Summary
Adjusting the credit lines of card users is a vital undertaking. It is crucial to establish an effective approach for credit card organizations to find out the proper amount of credit to advance to clients. Most of the related research tends to focus on the estimation of credit card users’ default. A holistic and heuristic strategy that explores the credit line that maximizes the net profits of credit card companies is fundamental. For example, regression models can be employed to find the probability of default of customers and customers’ current balances as a function of the credit line. The outcomes are then used to estimate the net profit.
A rigorous research study can significantly contribute towards a viable strategic guideline in the management of credit lines for firms that deal with cards. Even though the concept of functional integration has been embraced across the board, little progress has been registered when it comes to credit employment. Some of the factors that have contributed towards the latter poor score in credit include limited understanding of the concept of credit, declining market performance, and lack of effective objectives that can be pursued by an organization.
Management of Risks and Credit in organizations
Credit service, therefore, creates an identifiable value which might be called “possession utility.” Such utility is not limited to the field of consumer buying but is also involved in mercantile transactions and lending. Little Interest in Credit Work
The notions that “sellers give credit,” that credit service is not distinct from the commodities to which it is applied, that credit service has no value, and that the offering of credit service is primarily a finance function—^these are fallacies which have hindered the development of credit theory and practice.
For many years credit was regarded as an unproductive business activity, and the position of Credit manager was at a much lower level than other marketing roles. Generally credit management has been assigned to the finance and bookkeeping departments and has been concerned mainly with allocation and utilization of working capital, turn-over of receivables, sources of long-term and short-term funds for carrying receivables, credit loss ratios, cash discount tactics, economic indicators of the quality of receivables, and the like.
Anecdotal evidence on credit scoring has pointed to possible manipulation that may increase the credit scores of borrowers without any real improvements in their creditworthiness (see, e.g., Foust and Pressman 2008). In theory, score manipulation has a minimum impact in terms of our metric if its occurrence were to be uniformly distributed. However, this is unlikely: A more probable scenario is one in which manipulation is more likely to occur at low levels of credit scores. Moreover, most anecdotal accounts argue that such manipulation increased over time. Therefore, if credit score manipulation affects default rates, it is most likely to be reflected in our results at low levels of FICO and for later cohorts.
More importantly, evidence of manipulation of credit scores should be reflected in anomalous behavior in terms of our parametric measure – a measure that controls for other characteristics on the origination. However, the evidence shows the opposite: parametric measures of FICO performance show improvement at all levels of FICO. This result is fairly robust and holds for multiple variations of credit score groupings. In light of this, we conclude that the evidence from our data does not reflect any anomalous behavior that would suggest that such manipulation was widespread. That is not to say that such instances of manipulation did not occur, but simply that given our large sample size, score manipulation would have to be fairly widespread to affect our results.
Portfolio guidelines and investment policy statements allow investors to communicate to portfolio managers their overall returns objectives and appetite for risk. By necessity, these guidelines need to be general enough to cover a wide range of possible assets and communicate objectives in a relatively concise way. Standardized tools have helped to facilitate this communication and for fixed income portfolios, the most common of these are credit ratings. Cantor et al. (2007) document that close to 80% of portfolio managers and fund sponsors explicitly rely on credit ratings in their portfolio guidelines.
In this paper, we explore the mapping between rating categories assigned to a bond at issue and the yield that is ultimately required by investors. In particular, we ask whether there are industry differences in the yields required for bonds that are assigned the same credit rating. Portfolio guidelines, financial regulations, and rating agencies themselves generally make no distinction across industries, creating a bond rated “A” as bearing the same amount of risk as a similarly rated bond, regardless of the issuer’s industry. Yet, if ratings are noisy or imperfect assessments of how investors view the risks of a This is generally the case for both issue and issuer ratings, however, there are a small number of cases in which a rating agency may provide a separate category of rating that is particularly relevant for firms in that industry. Examples include ratings with the designation “F” issued by S&P to assess the creditworthiness of a fixed income portfolio or “Insurer Financial Strength” ratings designed to assess the ability of an insurer to meet policy obligations rather than debt obligations.
