台灣證券市場權益風險溢酬之估測_比較GARCH-M、移動視窗和MIDAS模型

Abstract

本篇論文使用Ghysel (2005)所提出混合數據抽樣模型(Mixed Data Sampling)，探討台灣證券交易市場的權益風險溢酬，在引入市場報酬的條件期望值和條件變異數下，以跨期資本資產定價模型為基礎來進行估計與預測。樣本期間自2006年1月至2010年12月，以股票報酬之條件變異數作為風險替代變數，預測對象是以月頻率為單位，觀察資料則為日/週頻率，針對不同的權重函數和波動因子，進行和GARCH-in-mean模型與移動視窗模型之比較。 實證結果發現：(1) 此樣本期間之風險和權益風險溢酬有負向關係存在。(2) MIDAS模型在時間序列資料的迴歸估計能力較顯著，其次是移動視窗法，且樣本外資料的預測誤差偏小，表示預測能力良好。(3) 根據不同的波動因子和抽樣頻率，以日頻率報酬資料的平方多項式有著較顯著估計結果。有別於傳統研究方法，混合數據抽樣模型最大特點為採用不同頻率資料，配適出最佳迴歸模型，以此估計證券市場的條件變異和風險報酬。

This paper investigates risk premiums of Taiwan Stock Exchange Capitalization Weighted Stock Index (TAIEX) by using Ghysel’s mixed data sampling (MIDAS) model which is a new regression regarding volatility estimation. We study the intertemporal relation between conditional mean and conditional variance of the aggregate stock market return. Compared with various approaches such as GARCH-in-mean, rolling window and MIDAS models, we find that: (i) We support for a negative relation between risk and equity risk premium in TSEC weighted index during the period 2006 - 2010. (ii) MIDAS is more convincing in predicting regression for sampled time-series data. (iii) Empirical results show out-sample forecasting ability of MIDAS model also performs well. Specifically, it has smaller forecasting error. (iv) Under MIDAS model of different volatility predictors and different sampling frequencies, a squared premium polynomial with daily frequency data has better estimation.