Home › Guides › CPI vs Core CPI: What's the Difference?
CPI is the most common U.S. inflation measure. 'Core' CPI drops food and energy so price spikes don't muddy the signal.
CPI stands for the Consumer Price Index, and it's the most widely-cited measure of U.S. inflation. The Bureau of Labor Statistics, an agency inside the Department of Labor, tracks the prices of about 80,000 different goods and services each month - everything from a gallon of gasoline to a haircut to a year of college tuition - and asks the question: how much more (or less) does it cost a typical household to buy the same basket of stuff this month compared to last month, or compared to a year ago? That percent change is the inflation rate.
CPI is the most common U.S. inflation measure. 'Core' CPI drops food and energy so price spikes don't muddy the signal.
CPI stands for the Consumer Price Index, and it's the most widely-cited measure of U.S. inflation. The Bureau of Labor Statistics, an agency inside the Department of Labor, tracks the prices of about 80,000 different goods and services each month - everything from a gallon of gasoline to a haircut to a year of college tuition - and asks the question: how much more (or less) does it cost a typical household to buy the same basket of stuff this month compared to last month, or compared to a year ago? That percent change is the inflation rate.
There are two flavors of CPI you'll see in the news: "headline CPI" and "core CPI." Headline CPI includes everything in the basket, including food and energy. Core CPI strips food and energy out. Why? Because food and energy prices swing wildly month to month for reasons that have little to do with the underlying economy - a hurricane in the Gulf can spike gasoline prices for a few weeks, a frost in Florida can spike orange juice prices, a war in the Middle East can spike crude oil. Stripping those volatile components out gives a cleaner read on what's happening underneath.
Both measures matter, just for different purposes. Headline CPI is what you actually feel as a consumer - if your gas bill is up 30% and your grocery bill is up 8%, that's what's emptying your wallet. Core CPI is what economists and the Federal Reserve focus on, because it tells you whether the trend in prices is accelerating or decelerating in a sustained way.
A real-world example: in summer 2022, headline CPI peaked at 9.1% year-over-year - the highest in forty years. But that headline was inflated by gasoline at $5/gallon and food prices spiking from supply-chain disruption. Core CPI peaked lower (around 6.6%) and lagged a few months behind. As gasoline came back down through 2023, headline fell faster than core. By 2024 both were back in the 3% range, with core still higher than headline because shelter costs (a big chunk of core) were slow to come down.
The BLS publishes CPI on a pre-announced monthly schedule, generally in the second or third week of the following month at 8:30 AM Eastern (so the April reading comes out in mid-May, the June reading in mid-July, etc.). The release moves markets meaningfully - a CPI surprise of 0.2 percentage points can shift the 10-year Treasury yield by 10–20 basis points in seconds.
BLS CPI: basket of urban consumer prices, Laspeyres formula, weights updated annually since 2023.
The CPI is calculated by the Bureau of Labor Statistics (BLS) using a modified Laspeyres formula. The basic idea: define a fixed market basket of goods and services that represents what a typical urban consumer buys, then track the cost of that basket over time. Total cost this month divided by total cost in a reference period, expressed as an index. Inflation is the percent change in the index.
BLS staff collect about 80,000 individual price observations each month from approximately 23,000 retail outlets across 75 urban areas, plus a rental-housing panel of roughly 50,000 units surveyed on a six-month rotation (several thousand units priced in any given month). Some prices are collected in person; some by phone; some scanned automatically from web sources. The collected prices are aggregated into about 200 elementary-level price categories ("item strata"), then weighted together using expenditure weights derived from the Consumer Expenditure Survey (CEX, also run by BLS).
The CEX-derived weights tell us what fraction of a typical consumer's spending goes to each category. Shelter (rent and owners' equivalent rent) is about 33% of the headline basket. Food is about 13%. Energy is about 7%. Transportation is about 15%. Medical care is about 8%. Apparel is about 2.5%. The weights were revised every two years from 2002 through 2022; since the January 2023 index, BLS updates them every January using a single calendar year of CEX data, with a two-year lag - the 2024 weights, for example, were based on 2022 consumer spending.