Particular industry and systematic differences may exist in the yield required for bonds within the same rating category. These differences may allow a portfolio manager to introduce higher-yielding bonds into their holdings while remaining within their set rating constraints. As suggested by Becker and Ivashina (2013) this additional yield may be particularly attractive during extended periods of low-interest rates. The specific industry comparison we focus on in this paper is the yield on bonds issued by financial institutions versus non-financial firms. We are motivated to focus on the finance industry for several reasons. First, financial firms are frequent issuers of debt. In our sample, covering the years 1967–2012, approximately 40% of public bonds are issued by financial firms. The sheer number of issues by this industry makes understanding the determinants of yield particularly relevant and implies that it would be hard for a fixed income portfolio manager to avoid investing in this debt. Second, previous research has documented a higher degree of noise in the credit assessments assigned to banks than other firms.
Therefore, we embarked on this study to determine the efficacy and usage of credit scoring among some of the riskiest loans in recent history. To this end, this paper has introduced a simple yet effective measure for evaluating the performance of credit scoring. As mentioned earlier, the advantage of using such a measure is twofold. First, it lends itself to both nonparametric and parametric measurements. Second, it minimizes the impact of situational factors.
Using this measure, we find that credit score performance is robust to both high- and low-default environments. However, evidence suggests that some of the increase in credit scores over the cohorts can be explained as adjustment for the increased riskiness in other attributes on the originations. This was particularly true for low levels of credit scores – resulting in a sharp deterioration of credit score performance in terms of our nonparametric measure. Significantly, once we control for other (riskier) attributes in the origination, our parametric measure of credit score performance shows improvement over the cohorts. This would suggest an increased emphasis on credit scoring – not only as a measure of credit risk but to offset risk on other origination attributes.
Adjusting the credit lines of card users is an important issue. It is essential to establish an optimized approach for credit card companies to identify the proper amount of credit to offer for their customers. Most of the related research concentrated on the prediction of credit card users’ default. Our contribution is a consideration of a holistic and heuristic approach that looks at the credit line that maximizes the net profits of the credit card companies. We first apply regression models to find the probability of default of customer and customer’s current balance as a function of the credit line. Next, we use a regression tree to identify groups of customers assigned with the same credit line. The results are then used to formulate the net profit and a genetic algorithm is used to find optimally adjusted credit lines for each group of customers. It is expected that our study can contribute to presenting strategic guidelines for the management of credit lines for card companies.
Assigning proper credit limits is one of the most critical tasks of credit card companies. Credit card companies usually assign an initial credit line based on the customer credit score. The initial credit line can be adjusted when required by a customer. It is essential to establish proper credit lines that can maximize the net profit of the credit card company. However, the existing methods that are used by credit card companies to adjust credit lines are not very systematic. Adjusting a credit line without proper reason or basis may put the credit card company at risk of losing business and at the very least decrease their profit.
There has been some effort to find the relationship between credit limit and the response of the credit card users. Dey and Mumy have examined the determinants of credit limits on credit cards. The authors have found evidence that suggests a positive relationship between the proxies (LOGINCOME, SELF-EMPLOYED, AGE, and CREDITRATE) of borrower quality and the approved borrowing limits on credit cards. Gross and Soulless investigated consumer response to changes in credit supply using the data set collected on credit card accounts from user in the U.S. Based on the results of their study, increases in credit limits seem to generate an immediate and significant increase in debt. This response is especially true for people who used near their credit limit. However, people who used well below their credit limit also experienced a significant increase in debt. Song has shown that the credit levels and needs of customers are reflected in the credit line, and that these factors can effectively increase credit card usage. By determining the limit value at which profit can be maximized by taking into account the response to the limit exhibited by both good and bad customers, the author provided the appropriate credit limits for different customer groups.
None-the less, data is still lacking in regard to developing an optimized strategy for credit card businesses to manage lines of credit. Most of the related research concentrated on the prediction of credit card users’ default. Our contribution is a consideration of a holistic and heuristic approach that looks at the credit line that maximizes the net profits of the credit card companies. The purpose of this study is to develop a strategy that will maximize the net profit of credit card companies by adjusting the credit lines of credit card customers. In order to determine an appropriate credit line, we assessed the expected profit as a function of the credit line for customer groups and applied data mining algorithms to find the probability of default (PD). Using this information and the current balance, we grouped customers by their initial credit line.
