표본선택 모형을 이용한 LTV 추정 = LTV Estimation with a Sample Selection Model

다국어 초록 (Multilingual Abstract)

It is observed that a strong growth of housing price around Seoul metropolitan area since year 2000 led to pro-cyclical residential mortgage credit. However, the global financial crisis in 2008 caused rising mortgage delinquencies and declining expectation on housing price growth and thus this downturn deepened the negative cycles in housing market activities. Because mortgage loan is linked to the asset formation by increasing leverage, it has a fundamental problem to be vulnerable to the fluctuation of housing price. In this sense, the LTV (Loan to Value) ratio recently emerges as a main issue for maintaining macro-prudence and risk-management of financial institutions. For financial institutions, it is more important to understand and manage the quality of LTV in that higher LTV is associated with an adverse effect on financial and operational restructuring. When we construct an LTV decision model, it should be noted that only partial proportion of home buyers are using mortgage loan. According to 2011 Household-Finance Survey, 34% of homeowners are currently mortgage borrowers. Corresponding to the observed housing finance phenomenon, this paper provides the simultaneous estimation of mortgage loan probability and LTV ratio using a sample selection model. Unlike previous LTV-related researches, our empirics proposes an LTV model controlling various borrower`s socio-economic characteristics. For this purpose, we employ the Household- Finance Survey in 2011 which includes a representative sample of 10,000 Korean households. This survey has much information on household asset, liability and household-level micro financial soundness and therefore this data can be utilized to establish housing and real estate policy by government. Our empirical analyses find that the marginal effects of several variables on outstanding LTV are overestimated by OLS estimation because the conventional approach is not taking a selection bias into account. According to estimates of the sample selection model, LTV has strong positive relationship with household income whereas LTV has negative relationship with borrower`s age and net wealth level. More specifically, empirical results show that the marginal effect of income variable on LTV ratio is estimated lesser by 28% compared with the OLS estimate for the same variable. By post-estimation prediction, a 35-year-old borrower`s LTV level is 30.6% and this value is declining to 24.7% for a 65-year-old borrower. This prediction is in line with our intuition based on life-cycle effects. Finally, using our empirical estimation, we provide a simulation work of predicting the mortgage demand of FTB (First-Time home Buyer). Assuming 10% of Chensei households at aged 31~40 switches to home owners, the expected demand of mortgage loan for one year is calculated to be KRW 5.5 trillion.