동태적 모형 선택법을 이용한 주택가격 예측 변수 분석

This paper presents a time series forecasting model of the Korean housing prices using the dynamic model selection. This methodology is attractive in that not only model coefficients but also the entire forecasting model (i.e., predictors) are allowed to change over time. It is found that dynamic model selection performs better than other alternative popular forecasting models including conventional regressions, random walk, autoregressive models, and time-varying parameter models. When looked at which sets of predictors are relevant for forecasting in each period, there are noticeable changes in predictors over forecast horizons. For short horizons, housing prices seem to be largely affected by housing market developments and business cycles; however, GDP and consumption seem to play a lesser role during housing market booms. For medium-run horizons, monetary aggregates and financial market variables appear to replace the aforementioned variables and have more predictive power, indicating that portfolio adjustment in asset markets may be a critical factor in housing price changes.