The PD was estimated using a logistic regression in which the credit line was used as one of the predictor variables. We estimated the current balance using the multiple regression models and used a regression tree to identify homogeneous groups assigned with the same credit line. The expected net profit was then obtained as a function of the initial credit line of the individual group, as well as the current balance and PD. we utilized a genetic algorithm (GA) to find the optimal credit line that maximizes the net profit of the credit card company. It is expected that an optimal strategy derived from adjusting lines of credit can help credit card companies better manage their consumers and costs.
Credit line
Credit card companies provide convenient payment options to their customers and allow postponement of payment by granting a line of credit. This is why the credit card business has similar credit risks to that of companies in the mortgage business. The current balance of a credit card user directly affects the profit of a credit card company. In this regard, the credit line assigned to a customer, which is closely related to the current balance of the card user, is very important. Therefore, two important factors to consider when increasing or lowering the credit lines of customers are the current balance and credit score. Oman and Cheema manipulated credit limits and the credibility of card users. They posed hypothetical purchase opportunities to their test subjects. The authors mentioned that when user’s credibility is high, the effects of the credit limit on spending will be strong while if the credibility is low then the effects of the credit limit will be attenuated.
Ko et al. analyzed how changes in credit limit affect household consumption and default patterns using data of Korean credit card users. Through empirical analysis, the authors have shown that changes in credit limits can critically affect how people with low credit ratings make purchases. In their study, it was found that credit line usage, limit violation, and cash flow all exhibited abnormal patterns approximately 12 months before a default event. This observation shows that account activity provides a real-time window into the borrower’s cash flow and thus explains why banks have an advantage in allowing certain types of customers to go into debt. Using trended Brownian motion to characterize the cash needs of a borrower over time, Stan house and Ingram derived a probability density function to determine the time to depletion of a bank credit line as well as the time to exhaust the sources of liquidity that fund the loan.
Financial optimization models
Genetic algorithms have been frequently applied to financial optimization models. Korhonen presented two-stage genetic programming (GP) approach that can be applied to the management of a bank’s assets and liabilities. The model included three one-year planning periods with multiple scenarios to describe uncertainty, changing priorities, and multiple goals in terms of expected profit, risk, liquidity, capital adequacy, growth, customer relationships, and other aspects of a bank’s operation. Zopounidis and Doumpos proposed a methodology that combines the preference disaggregation approach (a multicriteria decision aid method) with a decision support system. This system incorporates a plethora of financial modeling tools, along with powerful preference disaggregation methods that lead to the development of additive utility models for the classification of Alternatives that could be considered as predefined classes. Saunders et al. presented portfolio credit risk management using factor models, with a focus on the optimal portfolio selection based on the tradeoff between expected return and credit risk.
By using the Central Limit Theorem for the large portfolio approximation, the authors show how the results on the large portfolio approximation can be used to reduce significantly the computational effort required for credit risk optimization. Using the consequences of extreme dependence on the risk of large heterogeneous credit portfolios, Bass bamboo et al. built algorithms to efficiently estimate the risk of credit portfolios via Monte Carlo simulation.
Killough and Souders developed a GP model for public accounting firms, and Lawrence and Reeves developed a zero–one GP model for capital budgeting in a property and liability insurance company. Bhaskar and McNamee discussed the nature of multiple objectives in accounting, and Farnand Waung presented a multiple criteria Markovian process system for pension funds and manpower planning. In this category, for example, Ashton and Atkins introduced a multi-criteria model to take advantage of both simulation models as financial statement generators and mathematical programming as a flexible search tool. Vinson presented a stochastic GP model to deal with uncertain exchange rates and other barriers to free capital flows, and Eom et al. introduced a GP model-based multiple criteria decision support system for global financial planning in a multinational corporation to allow managers to satisfy the multiple financing goals and to effectively analyze the trade-offs among costs, foreign exchange risks, and political risks  various fees and interests to be the revenue yielded from card users.
We do not use the (1-PD) for deriving the revenue because the various fees and interests occur when credit card users use their credit cards. PD is defined as the probability of a credit card user defaulting on their loan. In previous research studies, the mortality default trade model of Altman and the aging approach of Asquith were used to analyze the bank loan market. These mortality default rate models are designed to derive actuarial-type probabilities of default using past data on bond defaults based on the credit grade and the years to maturity.
