The Elusive Quest for Preserved Quantities in Financial Time Series: Making a Case for Intraday Trading Strategies

Abstract

In the context of the supervised learning problem for time series forecasting, we focus on financial time series and use the currency pair EURUSD to highlight issues that arise when daily data are utilized for one-day forecasts of currency exchange rate moves. In light of our results for forecast horizons of one day or more, we take a closer look at the EURUSD time series data to get a better understanding of typical intraday moves and their magnitude and how their potential can be harnessed for the development of consistently profitable trading strategies. By combining the results of our own numerical studies with published findings from the literature and illuminating them from a practical perspective, we motivate a simple intraday trading strategy for EURUSD that avoids some of the problems associated with longer-term forecasts.

Keywords: FX, Foreign Exchange, EURUSD, Machine Learning, Intraday Seasonality