The predictive power of economic indicators in consumer credit risk management
A study by Chase Manhattan Bank and Seton Hall University uses U.S. macroeconomic data and industry-wide credit card charge-off data to examine the predictive power of economic indicators in consumer credit risk management. The study finds that changes in economic variables, such as household debt service burden, unemployment rate, consumer confidence index, inflation rate, personal bankruptcy filings, and stock market returns, have strong forecasting power for changes in the consumer credit card charge-off rates. These indicators can be incorporated into credit risk modeling to enhance risk management in the credit card industry.
$1.72 trillion in U.S. consumer credit outstanding can give pause to the retail lender. This was the amount (excluding real estate mortgages) as of September 2002 and included $720 billion in revolving debt and $1 trillion in nonrevolving debt. Revolving debt–credit repeatedly available up to a specified amount as periodic repayments are made–carries its own red flags, having grown from 15% of the total consumer credit outstanding in 1980 to 42% in 2003. In the U.S., most revolving debt is in the form of unsecured credit card loans. With the rising level of credit card debt, charge-offs on credit card loans also increased dramatically–from 2.0% in 1985 to 6.4% in 2002. Tremendous growth in credit card debt, coupled with a substantial decline in consumer credit quality, signals the need for intense risk management focus on the evaluation of a new applicant for credit, increases or decreases in credit lines, transaction authorizations, balance consolidations, collections of credit card debt, and other related account servicing and management practices.
Managing Credit Card Risk
As for any business, credit risk management in the consumer credit card industry aims to measure and control the risk exposure to achieve maximum profits while limiting exposure to defaults, that is, charge-offs. Credit card issuers typically integrate business strategic analysis with legal and regulatory constraints to establish credit policy and guidelines. Credit policy helps an institution to develop strategies consistent with the profitability expectations of the institution within an expected level of asset quality.
At the heart of consumer credit risk management is measuring and controlling the probability of default (PD); yet consumer risk has yet to receive the level of attention as its commercial counterpart.
The corporate bond market has provided an important function of market discipline on corporate risk-taking behavior.
* Ratings agencies (such as S&P, Moody’s, and Fitch) constantly evaluate the credit risk of companies and downgrade the credit ratings of their debt instruments when increased level of credit risk is observed.
* Debt market investors react to a downgrade with a higher required yield and lower pricing on the traded debt instruments.
The relationship between the default risk in the corporate debt market and general economic environment has been widely researched, resulting in the general consensus that macroeconomic conditions directly affect the credit-rating transitions and PD. In the consumer credit area, the role of the ratings agency and the debt market disciplinary function is replaced by the consumer credit bureau reporting agency. Financial institution and other retail lenders supply account and payment information about their customers on a regular basis. The agency obtains additional information from federal, state, and county courts, then combines the credit record and public record to create a credit profile for each consumer. Card issuers rely on credit bureau scores, along with customized in-house data, because the information is based on updated, objective, comprehensive information, such as transactions data, debt to income burden, and payment history. The applications of credit bureau scores range from credit risk evaluations to account performance prediction.
There’s always a “however.”
However, as the overall economy goes through good times and bad times, consumers’ default risk behavior could be affected accordingly. Even though future economic perspectives are often informally recognized by risk managers in strategic decision making, they are not explicitly modeled in the credit risk management process. This leads to the question: How do different macroeconomic scenarios affect the credit card portfolio’s risk exposure? In good economic conditions, credit card charge-off tends to be low; in bad economic conditions, credit card charge-off tends to be high. It seems common sense that average consumer credit quality should decline as economic conditions deteriorate. If true, the next question concerns the most predictive economic indicators and the lag time. Then the challenge is to incorporate these economic indicators in credit risk modeling.
The correlations between quarterly changes in charge-off rates and lag growth in various economic variables are illustrated in Table 1. (1) There are significant correlations between quarterly change in the credit card charge-off rate and the following lag economic variables:
* Lag one quarter growth rate in household debt service burden (positive); lag one quarter change in employment (positive).
