Strategies for Predicting Equipment Lease Defaults
Executive Summary
The present-day equipment leasing and finance industry, characterized by stiff competition, diminishing profit margins, and exigent regulatory oversight, must take advantage of the best that contemporary business research has to offer, especially applied research aimed at helping companies become more agile and efficient in their day to day operations.
Recent advances in the area of forecasting, for example, can be of significant value for lenders/lessors in this industry, whose performance depends greatly on their ability to avoid poor credit risks. Indeed, given the magnitude of financial resources involved in such decisions, even small improvements in forecast accuracy can add substantially to a company's bottom line.
Results of the present research offer evidence of an opportunity for marked improvements in default forecasting accuracy when key predictor variables and an appropriate statistical classification model are used to identify credit risks. Most interestingly, the composite PayNet Rating Score, a metric based upon conventional demographic variables and a company's prior leasing/ borrowing history, has proven to be an outstanding predictor of credit risk in all three of the classification models investigated. Other predictor variables from the PayNet Database that appear to yield good classification accuracy in these models include past-due history, number of open contracts; and to a lesser extent, geographic region, public/private ownership, government/ non-government contracts, and ease of credit access. Most surprising, however, has been the superior, comparative performance of the linear discriminant model which, in many circumstances, has yielded classification accuracies as much as 10 percent better than the competing logistic regression and neural network models. This is not an inconsequential result (statistically or practically speaking), in an industry where analysts are struggling to improve classification accuracies by just 1 or 2 percent. In short, we would advise credit analysts in the industry to seriously consider modeling their credit risks (loan/lease default) with a linear discriminant model; using the PayNet Rating Score, history of Past Due Experience, and Number of Open Contracts as independent variables. Programming this procedure is SPSS is straightforward.
Table of Contents:
- 5 Executive Summary
- 7 The Data We Have Used
- 8 Research Methodology
- 9 Statistically Significant Variables and Three Predictive Models
- 13 Some Comments and Conclusions
- 14 References
- 15 About the Authors