樣本外匯率預測：使用IPCA

Abstract

本文利用Kelly, Pruitt, and Su (2019) 提出的工具變量主成分分析(Instrumented Principle Component Analysis, IPCA) 作為實證模型，對G10國家貨幣的超額報酬進行分析。IPCA是一種線性隱因子模型（linear latent factor model），允許研究者利用資產報酬以外的資產特徵估計風險因子（risk factor）和動態因子載荷（dynamic factor loadings）。本文以擴展窗口（expanding window）的方式遞歸地對樣本外資料進行預測，實證結果顯示IPCA模型對貨幣超額報酬在樣本外的預測表現優於隨機漫步模型（random walk model）和傳統上以PCA估計的線性隱因子模型。特徵重要性分析結果顯示利率相關變數如短、中、長期公債殖利率差額對貨幣超額報酬最為重要。同時本文也發現IPCA模型不止對貨幣超額報酬有很好的預測能力，對匯率一般報酬的預測表現也優於隨機漫步模型。

This paper utilizes the Instrumented Principle Component Analysis (IPCA) proposed by Kelly, Pruitt, and Su (2019) as an empirical model to analyze the excess returns of G10 currencies. IPCA is a linear latent factor model that allows researchers to estimate risk factors and dynamic factor loadings using characteristics information other than asset returns. This paper employs an expanding window approach to recursively forecast out-of-sample data. The empirical results show that the IPCA model outperforms the random walk model and the traditional linear latent factor model estimated via PCA in predicting currency excess returns. The results of the feature importance analysis indicate that interest rate related variables, such as the yield spread of short-term, medium-term, and long-term government bonds between a given country and US, are the most important for currency excess returns. Additionally, this paper also finds that the IPCA model not only exhibits good predictive ability for currency excess returns but also outperforms the random walk model in predicting general exchange rate returns.