本文採用「住宅需求動向調查」2006Q4 至2007Q3 之已購屋者問卷，以二元羅吉特模型之機率預測模型，分析投資者與自住者在住宅屬性、購屋搜尋行為與購屋者屬性之差異，並探討不同機率界限下，預測可信度最高之最適機率界限值。實證結果顯示，購買區位在中心都市、高單價、小面積產品、預售屋、中古屋與法/ 銀拍屋之購屋者，其為投資目的機率較高。此外，投資者具有搜尋時間短、年齡較長、男性、無固定職業及家庭平均月收入較高等特徵。本文運用貝氏定理得出當機率界限值為0.7 時，模型預測可信度最高 ( 投資者達72.22%，自住者80.06%)。以2007Q4 的資料作樣本外驗證，投資者預測可信度為65.52%，自住者預測可信度為84.51%。最後，模擬不同權重下，應貸而未貸以及違約與提前清償的誤差，對於金融機構個別可能造成的損失，模擬結果同樣顯示0.7 為損失最少的機率界限值。

This study investigates the government-conducted “Housing Demand Survey” from 2006Q4 to 2007Q3 to establish a binary logit model to analyze the housing attributes, search behaviors, and investors and owner-occupiers’ personal characteristics for financial institutions prediction in determining house buyer motivation and reducing the loss incurred when payment is not advanced or is defaulted. The empirical results show that the locations, prices, sizes, and types of houses are the dominant factors affecting investment probability. Investors end to select houses located downtown with high unit prices and small in size. They also tend to buy pre-sale, resold, and auction houses. The individual attributes of buyers, such as gender, age, occupation, and income, also play an important role in investment decisions. The proposed model has the highest prediction accuracy when the probability cutoff poin is 0.7 and the investors’ hit rate is 65.52 % in contrast to the owneroccupiers’ 84.51%, as indicated by the out-sample test. This study also determines that financial institutions incur the least loss with a cutoff point of 0.7 under different weights, when payment is not advanced or is defaulted.