Headline CPI is published as the FRED series CPIAUCSL (seasonally adjusted) or CPIAUCNS (not seasonally adjusted). Core CPI is published as CPILFESL - the "core" in the FRED ticker means "less food and energy." The seasonally adjusted series corrects for predictable seasonal patterns (apparel prices drop in end-of-season clearance months, gasoline prices rise into the spring driving season, etc.) and is the version typically cited in news coverage. The non-seasonally-adjusted version is used for contracts (Social Security COLA, TIPS principal indexation) and for cross-period comparisons that need raw data.
Shelter is the single most important category and the most-debated. Within shelter, the largest piece is Owners' Equivalent Rent (OER), an imputed estimate of what a homeowner would pay if they rented their own home. OER is derived from the rents observed in the rental market, applied to owner-occupied housing units using a complex matching procedure. The crucial caveat: OER lags actual market rents by approximately 12–18 months, because the BLS sample includes both new leases and existing leases (which only reset annually for most renters). When market rents spike, OER takes a year or more to fully reflect the move; the same is true when market rents soften. This lag is the single biggest source of CPI inertia and a frequent commentary topic during turning points.
Alternative inflation measures sit alongside CPI. Personal Consumption Expenditures (PCE) inflation is the Fed's preferred target measure - see the PCE vs Core PCE explainer. The Trimmed Mean PCE, Sticky CPI, and Median CPI are alternative measures designed to filter out short-term noise from the cross-section of component prices - see the Inflation Measures Overview. The market-implied inflation rates derived from TIPS - see Breakeven Inflation. Each answers a slightly different question; CPI is the most directly-felt and most widely-cited.
What moves CPI from month to month: short-term, energy prices (gasoline, natural gas, electricity) and food prices (groceries) cause the largest visible swings. Medium-term, shelter costs and core services drive the trend. Long-term, wage growth, productivity, monetary policy, and inflation expectations are the structural drivers. The decomposition Fed staff and dealer research teams produce after each release breaks the print into goods (sensitive to supply shocks, tariffs, FX), services excluding shelter ("supercore" - wage-driven), and shelter (lagged housing market).
CPIAUCSL (headline) / CPILFESL (core); CPI-U vs CPI-W vs C-CPI-U; OER drives shelter inertia.
CPI families: CPI-U (Urban Consumers - the headline number, covering ~93% of population) is what's typically cited as "CPI." CPI-W (Urban Wage Earners and Clerical Workers - covers ~30% of population, narrower sample) is used for Social Security COLA, federal civil-service retirement, and military retirement adjustments. C-CPI-U (Chained CPI for Urban Consumers) uses a superlative Törnqvist chained formula that captures substitution across categories the way PCE does (initial monthly values are preliminary and are finalized about a year later); typically runs 0.2–0.4 percentage points below CPI-U annually.
Underlying-inflation alternative measures: (1) Core Sticky-Price CPI (Atlanta Fed, FRED ticker CORESTICKM159SFRBATL) - only includes components whose prices change less frequently than every 4.3 months, capturing the slow-moving structural trend; (2) Median CPI (Cleveland Fed, MEDCPIM158SFRBCLE) - the 50th-percentile component change each month, robust to outliers; (3) 16% Trimmed-Mean CPI (Cleveland Fed, TRMMEANCPIM158SFRBCLE) - drops the top 8% and bottom 8% of components weighted by expenditure share; (4) Trimmed Mean PCE (Dallas Fed, PCETRIM12M656N) - analogous to trimmed CPI but on PCE data.
Core PCE vs Core CPI: historically Core PCE runs ~0.3–0.5 pp below Core CPI on a year-over-year basis. Three drivers of the gap: (1) different formulas - PCE is Fisher chain-weighted, CPI is modified Laspeyres; (2) different basket weights - PCE weights based on quarterly NIPA data, CPI weights based on annual CEX data with a lag; (3) different scope - PCE includes employer-paid medical insurance and certain financial services that CPI excludes; CPI's OER methodology and PCE's housing methodology differ in detail.
