A Penalized Regression Based Repeat Sales Price Index Estimation

Transaction-based HPI (house price index) in a low transaction-volume area inferred by the repeat sales price model frequently suffers from high volatility of the estimated indexes because of the thin file of the house prices traded at least over two times during the whole periods of interest. We tackle the problem by using a penalized regression for constructing repeat sales indexes which are induced from smoothed regression coefficients for some properly selected regularization parameter. As a regularization term we consider a ridge type penalty materialized with the difference of two adjacent coefficients in order to make the estimated regression coefficients smoothed enough to induce the corresponding stable house price indexes. The proposed method is applied to the real data set and the results show its superiority to the ordinary repeat sales model especially for the house price indexes in a thin file area.