Kamakura Fourth Generation Model Reaches 99.0% Accuracy
HONOLULU, March 15, 2006: Kamakura Corporation released results today of a study showing that Kamakura’s fourth generation “reduced form” credit models were more accurate in predicting defaults of public companies than the debt ratings issued by a major rating agency. Reduced form models predict corporate default using a mix of financial ratios, macroeconomic factors, and stock price inputs. Kamakura has previously released accuracy studies for the entire public company universe, but the new study focused on those public companies with debt ratings from 1990 to 2004. The Kamakura study shows that the ROC accuracy ratio, a standard measure of credit model performance, was 99.00% for Kamakura’s fourth generation reduced form model, compared to 96.44% accuracy for the rating agency debt ratings. The Kamakura report shows that reduced form models are more accurate at all monthly forecasting horizons from 1 month to five years. The study is being incorporated in the Kamakura Risk Information Technical Guide, which documents the credit model tests required by the Basel II capital accords from the Basel Committee on Banking Supervision.
“Kamakura’s fourth generation Jarrow-Chava reduced form model dominates ratings in forecasting corporate failure at all forecasting horizons tested out to five years,” said Warren Sherman, Kamakura President and Chief Operating Officer. “The new study shows that Kamakura reduced form models have 10% higher accuracy than rating agency ratings even when forecasting default in 60 months for those companies who have survived for the previous 59 months. In month 60, the reduced form models have an 83% accuracy ratio compared to 73% accuracy for agency ratings. The 99.00% accuracy of the reduced form models in predicting short run default shows why credit market participants are moving away from 30-year old Merton credit model technology in such a major way.”
The chart below, excerpted from the Kamakura study, shows the ROC accuracy ratio for the Jarrow-Chava reduced form model and debt ratings at every forecasting horizon analyzed in the study. The ROC accuracy ratio for month 30, for instance, is the accuracy of a default prediction for month 30 for those companies who have survived the first 29 months. The Kamakura study was based on 230,000 observations of companies with debt ratings from 1990 to 2004.