availability of loanable funds, would affect banks′ ability to lend but would have no impact on credit demand. We followed this approach, including as LC a subset of loanable funds over which banks have little discretionary power, thereby constituting an exogenous determinant of credit supply.

A second specification issue involves variables reflecting the macroeconomic and business environment, since one expects credit demand to be positively related to them, and credit supply to respond to these variables to the extent that they affect the riskiness of loans. As in previous studies, we included manufacturing production indices (MANUF), GDP measures (y), the output gap (GAP),9 the expected inflation rate (INFE)10 as a measure of macroeconomic stability, and the stock market index (STKMKT). As discussed in Ghosh and Ghosh (1999), the last variable may also reflect the availability of alternatives to bank credit—equity finance in particular—from the demand side. Thus, the stock market index will have a positive coefficient if the macro conditions effect dominates, or a negative one if the substitution effect dominates.

Third, as in previous studies, we included interest rates on deposits (id) and on government securities (ig) as proxies for the opportunity cost of bank credit either from the demand or supply side. Fourth, in contrast to previous studies, for Peru and Mexico we included the country–specific JP Morgan EMBI price,11 which we expected to have dual effects similar to those of the stock market index. A positive impact on credit demand would arise if the dominant effect was an improvement in macroeconomic perspectives, while a negative effect would arise when EMBI signaled an increase in foreign investors′ willingness to lend to domestic residents, thus drawing customers away from bank credit.

Finally, also in contrast to previous studies, we included two additional variables specific to the supply function: the ratio of nonperforming loans to total loans (NPL), and the ratio of loan–loss provisions to nonperforming loans (PROV).12 The former reflects past credit risk and may signal financial difficulties in the banking system, while the latter reflects the severity of regulations on risk-taking in lending activities. We expected banks to diminish credit supply in response to mounting credit risk and/or increasing provisions being imposed upon them.13 These variables also provided us with additional identifying distinctions between the supply and demand functions.

III. Estimation Results

In the next sections we describe the estimation results for each of the three countries, as shown in Tables810. In all cases, the dependent variable was the natural log of real credit to the private sector, LRCRED, with subtle differences from country to country, as we explain below. We report the estimated parameters with their t–statistics, the value of the log–likelihood function, and the R2 for the initial OLS estimations of the supply and demand functions. Since the dependent variable was found to exhibit a unit root in all three cases, we conducted tests to determine whether real credit and its predicted value formed a cointegrating vector.14 In virtually all cases we rejected the null hypothesis of no cointegration, at least at the 5 percent level.15 Thus, even though real credit has a unit root, it is appropriate to estimate the model in levels.


We used both monthly and quarterly data. Monthly data were available from September 1992 to March 2001, while quarterly information was available from the last quarter of 1991 to the second quarter of 2001. Quarterly estimations allowed us to use the quarterly GDP directly. We do not report OLS estimations, which in most instances provided coefficients that were statistically significant, and with the expected sign. In order to partially deal with endogeneity problems, we use lagged values of provisions, nonperforming loans, manufacturing output, stock market indices, and real lending capacity. Most of the variables are also taken in natural logs, so that coefficients can be interpreted as elasticities.

In Table 8 we report the results of the disequilibrium ML estimation. In the first three columns we present the results for DMBs, and in the last four columns we present those for the entire financial system, adjusted by loan write offs to offset accounting changes that may not reflect the true flow of bank credit to the private sector.16 All estimations yield some common results. Various measures of macroeconomic conditions appear to be significant in the supply and demand functions. In particular, lagged manufacturing output is positively related to both supply and demand in all monthly estimations, and real GDP is positively related to loan demand in the quarterly estimations. The stock market index appears to reflect present and future economic conditions, rather than equity finance as a substitute for bank credit; thus, its coefficient in the demand function is positive and significant across monthly and quarterly estimations. The output gap exhibits a negative coefficient, similar to the result obtained by Ghosh and Ghosh (1999)

