A take on "Moody's Bond Ratings 1980" Analysis
Supervised Learning: Linear Discriminant Analysis
Linear discriminant analysis (LDA), normal discriminant analysis, or discriminant function analysis is a generalization of Fisher's linear discriminant, a method used in statistics and other fields, to find a linear combination of features that characterizes or separates two or more classes of objects or events. Moody’s Bond Ratings are intended to characterize the risk of holding a bond. These ratings, or risk assessments, in part determine the interest that an issuer must pay to attract purchasers to the bonds. The ratings are expressed as a series of letters and digits.
LDA correctly classified BBA most accurately with 81.8% in training data and 84.6% in the whole data set. The least accurately classified ratings for training is AA at 45.4% and AAA at 44.4% for the whole data set.
All of the ratings were classified correctly by LDA except for AA. 1 out of 2 was predicted correctly for AA.
For both the training and whole data set, all three A-Rating category (A, AA, AAA) were classified poorly below 60%.
I would suggest that one use LDA to classify the terrible bonds (B, BA, BAA) instead of the best bonds, as it predicts it better. Then separate the B-ratings from the rest of the data, the re-run another LDA just on the A and C ratings.
As for the variables, LTDCAP - Long-term debt to capitalization, seems to be the most influential to LD1 for all data sets.
Appendix
Training & Validation
