A successful loan default prediction model for small business

Credit & Financial Management Review,  Fourth Quarter 2000  by Yegorova, Irena,  Andrews, Bruce H,  Jensen, John B,  Smoluk, Bert J

• E-mail
• Print
• Link

Abstract

This study contributes to the credit risk management literature by implementing a credit scoring model using commonly available data on small business loans made by an economic development lender based in Maine. A total of 117 variables representing loan characteristics are initially examined, and a series of practical screening methods are used to isolate the more statistically relevant variables for predicting loan default. Only the most statistically significant variables with an economically "correct" sign are then used to build a binary logistic regression model. Three ratios, the current liabilities/current assets, the sales/gross margin, and the equity/working capital are found to be highly significant in predicting loan default. The resulting model correctly predicted 87% of bad loans.

Introduction

One element supporting the foundation for strong economic growth is the efficient allocation of capital to productive small business borrowers. This allocation process depends largely on the amount of information available to lenders about the borrower, see Saunders (1994, p.166). When information is not equally shared, information asymmetries between lenders and borrowers make lending to small businesses very risky.' This paper seeks to implement practical methods in credit scoring that are designed to help lenders manage credit risk by reducing information asymmetry.

The motivation for this study is described in Section I. Section II discusses the relationship between information asymmetry and credit scoring. Section III examines the relevant literature on credit scoring models and focuses on the statistically significant variables in predicting loan defaults found in previous studies. A description of the data set used in this paper is provided in Section IV. Section V describes the methodology used and a practical screening method employed to reduce the number of variables to a manageable level. This section can benefit small business borrowers in that it shows how lenders go about evaluating credit risk. Lenders to small business will also benefit from the description of the methods used in developing a credit scoring model and the types of variables employed. Section VI employs binary logistic regression to determine the statistical importance of each variable in the reduced data set and then`to construct a final model. This insight may be important to small businesses seeking loans because it allows them to focus on managing and maintaining a few critical financial ratios necessary for obtaining credit. Section VII summarizes the paper.

Motivation for the Study

This paper was inspired by a regional economic development lender that sought help in assessing the credit quality of borrowers and requested assistance in developing a systematic and a more accurate method of making lending decisions. The portion of the activity of this lender under study is "micro" with a median loan size of approximately $64,100. This organization targets innovative, job-generating manufacturers in the state of Maine. Borrowers include small businesses in the natural-resource, telecommunications, social services facilities, manufacturing, and health-care industries. Much of their micro-lending is to businesses with limited resources, women, and low-income and minority entrepreneurs, and young businesses. This lender obtains both private and public funds with the objective of providing gap financing as a supplement to local bank loans and owner supplied equity. Public funding comes from a variety of federal, state, and local organizations who seek to create jobs in the small business community and promote harmony with the natural environment. Much, like a bank, this economic development agency has limited funds and seeks to reduce the probability of making bad credit decisions. Poor credit decisions result in less resources for future projects. Previous to this study, the lender would collect numerous financial data from prospective borrowers and then subjectively make loans with little statistical guidance.

Information Asymmetry and Credit Scoring

Information asymmetry, which affects both lenders and small business borrowers, may hinder the efficient allocation of capital. Lenders to small business need to quickly assess the creditworthiness of prospective borrowers so as to reduce the probability of issuing bad loans while attempting to maintain their own profitability. Stiglitz and Weiss (1981) argue that information asymmetry may result in adverse selection in that only the most risky small businesses will seek loans from financial institutions, especially when higher interest rates are used as a form of rationing credit. Consequently, small businesses need to gain the perspective of lenders and credit scoring models in order to increase their chances of obtaining much needed loans. The credit scoring model implemented in this paper employs binary logistic regression to estimate the probability of loan default based on historical data readily available to the lender. These data are routinely supplied to the lender by the borrower and are used in making lending decisions. Included are commonly used economic variables in managing small businesses such as financial ratios, owner characterisitcs, and loan type. The credit scoring model relies upon systematic relationships between these variables in order to estimate the probability of default.