Prasanna K · Economic research
What was known before the release?
Inspect publication-aware CPI forecasts, later revisions and an unchanged final test. This is a retrospective old-base study for July 2024–December 2025.
The complete held-out period
Solid navy: selected target; dashed green: displayed forecast. All 18 dates remain in the comparison.
Read all 18 monthly values
The table shows the same dates, forecasts and targets as the chart. The selected month is marked.
| Month | Forecast (%) | Target (%) | Error (pp) |
|---|
Evidence available at the cutoff
- Forecast input vintage
- Published at
- Source
- Open official release
Keep the original decision
Development selected persistence. On the primary first-provisional final targets, its RMSE is 0.8185 percentage points, versus 1.6901 for the fixed AR(1). The paired squared-error contrast, AR(1) minus persistence, is 2.1865 pp²; conditional 95% three-month block interval [0.6137, 3.8166].
The interval describes average retrospective loss in a short dependent panel. It is not a probability range for an inflation print. Display controls do not change the selected model or refit coefficients.
Four unavailable lag-final revisions slightly improve apparent AR(1) MSE. Four later-final target revisions separately change the estimand. Both diagnostics preserve the first-release primary result.