Credit Card Portfolio Research Articles

Credit card loss forecasting: Some lessons from COVID

AbstractModels developed by banks to forecast losses in their credit card portfolios have generally performed poorly during the COVID‐19 pandemic, particularly in 2020, when large forecast errors were observed at many banks. In this study, we attempt to understand the source of this error and explore ways to improve model fit. We use account‐level monthly performance data from the largest credit card banks in the U.S. between 2008 and 2018 to build models that mimic the typical model design employed by large banks to forecast credit card losses. We then fit these on data from 2019 to 2021. We find that COVID‐period model errors can be reduced significantly through two simple modifications: (1) including measures of the macroeconomic environment beyond indicators of the labor market, which served as the primary macro drivers used in many pre‐pandemic models and (2) adjusting macro drivers to capture persistent/sustained changes, as opposed to temporary volatility in these variables. These model improvements, we find, can be achieved without a significant reduction in model performance for the pre‐COVID period, including the Great Recession. Moreover, in broadening the set of macro influences and capturing sustained changes, we believe models can be made more robust to future downturns, which may bear little resemblance to past recessions.

Machine Learning model based on the Light Gradient Boosting Machine to predict the probability of default in customers of the Credit Card portfolio

Bank loans are a widely used means of payment in recent times, more and more people are accessing products such as credit cards, loans, etc. Banks have implemented classic prediction models, the vast majority based on logistic regression since it allows great interpretability for the business and the effect of the model variables. The purpose of this research is to perform a predictive analysis on the probability of customer default in the credit card portfolio using a risk score. The dataset used is the so-called default of credit card clients Data Set from the UCI Machine Learning DB, the approach is quantitative and the methodology is descriptive analytics, techniques based on gradient boosters will be used to make the prediction, among the trained algorithms We have Logistic Regression with WOE, CatBoost, As a result, the light gradient enhancement machine (LightGBM) tuned with a Bayesian search was obtained, obtaining a GINI of 57.4, which improves by +6 points to the Logistic Regression with Woe and by +3p to XgBoost and CatBoost. Finally, obtaining the Gain and Shapley values made up for the lack of interpretability of the variables, allowing better decision making when evaluating clients. Likewise, as future work, it is intended to add unstructured variables that allow the Model's indicators to be improved.

Modeling Recoveries of US Leading Banks Based on Publicly Disclosed Data

The credit risk management process is a critical element that allows financial institutions to withstand economic downturns. Unlike the methods regarding the probability of default, which have been deeply addressed after the financial crisis in 2008, recovery rate models still need further development. As there are no industry standards, leading banks are modeling recovery rates using internal models developed with different assumptions. Therefore, the outcomes are often incomparable and may lead to confusion. The author presents the concept of a unified recovery rate analysis for US banks. He uses data derived from FR Y-9C reports disclosed by the Federal Reserve Bank of Chicago. Based on the historical recoveries and credit portfolio book values, the author examines the distribution function of recoveries. The research refers to a credit card portfolio and covers nine leading US banks. The author leveraged Vasicek’s one-factor model with the asset correlation parameter and implemented it for recovery rate analysis. This experiment revealed that the estimated latent correlation ranges from 0.2% to 1.5% within the examined portfolios. They are large enough to impact the recovery rate volatility and cannot be treated as negligible. It was shown that the presented method could be applied under US Comprehensive Capital Analysis and Review exercise.

An interpretable Comprehensive Capital Analysis and Review (CCAR) neural network model for portfolio loss forecasting and stress testing

This paper proposes an interpretable nonlinear neural network (NN) model that translates business regulatory requirements into model constraints. The model is then compared with linear and nonlinear NN models without the constraint for Comprehensive Capital Analysis and Review (CCAR) loss forecasting and scenario stress testing. Based on a monthly time series data set of credit card portfolio chargeoffs, the model outperforms the benchmark linear model in mean squared errors, and the improvement increases with network architecture complexity. However, the NN models could be vulnerable to overfitting, which could make the model uninterpretable. The constrained NN model ensures model interpretability at a small cost to model performance. Thus, it is insufficient to measure the model’s statistical performance without ensuring model interpretability and clear CCAR scenario narratives.

