Information Spillovers and Predictable Currency Returns: An Analysis via Machine Learning
43 Pages Posted: 3 Feb 2019 Last revised: 26 Oct 2020
Date Written: January 4, 2019
This paper employs the post — Least Absolute Shrinkage and Selection Operator (post — LASSO) to make rolling 1-month--ahead currency excess return forecasts using all other currencies' lagged forward discounts as candidate predictors. The trading strategy of buying (selling) quintile currency portfolios of the high (low) post — LASSO forecasts yields a monthly excess return of 1.05% for the 48-currency sample. The results do not change even after controlling for various predictors. The return predictive power of the post — LASSO comes from two sources. First, it identifies the origin currencies of information spillovers in the FX market, which are sparse and time-varying. Second, it incorporates cross-sectional variations in currencies' predictive relations with the origin currencies.
Keywords: Foreign Exchange, Currency Return, Post — LASSO, Information Spillovers
JEL Classification: G12, G14, G15, F31, F37
Suggested Citation: Suggested Citation