本文應用人工智慧領域新發展的遗傳規畫重新對林祖嘉等(1994)自然空屋率的估計進行研究。雖然遗傳規畫所得到的估計值舆論文研究相似，但是對自然空屋率高峰期的出现卻有不同。另外，遗傳規畫也發现空屋率背後的模型，可能有穩定性的問題。這點也許與房地產市場在資料的涵蓋期間，正好是由衰退步入繁榮的變化有關。 This paper discusses the applications of genetic programming to the empirical study of the natural rates of vancancy in Taiwan's housing market. The genetic programming paradigm, a new approach developed in artificial intelligence, is an automatic model search process and is very promising in treating the issue of model selection. By using the model in Lin et. al. (1994) as a benchmark, we explore the advantages of this approach by demonstrating two things: firstly, how genetic programming can be used to investigate the robustness of a given model; secondly, how genetic programming can be used to detect the potential nonlineraity and structural stability in the model. Our findings are two-fold. First of all, using the data running from 1981 to 1988, we find that the 2SLS model considered in Lin et. al. is pretty robust at least in the sense of linearity. However, if we exclude 1981 and add 1989 to our data set, the model is not robust any more. Our further analyses suggest that business cycles in the housing market might affect our estimate of natural vancancy rates and should be taken into account in future studies.