Where inflation goes over the next six months.
A transparent statistical projection of the CPI index, extended six months past the last official print, with 80% and 95% confidence ranges. Pick a series, pick a method, and download the whole thing for your own model.
Backtests every method against past releases, weights them by realised accuracy, removes their historical bias and sizes the bands from actual past errors.
BLS CPI-U · index level with 6-month projection
Actual through Sep '26 · shaded 80% and 95% confidence ranges · Learned ensemble
Projected year-over-year inflation
Implied YoY % for each forecast month, with the 80% range
Forecast table
Six months ahead · index level and confidence ranges · BLS CPI-U
| Month | Index | 80% range | 95% range | YoY |
|---|---|---|---|---|
| Oct '26 | 337.9 | 336.3 – 339.4 | 335.5 – 340.3 | +4.02% |
| Nov '26 | 338.8 | 336.1 – 341.5 | 334.7 – 342.9 | +4.53% |
| Dec '26 | 339.7 | 336.2 – 343.2 | 334.4 – 345.1 | +4.81% |
| Jan '27 | 340.5 | 336.4 – 344.7 | 334.2 – 346.9 | +4.68% |
| Feb '27 | 341.3 | 336.9 – 345.8 | 334.6 – 348.1 | +4.43% |
| Mar '27 | 342.1 | 337.4 – 346.8 | 335.0 – 349.4 | +3.60% |
AI outlook for the next six months
An AI analyst reads the same evidence the statistical model uses — every recent print, seasonal patterns by calendar month, category momentum in energy, housing and food, market and survey expectations — plus a scorecard of how each method has actually performed against past releases, then produces a judgement-adjusted path with 80% ranges.
Scored against 49 past releases · best simple method Damped trend at ±0.56% average error
How each method performed against past releases
Rolling-origin backtest over 49 past releases · lower error earns a bigger weight in the learned ensemble
| Method | Avg error | Bias | Weight |
|---|---|---|---|
| 12-month drift | ±0.65% | +0.19% | 27% |
| 24-month drift | ±0.72% | +0.27% | 22% |
| Damped trend | ±0.56% | +0.05% | 36% |
| 6-month momentum | ±0.84% | +0.10% | 16% |
How this forecast is built
The index is projected forward in log space. The learned ensemble replays past releases, scores each simple method on the errors it actually made, weights them by accuracy and strips out their historical bias.
In learned mode the bands are the spread of the ensemble's own past errors at each horizon (0.36% one month out); other methods use recent monthly volatility scaled by the square root of months ahead. 80% and 95% coverage.
These are statistical extrapolations, not official projections or advice. Shocks to energy, housing or policy can move prints well outside the ranges shown.