* Lag one quarter and lag four quarter changes in consumer confidence index (negative).
* Lag three quarters growth rate in personal bankruptcy filings.
* Lag three quarters growth rate in personal bankruptcy filings (positive).
* Lag one quarter and lag two quarter GDP growth rate (negative).
* Lag two quarters and lag four quarters return in S&P 500 stock market index (negative).
In contrast, the change in other consumer loan charge-off rates does not have significant correlation with most of the lag economic variables, except for the four-quarters-lag growth rate in household debt service burden. Other consumer loans are typically secured loans with collaterals, and thus are less sensitive to changes in economic conditions. Preliminary correlation analysis seems to indicate that macroeconnomic variables are indeed more predictive of credit card loan default risk than the default risk on other consumer loans.
Model Selection and Results
Three alternative predictive models for the quarterly change in credit card charge-off rate were compared, using various economic variables (see Table 2).
1. The first model includes four autoregressive terms and four lags for each of the following economic variables: growth rate in household debt service burden; change in unemployment rate; change in consumer confidence index; inflation rate; growth rate in personal bankruptcy filings; growth rate in GDP; and return in S&P 500 stock market index and GDR.
2. The second model includes only lag economic variables that have significant correlations with the quarterly change in credit card charge-off rates as measured in Table 1.
3. A stepwise regression is performed in the third model to select the “best” subset of variables from those included in model 1.
The first model explains 86% of the time series variations in the quarterly change in credit card charge-off rate, while the second and third models explain 71% and 80%, respectively. Although the first model has the highest R-square, it is at the cost of including a larger number of predictive variables. After adjusted for the number of predictive variables used in the models, the third model shows the highest adjusted R-square, the lowest regression standard error, and the highest F statistics.
In examining the degree of multi-collinearity (moving along the same line) in the three models using the variance inflation factor (VIF) statistics, a large VIF implies that the variable is redundant with other independent variables in the model. The first, second, and third models have average VIF of 3.02, 1.53, and 1.48, respectively. This indicates that the stepwise model (model 3) has the least exposure to multi-collinearity.
Another commonly used model selection method is the Alkaike information criterion (AIC). The notion of the information criterion is to provide a measure of information that strikes a balance between the goodness of fit and parsimonious specification of the model. The AICs for the first, second, and third models are 0.667, 0.833, and 0.446. Model 3 has the smallest AIC and hence is the optimal model. Finally, we conduct an out-of-sample forecast for the period between the first quarter of 1999 and the third quarter of 2002 using the model estimates from 1986 to 1998. The third model has the lowest mean absolute forecasting error, and hence is the best predictive model among the three.
In summary, various model indicators (adjusted R-square, F statistics, VIE AIC, and out-of-sample mean absolute forecasting error) consistently suggest that the third model best predicts the quarterly change in credit card loan charge-off. The predictive variables from the third model (see Table 2) are as follows:
1. Lag one quarter growth rate in household debt service burden (positive sign).
2. Lag two quarters growth rate in household debt service burden (negative sign).
3. Lag four quarters growth rate in household debt service burden (positive sign).
4. Lag one quarter change in unemployment rate (positive sign).
8. Lag one quarter change in consumer confidence index (negative sign).
6. Lag two quarters change in consumer confidence index (positive sign).
7. Lag one quarter inflation rate (negative sign).
8. Lag two quarters inflation rate (positive sign).
9. Lag three quarters inflation rate (positive sign).
10. Lag four quarters growth rate in personal bankruptcy filings (positive sign).
11. Lag two quarter return in S&P 500 stock market index (negative sign).
Overall, a higher growth in household debt service burden indicates a higher level of consumer indebtedness and thus predicts a higher growth in the charge-off rate for credit card loans. Though there is a negative coefficient with the two-quarters-lag growth in debt service burden, it is much smaller and less significant than the positive coefficient associated with the one-quarter-lag growth in debt service burden. The four-quarters-lag growth in household debt service burdcn also has a large and significant coefficient with the quarterly change in credit card loan charge-off rate. As for the change in employment rate, its one-quarter-lag has a significant and large positive impact on the change in credit card loan charge-off rate. This indicates that weakness in the labor market has a substantial and immediate impact on the consumer credit quality and default risk. A decline in the previous quarter’s consumer confidence index also leads to a higher growth in credit card charge-off rate, although this negative relationship is partially offset by a positive relationship with the two-quarters-lag change in consumer confidence index. The conflicting signs indicate a sharp initial reaction followed by a partial correction.