From incurred loss to current expected credit loss: a forensic analysis of the allowance for loan losses in unconditionally cancelable credit card portfolios

From incurred loss to current expected credit loss: a forensic analysis of the allowance for loan losses in unconditionally cancelable credit card portfolios

Loss rate forecasting framework based on macroeconomic changes: Application to US credit card industry

A major part of the balance sheets of the largest U.S. banks consists of credit card portfolios. Hence, managing the charge-off rates is a vital task for the profitability of the credit card industry. Different macroeconomic conditions affect individuals’ behavior in paying down their debts. In this paper, we propose an expert system for loss forecasting in the credit card industry using macroeconomic indicators. We select the indicators based on a thorough review of the literature and experts’ opinions covering all aspects of the economy, consumer, business, and government sectors. The state of the art machine learning models are used to develop the proposed expert system framework.We develop two versions of the forecasting expert system, which utilize different approaches to select between the lags added to each indicator. Among 19 macroeconomic indicators that were used as the input, six were used in the model with optimal lags, and seven indicators were selected by the model using all lags. The features that were selected by each of these models covered all three sectors of the economy. Using the charge-off data for the top 100 US banks ranked by assets from the first quarter of 1985 to the second quarter of 2019, we achieve mean squared error values of 1.15E−03 and 1.04E−03 using the model with optimal lags and the model with all lags, respectively. The proposed expert system gives a holistic view of the economy to the practitioners in the credit card industry and helps them to see the impact of different macroeconomic conditions on their future loss.

Default avoidance on credit card portfolios using accounting, demographical and exploratory factors: decision making based on machine learning (ML) techniques

Effective and thorough credit-risk management is a key factor for lending institutions, as significant financial losses can arise from the borrowers’ default. Consequently, machine learning methods can measure and analyze credit risk objectively when at the same time they face increasingly attention. This study analyzes default payment data from a credit cards’ portfolio containing some 30,000 clients from Taiwan with twenty-three attributes and with no missing information. We compare prediction accuracy of seven classification methods used, i.e. KNN, Logistic Regression, Naive Bayes, Decision Trees, Random Forest, SVC, and Linear SVC. The results indicate that only few out of most of the typical variables used can adequately analyze default characteristics in terms of lending decisions. The results provide effective feedback to credit evaluators, lending institutions and business analysts for in-depth analysis. Also, they mention to the importance of the precautionary borrowing techniques to be used to better understand credit-card borrowers’ behavior, along with specific accounting, historical and demographical characteristics.

Exposure at default models with and without the credit conversion factor

The Basel II and III Accords allow banks to calculate regulatory capital using their own internally developed models under the advanced internal ratings-based approach (AIRB). The Exposure at Default (EAD) is a core parameter modelled for revolving credit facilities with variable exposure. The credit conversion factor (CCF), the proportion of the current undrawn amount that will be drawn down at time of default, is used to calculate the EAD and poses modelling challenges with its bimodal distribution bounded between zero and one. There has been debate on the suitability of the CCF for EAD modelling. We explore alternative EAD models which ignore the CCF formulation and target the EAD distribution directly. We propose a mixture model with the zero-adjusted gamma distribution and compare its performance to three variants of CCF models and a utilization change model which are used in industry and academia. Additionally, we assess credit usage – the percentage of the committed amount that has been currently drawn – as a segmentation criterion to combine direct EAD and CCF models. The models are applied to a dataset from a credit card portfolio of a UK bank. The performance of these models is compared using cross-validation on a series of measures. We find the zero-adjusted gamma model to be more accurate in calibration than the benchmark models and that segmented approaches offer further performance improvements. These results indicate direct EAD models without the CCF formulation can be an alternative to CCF based models or that both can be combined.

Forecasting credit card portfolio losses in the Great Recession: a study in model risk

Forecasting credit card portfolio losses in the Great Recession: a study in model risk

Credit Risk Modeling in Segmented Portfolios: An Application to Credit Cards

The Great Recession offers a unique opportunity to analyze the performance of credit risk models under conditions of economic stress. We focus on the performance of models of credit risk applied to risk-segmented credit card portfolios. Specifically, we focus on models of default and loss and analyze three important sources of model risk: model selection, model specification, and sample selection. Forecast errors can be significant along any of these three model-risk dimensions. Simple linear regression models are not generally outperformed by more complex or stylized models. The impact of macroeconomic variables is heterogeneous across risk segments. Model specifications that do not consider this heterogeneity display large projection errors across risk segments. Prime segments are proportionally more severely impacted by a downturn in economic conditions relative to the subprime or near-prime segments. The sensitivity of modeled losses to macroeconomic factors is conditional on the model development sample. Models estimated over a period that does not incorporate a significant period of the Great Recession may fail to project default rates, or loss rates, consistent with those experienced during the Great Recession.