for Indonesia, and may reflect the fact that firms increase (decrease) their demand for external financing over own resources in bad (good) times. Since this variable was only available for a shorter time period, we excluded it from all but regression (1). The deposit interest rate performed better than the government interest rate as a proxy for the opportunity cost of bank credit from the demand side. As the deposit rate increased relative to the lending rate, it tended to have a positive impact on credit demand. Finally, as in previous studies, we found lending capacity to be positively and significantly related to credit supply, with an elasticity approaching unity in quarterly regressions. We used a summary measure of real lending capacity (RLC), defined as deposits minus reserves,17 as well as including deposits and reserves separately. The results overall were similar, but we observed better performance, primarily in the demand function when we used the summary measure.

Some differences also arise across equations. Expected inflation tended to be slightly negatively related to credit demand, but only exhibited a large tstatistic in regression (4), a monthly estimation for adjusted financial system credit. Monthly estimations were noticeably better at capturing the sensitivity of demand and supply to real or nominal lending interest rates. In particular, lending interest rates were rarely significant in the supply function when using quarterly data, and the results for other variables tended to improve when this variable was excluded from the supply equation. Finally, although the measures of credit risk and regulation always exhibited the expected negative sign in the supply function, both were rarely significant at the same time. The interaction variable (NPL*PROV, or the ratio of loan–loss provisions to total loans) was negative and significant throughout, however, thus showing that a combination of credit risk and regulatory power had a negative impact on the willingness of banks to provide credit.


There are important differences in the Mexican case with respect to the Colombian estimations. On the one hand, we included JP Morgan′s EMBI price as an explanatory variable in the demand function. This allows us to test whether, particularly after the 1994 “tequila crisis,” greater availability of foreign financing permitted the private sector to grow in spite of severe problems in the financial sector. If this was the case, then we would expect a negative coefficient in the demand function; as the index increases, foreign financing becomes more easily available and, other things constant, demand for credit from domestic banks should decline. A second difference is that we were unable to assemble the dataset for the period prior to 1993 while for the post–1993 period, there are variables for which more than one source is available.

In Table 9 we report six different estimations, for two different time periods, and using two different dependent variables. For regressions (1)–(3), the dependent variable is the stock of private credit from deposit money banks as reported by the IFS, we used monthly data for 1993:12 to 2000:12. This time period allows us to capture part of the pre–1995 lending boom as well as the subsequent downturn. Similar to the Colombian case, for regressions (4)–(6) we used an adjusted credit series constructed by the Banco de Mexico, which incorporates the effects of two major debt restructuring programs,18 one consisting of a partial write off of bank debts funded, and another consisting of a government purchase of nonperforming loans. The time period is 1997:02–2001:05, and thus includes only the downturn phase. In most estimations, we used real GDP in order to proxy for aggregate macroeconomic conditions. Since GDP is available quarterly, we used the same real GDP level for each month within the quarter. In the last estimation, regression (6), we used the monthly index of manufacturing production.

Table 8.

Colombia: Credit Demand and Supply

(Maximum likelihood disequilibrium estimation)

article image
Notes: t–statistics shown in parentheses. Significance levels of 5 percent (**) and 10 percent (*) indicated. The letter L at the beginning of a variable name denotes natural logarithm.

Adjusted credit is defined as the stock of credit plus loan write–offs. In addition, in regression (4) two small state–owned financial institutions are excluded.

Defined as deposits plus foreign liabilities minus reserves.

There are two sources of data for nonperforming loans. The first, which we used in regressions (1)–(3), was obtained from the World Bank. This series is quarterly, and was available from 1992:4 to 2000:4. Monthly series for both nonperforming loans and the loan–loss provisions were obtained from the Banco de Mexico, but were available only starting in 1997:01, and so were used in regressions (4)–(6).

In all six estimations, the EMBI price appears to be negatively and significantly related to credit demand, reflecting the degree of substitution for Mexican firms between domestic and foreign financing. As expected, demand for loans is negatively related with the lending interest rate, and the respective coefficient is statistically significant in all but one estimation. Finally, the demand for financial sector loans is positively associated with economic activity, either proxied by real GDP or manufacturing output. However, this association is significant only in two of the six estimations reported.

