Magazine article The RMA Journal

Using Data Envelopment Analysis to Evaluate Loans

Magazine article The RMA Journal

Using Data Envelopment Analysis to Evaluate Loans

Article excerpt

This article explains how data envelopment analysis can be used as a decision support system to screen potential defaulters on consumer loans. The authors assert that DEA offers several benefits over traditional loan evaluation techniques, a major advantage being that it clearly identifies the factors contributing to the loan evaluator's decision to accept or reject the loan.

Today's tool kit for evaluating consumer loan applications is bulging: statistical models, credit-scoring models, neural networks, expert systems, neuro-fuzzy systems, and case-based systems, as well as more basic experience-based rules. The choice depends on the complexity of the institution, as well as the size and type of the loan. Analytical models, such as credit-scoring systems, largely use historical data and the probability of default to predict a loan applicant's relative creditworthiness.

But the credit-scoring model does not completely eliminate the human element. The selection of cutoff scores is a subjective decision. Expert systems, neuro-fuzzy systems, and case-based systems are similarly subjective in nature, as is the evaluation of applicants who have scores between the accept and reject levels.

In the past 10 to 15 years, lenders have come to embrace automated decision-making and modeling to speed loan decisions and manage credit risk. Despite the application of these techniques, loan evaluators increasingly feel the need to implement transparent and consistent decision processes. In short, they are looking for procedures that support their decision to grant or deny a loan.

What Is Data Envelopment Analysis?

Data envelopment analysis (1) is a technique used to assess the productive efficiency of homogenous operating units such as schools, hospitals, banks, or utility companies. It is a powerful technique for measuring performance because of its objectivity and ability to handle multiple inputs and outputs that can be measured in different units. The DEA approach does not require specification of any functional relationship between inputs and outputs or a priori specification of weights of inputs and outputs. DEA provides gross efficiency scores based on the effect of controllable and uncontrollable factors. Figure 1 illustrates a decision support system using data envelopment analysis.

DEA uses a number of variables to determine how good a loan is. A loan application includes information such as the applicant's age, housing, address time, total income, number of credit cards, number of dependents, job time, other loan obligations, total debt, monthly rent/mortgage payments, number of inquiries for an applicant, and credit rating. Using this information, loan evaluators can calculate the ratio of the total payment to the applicant's total income (ratio 1) and the ratio of the total debt to the applicant's total income (ratio 2). The two ratios, along with the applicant's credit rating, are the factors that affect the decision to accept or reject a consumer loan application.

With these three factors as inputs, the DEA-based decision support system calculates an efficiency score for a loan application. This score is a relative value computed by comparing the given loan to a pool of well-performing loans that serve as a benchmark for the loan under evaluation. Each loan is evaluated against either an existing loan or a hypothetical loan with an identical set of inputs or outputs that is constructed as a combination of good loans. By using the existing good loans as a "role model," DEA not only helps differentiate good credit risk loans (efficient loans) from bad credit risk loans (inefficient loans), but also brings out the reasons why an applicant may be unreliable. This helps loan officers justify their decisions to grant or deny the loan from a legal and regulatory viewpoint.

Using DEA for Loan Evaluation

Traditional computation techniques require a precision yes/no or right/wrong. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.