A. Introduction

1. The effects on macroeconomic and financial shocks on the banking system are of great interest to policymakers. Given the important linkages between the real and the financial sectors, particularly during volatile periods, a quantification of these effects may prove useful to anticipate potential changes in the level of risk faced by financial institutions. An option for estimating such effects is a modeling framework of banking system risk, combined with econometric models incorporating relevant macroeconomic and financial sector variables.

2. This paper uses contingent claims analysis (CCA) to assess risks to the Colombian banking sector. The CCA approach is based on the estimation of the default probability by an entity on its obligations, and is widely used by rating agencies to assess creditworthiness in the corporate sector. The paper builds on recent work using the CCA methodology undertaken in other Latin American countries (e.g., Gray and Walsh, 2008; Souto, 2008). It uses Moody’s-KMV estimates of expected default frequencies (EDFs) for Colombian banks, which are constructed with market-based data.² The advantage of this indicator is that it incorporates a forward-looking, collective view of risk by market participants. Such a forward-looking element cannot be captured by traditional balance sheet measures of bank vulnerabilities. In addition, EDFs directly incorporate the effects of volatility on risk, and thus better capture the nonlinearities that are particularly important during periods of distress.

3. The paper estimates the effects of changes in selected macroeconomic and financial variables on default probabilities for a sample of Colombian banks. The sample includes five banks for which market-based default probabilities are available. These account for over one-half of the Colombian banking sector. The paper estimates step-wise regressions for both individual banks and the aggregate system, as well as panel regression for the pooled data for individual banks. Reflecting the heterogeneity of the financial institutions included in the sample, results from step-wise regressions differ significantly across banks, although they generally show the vulnerability of banks to changes in key economic activity variables and financial market conditions. The paper also tests for Granger causality between default probabilities and a traditional indicator of bank credit vulnerability (the ratio of non-performing loans to total loans—NPL ratio) for the banking sector as a whole. The results show that EDFs are a leading indicator of NPL ratios.

4. Our findings are comparable to those from similar studies undertaken for other emerging markets in Latin America. Using market data, Gray and Walsh (2008, on Chile), and Souto, Tabak, and Vazquez (2008, on Brazil) find that bank soundness is significantly related to macro-financial variables, while also finding evidence of heterogeneity between banks. Blavy and Souto (2008, on Mexico) using risk indicators estimated with book value data for the Mexican banking sector find equally consistent results. In a similar vein, Souto (2007) finds that CCA indicators capture well the episodes of bank stress in Uruguay.

5. The paper is organized as follows. Section B presents a brief description of the CCA methodology. Section C provides some background on the Colombian banking system. Section D discusses the data and results of the CCA analysis. Section E presents some concluding remarks.

B. The CCA Framework

6. The CCA offers some distinct advantages over other approaches to vulnerability analysis.³ First, the CCA takes balance sheet information and combines it with current and forward-looking financial market prices to compute risk-adjusted, marked-to-market balance sheets (i.e. asset values). Using financial market price information to derive forward-looking risk-adjusted balance sheets is a significant advantage compared to an analysis based on past balance sheet information. Second, the CCA distinguishes itself from other vulnerability analysis in that it incorporates market volatility when estimating credit risk. Volatility is crucial in capturing nonlinear changes in risk, especially during times of stress, when small shocks can gain momentum and trigger systemic repercussions.

7. Under the CCA model, a firm’s equity can be viewed as a (junior) contingent claim on the residual value of its assets. In the event of default, all the firm’s assets are used to pay the senior stakeholders (e.g. debt holders); otherwise equity holders receive the difference between the value of assets and debt. Thus, the equity of the firm can be seen as a call option on the residual value of the firm’s assets. This framework enables a rich characterization of a firm’s (or sovereign’s) balance sheet.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

• Download Figure
• Download figure as PowerPoint slide

• Download Figure
• Download figure as PowerPoint slide

8. With information on the market value and volatility of equity and the value of debt, it is possible to estimate the implied value for assets and volatility through the Black and Scholes option formula. Firms are assumed to default whenever the value of assets fall below a given “distress” barrier. It is then possible to estimate a set of credit risk indicators, including distance-to-distress, default probability, credit spread, and expected losses in the event of default.