Forecasting Credit Card Portfolio Losses in the Great Recession: A Study in Model Risk

Credit card portfolios represent a significant component of the balance sheets of the largest US banks. The charge‐off rate in this asset class increased drastically during the Great Recession. The recent economic downturn offers a unique opportunity to analyze the performance of credit risk models applied to credit card portfolios under conditions of economic stress. Specifically, we evaluate three potential sources of model risk: model specification, sample selection, and stress scenario selection. Our analysis indicates that model specifications that incorporate interactions between policy variables and core account characteristics generate the most accurate loss projections across risk segments. Models estimated over a time frame that includes a significant economic downturn are able to project levels of credit loss consistent with those experienced during the Great Recession. Models estimated over a time frame that does not include a significant economic downturn can severely under-predict credit loss in some cases, and the level of forecast error can be significantly impacted by model specification assumptions. Higher credit-score segments of the portfolio are proportionally more severely impacted by downturn economic conditions and model specification assumptions. The selection of the stress scenario can have a dramatic impact on projected loss.

Forecasting and stress testing credit card default using dynamic models

We present discrete time survival models of borrower default for credit cards that include behavioural data about credit card holders and macroeconomic conditions across the credit card lifetime. We find that dynamic models which include these behavioural and macroeconomic variables provide statistically significant improvements in model fit, which translate into better forecasts of default at both account and portfolio levels when applied to an out-of-sample data set. By simulating extreme economic conditions, we show how these models can be used to stress test credit card portfolios.

Applying the zero-adjusted inverse Gaussian model to predict probability of default and exposure at default for a credit card portfolio

Applying the zero-adjusted inverse Gaussian model to predict probability of default and exposure at default for a credit card portfolio

Credit Risk Analysis of Credit Card Portfolios Under Economic Stress Conditions

We develop an empirical framework for the credit risk analysis of a generic portfolio of revolving credit accounts and apply it to analyze a representative panel data set of credit card accounts from a credit bureau. These data cover the period of the most recent deep recession and provide the opportunity to analyze the performance of such a portfolio under significant economic stress conditions. We consider a traditional framework for the analysis of credit risk where the probability of default (PD), loss given default (LGD), and exposure at default (EAD) are explicitly considered. The unsecure and revolving nature of credit card lending is naturally modeled in this framework. Our results indicate that unemployment, and in particular the level and change in unemployment, plays a significant role in the probability of transition across delinquency states in general and the probability of default in particular. The effect is heterogeneous and proportionally has a more significant impact for high credit score and for high-utilization accounts. Our results also indicate that unemployment and a downturn in economic conditions play a quantitatively small, or even irrelevant, role in the changes in account balance associated with changes in an account’s delinquency status, and in the exposure at default specifically. The impact of a downturn in economic conditions and, in particular, changes in unemployment on the recovery rate and loss given default is found to be large. These findings are of particular relevance for the analysis of credit risk regulatory capital under the IRB approach within the Basel II capital accord.

Arbitration Clauses in Credit Card Agreements: An Empirical Study

This article uses a newly available database of consumer credit card agreements to take the first, in‐depth empirical look at why credit card issuers use arbitration clauses. Based on a sample of credit card agreements made available by 298 issuers under the Credit Card Accountability Responsibility and Disclosure Act of 2009, it finds that while most credit card agreements include arbitration clauses, the substantial majority of credit card issuers (247 of 298, or 82.9 percent) do not use arbitration clauses in their credit card agreements. The article also finds that credit card issuers are more likely to use arbitration clauses when they (1) specialize in making credit card loans; (2) make riskier credit card loans; and (3) have a larger credit card portfolio. Conversely, issuers are less likely to use arbitration clauses when they are (1) mutually owned (i.e., credit unions) rather than shareholder owned (i.e., banks); and (2) are located in states in which class arbitration waivers are unenforceable. These empirical findings have potentially important implications for a number of timely policy questions, such as: What sorts of options are available to consumers who wish to obtain a credit card that is not subject to an arbitration clause? How might increased regulation of arbitration (whether by Congress or by the Consumer Financial Protection Bureau) affect the market for credit card loans? and How are businesses likely to respond to the Supreme Court's recent decision in AT&T Mobility LLC v. Concepcion?

