Table of Contents
- Strategies and segmentation
- Young accounts
- First-time delinquent
- Habitual delinquents
- Broken promises
- Data mining and behavior scoring
- Data and data preparation
Financial institutions face the challenge of managing the growing numbers of delinquent customers, while maintaining or cutting operational costs in the debt recovery process. In order to better understand the customers repayment behavior, it would be ideal to group the customers based on their socio-economic status and financial obligations. The approach to grouping customers to their respective profiles could be done through the use of data mining and scoring techniques. We are able to segment the customers and design an adequate collection strategy for each segment effectively. Within the new business models, there is a need for more proactive and effective collection and recovery strategies, by utilizing today’s flexible, responsive operational and IT systems to provide the agility and facilitate the respond to new and emerging risks in the market and empower the organization to deal with legacy risks inherited from earlier years.
In the retail banking industry, the key success factors to improving debt collection effectively are through designing collection strategies that best fit the customer’s payment behavior. The number of resources allocated for collection in any organization is very limited especially in retail banking where the number of delinquent accounts far outnumber the people in the collection team. In theory, a self-curing strategy is a strategy that provides a grace period for all customers to settle their outstanding balance without any direct action from the lender organization. Identifying these groups of customers will help optimize the limited resources to collect from other customer groups or segments. Despite the method of collection, the value of collection is the same therefore in order to avoid the preventable cost and customer irritation the employed segmentation strategies must maximize the number of self-cure accounts. Moreover, the strategy should be able to identify the accounts that need more aggressive approaches to avoid escalations to higher delinquency levels.
According to a recent survey by the Institute of International Finance (IIF), in emerging markets such as Asia’s and emerging Europe’s the rate of nonperforming loans (NPL) continues to deteriorate with the expectation of still being negative.
In emerging Asia, banks suffer from deterioration in NPLs in 2013 Q3, subsequent to the chain of deteriorating quarters. Furthermore, most banks participating in the survey expressed that they expect a further increase in the percentage of NPLs in their portfolios over the last quarter of 2013. As for emerging Europe, 2013 Q3 has brought another rise in NPL ratios. A recent Ernst & Young forecast estimated that NPLs would continue to increase around the world.
The rise in NPLs has made credit risk management an urgent issue. The huge impact of debt collection operation on the organizations’ financial results are increasingly realized by corporations. Therefore, debt collection evolved into a profit driver, but the key is to collect wisely. Regardless of how a collection recovery is achieved, the value is the same therefore it is important for the lenders to know their customers better in order to be able to optimize their collection strategy and maximizing the debt collected while minimizing the collection expense and maintaining the customers’ relationship. Self-curing is a strategy that provides a grace period for a customer in which to pro-actively pay off their outstanding balance before investing the organization’s resources to contact them to make a direct request for that payment.
Segmenting the customers into different queues is a common practice for tailoring an appropriate collection strategy for each segment. As a general principle choosing the action for each segment could be different in terms of Type, Tone, and Timing. Moreover, the traditional collection process is divided into two areas i.e. early-stage delinquents, and late-stage delinquents. The main objective in the former group is rehabilitating the customers while in the other group the focus is on recovering the balance as much as possible. In this paper, the focus of the author is on the early stage delinquents and the timing aspect of the collection strategy.
In theory, the self-cure strategy should be applied to all accounts given that they remain adequately liable to make a payment to compensate for the indirect costs of the self-cure strategy incurred by the opportunity to collect payments using the next best strategy. In practice, high-risk customers will be treated differently.
The expected benefit of a self-cure strategy is simply the expected number of payment agreements to be achieved as a percentage of all customers in the strategy – or the probability of payment. There is no direct cost associated with self-cure strategy therefore the problem needs to be looked at in a different way, by studying indirect cost.
The indirect costs of a self-cure strategy derived from the fact that the probability of recovery declines as the time to make contact with the customer increases. The cost of a self-cure strategy is therefore based on the rate at which the probability of payment from the next best strategy shrinks. If a customer is expected to make a payment when they are called on day one as they are called on day four, then there is no cost attached to the self-cure strategy for those first three days. Therefore, regardless of the extent of the probability of payment from the self-cure strategy no call should be made until day four. This is because, with no costs, any recovery made is value-generating and any recovery not made is value-neutral. However, if after the first three days a customer who has not been contacted begins to become less likely to make a payment when eventually called, costs start to accrue. The customer should remain in the self-cure strategy up to the point where the probability of payment from the self-cure strategy is expected to drop to a level lower than the associated drop in the probability of payment from the next best strategy, as illustrated in Figure 2.
Strategies and segmentation
Most lenders have advanced to the concept that given the limited number of resources, low risk customer can be left for a while to self cure and high risk customers should be contacted early. The collection strategies and actions must be chosen carefully to not alienate potentially profitable customer who might be going through a temporary stage of financial hardship. Targeting the high risk customers proactively makes sure that valuable low risk customer have a chance to self cure. By targeting accounts having lowest probability of self curing the collection costs in early stage delinquency can be radically reduced. Moreover, the lender may enjoy increase in level of customer satisfaction due to removing the collection treatment for low risk self cure group of customers.
Economic condition is one of the most important key factors which affect the strategy selection process. As economic condition changes, the customer behavior changes therefore the collection strategy should change accordingly. The challenge is to be agile to reflect these changes. Employing a behavior or behavior and balance scoring will facilitate the identification of those who are most likely to self cure. These scoring methods assists to predict the future performance of an account based on their past payment performance. These methods are popular and heavily used in pre-delinquency and early stage delinquency. Combining score and balance provides the lenders have the ability to calculate ‘Balance at risk’. Balance at risk is the multiplication of behavior score good/bad odds with the current balance. It is a good measure to choose the adequate strategy and manage the limited resources wisely.