9. At the same time, the CCA is only a tool to help understand credit risk, and certain caveats need to be kept in mind. First, as with any model that uses market information, the quality of the output of the CCA depends on how well market information reflects changes in fundamentals. If markets are imperfect, or react excessively to changes in fundamentals, the CCA indicators may overstate risk. It should also be noted that the CCA’s capacity to capture off-balance sheet risks is also imperfect, as it depends on how well these are incorporated into equity prices.

C. Some Background on the Colombian Banking System

10. Financial intermediation in Colombia has grown considerably in recent years and compares well with regional levels. The recovery from the crisis of the late 1990s and strong economic growth have contributed to a sizeable expansion of financial intermediation during the current decade. As of mid-2008, credit to the private sector was equivalent to 33 percent of GDP, over 10 percentage points up from the beginning of the decade, and slightly above the average for the largest economies in the region. Credit growth in Colombia was very strong in 2005∓07 but decelerated significantly over the last year, following substantial monetary tightening by the Banco de la República aimed at containing inflationary pressures and protecting financial sector stability.⁴ Consumer credit expanded particularly fast, and currently accounts for slightly under 30 percent of total credit to the private sector.

Credit to the private sector, 2000∓08

(In percent of GDP)

Citation: IMF Staff Country Reports 2009, 024; 10.5089/9781451808964.002.A001

1/ Simple average of Argentina, Brazil, Chile, Mexico, Peru and Venezuela.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Credit to the private sector, 2000∓08

(In percent of GDP)

Citation: IMF Staff Country Reports 2009, 024; 10.5089/9781451808964.002.A001

1/ Simple average of Argentina, Brazil, Chile, Mexico, Peru and Venezuela.

• Download Figure
• Download figure as PowerPoint slide

Credit to the private sector, 2000∓08

(In percent of GDP)

Citation: IMF Staff Country Reports 2009, 024; 10.5089/9781451808964.002.A001

1/ Simple average of Argentina, Brazil, Chile, Mexico, Peru and Venezuela.

• Download Figure
• Download figure as PowerPoint slide

11. The Colombian banking system exhibits somewhat high levels of concentration. The system is made up of 17 institutions, all but one privately-owned. In mid-2008, the 5 largest banks accounted for about 70 percent of the system’s total assets and liabilities. This level of concentration is comparable to that in other large Latin American countries.

12. Foreign presence in the sector is relatively modest. As of mid-2008, subsidiaries of foreign banks accounted for slightly over 20 percent of the system’s total assets and liabilities. While featuring a rising trend over the last few years, this level of foreign participation is significantly smaller than the average for the region, which currently exceeds 30 percent.

13. Financial soundness indicators for the banking system are solid. The ratio of non-performing loans (NPLs) to total loans has risen over the last couple of years—especially for consumer credit—but remains low at about 4 percent, and NPLs are well provisioned.⁵ The system is well capitalized, with risk-adjusted capital adequacy well above minimum requirements (9 percent) for most banks. Despite the recent rise in NPL ratios, indicators of profitability have remained strong, with returns on assets hovering around 2½ percent over the last year. Liquidity levels appear adequate, with short-term assets exceeding short-term liabilities by a comfortable margin. With the pace of economic activity decelerating rapidly from the high levels of 2006∓07,⁶ credit risk has become an important source of risk in the banking system (Banco de la República, 2008).

