Moody's Analytical Overview of 25 Leading U.S. Cities—Revisited

Abstract

Several attempts have been made to duplicate Moody's municipal bond ratings through the use of discriminant analysis. Based on cities and data published in Moody's Analytical Overview of 25 Leading U.S. Cities, Aronson and Marsden (A &M) report that they achieved 83% accuracy when classifying according to five bond rating groups and 95% accuracy when discriminating between cities with high and low bond ratings. They also indicate that the most important variable in this model is the percentage of a city's population that is black (PB). This article attempts to validate A&M's findings through the use of additional statistical techniques. By employing Lachenbruch's U-Method, it is found that the 83% and 95% classification accuracies reported fall to 25% and 70.8%. It is also found that by eliminating variables, the U-Method accuracy increases to 87.5% for the two-group model. The conditional deletion method supports A & M's findings concerning the relative importance of PB, and thejackknife technique indicates that, even though there is a relatively high variable-to-case ratio, variables in the two-group discriminant model are quite stable.