Although usually the behavioral scoring and concepts such as balance at risk are utilized to determine the risk factor for each customer but there also exist other concerns as well. Data availability and quality of data are considered a key ingredient for the success of any analysis. Traditional methods such as expert opinion and expert rule systems are not agile enough to cover the dynamics of collection. The introduction of data mining and machine learning to the collection domain empowers the lenders to respond quickly to economic and behavior changes. These methods constantly learn from historical data and discover new behavioral patterns as they emerge.
Moreover, there are some well-known groups of customers with higher risk factors that require special treatments. Among all, in this paper, the author investigates Young accounts, First-time delinquent, Habitual delinquents, and Broken promises further.
- Young accounts: accounts for less than six months old are considered risky due to a shortage of robust behavioral and payment historical data. These accounts are treated separately. Since there is not enough historical data to score these accounts, the bureau data, application score, and current account balance or amount due could be used to rank these accounts. It is proven that a phone call just before first payment prevents many first time delinquents and provides education about their responsibility and financial obligation. Moreover, early gentle reminders to those with moderate credit risk have also been shown to reduce the roll rate while still maintaining a positive relationship.
- First time delinquent: First-time delinquents are for those accounts that miss their first payment. Failing to meet the first payment is a critical risk sign. Defaulting on first payment in one hand can indicate a potentially fraudulent account while on the other hand it could reflect the lack of understanding about their responsibility and financial obligation. A similar but more comprehensive strategy should be adopted for these accounts.
- Habitual delinquents: These accounts are for those that are frequently in and out of delinquency. These are the accounts that are aware of the collection methods employed by the organization and are delicately weaving in and out of delinquency to their advantage. These accounts affects the days sales outstanding (DSO) and cash flow. Therefore an appropriate strategy is needed to improve the DSO and cash flow for these accounts
- Broken promises: A promise to pay may be made by a customer during the rehabilitation process, i.e. phone conversation, letter sent. The promise is for the payment of an agreed amount and date in the near future. The broken promise is when the delinquent customer does not pay the agreed amount in the agreed point in time. Typically a tolerance have been applied in respect of timing, i.e. one or two days, together with a tolerance set against the payment, i.e. payment of 95% of agreed amount considered as full payment. Therefore number of broken promises can be utilized to determine the strategy.
Data mining and behavior scoring
Data mining as an approach to support computer-based decision making is in fact not a merely new technology but borrows many algorithms from statistics, artificial intelligence and other fields. It is not the algorithms of data mining but the idea of automatically extracting nuggets of knowledge from large databases can be a revelation.
Classification methods are the most commonly used data mining techniques that are applied in the domain of behavior scoring to predict the risk level of customers. Statistical and machine learning methods, such as linear and logistic regression, decision trees, and etc. have been used for developing behavior scoring models.Data mining is not just a set of mining algorithms, but rather a process: A process, Figure 4, that aims at solving a definite problem or making a decision, employ different mathematical and computer techniques to examine the related data stored in large databases, finds a solution based on the discovered patterns in data and applies the solution to the predefined problem. Data miners are fond of saying that data mining is as much art as it is science. What they mean by this statement is that the data mining process is a scientific endeavor overlain with a significant amount of artistic practice. The process is not an exact science but the combination of the familiarity of the subject matter of a particular domain and discovering the best possible methodology in achieving the desired goals.
The probability that a customer will become delinquent during the predefined period of time will be estimated and the customers will be classified into different risk levels according to the estimated delinquent probabilities. This process is also known as risk assessment.
There are wide range of data mining algorithms and methods available and applied to behavior scoring but not all of them are practical in the commercial environment. Since the Equal Credit Opportunity Act (ECOA) is widely accepted across the industry, selecting a method that could easily justify the reason behind the score that is calculated is critical.
Data and data preparation
Scoring systems utilize information relating to the traditional 5Cs of credit:
- Character: The willingness to pay off debt.
- Capacity: The financial ability to repay debt.
- Capital and Collateral: Possessions or equities from which payment might be made.
- Conditions: Reflecting the general economic environment, or special conditions applying to the borrower or the type of credit.
For behavior scoring, that predicts how the existing accounts will perform, the credit information that happened after the credit is granted should also be included. Extra information that is collected in behavior scoring with regards to observation point A. A predetermined period of time before the observation point is selected as a performance period, for instance, the previous 12 months, then new variables that describe what happened in this period are added. ‘Outcome period’ is a period of time subsequent to the observation point A. The customer’s behavior is assessed at the end of ‘outcome period’, point B, and used as the dependent variable. The choice of time horizon is critical for behavior scoring. The normal practice seems to be from six months to two years. Thus any shorter horizon would be insufficient to determine the rates of delinquency and it does not reflect the whole population. The time horizon of more than two years would leave the system to be susceptible to population drift in which the distribution of the characteristics of the population change over time. Therefore, the population sample may significantly be different.
The benefits of adopting a self-curing strategy are self-evident. The reduction in the cost of collection, customer rehabilitation, increased customer satisfaction and many other direct and indirect benefits could be achieved by employing the appropriate strategy for each customer. In practice achieving these benefits depends on the careful selection of the right strategy, the ability to access to right data and analyzing the data effectively to target the right customer.
The use of analytics and data mining can make a significant distinction between the best-in-class and the average collection strategy. It is a matter of segregating different groups of customers within the population of delinquent customers and adopting the best strategy to optimize the collection effort for achieving the best collection result.