Non-performing loans

(In percent)

Citation: IMF Staff Country Reports 2009, 024; 10.5089/9781451808964.002.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Non-performing loans

(In percent)

Citation: IMF Staff Country Reports 2009, 024; 10.5089/9781451808964.002.A001

• Download Figure
• Download figure as PowerPoint slide

Non-performing loans

(In percent)

Citation: IMF Staff Country Reports 2009, 024; 10.5089/9781451808964.002.A001

• Download Figure
• Download figure as PowerPoint slide

Capital adequacy and profitability

(In percent)

Citation: IMF Staff Country Reports 2009, 024; 10.5089/9781451808964.002.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Capital adequacy and profitability

(In percent)

Citation: IMF Staff Country Reports 2009, 024; 10.5089/9781451808964.002.A001

• Download Figure
• Download figure as PowerPoint slide

Capital adequacy and profitability

(In percent)

Citation: IMF Staff Country Reports 2009, 024; 10.5089/9781451808964.002.A001

• Download Figure
• Download figure as PowerPoint slide

D. Implementing the CCA Approach for Colombia

Data

14. Available market-based EDF estimates for Colombia cover about half of the banking system over a relatively short time period. Moody’s-KMV estimates of EDFs are available for five banks (two domestically-owned institutions and three foreign subsidiaries), which as of mid-2008 accounted for about 55 percent of total banking system assets and liabilities. The data series starts in the fourth quarter of 2003. While data on EDFs are available on a monthly (and even daily) frequency, this paper uses a quarterly series, given that information for key macroeconomic variables affecting bank risk are available only on a quarterly basis.

15. Estimated EDFs are highly correlated with traditional measures of bank risk. The market-based EDF data show a decline in bank risk through the first quarter of 2006, and a subsequent upward movement of risk. These trends are strongly correlated with the evolution of the banking system’s non-performing loan ratio over the sample period.

EDF probablities and NPL ratios

(In percent)

Citation: IMF Staff Country Reports 2009, 024; 10.5089/9781451808964.002.A001

Source: Creditedge (Moody’s- KMV).

• Download Figure
• Download figure as PowerPoint slide

View Full Size

EDF probablities and NPL ratios

(In percent)

Citation: IMF Staff Country Reports 2009, 024; 10.5089/9781451808964.002.A001

Source: Creditedge (Moody’s- KMV).

• Download Figure
• Download figure as PowerPoint slide

EDF probablities and NPL ratios

(In percent)

Citation: IMF Staff Country Reports 2009, 024; 10.5089/9781451808964.002.A001

Source: Creditedge (Moody’s- KMV).

• Download Figure
• Download figure as PowerPoint slide

Main results

16. Stepwise regressions reveal that economic activity and financial market conditions have an important impact on EDFs, but with significant differences across banks. OLS stepwise regressions were ran over time series for individual banks and a set of macrofinancial variables (Table 1). We started with the full set of variables and allowed the regressions to extract the variables for which the covariate coefficient were found to be significant at the 10 percent level or greater. The optimal model specification differed across banks (Table 2). Among the most salient results are the following:

Table 1.

Variables Used in Regressions

Table 1.

Variables Used in Regressions

Dependent Variable:
	Bank KMV EDF
	System KMV EDF
Covariates:
Bank Specific		Bank System	Macrofin Domestic	Macrofin Foreign
- Bank Cred. Growth		- Syst. Cred. Growth	- IGBC	- VIX
- Bank NPL		- System NPL	- FX Rate	- S&P 500
		- Dom. Int. Rate	- Comm. Prices
		- CPI Inflation	- US Federal Rate
		- Unemp. Rate
		- Real GDP growth
		- EMBI Spread

View Table

Table 1.

Variables Used in Regressions

Dependent Variable:
	Bank KMV EDF
	System KMV EDF
Covariates:
Bank Specific		Bank System	Macrofin Domestic	Macrofin Foreign
- Bank Cred. Growth		- Syst. Cred. Growth	- IGBC	- VIX
- Bank NPL		- System NPL	- FX Rate	- S&P 500
		- Dom. Int. Rate	- Comm. Prices
		- CPI Inflation	- US Federal Rate
		- Unemp. Rate
		- Real GDP growth
		- EMBI Spread

Table 2.

Stepwise Regressions for Individual Banks

Notes: **, * means that the regression coefficients are statistically different than zero at the 1%, 5%, and 10% significance levels respectively.

Table 2.

