Core CPI vs Headline CPI (1957–2026): Three Episodes of Sustained Divergence and Three Different Outcomes

Chart: US Headline CPI vs Core CPI, Year-on-Year (1957–2026)

US Headline CPI vs Core CPI — Year-over-Year, Monthly, January 1957 to March 2026

Headline (red) and core (navy) move together more than 95% of the time. Only three episodes since 1957 saw a sustained gap above 2 percentage points — and the three resolved very differently.

Key Takeaway

The two series are highly correlated (0.926 contemporaneous correlation) — most of the time, “core” looks like “headline minus noise.” The interesting question is what happens in the rare months when they diverge persistently. The shaded episodes show that the answer has not been the same in each case.

Sources: BLS via FRED (CPIAUCSL, CPILFESL). Episode definition: gap_3mavg > 2pp for ≥3 consecutive months. Chart: Eco3min Research.

PNG
SVG
PDF
Free to use with attribution · Cite

How to Read This Chart

The red line is headline CPI year-over-year — the rate of change in the all-items consumer price index, including food and energy. The navy line is core CPI year-over-year — the same calculation applied to the all-items-less-food-and-energy index. Both series are seasonally adjusted (CPIAUCSL and CPILFESL on FRED) and use the same base period (1982–84 = 100). The three shaded regions correspond to the divergence episodes detected by the algorithm described in the Methodology.

Two observations dominate the visual. First, the two series track each other closely — they share trends, peaks, and troughs across the full 69-year sample. The pairwise correlation of year-over-year readings is 0.926. Second, the rare divergence episodes occur at moments of obvious supply-side stress: the OPEC oil embargo of 1973–74, the oil and food price spike of mid-2008 (followed almost immediately by the energy collapse of the Global Financial Crisis), and the 2022 collision of pandemic-era goods demand with the energy and food shock from Russia’s invasion of Ukraine. For context on how these inflation episodes related to the broader purchasing-power story, see our complete US inflation history dataset (1913–present).

What Core CPI Was Designed to Do

The standard interpretation of core CPI in macroeconomic policy circles is that it is a filter: by removing food and energy prices — the two CPI categories with the highest month-to-month volatility — it produces a smoother series that better approximates the persistent component of inflation. The Federal Reserve and academic researchers track core CPI (and the analogous core PCE) precisely because it is supposed to be less reactive to transitory shocks. In this view, when headline rises sharply and core does not, the appropriate inference is that the shock will pass without becoming embedded in wages, services prices, or expectations.

The data confirms this interpretation in most periods. Out of 818 months with computed year-over-year readings, 48% are “aligned” (gap within ±0.5 pp) and 46% are in “mild divergence” (between ±0.5 and ±2 pp). In only 3.4% of months does the headline-core gap exceed +2 pp on the 3-month average — and in only 2.1% does it fall below −2 pp. The filter is doing what it was designed to do most of the time: removing volatile components without creating large persistent divergences.

The Three Divergence Episodes

Applying a formal algorithm — gap_3mavg > 2 percentage points for at least 3 consecutive months — to the full 1957–2026 sample identifies three episodes. The same algorithm applied at higher thresholds isolates progressively fewer episodes; at the 3pp threshold, only the 1973–74 case qualifies. The 2pp specification is the choice that produces the most informative episode set while remaining narrow enough to exclude transient noise. The sensitivity to this choice is documented in the Threshold Sensitivity Analysis below.

Algorithmic Episodes — 1957–2026

The table shows the central empirical finding: the three episodes resolved in three structurally different ways. In the 1974 case, the divergence was the leading edge of a sustained transmission — core CPI rose by 7.12 percentage points over the following 24 months, reaching 10.31% by May 1975 (with an interim cyclical peak of 11.86% reached earlier, in February 1975). In the 2008 case, the divergence was a transient energy-driven spike in headline that resolved into a demand-driven downturn; core fell by 1.51 percentage points over the following 24 months. In the 2022 case, core was already at 6.03% — well above any historical “filter target” zone — when the energy and food shock pushed headline another 2.5 percentage points higher; core itself rose by only 0.6 pp from there before declining to 3.39% by 24 months out.

An honest reading of these three observations is that the headline-core gap is not, by itself, a reliable signal of future core inflation. The 1974 episode was followed by major transmission; the 2008 and 2022 episodes were not. The relevant difference between them appears to be the broader macroeconomic context — the labor market, the central bank’s policy stance, and the level of inflation expectations at the time of the shock — rather than the size or duration of the gap itself. This is also visible in the modest contemporaneous correlation between gap_3mavg and core_change_12m across the full sample: 0.343, far below the contemporaneous correlation between headline and core themselves (0.926).

Three Episodes, Three Outcomes — The Path of Core CPI

The most informative way to compare the three episodes is to plot the path of headline and core CPI from three months before each divergence began through the following 24 months. Each panel below shows the same window, on the same axis convention, with the episode duration shaded in gold. The differences in outcome become visually unambiguous.

