近年來隨著不動產實價登錄政策之施行,內政部將大量實際成交案件建構成不動產資料庫,如能善用其中之不動產資訊將可增進估價之效率,故建立可迅速且可靠評估不動產價值之估價模型日益重要。然而現行不動產估價中常使用之特徵價格法(Hedonic Price Method)為全域性之分析,缺乏對於空間與時間非均質性之探討,易使模型對真正價值之可解釋性不足,而地理加權迴歸則被認為可有效解決特徵價格法模型所採用之最小平方法(method of least squares)所出現之殘差項存在空間自相關的情況。 本研究主要目的為改善現行不動產特徵價格模型,使用內政部不動產實價登錄資料庫內之不動產交易資料進行分析,修正基於最小平方法所做出之特徵價格模型中假設整體空間均質性之瑕疵,並試驗在地理加權迴歸、地理與時間加權迴歸中加入傳統特徵價格模型之年份變數對模型成果之影響,並分別測試在固定帶寬與調適帶寬兩種模式下對於不動產價值之空間迴歸結果之差異。 綜觀研究成果,調適帶寬優於固定帶寬,且使用地理與時間加權迴歸時若於模型中加入年份變數將會對帶寬效果產生影響。此外,使用地理加權迴歸時若加入年份變數則可提高模型擬合效果,並可改善殘差項空間自相關的問題,而使用地理與時間加權迴歸時則以未置入年份變數之模型較佳。因此,地理加權迴歸並導入年份變數之模型,較地理與時間加權迴歸模型更為適用於本國不動產大量估價。
In recent years, with the implementation of the real estate actual price registration policy, the Ministry of the Interior has constructed a real estate database with large number of actual transaction cases. Real estate valuation can be improved if the database is properly used. It is important to establish a valuation model that can quickly and reliably assess the value of real estate. However, the Hedonic Price Method, which is often used in current real estate valuations, is a global analysis method that does not consider spatial and temporal heterogeneity. Therefore, it is difficult for the model to predict the true value. Conversely, the geographically weighted regression can resolve the problem of spatial autocorrelation in the residual term that is commonly seen in the method of least squares analysis adopted by the Hedonic Price Method. The main objective of this study was to improve the Hedonic Price Method. Using the real estate transaction database of the Ministry of the Interior for analysis, this study aimed to correct the defect due to the assumption of overall spatial homogeneity in the model based on the least squares method. Moreover, the study examined the effect of year variable on the model of geographically weighted regression (GWR) and geographically and temporally weighted regression (GTWR), and investigated the difference of the spatial regression results of real estate value under fixed bandwidth and adaptive bandwidth modes, respectively. The research results indicate that adaptive bandwidth mode is better than fixed bandwidth mode, and adding year variable to the model of GTWR will have an impact on the bandwidth effect. Furthermore, adding year variable to the GWR model can improve the fitting result, and resolve the problem of spatial autocorrelation in the residual term, while the GTWR model achieves better results without the year variable. Therefore, GWR model added with year variable is more suitable for large-scale valuation of domestic real estate than the GTWR model.