多元波動率因子模型於最佳投資組合選取之研究

Abstract

本文基於傳統兩階段的 Markowitz 投資組合選取理論，將其拓展成較容易時做的三階段投資組合選取架構。透過新加入的「投資組合建構」階段，我們可以巧妙地避開高維度共變異數矩陣估計與預測的問題，並妥善運用較為成熟的單元波動率模型。 此外，於本文我們運用了三種多元波動率因子模型，四種投資組合選取策略，以及兩種經過風險調整後的報酬率指標，進行最佳化投資組合的選取，並將其運用在兩組實務資料中。我們的資料包含匯率資料以及半導體類股資料，實證結果相當優異。 根據提出的交易策略，我們進行了縝密的分析。結果顯示：(1) 投資組合預期報酬的預測準確度並不是最重要的原因 (2) 我們的策略完全打敗傳統的最小變異數投資組合與等權重投資組合。

In this thesis, we extend the traditional Markowitz's procedure to an easy-to-implement three-stage portfolio selection framework. By introducing the portfolio derivation strategy, we smartly avoid the problem of high-dimensional covariance matrix forecasting and leverage the maturity of univariate volatility models. Specifically, we apply 3 portfolio derivation strategies by factor volatility models, 4 portfolio selection strategies, and 2 risk-adjusted return portfolio selection measures. We implement these algorithms on foreign exchange rate dataset and the semiconductor stock dataset, leading to outstanding performances. We also conduct detailed analyses about our proposed trading strategies. The result suggests that (1) the forecast accuracy of portfolio returns is not the most important thing and (2) our proposed strategies outperforms traditional minimum-variance and equally-weighted portfolios.