Every month I watch the same thing play out. The Bureau of Labor Statistics (BLS) releases the Consumer Price Index (CPI) and a couple of weeks later the Bureau of Economic Analysis (BEA) releases the Personal Consumption Expenditures (PCE) price index. The numbers don’t match. Half of financial media grabs the higher one and the other half grabs the lower one. Readers are left wondering which number is real.
They both are. CPI and PCE measure overlapping slices of consumer inflation but they’re built with different coverage and different math. A gap between them doesn’t mean either one is broken. It means they were designed to answer different questions.
This matters for your household budget. It matters even more if you’re trying to read the Federal Reserve since the Fed states its inflation goal in PCE terms and not CPI.
Start with what each index is trying to measure
CPI tracks the prices urban households pay out of pocket. PCE sits inside BEA’s national accounts and it casts a much wider net. BEA says PCE measures spending by and on behalf of the personal sector which includes households and the nonprofits that serve them.
The words “on behalf of” are where a lot of the gap starts. Healthcare is the cleanest example. When your employer covers most of your health insurance premium or Medicare pays a hospital bill that spending shows up in PCE. CPI only picks up the part you pay yourself. PCE also counts rural households while CPI-U only covers urban consumers.
So PCE isn’t a picture of any one family’s bank statement. It’s an accounting of consumer spending across the whole economy. CPI isn’t a personal receipt either. It’s an average. Neither one replaces looking at where your own money goes.
Weights can move the number all by themselves
Here’s a simple way to see it. Picture two indexes that only hold two categories. One is housing and the other is everything else. Housing prices rise 5% and everything else rises 2%. Index A gives housing a 40% weight and Index B gives it a 20% weight.
Index A rises 3.2% and Index B rises 2.6%. The underlying price changes are identical. The entire 0.6 percentage point gap comes from the weights.
I made those weights up to keep the math clean. The real ones aren’t that different in spirit though. When BEA and BLS economists reconciled the two indexes in 2007 rent of shelter carried about a 32% weight in CPI compared to about 15% in PCE. When shelter inflation runs hot CPI feels it roughly twice as hard.
BEA splits the real gap between the two indexes into four buckets. There’s a formula effect and a weight effect. There’s a scope effect for things one index counts and the other doesn’t. The last bucket is “other effects” which covers items like seasonal adjustment and price differences. That framework turns the gap into something you can measure instead of something people argue about on X.
The numbers from that work are useful. From 2002 through 2007 CPI ran about 0.4 percentage points per year above PCE and almost half of that came from the formula difference. After adjusting for formula the difference in rent of shelter weights more than explained the rest while scope differences partly offset it.
That’s a long-run average though. It can’t tell you why the gap looks the way it does in any single month. For that you have to look at which categories drove the move in the period you’re talking about.
The math behind each index is different
PCE uses a Fisher-Ideal chain-type formula. CPI-U uses a modified Laspeyres formula. BEA identifies this as the formula effect in its reconciliation. The Fisher formula keeps up with what people actually buy as their spending shifts.
CPI isn’t frozen in time either. BLS started updating CPI weights every year beginning with January 2023 data after years of updating them every two years. Within categories CPI has used a geometric mean formula for most items since 1999 which allows for some substitution. Across categories the weights stay fixed for 12 months. BLS says directly that the CPI-U formula assumes consumers don’t substitute across item categories. It publishes a Chained CPI (C-CPI-U) that does.
Here’s what that looks like in real life. Say restaurant prices jump and you start cooking at home more. Your spending mix changed. An index that follows actual spending patterns will show less inflation than one that holds last year’s mix steady.
You’re still worse off. Cooking at home instead of going out is a downgrade even if it costs less. That’s the difference between what a statistic measures and how your wallet feels.
I’d stay away from the lazy take that CPI ignores substitution completely and PCE captures every choice a household makes. Neither is true. The official methods are more detailed than that and a two-category example can’t reproduce the published math.
Headline and core are a separate choice
Both indexes come in headline and core versions. Core strips out food and energy so you can see the underlying trend. It doesn’t take food and energy out of your life. BEA publishes both total PCE and PCE excluding food and energy.
This is where a lot of bad comparisons come from. If someone puts headline CPI next to core PCE they’re mixing two different index designs and two different baskets. Blaming the whole gap on CPI versus PCE is just wrong.
Time periods trip people up too. A monthly change isn’t a 12-month change. An annualized monthly rate is a third thing entirely. It tells you what a full year would look like if that one month repeated 12 times. It doesn’t tell you what already happened.
Before I call two numbers a disagreement I match the time period and the category coverage. I check whether both are seasonally adjusted and I check the release date. Only then does the leftover gap actually tell me something.
The Fed’s target is PCE
The Federal Open Market Committee (FOMC) defines its longer-run goal as 2% inflation measured by the annual change in the PCE price index. The FOMC revised its strategy statement in August 2025 and kept 2% as its definition of price stability. CPI still matters. A 2% CPI print just isn’t the same thing as the Fed hitting its target.
The Fed doesn’t react mechanically to one PCE release either. Policy depends on the whole economic picture and on how long price increases look likely to stick. The target tells you which benchmark the Fed measures itself against. It isn’t a formula that decides every meeting.
If you write about markets label your inflation number every single time. “Inflation hit 3%” doesn’t tell me enough to judge the claim. Say whether it’s CPI or PCE. Say whether it’s headline or core. Say what period it covers.
A lower rate doesn’t give you your money back
Say one index climbs from 100 to 120 over several years and another climbs from 100 to 118. Both show a big jump in the price level. If inflation cools after that neither index heads back toward 100. Prices just rise more slowly from the new higher level.
That’s why I push back when a soft PCE print gets framed as proof that consumers should feel fine. The inflation rate tells you the pace of price changes. Whether your purchasing power has recovered depends on how much your income grew compared to the cumulative price increase since your starting point. Even with the right index you still need the right comparison dates.
Keep track of which release you used
PCE gets revised. BEA updates past PCE data as better source data comes in and it publishes studies on how much its estimates move between the first print and the latest one. CPI works differently. The unadjusted CPI-U is not subject to revision once it’s published and only the seasonal factors get updated.
That matters if you track forecasts. A prediction should be scored against the data that was available on the release date and you should say which vintage you’re using.
My routine is simple. I identify the measure. I match the time frame. I look at the biggest components driving the move. Then I ask what the number actually says about the question I’m trying to answer. Use PCE when you’re talking about the Fed’s target. Pick one clear benchmark and stick with it when you’re talking about purchasing power. Neither number gets more useful by asking it a question it wasn’t built to answer.