A Study on Repeat Sales House Price Index Based on Penalized Quantile Regression

House price index (HPI) based on real transaction prices is commonly estimated by the repeat sales model which utilizes sales pairs traded over twice during the period of interest. Since the conventional repeat sales model is based on the ordinary multiple linear regression, it is affected considerably by outliers. In addition, it produces quite unstable HPI in case of insufficient sales pairs. To tackle these problems, a quantile regression based repeat sales model with an appropriate penalty function is suggested. The induced HPI from the proposed model shows much stability compared to the conventional methods. Moreover, the HPIs derived from the penalized quantile regressions for various quantile parameters provide us with valuable information which cannot be obtained from the existing repeat sales models based on ordinary multiple linear regression. As a result, the proposed method for constructing HPIs can be considered as an alternative to the conventional repeat sales price indices in that it provides very stable and informative HPIs.