Stepwise Regressions for Individual Banks

	Bank 1	Bank 2	Bank 3	Bank 4	Bank 5
VIX	0.002 **
IGBC			-0.004 ***	-0.002 **
FX Rate			0.009 **
Real GDP Growth					-0.046 **
Bank NPL				-0.132 *
Domestic Int. Rate		9.636 **
US Federal Rate		0.081 *			-0.181 ***
R-Square	31.2%	33.4%	68.6%	37.8%	55.0%
F-Stat	6.81 **	3.51 *	15.32 ***	4.25 **	8.54 ***

Notes: **, * means that the regression coefficients are statistically different than zero at the 1%, 5%, and 10% significance levels respectively.

View Table

Table 2.

Stepwise Regressions for Individual Banks

	Bank 1	Bank 2	Bank 3	Bank 4	Bank 5
VIX	0.002 **
IGBC			-0.004 ***	-0.002 **
FX Rate			0.009 **
Real GDP Growth					-0.046 **
Bank NPL				-0.132 *
Domestic Int. Rate		9.636 **
US Federal Rate		0.081 *			-0.181 ***
R-Square	31.2%	33.4%	68.6%	37.8%	55.0%
F-Stat	6.81 **	3.51 *	15.32 ***	4.25 **	8.54 ***

Notes: **, * means that the regression coefficients are statistically different than zero at the 1%, 5%, and 10% significance levels respectively.

• An increase in the Colombian stock market index (IGBC) is associated with a decrease in the individual KMV EDF for two of the five banks. This is consistent with the CCA framework, in which a higher equity/capital would result in an improved credit risk profile for banks. For one bank, an increase in risk aversion (as measured by the VIX index) is associated with a decrease in the KMV EDF. This would be consistent with the view that, as the degree of risk aversion increases, investors would shift their investment positions away from Colombia and into safer assets.

• For another bank, real GDP growth is associated with an improvement in the KMV EDF, reflecting the impact of real sector ‘good periods’ on the financial sector.

• The differences in the optimal specification are surprising. The fit of the equations is good, even though the same variable is rarely significant across different banks. These results mirror those of Gray and Walsh (2008).

17. A stepwise regression for the aggregate banking system also underscores these macro-financial linkages (Table 3). The results suggest that the Moody’s-KMV EDF indicator would improve (decrease) if: (i) real GDP growth increases; (ii) the Colombian stock market increases (revealing an increase in equity/capital); (iii) credit growth increases (potentially associated with the higher returns that banks can reap from these loans);⁷ and (iv) the U.S. S&P500 index decreases (in which case investors may look for better returns in other markets such as Colombia).

Table 3.

Stepwise Regression for the Banking System

Notes: **, * means that the regression coefficients are statistically different than zero at the 1%, 5%, and 10% significance levels respectively.

Table 3.

Stepwise Regression for the Banking System

	Bank System
Real GDP Growth	-0.009 **
System Credit growth (LC)	-0.002 *
IGBC	-0.003 ***
S&P 500	0.007 ***
R-Square	82.9%
F-Stat	14.55 ***

Notes: **, * means that the regression coefficients are statistically different than zero at the 1%, 5%, and 10% significance levels respectively.

View Table

Table 3.

Stepwise Regression for the Banking System

	Bank System
Real GDP Growth	-0.009 **
System Credit growth (LC)	-0.002 *
IGBC	-0.003 ***
S&P 500	0.007 ***
R-Square	82.9%
F-Stat	14.55 ***

Notes: **, * means that the regression coefficients are statistically different than zero at the 1%, 5%, and 10% significance levels respectively.

18. Panel regressions reveal similar results. In order to use all available market-based information for Colombian banks, we ran a linear dynamic panel-data model with fixed effects, based on the Arellano and Bond (1991) GMM estimator. The results (Table 4) show that: (i) an increase in the Colombian stock market index is associated with a decrease in EDFs, as predicted by the underlying framework; (ii) an increase in NPLs is also associated with a decrease in EDFs, consistent with the result found for credit growth in the stepwise regression for the aggregate baking system; and, finally, (iii) an increase in the interest rate is contemporaneously associated with an increase in EDFs. It is important to clarify that the interest rate used was the U.S. interest rate for the 3-month T-bill (proxy for the “risk-free” asset) plus the EMBI spread. Thus, an increase in this interest rate is consistent with a expected deterioration in the banks’ EDFs.

