類神經網路技術與特徵價格法於台北市房價預測結果之比較研究

Applying Artificial Neural Network and the Hedonic Equation in Real Estate Market in Taipei

許智淵(Chih-Yuan Sheu)

指導教授 ： 吳泰熙

共同指導教授 ： 李緒東(HsuTung Lee)

國立臺北大學/商學院/企業管理學系/碩士(2012年)

摘要

如何精確估計不動產價格一直為房地產研究領域的熱門話題。影響房地產價格形成因素眾多，變數選擇與估價方法之不同亦會影響估價的準確性。過往較常使用市場比較法、收益還原法、成本法與特徵價格法(迴歸模式)等進行估價，但上述方法各有其主客觀考慮不同之處，並不能準確有效地進行預估。隨資訊產業快速發展，利用電腦模擬人類思考模式而發展的類神經網路技術(artificial neural network, ANN)則逐漸運用於估價上。
產、官、學界過往針對房地產價格預測研究，主要針對透過不同估價方法與計量模型之建立，以進行房地產價格預測與檢驗。並探討房地產價格變化之相關變數對房地產價格影響程度。
本研究分別運用特徵價格法(迴歸模式)及類神經網路技術預測台北市房地產價格，試圖建立一套估價模型以進行房價之預測。實證結果顯示，資料樣本區分時，特徵價格法之預測準確性優於類神經網路。時間因素對預測準確性之影響並不顯著，但對公寓住宅則有顯著提升，整體預測準確性仍以特徵價格法優於類神經網路。以平均成交單價為依變數時亦有同樣之結果。影響台北市房價之主要因素仍以房屋自身因素如坪數、有無停車位、樓高比等因素較為相關，總體環境因素與鄰里環境影響房地產價格並不明顯。

關鍵字

類神經網路 ； 特徵價格法 ； 房價

並列摘要

Estimating a housing price precisely is always an important issue in real estate research. Normally, it is hard for real estate consumers or investors to evaluate their real estate purchases or investments. The value assessments in the past were made manually and wasted too much time. The housing price probably has lots of research bias. Due to the development of information technology, many researchers imitated the functioning of the human brain to develop the neural network. It was applied to real estate appraisal in the late 1990s.
The empirical results show that the hedonic price is better than ANN. The influence of time serious impact the Housing price is not significant. However, it will influence the apartment a lot. In unit price has the same results. Among all the influences on real estate prices, housing characteristics like feet, parking, and floors of the building have direct impacts on the structural changes of housing prices. Neighborhood attributes, and macroeconomic variables affects prices indirectly.
In this paper we build a mass appraisal model using the hedonic equation model and annual neural network (ANN), including housing characteristics, neighborhood attributes, and macroeconomic variables, and then apply a dataset with more than 10,000 real estate mortgages to construct a housing price equation. According to several statistics, including R-squared, the mean absolute percentage error (MAPE), and the firecast error (FE), the findings indicate that the estimation results of our pricing model are quite satisfactory. This implies that mass appraisal could be a good model to apply in Taiwan.

並列關鍵字

Artificial neural network ； Hedonic price ； Property price

參考文獻

林英彥 (2003)。不動產估價理論。臺北：文笙書局。

林祖嘉、馬毓駿 (2007)。特徵方程式大量估價法在台灣不動產市場之應用。住宅學報，16(2)，1- 22。

高明志 (1998)。類神經網路應用於房地產估價之研究。住宅學報，8，1-20。

曹昌勝 (2007)。房地產泡沫化模型之探討。國立台灣大學國際企業學系研究所未出版之碩士論文。

陳奉瑤、楊依蓁 (2007)。個別估價與大量估價之準確性分析。住宅學報，16(2)，67-84。