使用資訊熵分析房價成長風險

Abstract

本文以新型的方法資訊熵建構台灣房價風險值模型，模型設定以無條件分量迴歸做為房屋市場、條件分量迴歸歸作為房屋在金融市場的模型，兩者相減為房屋市場中潛在的不確定性，也就是房價風險。無條件分量迴歸為對傳統分量迴歸的拓展和補充。研究有關無條件分量迴歸的理論與方法正在逐漸完善中，本文旨在介紹模型並整理相關文獻。實證結果顯示，房價成長的變動具有不對稱性，且只能做部份預測。考慮到參數及非參數分配是否擬合台灣房價成長的結果，我們使用機率積分轉換 (Probability integral transform) 作為選定最佳預測的模型。最後根據結果，左尾在危機時的波動大於右尾在房價成長擴張期，我們發現此模型對於預測房價成長的下行風險相對於上行風險更具可行性。

The thesis studies the risk of house price changes in Taiwan, particularly its downside risk of house price growth by using information entropy as the new growth-at-risk framework. We used conditional and unconditional quantile regression to evaluate the housing market, where the conditional quantile is estimated based on the financial market information. The uncertainty of the housing market was calculated by deducting the conditional quantile regression results from the unconditional quantile regression results. We used a few parametric and nonparametric methods to fit model, and used probability integral transform tests to select the optimal density function that fits Taiwan's data. We found that housing price growth risks are time-varying, asymmetric, and partly predictable. We also found that the left tail of the future housing price growth distribution is responsive during crisis while the right tail does not show responsiveness during expansions. The result indicates that the model is more capable in forecasting downside risk as opposed to the upside risk.