時變或固定折現因子──以訊息萃取架構探討股價變動的因素

TIME-VARYING OR CONSTANT DISCOUNT FACTOR: THE ANALYSIS OF MOVEMENTS OF STOCK PRICES BY SIGNAL EXTRACTION FRAMEWORK

本文以折現因子之資本資產定價模型做為分析基礎，來探討時變折現因子是否比風險中立設定下固定折現因子的資本資產定價模型，對股價能夠擁有更佳的解釋力。在詮釋現實的資料上，我們分別將三種文獻上常用的時變折現因子模型拿來跟固定折現因子模型做比較，利用Durlauf and Maccini（1995）所建構的噪音比統計量來檢視何者的表現較佳。本文以台灣的加權股價指數月資料和美國的S&P 500指數月資料，分別對兩者做出比較和分析，實證結果顯示，時變折現因子模型的噪音比，在詮釋現實資料之表現上，優於固定折現因子模型；而由於直接用噪音比做比較有統計顯著上的不足，為了彌補這個缺口，我們設計了一個輔助性的統計頑強性的檢定程序，即利用「前樣本」對每一組相對應的模型做遞迴迴歸和滾動迴歸，以產生「後樣本」的t、Wilcoxon和DM統計量，來檢定對應組間噪音比差距是否具有顯著的差異。最後證據顯示，時變折現因子模型比固定折現因子模型更接近股市投資人行為的結論具有頑強性。

In this paper we explore the possibility that risk aversion leads to greater dispersion of stock prices than does risk neutrality based on the consumption capital asset pricing model. This is done by comparing the empirical performance of the asset pricing models with a time varying discount factor to the model with a constant discount factor. We employ the noise ratio method developed by Durlauf and Maccini (1995), to demonstrate the relative degree to which each null model approximates the data. The specification tests are applied to the data of Taiwan’s stock market as well as the S&P 500, and we find evidence that the asset pricing model with a time varying discount factor is superior to that with a constant discount factor. In order to make the test robust, we design a testing procedure of the t, Wilcoxon, and DM statistics to test the magnitude of the noise ratio difference between these models using a post-sample approach, which is generated by both recursive regression and rolling regression. The results show that there do exist some time-varying discount factor models that are superior to the benchmark for both indices.