近年來對於住宅價格模式之許多相關研究,大多使用特徵價格法、類神經網路、倒傳遞類神經網路等方法,獲得相當豐碩的成果,且由於住宅價格(包括特徵因子)常隨時間及空間產生變化,因此,對於具時間變化之住宅價格,常以結構性轉變的觀點進行探討;另外,透天住宅與店舖住宅兩者之住宅價格,存有許多差異,致使價格預測(評估)不易。因此,本文乃利用台南市東區1996年至2007年住宅價格簡訊之資料,首先以時間序列方法,分別進行透天住宅與店舖住宅價格預測模式之研究。時間序列方法主要應用在特徵價格模式係數及平均交易總價之歷年變化預測;最後,本文並將模式所得結果與傳統特徵價格模式進行比較。 本文獲得以年平均交易總價之時間序列模式(稱為模式一)及特徵價格係數之時間序列模式(稱為模式二),其中模式一只針對年平均交易總價進行預測,模式二可針對單一住宅進行預測。由本文案例結果比較顯示:(1)對2007年實際平均交易總價預測,透天住宅以模式一最佳,其誤差值為5.937%;對2007年實際平均交易總價預測,店舖住宅以模式二最佳,其誤差值為7.302%,均優於以傳統特徵價格法對透天住宅及店舖住宅之誤差值(分別為14.847%及49.864%);(2)對2007年透天住宅(204筆)及店舖住宅(46筆)之實際交易案例進行預測,以透天住宅之整體平均預測誤差而言,模式二(MAPE=26.888%)較傳統特徵價格法(MAPE=32.143%)為佳,以店舖住宅之整體平均預測誤差而言,模式二(MAPE=30.522%)亦較傳統特徵價格法(MAPE=66.375%)為佳。
Most of researches regarding the house prices model in recent years are applied with Hedonic Price Method, Artificial Neural Network and Back-propagation Neural Network, and have obtained quite significant results. The time-dependent house prices are often discussed from the angle of structural transformation because they (including characteristic factors) often change accompanying the time and space. Besides, the prices of town house and store house vary largely, so the price prediction (evaluation) is difficult. This research utilizes the data of house price index in East District of Tainan City from 1996 to 2007. Firstly, it probes into respective price prediction model of town house and store house through Time Series Analysis. This method is mainly applied to hedonic price model coefficient and the prediction of average house transaction price. Finally, this research compares results derived from the model with those of conventional hedonic price model. Two models are acquired in this research including the time series model of annual average transaction price (Model 1) and the time series model of Hedonic Price Coefficient (Model 2); Model 1 only predicts the annual average transaction price and Model 2 focuses on the prediction in respect of a single house. Results from the case study reveal that: 1. Comparing with the actual average transaction price of 2007, Model 1 acquires the best prediction for town house and the deviation value is 5.937%; Model 2 acquires the best prediction for store house and the deviation value is 7.302%. Both deviation values are better than those derived from the conventional hedonic price model in respect of town house and store house (14.847% and 49.864% respectively). 2. Conducting the prediction in respect of actual transactions in 2007 including 204 cases of town house and 46 cases of store house, Model 2 (MAPE=26.888%) is better than the conventional hedonic price model (MAPE=32.143%) as far as the average deviation of town house is concerned; Model 2 (MAPE=30.522%) is also better than the conventional hedonic price model (MAPE=66.375%) in point of the average deviation of store house.