Forecasting · 2025–2026
Forecasting the Effective Federal Funds Rate
Can Ridge or Random Forest beat a repeat-last-month forecast on an untouched future test period?
Beat a baseline a forecaster could actually use
Repeating the previous month’s rate was the benchmark because it is simple, transparent, and available at prediction time. Any model improvement had to survive a later, untouched test period.
Discard the leaked result and rebuild the experiment
- Compared Ridge and Random Forest with the repeat-last-month baseline.
- Found that the first pipeline allowed holdout information into training.
- Discarded those results and rebuilt the pipeline with time-safe preprocessing.
- Locked the design before evaluating 114 later months.
Ridge reduced RMSE by 16%
Ridge reached 0.1606 RMSE versus 0.1922 for the baseline. Random Forest was worse at 0.2368. The improvement was meaningful on RMSE, while the gain on MAE was much smaller.
114holdout months
0.1606Ridge RMSE
16%lower than baseline
A better forecast—and a more trustworthy research process
Ridge beat the baseline after I threw out the flawed first result and rebuilt the pipeline. The reported gain comes from the locked, leakage-free test—not the earlier experiment.