外匯報酬之流動性、動能及價值交易策略分析

本研究以探討流動性因子是否可解釋外匯報酬為主軸，比較兩種衡量流動性之指標Bid-Ask Spread與Corwin and Schultz (2012) 所建構之流動性因子，何者解釋外匯報酬能力較強。透過OLS迴歸及Fama-MacBeth兩步驟迴歸與定價誤差檢定，發現利用Corwin and Schultz(2012) 所建構之因子模型通過卡方聯合定價模型測試，其模型不存在定價誤差，故Corwin and Schultz (2012)所建構之因子模型較Bid-Ask Spread所建構之四因子模型更為適切。另一方面利用HML投組法將37國匯率資料透過貨幣高低價倍數(CS流動性指標)、貨幣前期超額報酬、貨幣五年累積實質匯率變化量(FXV值)之測度進行分類，分別建構出流動性、動能及價值交易策略因子。透過OLS迴歸及Fama-MacBeth兩步驟迴歸與定價誤差檢定，分別檢視二因子(市場因子及流動性因子)、三因子(市場、動能及價值因子)及四因子(市場因子、流動性、動能及價值因子)模型之外匯報酬定價能力。根據實證分析結果，三因子模型在變數解釋能力、調整後判定係數以及定價誤差為零之聯合卡方檢定皆表現的較二因子及四因子模型適切。

關鍵字

外匯交易；流動性交易策略；動能交易策略；價值交易策略； Fama-MacBeth兩步驟迴歸

並列摘要

This paper is mainly discussed whether the liquidity factor can explain the excess returns of foreign exchange, and compare the two liquidity factors constructed by Bid-Ask Spread and Corwin and Schultz (2012), which factors can explain the excess returns of foreign exchange properly. By using the OLS regression, Fama-MacBeth two-step regression and the test of pricing error, we found that only the factor model constructed by Corwin and Schultz (2012) passed the test of pricing error, which means that there was no pricing error in factor model constructed by Corwin and Schultz (2012) and is more suitable than the factor model constructed by Bid-Ask Spread. Further, we used the exchange rate data of 37 countries to construct the LIQ, MS, VALUE factors via HML (high minus low) methodology. Using the FX of high and low price multiples (CS index), last month of the FX excess return, and the FX five-year cumulative real exchange rate change (FXV) as the measurements. By the OLS regression and Fama-MacBeth two-step regression and test of pricing error, we respectively examined the pricing ability of two-factor model (market and liquidity factor), three-factor model (market, momentum and value factor) and four-factor model (market, liquidity, momentum and value factor). According to the empirical analysis, the three-factor model is the best model to explain the FX excess return, comparing with the two-factor and four-factor models. No matter the adjusted coefficient of determination, and the test of pricing error, three-factor model perform well indeed.