結合時空因子的大量估價模型應用： 桃園蛋白、蛋黃、邊緣區的分析比較Application of Mass Appraisal Models with Time and Space Factors: Analysis and Comparison of Core, Inner Suburbs, and Outer Suburbs of Residential Properties
本文以桃園市中古住宅大樓2018年至2020年的房產交易資料為對象，應用地理加權迴歸時間數列法(geographical weighted regression-time series, GWR-TS)建立房價特徵模型(Fotheringham et al., 2015)。首先，觀察特徵的隱含價格在蛋白、蛋黃、邊緣區的時、空分佈差異，再預測桃園市未來的房價，與傳統的最小平方法(ordinary least square, OLS)估計之特徵估價模型比較，並檢驗兩模型的預測表現。研究結果發現，蛋黃區、蛋白區及邊緣區房價的差異主要是受區域因素的影響。其次，住宅特徵的隱含價格會隨時間而變化，並受住房市場景氣影響，數值在空間中出現收歛的漣漪現象。實證結果證實GWR-TS模型的預測能力及配適度均優於OLS模型。
This paper studies the real estate transaction data of Taoyuan City from 2018 to 2020 and applies geographic weighted regression with time series (GWR-TS) approach (Fotheringham, Crespo, & Yao, 2015) in modelling the hedonic housing price. This paper observes the temporal and spatial differences of the implied prices in the core, inner suburb, and outer suburb and forecasts the future housing prices. Then the prediction performance of hedonic model applying GWR-TS and OLS are compared. The empirical results show that, first, the difference in housing prices in the core, inner suburbs, and outer suburbs is mainly affected by regional factors.
Second, the implicit price of the hedonic model changes over time. The values which may be related to the rapid cooling of the housing market appear to converge during study period, a phenomenon that is similar to ripples effect. Finally, the model prediction ability and goodnessof-fit of GWR-TS are superior to the traditional OLS hedonic model.
Key words: House prices, Spatial heterogeneity, Geographical weighted regression, Hedonic model