This thesis provides the analysis of adjustable rate mortgages which are prevalent on the residential mortgage market in Korea. It involves three main topics - prepayment and default modeling, valuation methods and empirical tests for valuation and sensitivity of various factors on ARM. Firstly, the prepayment models of mortgage were traditionally estimated by the aggregated data to reduce computational loan and turnaround time. The model here preserves all loan-level informations to estimate Discrete-time Multinomial Logistic Models for conditional probabilities of default and prepayment. The model can conform more closely to the way prepayment and default are actually measured in empirical test.
Secondly, this thesis analyzes the dynamics of the commonly used indices for ARMs and systematically compares the effects of their time-series properties on the interest-rate sensitivities of ARMs. Based on the interest rate analysis, the Monte Carlo Simulation was suggested to generate cash flows and evaluate ARMs. The reason is why securities like ARMs whose payoffs depend upon the actual path of the underlying state variables pose problems for standard backward-valuation technique. In other words, option-based pricing on ARM is not good due to interest path-dependency of cash flows.
Lastly, this thesis provides empirical test for the valuation of MBS samples whose underlying assets are adjustable rate mortgage. The results show that the prepayment and default probabilities significantly have an effect on the value of MBS. I find that the different dynamics of the ARM indices lead to significant variation in the interest-rate sensitivities on loans based on different indices. I also find that changing assumptions about contract feature, such as loan caps and coupon reset frequency, has a impact on the results.