Forecasting and stress testing the risk-based capital requirements for revolving retail exposures

This article presents a tractable and empirically sound technique for generating stressed probabilities of default (PDs), which are then used to derive loss rates for the provisioning of a bank's risk-based capital. This work is in response to the recent regulatory findings attributed to the Supervisory Capital Assessment Program (SCAP) stress tests of 2009, which revealed weaknesses in the existing regulatory and economic capital approaches. The SCAP projected losses of approximately US$82.4 billion in banks’ credit card portfolios for 2010, highlighting the need for better forecasting and stress testing of revolving retail exposures. This study proposes a timely model that will improve the ability of banks to determine the capital adequacy of revolving retail exposures. Using options theory, we discuss why an obligor may default and produce estimates of expected losses from our stressed PDs so as to determine loss provisions. This method relies on the simulation of PD distributions via changes in selected macroeconomic variables and the cardholder's debt-to-income ratio. The methodology offers the flexibility of being tractable and scalable to data in the issuer's credit card portfolio by geography and credit quality of the obligor.

Arbitration and Consumer Credit

This paper uses a newly available database of consumer credit card agreements to take the first, in-depth empirical look at why credit card issuers use arbitration clauses. Based on a sample of credit card agreements made available by 298 issuers under the Credit Card Accountability Responsibility and Disclosure Act of 2009, it finds that while most credit card loans outstanding (95.1%) are subject to cardholder agreements with arbitration clauses, the substantial majority of credit card issuers (247 of 298, or 82.9%) do not use arbitration clauses in their credit card agreements. The paper also finds that credit card issuers are more likely to use arbitration clauses when they (1) specialize in making credit card loans; (2) make riskier credit card loans; and (3) have a larger credit card portfolio. Conversely, issuers are less likely to use arbitration clauses when they are (1) mutually owned (i.e., credit unions) rather than shareholder-owned (i.e., banks); and (2) are located in states in which class arbitration waivers are unenforceable. These empirical findings have potentially important implications for a number of timely policy questions, such as: what sorts of options are available to consumers who wish to obtain a credit card that is not subject to an arbitration clause; how increased regulation of arbitration (whether by Congress or by the Consumer Financial Protection Bureau) might affect the market for consumer credit; and how businesses are likely to respond to the Supreme Court’s recent decision in AT&T Mobility LLC v. Concepcion.

Checking for asymmetric default dependence in a credit card portfolio: A copula approach

Traditional credit risk models adopt the linear correlation as a measure of dependence and assume that credit losses are normally-distributed. However some studies have shown that credit losses are seldom normal and the linear correlation does not give accurate assessment for asymmetric data. Therefore it is possible that many credit models tend to misestimate the probability of joint extreme defaults. This paper employs Copula Theory to model the dependence across default rates in a credit card portfolio of a large UK bank and to estimate the likelihood of joint high default rates. Ten copula families are used as candidates to represent the dependence structure. The empirical analysis shows that, when compared to traditional models, estimations based on asymmetric copulas usually yield results closer to the ratio of simultaneous extreme losses observed in the credit card portfolio. Copulas have been applied to evaluate the dependence among corporate debts but this research is the first paper to give evidence of the outperformance of copula estimations in portfolios of consumer loans. Moreover we test some families of copulas that are not typically considered in credit risk studies and find out that three of them are suitable for representing dependence across credit card defaults.

Decision Rule Extraction for Regularized Multiple Criteria Linear Programming Model

Due to the flexibility of multi-criteria optimization, Regularized Multiple Criteria Linear Programming (RMCLP) has received attention in decision support systems. Numerous theoretical and empirical studies have demonstrated that RMCLP is effective and efficient in classifying large scale data sets. However, a possible limitation of RMCLP is poor interpretability and low comprehensibility for end users and experts. This deficiency has limited RMCLP’s use in many real-world applications where both accuracy and transparency of decision making are required, such as in Customer Relationship Management (CRM) and Credit Card Portfolio Management. In this paper, the authors present a clustering based rule extraction method to extract explainable and understandable rules from the RMCLP model. Experiments on both synthetic and real world data sets demonstrate that this rule extraction method can effectively extract explicit decision rules from RMCLP with only a small compromise in performance.

Modeling and model validation of the impact of the economy on the credit risk of credit card portfolios

Since the global credit crunch, lenders have recognized how critical it is to assess the default risk of portfolios of consumer borrowing under different economic environments. We describe a Markov chain model for revolving consumer credit accounts based on consumers' behavioral scores that includes the impact of the economy on the risk migration of credit card accounts. We use two credit card data sets, one from Hong Kong and one from the UK, to validate the model, and hence provide empirical evidence to encourage lenders to use macroeconomic measurements to estimate the default risk in credit card portfolios. The models show how economic variables such as unemployment and price indexes have an impact, both directly, by changing the dynamics of the credit scores, and indirectly, by affecting how many credit card accounts become inactive or reactive themselves. The model also shows the difference in impact that economic changes have on transactors and revolvers.