The effect of lag inflation rates on the change in credit card charge-off rate is more complicated. Initially, a higher inflation rate lowers the real burden of the outstanding debt, and hence there is a significant negative relationship between one-quarter-lag inflation rate and change in credit card charge-off rate. However, as the credit card lenders revise APR upward to reflect the inflationary environment, interest charges increase dramatically and so do credit card charge-off rates. This accounts for large significant positive coefficients for both the two-quarters-lag and three-quarters-lag inflation rate. The effect of growth in personal bankruptcy filings on the change in credit card charge-off rate is significantly positive with four quarters lag, indicating the amount of time it takes for the bankruptcy filings to be fully reflected in the consumer credit card charge-offs. Finally, a lower return in the overall stock market reduces the net worth of households and also indicates weakness in economic perspectives; therefore we observe a significant negative relationship between the two-quarters-lag return in the S&P 500 stock market index and the change in credit card charge-off rate.
Given current and historical data on actual economic indicators, charge-off rates can be predicted for the next quarter (short term). In addition, this predictive model also can make use of forecasted economic data as input and predict longer-term charge-off rates for credit card portfolios. The shorter-term prediction can be done with greater accuracy while the longer-term prediction may be confounded by the noise of economic data forecasts.
Applications and Conclusions
Given the significance of the economic indicators in forecasting changes in credit card charge-off rates, there is a strong need for economic variables to be incorporated in the credit card underwriting and risk management practices. In particular, household debt service burden, unemployment rate, consumer confidence index, inflation rate, personal bankruptcy filings, stock market returns, and like variables have strong predictive power for changes in the consumer credit card charge-off rates. Credit decisions based on credit-scoring models developed using past credit history do not fully account for all the information available to credit card issuers, given that the economic environment may be substantially different from those that prevailed historically. This can be remedied quantitatively by enhancing the credit-scoring models currently used by card issuers with important economic indicators concluded from this study. Such augmentation of the existing credit-scoring models will allow lenders to determine the probability of default on credit card loans based on the predictive power of both past consumer credit behaviors and leading economic indicators.
Success in using credit risk models is measured nut only by the predictive power of the statistical models but also by the profit-and-loss performance of its credit card portfolios. Given the predictive power of business cycle indicators in consumer credit card charge-off behavior seen here, consistent applications of economic variables into various credit risk management areas (such as credit card account acquisition, account underwriting, credit line management, debt collection, transaction authorization, and balance consolidation) is highly recommended. A forecasted increase in credit card charge-off rates–based on leading economic indicators using the predictive model developed in this study–can guide credit card lenders to reduce exposure to the increased default risk or to achieve a better charge-off adjusted return on the credit card portfolio by strategically changing the credit score threshold required for account approval, credit line increase, credit line decrease, etc.
This predictive model can also be applied to various consumer market segments (such as subprime and prime sectors or subsectors with different credit score ranges) to formulate more detailed and effective credit card risk management strategies. Historical data on credit bureau scores, economic data, charge-offs, and return performance can be used to estimate predictive models concerning the profit and loss sensitivity of various credit card consumer sectors or subsectors to changes in economic indicators. These model estimates can then be applied to different segments to predict future charge-off and return performance for each sector or subsector, and thus allow credit risk managers to formulate optimal credit risk management strategies based on leading economic data.
Quarterly data on charge-off rates for credit card loans and other consumer loans comes from the Federal Reserve Board. Data on household debt service burden, unemployment rate, consumer price index, and GDP also come from the Federal Reserve Board. * Other economic data used includes personal bankruptcy filings from the Administrative Office of the U.S. Courts, a consumer confidence index from the University of Michigan, and the S&P 500 stock market index.
