貝氏多層次模型在台灣不動產市場估價之應用─以台北市住宅建物為例An Application of Bayesian Inference in the Real Estate Market – A Case Study of Taipei Collective Housing

在房地產價格估計的領域當中，特徵方程式是最常被應用來估計建物價格的工具之一，然因特徵估價法是建構在線性迴歸的基礎之上，對於建物特徵與建物價格關係的描述過於簡化， 同時實務上存在諸多無法量化的因素，致使模型容易產生異質變異的現象，而現有的非參數模型有時過於複雜，且使用上的限制亦多。針對上述問題，本文嘗試採用多層次貝式模型來彌補線性模型的缺陷，有別於多數研究將區位視為建物特徵之一的假設，本文由區位不同造成異質變異的角度切入，重新呈現建物特徵與建物價格的非單調性關係。實證結果指出多數的建物特徵對建物價格的影響，多因區位而產生變化，且時呈不同方向，同時在異質變異現象獲得舒緩後，建物價格估價的精確度亦獲得顯著提升。
關鍵詞：特徵方程式、貝氏分析、馬可夫鏈蒙地卡羅法

How to estimate housing prices precisely has always been an important issue in the real estate market. Most studies adopt parametric or non-parametric methods to deal with problems such as heteroskedasticity or non-monotonic phenomena which come from less influential attributes or from characteristics which can not easily be realized. Researchers have attempted to adopt certain methods such as non-parametric methods to recover from these failures but they still do not work well. This paper therefore tries to re-examine the issue of heteroskedasticity in the housing price model. By using data for collective housing-type buildings in Taipei, this study employs the Hierarchical Bayesian model to bridge the relationship between attributes and housing prices. By means of a random effect device, the location effect gives rise to a non-monotonic effect on regressors that affect housing prices. Besides, capturing the heteroskedasticity effects results in the Bayesian model providing a better estimation than OLS.
Key words: hedonic equation, Bayesian inference, Markov Chain Monte Carlo