農地價格指數之建構：應用臺灣實價登錄資料的大數據分析

在過去的文獻中，很多研究針對民宅、商用等不動產建構價格指數，有關研究相當豐富，而農地方面則十分缺乏。在臺灣方面，農地價格高居世界第一，農地面積亦迅速凋零，有鑑於前述問題，建構一套能夠有效衡量趨勢的農地價格指數是一個重要的問題。然而，過去的不動產假格指數建構方法只針對民用及商用建築，採用的方法、納入的變數未必適合於農地。所以，本言就企圖提供數種研究方法建構農地價格指數，比較優劣利弊，並供未來其他研究借鑑。至於方法的驗證，本研究納入臺灣鄉村社會學會（2019）所採用的臺灣實價登錄資料，分為都市與非都市進行驗證。總體來說，本研究提供了三種系列方法，分別是特徵價格法的修正、配對法以及拉索迴歸。經驗證後結果顯示，在特徵價格法的修正方面，以土地作為權重進行加權迴歸可以增進模型解釋力、降低均方誤差，而相關文獻也指出以標準差修正可以使推定量不偏；至於配對法方面，本研究採用最近配對法以及模糊配對法，透過t配對樣本檢定，結果顯示兩種方法皆能改善選樣偏差問題，但模糊配對法表現比最近配對法更好；而拉索迴歸方面，本研究則提供了兩種演算法，分別是double-selection與partialling-out。透過資料切分技巧進行訓練，本研究發現拉索迴歸相較於傳統特徵價格法，有更好的預測能力、解釋能力，並降低誤差。除此之外，拉索迴歸亦有變數篩選的功能。總結來說，本研所提供的三種研究方法，能夠補足臺灣過去沒有農地價格指數的研究缺口，並以新的方法解估計時所面臨的問題。

關鍵字

農地價格；實價登錄；特徵價格法；配對法；拉索迴歸

並列摘要

In past studies, there are a lot of researches focusing on property price index. However, few scholars concerned about farmland price index in Taiwan. Nevertheless, the quantity of farmlands is decreasing sharply. Moreover, the average price of farmland price is the most expensive in the world. Consequently, farmland is undoubtedly an important issue for modern Taiwan. Through constructing farmland price index, we can obtain some evidence and suggestion for policy decision. However, the past methodology may not be appropriate for farmlands, so we would like to offer three new methods to constructing price index. The results suggested that involving farm area in weighted least square hedonic can enhance the explaining ability and decrease the error. As to matching methods, they can eliminate selection bias successfully. Additionally, coarsened matching performed better than nearest matching. The last one is LASSO regression. LASSO not only predict better but also chooses variables which could affect farmland price. To sum up, these three methods perform well in some ways. This research may be a guideline when dealing with farmland price index in the future.