Regarding the supply of bank loans, in general it is associated in a positive and statistically significant manner with lending capacity, which we defined in two alternative ways: DEP1, equal to the sum of demand, time, and foreign currency deposits; and DEP2, which also includes money market funds.19 As expected, the supply of loans depends positively on their interest rates, and most of the regressions yielded a statistically significant coefficient.

We attempted to capture the opportunity costs of alternative uses of bank funds, by including the interest rate on government paper (ig) in the supply function. We obtain a positive, albeit not statistically significant, coefficient. The best estimations were obtained when ig was not included.

Table 9.

Mexico: Credit Demand and Supply

(Maximum likelihood disequilibrium estimation)

article image
Notes: t–statistics shown in parentheses. Significance levels of 1 percent (***), 5 percent (**), and 10 percent (*) indicated. The letter L at the beginning of a variable name denotes natural logarithm.

Defined as the stock of credit plus adjustments to incorporate private sector loan restructuring programs beginning in mid–1995.

This dummy variable takes a value of zero up until 1994:12, and then unity thereafter, thus testing for a structural change after the credit expansion period.

Regarding loan quality, the two groups of regressions shown in Table 9 must be interpreted separately. For the longer period (regressions (1)–(3)) we use the quarterly World Bank non–performing loans data, and are unable to control for provisions. We also interacted NPL with a dummy variable equal to zero until the end of 1994 and equal to 1 afterwards (D9501), so as to test for the possibility that regulation and supervision became tighter—and thus bank behavior became more sensitive to credit risk—after the financial crisis of late 1994 and early 1995.

Our results are consistent with a significant shift in bank behavior after the 1994 crisis. Prior to 1995, the supply of loans to the private sector was positively associated with the level of NPL. One possible interpretation of this result is that, in the context of weak regulation, banks were probably “gambling for resurrection,” by increasing their supply of credit when credit risk was rising.

Following the crisis, and given that we cannot reject the hypothesis that the sum of the coefficients of NPL and of the interactive variable (D9501 *NPL) is equal to zero, this type of adverse behavior would have ceased. Note, however, that according to this group of regressions, we do not obtain the expected negative relationship between credit supply and NPL, not even in the post–crisis period.

The results reported in regressions (4)–(6) of Table 9 seem much more comforting. For the shorter, post–crisis period, and using the redefined version of credit, in all three estimations we consistently obtain negative and statistically significant coefficients both for nonperforming loans as well as for provisions. As was mentioned above, unfortunately we were unable to obtain data on provisions for the period prior to 1997.


The regression results for Peru are shown in Table 10, which we divide into two pairs of regressions according to the credit series used. In regressions (1) and (2) we use the summary variable for real lending capacity (RLC), and in regressions (3) and (4) we disaggregate it into its components, deposits and reserves.

Several differences emerge with respect to the other two countries analyzed. First, the Peruvian case appears to be the only one in which expected inflation is a consistently significant variable explaining demand for credit. Private agents in Peru appear to extract more information about macroeconomic conditions from the evolution of inflation. Second, in contrast to the case of Mexico, the EMBI price is now positively related to credit demand, presumably because it reflects macroeconomic conditions more than the private sector′s ability to obtain financing abroad. Third, especially in regressions (1) and (2), credit demand is affected significantly by a wider set of macro indicators. Finally, it proved very difficult to arrive at supply and demand functions that both exhibited their expected signs for the lending interest rate. This may be related to the high degree of dollarization of bank activities in Peru, which makes it difficult to construct an appropriate lending interest rate covering both domestic and foreign currency operations,20 but it remains a puzzle.

Table 10.

Peru: Credit Demand and Supply

(Maximum likelihood disequilibrium estimation)

article image
Notes: t–statistics shown in parentheses. Significance levels of 5 percent (**) and 10 percent (*) indicated. The letter L at the beginning of a variable name denotes natural logarithm.