以決策樹之迴歸樹建構住宅價格模型──台灣地區之實證分析

Regression Trees for Housing Price Models: An Empirical Study on Taiwan

以往房地產特徵方程式之估計多採用複迴歸模型，後期學者開始使用半參數或無母數方法估計特徵方程式，本研究以決策樹中的Cubist迴歸樹作為房地產特徵方程式之估計模型，主要原因有三：其一，Cubist迴歸樹模型之設計符合房地產資料特性。其二，Cubist迴歸樹配適能力高且易於解釋。其三，當使用大量資料估計特徵方程式，Cubist迴歸樹相對於其他無母數方法運算上較有效率。本研究以台灣地區2002年至2004年間45，419筆房地產資料為研究樣本，以複迴歸模型為基準模型，研究發現迴歸樹之配適能力高於複迴歸模型，且並未有過度配適之問題。此外，特徵變數與房地產價格間存有非線性關係，個體變數較總體變數具有廣泛之解釋力。

The purpose of this paper is to use Cubist regression trees to estimate the hedonic equation, as the Cubist is expected to be more efficient than other nonparametric methods. In addition, the architecture of the Cubist is intuitive when applied to the housing price model. In this study, the regression method, which is frequently used in the estimation of the hedonic equation, is used as the benchmark model to be compared with. Based on 45,419 observations from the Taiwan area, it is found that the Cubist outperforms the benchmark model. Moreover, it is found that there are nonlinear relationships between the house prices and the characteristic variables. Finally, the micro characteristics exhibit higher explanatory power than the macro ones.