September PCE, translated from CPI and PPI.
The Fed's preferred gauge. By the time PCE prints, CPI and PPI for the same month are already out, and BEA prices most of PCE directly from them. So the call rebuilds core PCE the way BEA does: 37 spending categories, each priced from its CPI or PPI source and weighted by what Americans actually spend.
What August 2026 told us
The last release against what was expected, what drove it, and what it changes.
The print
| First print | Now | Consensus | Model | Key Data call | Surprise | |
|---|---|---|---|---|---|---|
| Core PCE | +0.2% | +0.25% | +0.3% | +0.22% | +0.30% | −0.1 pt |
| Headline PCE | +0.31% | +0.31% | – | +0.27% | – | – |
What it showed
- August 2026 core PCE prices rose +0.25% (headline +0.31%); consensus had core at +0.3%.
- Core PCE is 3.0% higher than a year earlier, versus 2.4% for core CPI. Unusually, PCE is running above CPI: the categories PCE weights more heavily (health care, financial services, priced partly from PPI and asset values) are rising faster than shelter, which dominates CPI.
- With September 2026 CPI and PPI already published, the bridge puts core PCE at +0.29%: core CPI's +0.32% less 0.03 pt for PCE's different weights and its PPI-priced health care, airfares and portfolio management (Cleveland Fed: +0.28%).
What fed the bridge
Same-month changes in the CPI and PPI pieces the bridge uses, percent.
Analyst notes
The Fed's gauge is running hotter than CPI. Core PCE rose 0.25% in August and is up 3.0% from a year ago, against 2.4% for core CPI. That reversal of the usual gap comes from what PCE weights more heavily: health care and financial services, where prices are taken partly from producer prices and asset values, have been rising faster than shelter.
September looks firmer. With September CPI and PPI already published, the bridge points to core PCE of about 0.27% (the Cleveland Fed nowcast is 0.28%). PPI health care and portfolio fees, lifted by higher stock prices, are the swing items.
Why it matters. A 3% core PCE rate with a policy rate of 3.75–4.00% leaves real policy only about one point restrictive, which limits the Fed's room to cut.
Where September 2026 should land
Odds of each outcome, sized by the model's own recent misses with fat tails for surprise months.
Core PCE model +0.29%
Headline PCE model +0.48%
What the model expects, piece by piece
Expected contribution to the headline, basis points.
Track record
Every month the model is refit using only data available at the time, then scored against the number first published, the same yardstick used for the consensus.
| Period | Releases | Model miss | Key Data call miss | Consensus miss | Cleveland Fed miss | Exact rounded print: model / call / consensus | Model closer / further | |
|---|---|---|---|---|---|---|---|---|
| Since 2023 | 43 | 0.059 | 0.037 | 0.037 | 0.083 | 53% / 65% / 65% | 13 / 30 | Even |
| Since 2019 | 75 | 0.060 | 0.044 | 0.044 | 0.081 | 44% / 59% / 59% | 27 / 48 | Even |
| Since 2014 | 128 | 0.055 | 0.041 | 0.041 | 0.069 | 48% / 61% / 61% | 42 / 86 | Even |
How it works
Bottom up: each of 37 PCE categories, covering 89% of core spending, is forecast from the CPI or PPI series BEA uses for it (owners' and tenants' rent, doctors, hospitals, nursing homes, airfares, portfolio management, restaurants, vehicles, recreation and more), then weighted by last month's spending shares. The rest, mostly imputed bank services and nonprofits, follows its own trend.
A second view forecasts the gap between core PCE and core CPI as first published. The call averages the two. The backtest feeds each month's CPI inputs as first published, from BLS's release archive and, going forward, a record the site keeps of every first print, so it reflects what the model would have seen at the time.
Professional forecasters do this translation with BEA's full detail and first-release PPI data, so the consensus is very accurate here; the Key Data call follows it unless the model has earned a record of beating it.
The Key Data call: once Wall Street's consensus is published, the model's view is combined with it. The combination starts from the consensus and moves toward the model by an amount learned from past releases, also weighing last month's surprise and revision. It is used only while it has beaten the consensus over the previous three years of releases, scored out of sample; otherwise the call is the consensus.