Usually, we derive PD by fitting a logistic regression model representing whether a credit card user falls behind in their loan payments using X variables representing customer account activity, credit bureau score, credit line, and utilization of the loan When observing the trend of the current balance according to a customer group, it was discovered that defaulted customers and non-defaulted customers have different trends. Due to this reason, two different multiple regression models were fitted to predict the current balance for both the defaulted (bad) and non-defaulted (good) customers using a credit line, along with the other variables. We added a squared term to the credit line to represent the nonlinear relationship between the credit line and the current balance. As a final step, a regression tree is used to classify the customers in terms of the credit line. The target variable used to classify customers is the credit line, and the predictors are described in Table 1. The PD typically needs to be calibrated for each group obtained from the regression tree based on a credit line to be matched with a more realistic PD that would occur in an areal environment.
(FI) Genetics is a biological term. Biologically, the genes of a good parent produce better offspring. These types of algorithms encode a potential solution to a specific problem using a simple chromosome-like data structure and apply recombination operators to these structures in such a way as to preserve critical information. After developing the foundations of the GA approach by Holland, many researchers used the algorithms to resolve their problems such as the generalized assignment problems, the set covering problems, the scheduling problems, and the producing good routing problems.
First, we estimated the default probability of individual card users using logistic regression analysis and used the credit limit as one of the predictor variables. The data used for this study was obtained from the ‘Strategy Designer Sample Files’ belonging to Fair Isaac Corporation (FICO). The data set has 13 predictor variables describing the status of the card user and three target variables that are used for the regression tree, logistic regression, and multiple regression, respectively. All of the variables are assessed on an interval scale except delinquency, which uses an ordinal scale. The raw data consists of a snap shot of one month’s activity for 50,000 customer accounts. The x5 (Delinquency) factor indicates whether a customer has had a delinquent experience. The x5 is composed of 36,571 ‘0,’ 12,029 ‘1,’ and 886 ‘2.’ A ‘0’ indicates that the person had no delinquencies, and a ‘1’ means that they had less than 30 days of delinquent activity. In addition, a ‘2’ represents more than 30 days of delinquency.
Credit card companies use various fees for these types of delinquencies. However, we could not consider variables such as the interest on revolving credit or delinquency interest since the data used in this study was only collected over one month. In addition, information such as the late fees and payment protection fees were also not considered, although they would not have had any significant effects due to their low frequency of occurrence or distinct characteristics. Traditionally, credit scoring systems aid the decision of whether to grant credit to an applicant or not, by estimating the probability that an applicant will default (PD). As a first step, we use the idea of credit scoring system. Logistic regression analysis was performed in order to predict the PD. The target variable is binary, which identifies whether or not a customer experienced a default.
Defaulted customers were defined as those who had experienced more than one month of delinquency and were coded as belonging to the ‘1’ group. The other customers, those rated as ‘0’ or ‘1’ for factor x5, were coded as ‘0.’ However, since the rate between ‘1’ (886) and ‘0’ (48,600) is imbalanced, we performed a random oversampling for the ‘1’ group. Ultimately, the total number of data points used in our study was 97,145 cases, where ‘0’ and ‘1,’ respectively, consisted of 48,600 and 48,545 cases after the cases with an outlier value of x8, x9, x10, and x11 were eliminated. We use the 60% of the sample as training data set and 40% as validation set. The results of the logistic regression model used to determine the default after selecting the independent variables.
Over the last couple of decades, technological advances and private arrangements of information sharing have increased the use of credit scoring in almost all forms of loan origination (Altman and Saunders 1998; Berger et al. 2005). However, the use of credit scoring is not without its limitations (Mester 1997; Avery et al. 2000). Still, ‘a good model should be able to accurately predict the average performance of loans made to groups of individuals who share similar values of the factors identified as being relevant to credit quality’ (Mester 1997, p. 11). Despite the limitations of credit scoring, most approval processes continue to use credit scores as a measure of borrower creditworthiness at the time of loan origination (Avery et al. 2003; Brown et al. 2009). The continued importance of credit scoring in loan approvals merits careful study of the use and performance of such metrics.
