Comparing EURUSD Forecast Errors Across Multiple Time Frames

Summary

The article tests whether forecasting performance differs across 11 EURUSD time frames, using equal-sized samples of historical and future closing prices and a 20-step forecast horizon. It reports that models had lower errors on monthly and hourly data than on the other tested frames, a result the author considers counterintuitive. Exploratory analysis includes cross-time-frame correlations, return distributions, Granger causality tests, and time-series warping to examine possible relationships between hourly and monthly data.

The proposed trading system forecasts anticipated monthly price levels with a tuned model exported to ONNX, then executes on an hourly chart using technical entry timing and either predicted reversals or moving averages for exits. The article includes backtest and walk-forward examples, but its conclusion is limited to EURUSD and the chosen sample, model, and frames. It acknowledges that results may differ across markets and that broader frame searches or combinations could change the findings; low forecast error alone does not establish durable trading profitability.

Key ideas

• The study compares forecasts over 11 EURUSD time frames using 400 observations per frame and a 20-step horizon.
• The reported lowest model errors occur on monthly and hourly data.
• Return-based analysis includes Granger causality testing and time-series alignment between hourly and monthly observations.
• The example system forecasts on the monthly frame and executes trades hourly with technical entry timing.
• Findings are specific to the tested EURUSD data and modeling choices and may not generalize to other markets.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.