Forecast Drift
Forecast drift is a gradual change in forecast behavior or calibration over time, often due to changing environments or shifting question mixes.
Definition
Forecast drift is a slow change in your forecasting pattern over time. Drift can show up as a change in average confidence, a shift in calibration, or a change in performance on certain categories.
Common causes
• Changing base rates (base rate shift)
• A different mix of question difficulty
• New information sources or new habits
• Market regime changes in the underlying domain
Why it matters
Drift can quietly degrade performance. Rolling evaluation, such as a rolling window, helps detect it early.
Related
Drift is linked to regime change and to the stability of benchmarks across time.