2次判別分析에 의한 CP等級分類

Commercial Paper Ratings Classification by a Quadratic Discriminant Analysis

The purpose of this study is to determine the financial variables that are most important in explaining the commercial paper ratings and to develop a discriminant model that can forecast actual commercial paper ratings. Commercial paper ratings of 109 manufacturing companies were used as published in Fall 1991 issues of Korea Investors Service. Since the dispersion matrices were found to be unequal at the 0.10 significance level, a quadratic, as opposed to linear, classification technique was used. The quadratic discriminant function was incorporated the 20 financial indexes. To reduce the multicollinearity problems which could hamper the combined predictive ability of financial variables to discriminate among commercial paper issuers in different rating groups, a stepwise inclusion procedure was employed. By sequentially selecting the variables with the highest F-value, the resulting Wilks' Lambda was minimized, thereby maximizing discrimination among rating groups. The final variables selected in the quadratic discriminant model using interval data were return on equity, net income, times interest earned, equity, equity turnover, and capital stock turnover. The accuracy of the model were was high, having a classification success rate of 79.45% for the 73 commercial papers in the estimation sample. The result of applying the model on the validation sample of 36 commercial papers was extremely successful with 91.67% of classification accuracy. Rank discriminant analysis was used to mitigate the non-normality ploblem as well as ratios consistent with the rating agencies. The final variables selected in the rank transformation quadratic discriminant model were net income, coefficient of debt repayment, equity, and capital stock. The accuracy of the model had a classification success rate of 68.49% for the estimation sample. The result of applying the modle on the validation sample was successful with 86.11% of classification accuracy. The results of this study show that models based on financial indexes can predict commercial paper ratings with good accuracy and therefore maybe a useful tool for rating agencies, at least as an initial screening device. Examining which financial variables provide insight in forecasting the downgrade or upgrade of a firm's commercial paper rating would be of further interest.