Rolling Window
A rolling window evaluates performance over a moving time period (for example the last 30 days). It helps track improvement and detect drift over time.
Definition
A rolling window is a time based evaluation method where you compute metrics (like Brier score) over the most recent fixed period, then roll forward over time.
Examples
• Last 30 days Brier score
• Last 90 days Brier skill score
Why it matters
Rolling windows show whether your skill is stable, improving, or deteriorating. They also help detect changes in calibration when the environment shifts.
Common pitfalls
Small samples: short windows can be noisy. Pair with longer windows and report N.
Changing mix: if the types of questions change, rolling metrics may move even if skill is unchanged.
Related
Rolling windows connect to forecast horizon, out of sample evaluation, and benchmarks.