Rated municipal bonds: an analysis, classification and extension

Abstract

The purpose of this study was to (1) identify the variables important in analyzing a general obligation municipal bond's investment quality using improved statistical techniques, (2) develop models that accurately classify municipal bonds into investment quality versus non-investment quality using variables identified in the literature and by practitioners, and (3) evaluate the stability of the variables in predicting investment quality over time.

A review of the literature revealed over 170 variables thought to indicate a bond's investment quality. Because banks hold a large percentage of municipal bond issues and some perform their own credit analysis, bond analysts were queried to determine what variables they use in making investment decisions.

Eight logit models were developed. Four models were based upon variables identified in the literature; four by practitioners. With each group, models were developed using sized and non-sized variables and principal components developed from the raw variables as input. Data from a random sample of general obligation municipal bonds issued in 1982-1985 were used to evaluate the models.

The non-sized PCA literature model correctly classified 96.47 percent of the sample municipal bonds. The sized PCA literature model correctly classified 91.49 percent; non-sized PCA practitioner, 92.47 percent; and the sized PCA practitioner, 88.30 percent. The models developed using raw data were, in all cases, less successful in classifying bonds as investment quality. The results for the non-sized raw data literature model was a correct classification of 92.11 percent; raw data sized literature model, 90.32 percent; raw data non-sized practitioner model, 87.36 percent; and raw data sized practitioner model, 85.06 percent.

The most significant variables as seen in the eight models were CHGPOP and PERCAP. As noted, the literature models outperformed the related practitioner models. Also, the use of principal components as inputs improves the ability to classify the bonds. Lastly, the variables determinative of investment quality are not stable over time.