There are important reasons to focus on FICO performance in the subprime mortgage market. While FICO scores have been an integral part of the prime mortgage approval process, they played an essential role in extending credit beyond the prime segment to some the riskiest consumer loans in recent times, namely, U.S. subprime mortgages (Fishelson-Holstine 2005). Subsequently, high default rates of subprime mortgages have raised important questions about the efficacy and usage of credit scoring in loan origination (Demyanyk 2008).
At the same time, examining FICO performance for the subprime segment allows us to study credit scoring performance at the lower end of the credit score range. Stated differently, it helps determine whether FICO scores can be a stable metric for credit risk in lending to credit-impaired borrowers. In this sense, it would also help increase our understanding of the challenges faced in lending to credit-impaired borrowers.
Our results reveal that, for the most part, the performance of credit scoring as a measure of (relative) credit risk remains fairly stable. The results provide little evidence of deterioration in the performance of FICO scores as rankings of borrower ex ante credit risk. However, they also suggest a pattern in which credit scoring was likely used to offset other riskier attributes on the origination – leading to an unconditionally higher rate of default, especially on originations with low credit scores. FICO scores were first recommended for use in mortgage lending by Fannie Mae and Freddie Mac back in 1995. In the subsequent years, they became a significant factor in the development of the US subprime mortgage market (Fishelson-Holstine, 2005).
Foust and Pressman (2008) document the industry view on how the FICO scores were ‘being blamed for failing to flag risky home-loan borrowers.’ From a policy research perspective, Demyanyk (2008) uses the performance of subprime mortgages to express doubts about the effectiveness of FICO scores. To be sure, we do not address the question whether lending to credit-impaired borrowers is a desirable objective from the standpoint of social welfare (see Bolton and Rosenthal 2005 for a discussion).
Our objective is somewhat modest: If financial inclusion were to be the desired policy objective, we examine whether FICO scores could be reliably used a metric of credit risk in lending to credit-impaired borrowers. In such cases, increases in a borrower’s credit score occur without any increase in the borrower’s creditworthiness. We discuss this issue in greater detail improvement in the distribution of credit scores on subprime originations between the 2000–2002 cohorts and 2004–2006 cohorts. Second, as has been well documented, there was an increase in the proportion of subprime originations with riskier attributes over this period (Mayer et al. 2009).
Interestingly, we find that an origination with a riskier attribute (such as lower documentation and higher LTV) was more likely to have a higher FICO score in the 2004–2006 cohorts than a similar origination in 2000–2002 cohorts. Stated differently, we find that, among the various attributes on the origination, the apparent trade-off between the credit score and other attributes (e.g., LTV) grew larger over time. Finally, we show that there was an overall increase in FICO scores from the earlier cohorts to the later cohorts even after adjusting for other origination attributes. This result indicates that the increase in credit scores across cohorts might be interpreted both in terms of adjustment for the increased riskiness in other origination attributes and the increased strength of adjustment to such riskier attributes.
In sum, this pattern is suggestive of an increased emphasis on credit scoring to offset other attributes on the origination. We develop a simple measure of FICO score performance in terms of ex post loan performance. It is important to mention here that both academics and practitioners alike view FICO scores in terms of rankings of borrowers rather than absolute metrics that provide time-invariant estimates of the probability of default.
Continuing in this vein, we propose measures of FICO scores performance that determine whether such credit score ‘rankings’ are maintained in terms of observed loan performance. Any metric of FICO score performance should ideally account for factors that confound the effect of credit scores on loan performance. This becomes especially relevant in comparing loans that perform under different macroeconomic conditions.
Bibliography
Bartels, Robert. “Credit management as a marketing function.” The Journal of Marketing (1964): 59-61.
Purda, Lynnette, Fatma Sonmez, and Ligang Zhong. “Financial Institution Credit Assessment and Implications for Portfolio Managers.” Journal of International Financial Markets, Institutions and Money (2015).
Sengupta, Rajdeep, and Geetesh Bhardwaj. “Credit Scoring and Loan Default.” International Review of Finance (2015).
Sohn, So Young, Kyong Taek Lim, and Yonghan Ju. “Optimization strategy of credit line management for credit card business.” Computers & Operations Research 48 (2014): 81-88.