결합부도예측모형을 이용한 항공산업 자금운용의 위험관리방안

Abstract

We propose the alternative model combining financial variables and market variables which is estimated in two stages for the risk management of the fund application of the aviation industry. At the first stage, we estimate the EDFs for the listed banks using stock market information or PDs from credit spreads using bond market information as dependent variables, and select candidate predictor variables from CAEL financial ratios. At the second stage, we estimate thelogistic regression model combining these financial variables as predictors and EDFs and PDs from credit spreads as dependent variables. The combined model settles down the problems such as the data restriction and default concentration in the case of the model based on the historical default cases, and the data limit to the listed enterprises or market inefficiency in the case of the model based on the market information. We estimate two combined models with respect to EDFs and credit spreads, and compare theperformance with the traditional model based on the historical default cases. We validate the respective models with respect to the discriminatory power. We find that in the case of savings banks 1) the CAEL variables related with asset soundness and capital adequacy are selected for both logistic and combined models, 2) although the combined model uses 10% out-of-sample for the model estimation, its performance is not less discriminatory than the logistic model using all in-samples, 3) its sensitivity for forecast errors are better than the logistic one. In the case of banks, due to the small size of default cases, the combined model is only used for the analysis and variables related with asset soundness and earnings are selected for the dependent variable based on EDFs, but capital adequacy represented by BIS ratio is added for the dependent one based on credit spreads. Every selected variable is very significant and persistent through every estimation period. In the validation of model, the model is tested with likelihood ratio test. Partial model is significant through every estimation period. We find high discriminatory power of both the combined models compared with the traditional models in validation.