Banks Neglect Use of Models in Loan Analysis
Rose, Sanford, American Banker
a survey of credit practices at the 100 largest banks in the country reveals that most institutions have been slow to exploit the potential of credit models in evaluating loans and in helping to determine appropriate loans prices and loss reserves.
Although a number of banks have recently adopted computer credit models, their overall penetration remains low. Only 22% of the 41 banks that responded to the survey employ them, and then only as adjuncts or secondary inputs to the loan-evaluation and loan-pricing process.
Despite growing evidence that credit is too important a matter to be left in the hands of loan officers, most leading banks still base their credit, reserving, and pricing decisions on the all too fallible judgments of these officers.
The view of many survey participants is that credit modeling is still an infant and unproved technology. But this is almost certainly an incorrect judgment. Automated credit evaluations have progressed to the point where they are, in many cases, more accurate, more consistent, and very much cheaper than traditional, labor-intensive methods.
Improving Rating Systems
Models can be used to greatly enhance the utility of bank risk-rating systems by providing superior estimates of the expected probability of default in each risk grade. Based on these probabilities, banks can compute a given loan's expected loss, or risk charge, which can then be incorporated directly into the loan price.
It is significant that while most of the banks surveyed by Oliver, Wyman & Co. have loan risk-rating systems, a quarter do not use these systems in any shape or manner for price setting, and only about half employ them to help set loss reserves, which should be equal conceptually to the amount of the computed risk charges. Very probably, this relative underutilization of risk-rating systems is traceable to widespread doubts as to their accuracy -- doubts that can be greatly dispelled by the judicious use of the better models.
Oliver, Wyman believes that it is now possible to introduce an automated grading and "early warning" monitoring system for loans to public corporations based on a model that measures default probabilities by tracking the volatility of equity prices. Such a model is available from our joint-venture partner, the KMV Corp. of San Francisco.
It provides real-time estimates of changing default potentials and has been proved to signal credit difficulties on average one to three years before they were recognized by the rating agencies. It also provides a means of determining the variance of losses from their expected magnitudes, enabling banks to allocate capital to loans in proportion to their risk of unexpected loss.
KMV has another model that establishes the default potential of private middle-market borrowers based on financial statement data, account behavior information, and sectoral economic data. Though by no means as accurate as that used for public corporations, this automated tool has been found to be superior to most of the evaluation methods on which the leading banks now rely. It therefore provides a sounder foundation for the estimation of default likelihoods and risk charges.
Reactive, Not Proactive
While the comparative neglect of models circumscribes their use of in-house risk-rating systems, the majority of survey respondents still find some value in these systems. …