All time series are checked for unit roots using the augmented Dickey-Fuller test and are found to be nonstationary. We compute a first differencing of these variables (i.e., either changes or percentage changes) to induce stationarity. Quarterly data is available for all series from the first quarter of 1986 to the third quarter of 2002. During the sample period, the average charge-off rate is 4.16% for credit card loans and only 0.93% for all other consumer loans, while the average quarterly increase in charge-off rate is 0.063% for credit card loans and 0.013% for other consumer loans.
* Charge-offs, which are the value of loans removed from the books and charged against loss reserves, are measured net of recoveries as a percentage of average loans and annualized. Other consumer loans do not include residual real estate loans and credit card loans. The household debt service burden is an estimate of the ratio of debt payments to disposable personal income. Debt payments consist of the estimated required payments on outstanding mortgage and consumer debt.
(Source: Federal Reserve Board)
Change in Credit
Lag Charge-Off Rate
Auto-correlation 1 -0.328 ** (0.006)
2 0.086 (0.485)
3 0.233 * (0.058)
4 -0.195 (0.116)
Correlation with Lag Growth 1 0.416 ** (0.000)
Rate in Household Debt 2 -0.175c (0.146)
Service Burden 3 0.281 ** (0.018)
4 0.179 (0.139)
Correlation with Lag Change 1 0.446 ** (0.000)
in Unemployment Rate 2 -0.115 (0.343)
3 0.047 (0.702)
4 -0.016 (0.893)
Correlation with Lag Change 1 -0.436 ** (0.000)
in Consumer Confidence 2 0.112 (0.355)
Index 3 -0.002 (0.985)
4 -0.221 * (0.065)
Correlation with Lag Inflation 1 -0.193 (0.109)
Rate 2 0.197 (0.102)
3 0.240 ** (0.046)
4 0.031 (0.797)
Correlation with Lag Growth 1 0.129 (0.288)
Rate in Personal Bankruptcy 2 -0.030 (0.808)
Filings 3 0.310 ** (0.009)
4 0.117 (0.334)
Correlation with Lag Growth 1 -0.243 ** (0.043)
Rate in GDP 2 -0.207 * (0.086)
3 -0.102 (0.401)
4 0.056 (0.646)
Correlation with Lag Return 1 0.070 (0.563)
in S&P 500 Stock Market 2 -0.290 ** (0.015)
Index 3 0.170 (0.159)
4 -0.236 ** (0.049)
Change in Other
Lag Charge-Off Rate
Auto-correlation 1 -0.469 ** (0.000)
2 0.234 * (0.055)
3 0.036 (0.772)
4 -0.002 (0.987)
Correlation with Lag Growth 1 0.140 (0.249)
Rate in Household Debt 2 0.190 (0.115)
Service Burden 3 0.089 (0.466)
4 0.238 ** (0.047)
Correlation with Lag Change 1 0.168 (0.165)
in Unemployment Rate 2 -0.041 (0.735)
3 -0.073 (0.551)
4 -0.093 (0.445)
Correlation with Lag Change 1 -0.003 (0.980)
in Consumer Confidence 2 -0.017 (0.888)
Index 3 -0.018 (0.884)
4 -0.003 (0.981)
Correlation with Lag Inflation 1 0.080 (0.509)
Rate 2 0.011 (0.928)
3 0.163 (0.178)
4 -0.062 (0.611)
Correlation with Lag Growth 1 0.084 (0.490)
Rate in Personal Bankruptcy 2 0.173 (0.151)
Filings 3 -0.009 (0.943)
4 0.023 (0.850)
Correlation with Lag Growth 1 -0.136 (0.262)
Rate in GDP 2 0.063 (0.606)
3 -0.028 (0.817)
4 0.148 (0.221)
Correlation with Lag Return 1 0.026 (0.831)
in S&P 500 Stock Market 2 -0.065 (0.592)
Index 3 -0.002 (0.988)
4 0.073 (0.550)
Note: The significance level of correlation coefficient is
in parentheses. ** Significant at 5% * Significant at 10%
Comparison of Three Predictive Models for the
