Inflation Monitor
Data through 17 Oct 2026 (Truflation) and September 2026 (CPI)
Data snapshot: 18 October 2026
Truflation, latest
3.08%
17 Oct 2026 | +0.72 pp vs 30 days ago
Official CPI, latest print
3.7%
September 2026 y/y (NSA) | SA m/m +0.6%
Model estimate: October 2026 CPI
4.0% implied y/y
Empirical 80% band: 3.8-4.2% | Conditional SA m/m +0.39%
Residual scenarios above last CPI
92%
Illustrative frequency, not calibrated probability | vs 2026-09 3.74% | model: Truflation + gasoline (OLS)
Custom range
Overlays
Loading charts...
Truflation vs official CPI, year-on-year % (same scale)
CPI is plotted by reference month, not its later publication date; Truflation is daily. Drag the slider to zoom. Grey bands: NBER recessions.
Gap: Truflation minus CPI (pp)
Shaded: divergence episodes (|gap| >= 1pp for 3+ months). The housing concepts, expenditure weights and coverage differ between the two indexes.
Do they move together? Rolling correlation
Fixed headline NSA comparison, using 24 available pairs. Missing calendar months remain missing.
October 2026: model-implied YoY with empirical bands
Cutoff 2026-10-18; model uses Truflation through 2026-10-16. Bands come from each model's own last 36 available test errors. Bias can shift a band away from its point. The highlighted model has the lowest live-era RMSE in the comparison sample.
What drives the preferred nowcast
Starts from the last official print and adds the estimated effect of the change in Truflation (and gasoline) since that month.
Historical accuracy, 2015-2026 (matching forecast day)
Expanding-window estimates using release-date and publication-lag filters. See Data & method for timing assumptions.
Delayed CPI releases: error by length of delay
Simulated shutdown: the last usable CPI print is 1-6 months old. Anchoring on that print and adding the Truflation change halves the error of carrying the stale print forward.
Nowcast vs outcome, 2019 onward
Each dot simulates a forecast using only CPI outcomes published before its cutoff.
Forecast error by day of the month
Early in the month the previous CPI print is often not out yet - that is where Truflation adds most.
Error metrics (pp of the 12-month rate)
DM = Diebold-Mariano test vs random walk (HLN-corrected); direction hit = sign of the change when the actual move is >= 0.05pp.
Why the levels differ: gap vs headline and vs CPI ex-shelter
Headline and ex-shelter comparisons describe how housing coverage relates to the gap. Weights, energy and timing also differ; this is a descriptive comparison.
CPI components vs Truflation
US regular gasoline, USD/gal (EIA weekly)
Brent crude, USD/bbl (EIA daily)
Monthly panel (table view of every chart series)
Sortable, filterable, exportable to CSV.
Method notes
- Truflation reports YoY rates. Its monthly value is the arithmetic mean of daily rates; the CPI basket and index formula differ.
- The headline forecast targets NSA CPI YoY. SA headline, core and ex-shelter series are available in Comparison. CPI observations enter the fit after their listed release date.
- The historical test uses current data vintages and conservative lags: Truflation date + 2 days, gasoline date + 3 days. The supplied release calendar has not been verified row by row. Truflation history before its December 2021 launch is back-cast.
- Model selection and evaluation use the same 2022+ sample. The nominal 80% bands are past-error quantiles; empirical coverage is reported in the model table. Residual scenario frequencies and exploratory DM tests are not calibrated trading probabilities.
- October 2026 YoY is a model-implied proxy because the October 2025 CPI base is missing. The conditional SA m/m translation uses a geometric interpolation of that base and the latest SA-minus-NSA gap. Its coverage has not been independently backtested.