Testing FRED Dollar Data for EURUSD Forecasting

Summary

The article tests whether Federal Reserve economic data can improve forecasts of EURUSD. It compares ordinary daily market quotes with three FRED series: the broad dollar index, US bond interest rates, and expected inflation. After expressing the data as annual percentage changes, the author reports a correlation near -0.9 between EURUSD and the dollar index, and a correlation near -0.7 between the current index and EURUSD’s value 20 days later. These relationships motivate using the macro data as model inputs.

Three deep neural network regressors are compared using market data, FRED data, and their combination. Time-series cross-validation preserves chronological order, and parameter tuning improves validation performance over the default model. However, market quotes alone produce the lowest prediction error; feature selection also favors none of the FRED series, and residuals appear poorly behaved. The article concludes that these features did not provide a demonstrated forecasting edge in this experiment. Results are limited to the selected data, transformations, models, and test setup; the author suggests trying other transformations or models.

Key ideas

• Annual percentage changes reveal a strong inverse relationship between the broad dollar index and EURUSD in the analyzed data.
• The current dollar index is reported to correlate negatively with EURUSD’s value 20 days later.
• The study compares market-only, FRED-only, and combined inputs using neural network regressors.
• Chronological cross-validation is used to preserve time order during model evaluation.
• Despite promising correlations, adding FRED data worsens prediction results relative to ordinary market quotes.

Tags

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