不動產預售代銷來客購買機率之預測模型建立-以台北市為例

Abstract

購屋的決策行為近年來逐漸受到學界重視，然而國內的相關文獻多是從已購屋者，或是從購屋意願等單一角度去探討購屋考量因素、個人特徵對購屋意向的影響、搜尋方式對購屋行為的影響等，鮮有研究係針對顧客的購買機率或預測模型的概念進行探討。雖然專家可以憑藉著個人的經驗來判定來客的購買可能性，業界亦普遍認為瞭解來客的購買機率是十分重要的，但至今未有研究建立預測模型於不動產領域之來客購買機率。 不動產交易雖然涉及金額龐大，但仍屬於消費者行為的範疇中，而消費者行為乃是針對特定產品與情境所產生的，因此相異之產品所引發的購買行為就會不同，甚至是同樣的產品其消費者行為也會隨不同的使用者而異。回顧過去的相關研究結果發現，消費者者對預售屋、新成屋與中古屋這三者的購屋偏好各異，因此不同的不動產市場並不宜合併討論。又隨著研究對象的不同、研究結論亦不相同，若僅針對已購屋者的購屋意向進行研究，則可能因資料分析的侷限，而無法有效地預測未來的購屋決策。因此本研究參考一家上市不動產代銷業者之已購屋者與未購屋者資料，分析該公司從2012年到2013年於台北市所承銷之預售屋推案，藉由Logistic迴歸模型建立來客購買機率預測模型，並分別就推案之基地坪數大小、房屋坪數大小與推案坐落之行政區這三方面進行探討，最後以Leave-One-Out交互驗證法收集模型之預測結果、計算二維接受者操作特性曲線(Receiver Operating Characteristic Curve, ROC-Curve)之曲線下面積(Area Under Curve, AUC)作為預測模型的驗證。 不動產的特性多元、來客的購買行為亦十分地複雜，本研究雖然無法較全面地反映不動產購買之相關特徵，可能因此使得本研究所提出之預測模型的類R2略微偏低，但是從交互驗證的結果顯示，本研究所提之各個預測模型的排序能力多達到可以被接受的標準、亦符合專家學者們的接受水準，代表來客之購買機率預測模型於預售屋市場是具有其可行性與貢獻性的。

Research of house buying in the real estate industry is gradually catch people’s attention in recent years. However, most research is focusing on the preferences of final buyers, the relationship between consumer’s preferences and house properties, the impact of different searching methods in house buying and so on. Topics like purchasing probability or forecasting model in house buying are still waiting us to discover. Although experts can guess a rough percentage for each consumers through their experiences, and it generally reaches an agreement that it is very important to know the probability of a consumer to buy a house, there is no research in Taiwan to establish a forecasting model in realestate industry to predict the probability of a consumer to buy the house until now. The transaction amount of house buying is relatively large, but it is still in the scope of consumer behavior. That is to say, different products may cause different purchasing behaviors; even the same product may lead to different results according to different consumers. When we looked back to what we could know from the historical research, we found out that consumers would change their preference procedure to different types of house, such as pre-sale, new built, and existing house for many years. Therefore, we should not put them together to discuess. On the other hand, if we only looked from one side, like buyers’ point of view only, and ignored other consumers who did not buy the house in the end, it is possible that we could not predict the future market efficiently due to the limitation of data analysis. Consider what mentioned before, this research is aimed to a listed real esate agency company, collecting both the buyers and the consumers who didn’t buy the pre-sale house form 2012 to 2013 in Taipei, and finally establishing a forecasting model by logistic regression model in pre-sale house market, which is verified through Leave-One-Out Cross-Validation method. Real estate has various characteristics, and consumer behavior is also very complicated. This research may not able to completely explain it, which may lead to low R-square for the models, but it can be proved that most models in this research is acceptable according to AUC results and meet the expectation of experts. In conclusion, the forecasting model in pre-sale house buying is feasible and has crucial contribution to Taiwan’s real estate industry.