Shelter mechanics: shelter category includes Rent of Primary Residence, Owners' Equivalent Rent (the largest component, applied to owner-occupied housing), Lodging Away From Home (hotels), and Tenants' and Household Insurance. Rent of Primary Residence is computed from actual rent observations on the rental sample. OER is computed by matching owner-occupied housing units to comparable rental units and applying the rental rate change to the owner unit. The OER methodology was the subject of substantial debate during the 2022–24 disinflation as the OER lag held headline shelter CPI elevated long after market-rent data (Zillow ZORI, Apartment List, CoreLogic SFRI) showed rent growth deceleration.
The lag dynamics: BLS rental sample is panel-based. Each unit in the sample is observed every 6 months. When a unit's rent changes (whether at lease renewal or in-tenancy), the change shows up in BLS data 0–6 months later. The full pass-through from a market-rent regime change to OER takes 12–18 months. New-lease rent indices (like the BLS New-Tenant Rent Index, NTR, published since 2022) lead OER by 6–12 months and are the closest leading indicator of where OER will go.
Geometric-mean revision (1999): BLS adopted geometric-mean aggregation within most item strata (rather than arithmetic mean) starting in 1999. This addressed one component of the substitution bias identified by the Boskin Commission. The change reduced measured CPI inflation by approximately 0.2 pp annually.
Hedonic adjustments: for certain product categories where quality changes rapidly (computers, televisions, smartphones, some apparel), BLS uses hedonic regression; new vehicles instead get manufacturer cost-based quality adjustments to separate price change from quality change. The hedonic adjustment can produce CPI components that decline even when nominal prices rise, because the per-quality-unit price is falling. Hedonic methodology is documented in BLS's Handbook of Methods, Chapter 17.
Seasonal adjustment: the BLS X-13-ARIMA-SEATS procedure produces the seasonally adjusted series (CPIAUCSL etc.). Annual seasonal-factor updates happen each February. Revisions to seasonal factors can produce revisions to historical seasonally adjusted CPI, though the non-seasonally adjusted series doesn't get revised.
For research, the BLS CPI Handbook of Methods (Chapter 17, https://www.bls.gov/opub/hom/cpi/) is the comprehensive primary source. Federal Reserve staff inflation analyses (Federal Reserve Bulletin, Federal Reserve Staff Working Papers) are useful for policy-relevant decompositions. The Cleveland Fed's CPI page provides the Trimmed Mean and Median series along with methodology.
Laspeyres vs Fisher; CEX reweighting (annual since 2023); Boskin Commission ~1.1 pp upward bias estimate (1996).
The Boskin Commission (Advisory Commission to Study the Consumer Price Index, 1995–96) was a Congressionally-appointed group chaired by Michael Boskin that examined the accuracy of CPI as an inflation measure. The Commission's final report (December 1996) concluded that CPI overstated true inflation by approximately 1.1 percentage points annually, citing four bias sources: substitution bias (consumers shift to relatively cheaper goods within and across categories), outlet bias (consumers shift to cheaper retailers like Walmart), new-product bias (CPI sample lags behind market adoption of new product categories), and quality-change bias (CPI inadequately captures quality improvements in existing products).
BLS has implemented partial fixes for several of these biases since 1996. Geometric-mean aggregation within item strata (1999) addressed substitution bias within categories. More frequent CEX-based reweighting (biennial starting 2002; annual since January 2023, using a single calendar year of expenditure data). Hedonic regression for major categories (PC, TV, smartphones, apparel) addressed quality bias for those specific items. The C-CPI-U series (Törnqvist formula, published since 2002) directly addresses across-category substitution. Estimates of remaining bias are debated; the most-cited follow-up studies, Lebow & Rudd (2003) and Moulton (2018), put the remaining upward bias at roughly 0.8–0.9 percentage points annually, with other estimates lower - below the Boskin figure but still meaningful.
