An Empirical study on K-MBS prepayment patterns
This study aims to analyze the determinants of prepayment risk in K-MBSs (Korean Mortgage Backed Securities) issued by KHFC (Korea Housing Finance Corporate). In total data used, 205 K-MBSs issued from 2004 to 2016 backed by 1,752,299 underlying assets that are all FRMs (Fixed Rate Mortgages). 14,573 observations (or month-security combinations) were recorded.
This research assumes that the five main causes of the risk are refinance incentive caused by the change of interest rate, cash-out refinance, housing turnover, curtailment, and default that the KMBS backed by FRMs has very low default risk. Four independent variables, market (interest rate) spread, change rate of apartment purchase price, aging in month, and seasonal dummy, are estimated for the effects on latest 12-month CPR (Conditional Prepayment Rate).
Furthermore the study tries to control unobserved factors and the unbalanced panel of the study that is possible to have time heterogeneity caused by different issuance dates. In terms of methodology, a panel regression Fixed Effects model and a Two-way Fixed Effects model with month dummy and year dummy variables are considered. The study also assumes that Korean MBS market is lemon market caused by the asymmetric information. It can be considered as unobserved historical factors on dependent variable and independent variables. Therefore the study employs one and two-month lagged dependent variables as proxy independent variables and one and two-month lag effects on the two main explanatory variables, the market spread between the mortgage rate of K-MBS and the prime mortgage rate in market and the logarithm changing rate of apartment purchase price for controlling these unobserved factors.
Overall, all explanatory variables except seasonal dummy are significant. Based on the result of adjusted R2, AIC, and F-test for H0: ui = 0, the most significant models are the 2nd order autoregressive (AR2) CPR model without lagged independent variables and the model with year dummy variables. The result summary of the AR2 CPR models is here,
1) When the market spread increases 1% point, CPR increases 0.57 ~ 1.07% point.
2) When ΔHPI of Apt. increases 1% point, CPR increases 0.4433 ~ 0.6405% point.
3) AR2 CPR model implies that 1% point increase in CPR one month ago affects 1.4519 ~
1.5375% point increase in the CPR this year and that of CPR two months ago leads to an estimated 0.4732 ~ 0.5579% point drop in the CPR this year.
The time lag effects on CPR are similar in all six AR1 (1st order autoregressive) CPR models and all six AR2 CPR models regardless of F-test so that the time lag effects on CPR exist in this research. It is also considered that there might be dispersion of the time lag effects on CPR and the two main explanatory variables even though the models are jointly significant in p-value of F-statistic. Arellano-Bond estimation of ‘Dynamic panel regression models’ is applied to the optimal model for controlling these unobserved historical factors. Comparing with the result of non-dynamic panel regression model based on the study assumption, one and two-month lagged CPR show the similar effects so that the study assumption is reasonable.
Consequently, the pool-level of this study with some controlling measures about the unobserved factors in panel dataset has an advantage that the study method estimates easily and directly the main four variables on CPR. This study will enhance the understanding of the prepayment factors of K-MBS and will be the basis for pricing K-MBS.
Click the button and follow the links to connect to the full text. (KDI CL members only)
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.