Forecasting Emerging Market FX Spot Rates: An AR(1) approach

Abstract

This paper outlines a method to forecast FX spot rates. The data set consists of the Bloomberg FX spot rates for emerging markets as defined by Bloomberg. The in-sample data set consisted of weekly FX spot rates for ten Emerging markets, from August 2013 to March 2019. The out sample spanned March to November 2019. PACF tests revealed that the most appropriate model would be an AR(1). After applying the AR(1) model to the data a combination of AIC and Log-Likelihood criteria as well as a sigma squared measure were applied to determine the spot rates with the best fit. 3 spot rates remained that had the best fit relative to the other spot rates given that the sample sizes were identical. Applying the relevant AR(1) models to the out-sample data highlighted that using a long-only approach, to avoid short side risk, produced negative returns in all three FX spot rates. The use of an out-sample to test the applicability of the AR(1) forecast supplements the within model criteria: AIC, Log-likelihood and sigma squared. The out-sample results highlight that in practice an AR(1) model may not necessarily produce positive returns in Emerging FX markets.