신용협동조합 도산예측에 관한 실증적 연구

Abstract

After the 1997 Asian financial crisis, the rules and business practices in Korean financial industry underwent major changes. The restructuring of financial institutions including banks took place on a previously unimaginable scale. Various advanced new financial regulatory schemes were adopted. Financial institutions engaged in active competition to expand their asset sizes in an attempt to increase market share and preoccupy business opportunities. These changes resulted in a shift in financial industry structure, Big-sized financial companies acquired more competitive power while small-sized ones lost their business ground. Because the low-income people were the main consumers of these small-sized financial companies, this reorganization caused a weakening of financial functions for such needy people. Credit unions, ones of these small-sized financial companies, also experienced severe restructuring. Prior to the financial crisis, there were 1,666 credit unions at the end of 1997. However, after restructuring the number has decreased to 1,007 at the end of 2007 when the restructuring is done. Also 4.8 trillions KRW worth of public funds was injected for the restructuring of credit unions. However, the business environment for credit unions has become worse after the financial crisis because of the intensified competition in the financial sector, as well as the local economic downturn. Credit unions have provided financial accomodations for urban low income families, farmers, fishermen, and served as a buffer against private loans which lend money at very high interest rates. In that matter, sound management and prevention of insolvency of credit unions are critically important for financial system stabilization and consumer protection. Although forecasting bankruptcy of credit unions at the early stage is in the best interest of both financial supervisory authority and consumers, no significant efforts have been given on this issue. The academic society has not been unconcerned with due to either the lack of samples or research data. This research is the first in Korea that studies a failure prediction model for credit unions. Based on the data from 2001 to 2007, in which reliable data can be available, all credit unions are divided into two groups. Credit unions, either restructured or liquidated, are classified as "insolvent group" while others are categorized into the "solvent group". Quarterly financial data and character variables are obtained from 2001 to 2007 for the solvent group. On the other hand, same data and variables for the insolvent group are collected in the period between forth quarter and first quarter prior to the time of failure. Generally, when the variation of the financial ratios used as the explanatory variables is high, the outliers substantially affect a failure prediction model's prediction ability and thus reduce the predictability of the model. Another problem is that the level of influence on the failure probability of each of the regression coefficients is not easily comparable. This study is different from previous studies in that it converts individual independent variables into failure probability value through preliminary univariable analysis. Then, it uses the converted values as the final explanatory variables for the failure prediction function. This method will allow us to compare the level of influence of each of the regression coefficients and to eliminate the abnormal effect of the outliers. For the preliminary univariable analysis, I have considered 25 variables as the potential explanatory variables. After converting process through the univariable analysis, only 14 variables are chosen as the final explanatory variables. By applying stepwise logistic regression analysis on the 14 candidate variables, the following variables are found to be significant: net capital to asset ratio as capital adequacy index, the ratio of net substandard or below loan ratio and overdue loan ratio as asset quality index, and return on asset ratio as the profitability index. Besides above indices, total asset and character variable that represents unions located in the metropolitan areas are also found to be significant. It is found that total asset and net capital to asset ratio have the most association with the probability of failure. As asset size gets larger and net capital to asset ratio gets higher, the probability of failure becomes smaller. Therefore, in order to lower probability of failure and secure proper management of business, it is important to have certain level of asset size and capital adequacy. The AUROC and K-S statistics in the converted model which measure the predictability of model are shown to be 0.912 and 0.696, respectively. They are higher than those in the not-converted model, which the AUROC and K-S statistics are 0.880 and 0.611, respectively. Also, both in-sample and out-of-sample validation verifications of the converted model yield satisfactory results. Therefore it is proved that estimating the model using converted variables through univariable analysis results in a better failure prediction than one using not-converted variables. This study also finds that many of the CAEL evaluation indices are significant for the failure prediction. It means that the management status evaluation indices commonly used by the financial supervisory authority function can be useful as an early warning system to some extent. Considering the significant influence of total asset variable in the estimation of the model, it seems necessary to include total asset variable as an evaluation index. Another implication of this study is that it is necessary to allocate different weight on individual evaluation indices according to the level of influence that it has on the probability of failure. This study differs from previous studies in that it uses converted variables though univariable analysis in estimating the failure prediction model. By applying this method, it becomes possible to eliminate the effect of outliers and compare the level of influence of the regression coefficients. As for the validation verification method, I used the AUROC and K-S statistics instead of the accuracy ratio which is generally used in the previous studies. This method suggests that financial institutions can use advanced risk consulting techniques such as the Moody's in predicting the failure probability. In the future, the study needs to be developed and supplemented further, reflecting financial and non-financial variables that affect the insolvency of credit unions.