Treasury TIPS principal indexation: TIPS use CPI-U NSA (Non-Seasonally Adjusted) with a 2–3 month lag. The reference CPI for any given month is the CPI-U NSA published for the third-prior month (e.g., the December reference is based on the September CPI release published in October). The principal-indexation lag is structural - it's needed to ensure that TIPS settlements use a CPI value that has actually been published before the settlement date. The lag creates a small seasonal component in TIPS pricing that arbitrageurs trade around.
Social Security COLA: uses CPI-W from Q3-to-Q3 (Q3 of current year minus Q3 of prior year, divided by Q3 of prior year). The COLA is announced each October following the September CPI release. CPI-W's narrower sample (urban wage earners and clerical workers - ~30% of population) tends to weight slightly more heavily toward energy and food than CPI-U, so COLA-adjusted Social Security benefits can drift from cost-of-living for retirees (whose consumption mix is older and more medical-care-heavy than CPI-W's reference population).
Medical-care methodology: CPI medical care includes out-of-pocket health-care spending - what consumers pay directly. It excludes employer-paid health insurance (which PCE includes) and government-paid medical care (Medicare, Medicaid). The exclusion of employer-paid medical from CPI is one reason PCE inflation is structurally different from CPI inflation - PCE captures the full cost of medical care to society, CPI captures only the consumer-paid portion.
Shelter methodology controversies during 2022–24: as market rent growth decelerated rapidly from late 2022, the OER lag held CPI shelter inflation elevated. The CPI shelter component was running 6–8% year-over-year through most of 2023 (peaking at 8.2% in March 2023) while market-rent indices (Zillow, Apartment List, CoreLogic) showed new-lease rent growth at 0–2%. Fed staff explicitly discussed the OER lag as a temporary distortion to disinflation prospects. The lag began to unwind in 2024 as the older high-rent observations rolled out of the year-over-year window.
New Tenant Rent (NTR) Index: BLS launched the NTR experimental series in 2022 (Adams, Loewenstein, Montag, and Verbrugge, 'Disentangling Rent Index Differences'; Cleveland Fed Working Paper 22-38 / BLS Working Paper 555) that isolates the new-lease subset of the BLS rental sample. NTR leads OER by 6–12 months because it captures the market-rent regime change before existing leases reset. NTR fell sharply through 2023–24, providing a leading indicator of subsequent OER deceleration.
CPI-vs-PCE divergence sources documented in Bullard (2014) and McCully-Moyer-Stewart (2007) decomposition papers: roughly 0.3 pp from formula, 0.2 pp from weights, 0.2 pp from scope. The exact decomposition varies by year and economic regime.
Seasonal adjustment risks: the February 2023 seasonal-factor revision materially altered the late-2022 inflation trajectory and caught markets off guard; the February 2024 revision, closely watched as a result, proved negligible. BLS provides extensive documentation of seasonal-factor methodology in the Handbook of Methods. The X-13-ARIMA-SEATS procedure used for seasonal adjustment can produce noticeable revisions in turning-point periods.
Research literature for methodology: Boskin et al. (1996) is the foundational policy document. Hausman (2003), Lebow & Rudd (2003), Diewert & Fox (2018) have updated bias estimates. Reinsdorf (1993) on outlet substitution bias; Moulton & Moses (1997) on quality-change bias. The BLS Handbook of Methods (CPI chapter) and the international Consumer Price Index Manual (published jointly by the ILO, IMF, OECD, UNECE, Eurostat, and the World Bank) provide methodology standards.
For practitioner analysis: the BLS monthly CPI release includes detailed tables for component-level inflation, which are essential for goods-vs-services-vs-shelter decomposition. Supplemental files on the BLS CPI website carry additional component-level detail (the standalone CPI Detailed Report publication was discontinued in 2017). The Federal Reserve Bank of San Francisco's Cyclical and Acyclical Core PCE Inflation decomposition and the Atlanta Fed's underlying-inflation dashboard provide complementary cross-sectional inflation views.