Table 4.

Dynamic Panel Results

Notes: ***, **, * means that the regression coefficients are statistically different than zero at the 1%, 5%, and 10% significance levels respectively.

Table 4.

Dynamic Panel Results

	Bank EDF
Bank EDF (lag1)	0.185 ***
Domestic Int. Rate	1.670 ***
IGBC	-0.002 ***
Bank NPL	-0.015 ***

Notes: ***, **, * means that the regression coefficients are statistically different than zero at the 1%, 5%, and 10% significance levels respectively.

View Table

Table 4.

Dynamic Panel Results

	Bank EDF
Bank EDF (lag1)	0.185 ***
Domestic Int. Rate	1.670 ***
IGBC	-0.002 ***
Bank NPL	-0.015 ***

Notes: ***, **, * means that the regression coefficients are statistically different than zero at the 1%, 5%, and 10% significance levels respectively.

19. Causality tests reveal that the system EDF is a leading indicator of the system NPL ratio, with a one-quarter lag. Granger causality tests fail to reject the hypothesis that the system EDF cannot explain the system NPL at the one percent confidence level.⁸ At the same time, the tests indicate that the NPL ratio does not explain future movements in the EDF. These results suggest that the system EDF provides useful information beyond what is contained in the NPL ratio and can thus be useful in early warning systems.

Table 5.

Results for Granger Causality Tests

Notes: ***, **, * means that the regression coefficients are statistically different than zero at the 1%, 5%, and 10% significance levels respectively.

Table 5.

Results for Granger Causality Tests

Equation	Excluded	Chi-Square
System EDF	System NPL	16.36 ***
System NPL	System EDF	3.53

Notes: ***, **, * means that the regression coefficients are statistically different than zero at the 1%, 5%, and 10% significance levels respectively.

View Table

Table 5.

Results for Granger Causality Tests

Equation	Excluded	Chi-Square
System EDF	System NPL	16.36 ***
System NPL	System EDF	3.53

Notes: ***, **, * means that the regression coefficients are statistically different than zero at the 1%, 5%, and 10% significance levels respectively.

E. Concluding Remarks

20. Contingent claims analysis can be a useful tool to assess risk in the Colombian financial sector. When compared to traditional measures of bank risk, EDF estimates capture well developments in banking system over the period for which EDF market-based data are available. Importantly, Granger causality analysis shows that the EDFs are a leading indicator of traditional measures of bank risk. Given the availability of EDF estimates on a high frequency, CCA could thus be a helpful tool in improving the monitoring of the financial system’s health.

21. Empirical estimates show that macroeconomic and financial shocks have an important bearing on banking sector vulnerabilities. Econometric results show that a positive shock to economic growth reduces risk for the banking system as a whole. Interest rate increases and downward movements in the domestic stock market are associated with a rise in bank risk. Results for individual banks vary widely, reflecting the heterogeneity of the Colombian banking sector. However, financial institutions in general show vulnerability to changes in key domestic macroeconomic variables, as well as to changes in domestic and international financial market conditions. These results are consistent with findings for other Latin American countries in studies using CCA.

22. There is scope to improve and extend the application of the current analysis of the Colombian banking system. The quality of the estimates for the relationship between macroeconomic and financial market variables, and banking sector risk could be strengthened through the use of longer time series. This could be done by using higher frequency data, which are readily available for all variables included in the various models, except for economic growth. The latter could, however, be proxied by combined forward-looking indicators of economic activity currently produced on a monthly basis in Colombia (e.g., industrial production, retail sales, etc.).⁹ Another approach could be to use principal component analysis to summarize the impact on bank risk of changes in diverse macrofinancial variables, building, for example, on work done for Chile by Gray and Walsh (2008).