주택담보대출 연체율의 분포적 특성을 고려한 미시 스트레스 테스트 - 이항분위회귀 모형을 이용하여 -

초록

This study aims to offer a new empirical methodology, the binary quantile regression model,in modeling residential mortgage delinquency. Although it is well-developed for a continuousdependent variable, the quantile regression technique for a discrete response variableis not yet firmly settled. In this paper, we develop a parametric binary quantile regressionmodel to estimate the risk of mortgage delinquency in Korea, which is implemented througha Markov-Chain Monte Carlo (MCMC) algorithm. Based on a borrower level mortgage performancedata set, our empirical results show that: the main drivers of mortgage default observedin the U.S., such as LTV, DTI, and consumer credit ratings, are all important explanatoryfactors in Korea; the magnitudes of impacts of those independent variables varyacross four quantiles defined (50th percentile, 90th percentile, 95th percentile, and 99th percentile);and, while DTI was shown to have a larger impact on the delinquency risk underthe conventional logistic model and the 50th quantile regression, the effect of LTV magnifiesin the higher-risk quantiles. As a subsequent step, we perform a stress testing by definingthree severe economy-wide stress events, and demonstrate that the logit model based delinquencyprediction generates significantly under-biased estimates compared to the outcomesof the quantile regressions. For example, the 95th quantile regression predicts the delinquencyrate that is 2.74 times higher than that from the logit model assuming a same stressscenario.

키워드