Forecasting Bid-Ask Spreads in Foreign Exchange: Analysis and Machine Learning Prediction

Abstract

The foreign exchange (FX) markets are some of the most liquid in the world, yet this liquidity is not spread evenly across currencies or market conditions. Factors like the time of day or week factor heavily in the transaction cost of exchanging currencies. The worst times to trade can be 10 or 100 times more expensive than the best, with liquidity concentrated with some predictably over time.

Forecasting the likely cost of trading currencies (measured by bid-ask spreads) is thus critical for effectively trading in these markets. This work analyses these transaction costs and the use of machine learning to accurately forecast them.

We show that by using supervised machine learning algorithms with appropriately chosen features, bid-ask spreads can be forecast reliably in the FX market.

Keywords: Transaction cost, bid-ask spread, forecasting, time series, machine learning, prediction, FX, forex, foreign exchange

JEL Classification: C01, C53, C58, G12