Change in Credit Card Loan Charge-Off Rate (CORCC_C)
Predictive Models Model 1 Model 2
Independent Variables Four Lags of Lag Economic Indicators
(Predictors) Various Economic that Have Significant
Indicators Correlation with the
Standard Error 0.2896 0.3331
of the Regression
R-Square 85.86% (highest) 70.60%
Adjusted R-Square 72.15% 62.67%
F-value 6.26 ** 8.92 **
Alkaike Information 0.6673 0.8334
Average Variance Inflation 3.02 1.53
Mean Absolute 0.2854 0.2751
Predictive Models Model 3 (best)
Independent Variables Stepwise Selection of Lag
(Predictors) Economic Indicators
Standard Error 0.2762 (lowest)
of the Regression
Adjusted R-Square 75.36% (highest)
F-value 16.76 ** (highest)
Alkaike Information 0.4457 (lowest)
Average Variance Inflation 1.48 (lowest)
Mean Absolute 0.2207 (lowest)
** Significant at 5% * Significant at 10%
(1) Since our purpose is to find the predictive power of economic indicators in credit card risk management, we use only lag economic variables, rather than concurrent economic variables.
Athreya, K., “The Growth of Unsecured Credit: Are We Better Off,” Economic Quarterly Federal Reserve Bank of Richmond, Summer 2001, vol. 87, No. 3, pp.11-33.
Bangia, A., F.X. Diebold, T. Schuermann, “Ratings Migration and the Business Cycle, with Applications to Credit Portfolio Stress Testing,” Penn Institute for Economic Research, Working Paper No. 01-004, 2000.
Carey, M., “Credit Risk in Private Debt Portfolios,” Journal of Finance, August 1998, pp. 1363-1387.
Crouhy, M., D. Galai, and R. Mark, “Prototype Risk Rating System,” Journal of Banking and Finance, January 2001, pp. 47-95.
Fama, E., “Term Premiums and Default Premiums in Money Markets,” Journal of Financial Economics, 1986, vol. 17, no. 1, pp. 175-196.
Ferri, G., L.G. Liu, and G. Majnoni, “The Role of Rating Agency Assessments in Less Developed Countries: Impact of the Proposed Basel Guidelines.” Journal of Banking and Finance, January 2001, pp. 115-148.
Fridson, M., C. Garman, and S. Wu, “Real Interest Rates and the Default Rates on High-Yield Bonds,” Journal of Fixed Income, September 1997, pp. 27-34.
Grasa, A. A., Econometric Model Selection: A New Approach, Kluwer Publishers, 1999.
Lawrence, D., and A. Solomon, Managing a Consumer Lending Business, Solomon Lawrence Partners, 2002.
Mays, E., (ed.), Credit Risk Modeling: Design and Application, Glenlake Publishing Company, 1998.
Monfort, B. and C. Mulder, “Using Credit Ratings for Capital Requirements on Lending to Emerging Market Economies–Possible Impact of a New Basel Accord,” International Monetary Fund, Working Paper, 2000.
Nelson, R.W., Credit Card Risk Management, Warren Taylor Publications, 1997.
Nickell, P., W. Perraudin, and S. Varotto, “Stability of Rating Transitions,” Journal of Banking and Finance, vol. 24, no. 1/2, 2000, pp. 203-228.
Xu can be reached by e-mail at email@example.com. Liu can be reached at Jiong.Liu@chase.com.
[c] 2003 by RMA. Jiong Liu is AVP of Knowledge Management, Chase Regional Banking, J.P. Morgan & Chase Co. Xiaoqing Eleanor Xu is Associate Professor of Finance, W. Paul Stillman School of Business, Seton Hall University.
COPYRIGHT 2003 The Risk Management Association
COPYRIGHT 2005 Gale Group