Headline and Core CPI in the 24 Months After Each Divergence Began

In one episode, core caught up to headline. In another, core stayed flat while headline collapsed. In the third, both fell from already-elevated levels.

Key Takeaway

The three panels show three structurally different stories. In 1974, the divergence was the start of a transmission cascade — core caught up and exceeded the original headline level. In 2008, the divergence was the leading edge of a demand collapse — both series fell together. In 2022, the divergence was a secondary impulse on top of an already-elevated core — limited additional transmission, followed by disinflation in both series.

Sources: BLS via FRED (CPIAUCSL, CPILFESL). Window: t−3 months to t+24 months from episode start. Chart: Eco3min Research.

PNG
SVG
PDF
Free to use with attribution · Cite

Threshold Sensitivity Analysis

One legitimate concern with any episode-based analysis is that the episode count is mechanically driven by the threshold and minimum-duration choices. To address this, the heatmap below shows how many episodes the algorithm identifies under each combination of gap threshold (1.0 pp to 4.0 pp) and minimum duration (2 to 12 consecutive months).

Episode Count by Threshold × Minimum Duration

The 1974 episode is robust to every parameter choice. The 2008 and 2022 episodes are only visible at thresholds at or below 2.5pp.

Key Takeaway

The 1974 episode is detected in every cell of the matrix — it persists at any combination of threshold up to 4.0 pp and any minimum duration up to 12 months. The 2008 and 2022 episodes disappear from the count above the 2.5 pp threshold, reflecting their shorter duration (4 and 5 months) and lower peak gaps (2.8 and 2.7 pp).

Sources: BLS via FRED (CPIAUCSL, CPILFESL). Chart: Eco3min Research.

PNG
SVG
PDF

The choice of 2.0 pp × 3 months as the primary specification is documented but is not the only defensible choice. At 1.5 pp × 3 months, six episodes qualify (adding episodes from 1979–80, 2000–01, 2005, and 2011). At 3.0 pp × 3 months, only the 1974 case qualifies. The choice of 2.0 pp produces the most informative comparison set — three episodes with three different macroeconomic contexts — without expanding to noise-driven detections at lower thresholds. The pre-Volcker period (before 1985) contains a disproportionate share of the gap > +2pp observations: 5.2% of pre-1985 months versus 2.6% of post-1985 months, consistent with the standard interpretation that the post-Volcker inflation-anchoring regime reduced the frequency of large persistent positive divergences.

Forward Distribution: Change in Core CPI by Current Gap Regime

The most rigorous way to ask “what does the gap predict?” is to compute the distribution of forward changes in core CPI, conditional on the current value of the gap. The table below bins every month in the sample by its current gap_3mavg value, then reports the median and interquartile range of the change in core CPI YoY over the following 6, 12, and 24 months.

Methodological note: Forward returns are computed on overlapping monthly observations, so adjacent rows in the same gap regime are not independent. The 12-month change is measured from observation date t to t+12 calendar months in the same calendar-month sample. Sample sizes for the extreme regimes (n=28 for gap > +2pp; n=17 for gap

Past distributions are not predictive of future outcomes. Regime-conditional statistics describe historical patterns observed in a small number of episodes, not expected forward changes.

Historical Turning Points: The Three Episodes in Detail

May 1973 – August 1974 — The Oil Embargo and the Onset of the Great Inflation

The 1973–74 episode is the longest and largest in the dataset. At the start of the window (May 1973), headline CPI YoY stood at 5.5% and core CPI at 3.2% — a gap of 2.34 percentage points. The OPEC oil embargo, which began in October 1973, produced an energy price shock that propagated through the headline index over the subsequent months. The 3-month average gap peaked at 4.48 percentage points in March 1974, with headline at 10.1% and core at 5.8%. By August 1974, core had risen to 9.85% — the divergence had transmitted. Twelve months out from the May 1973 start, headline was at 10.71% and core at 7.06%; twenty-four months out, core reached 10.31%, with a peak of 11.86% in February 1975. The 1974 episode preceded the second leg of the Great Inflation — for the broader inflation context, see our complete US CPI history dataset and the analysis of how this period interacted with Federal Reserve policy decisions.

July 2008 – October 2008 — Energy Spike Then Demand Collapse

The 2008 episode is the shortest of the three (4 months) and the clearest case of the filter operating as designed. The episode began with headline CPI YoY at 5.50% in July 2008 — driven by the WTI crude oil price spike above $140 per barrel — and core CPI at 2.46%. The 3-month average gap reached 2.80 percentage points in August 2008. Within four months of the episode end, the headline-core gap had collapsed and then reversed: by January 2009, headline YoY was −0.11% while core was at 1.67% — a gap of −1.78 pp. The oil price collapse from $140 to under $40 between July and December 2008 mechanically pulled headline down. Core CPI did not transmit: twelve months after the July 2008 start, core stood at 1.53% (a decline of 0.94 pp from the start), and twenty-four months out it had fallen to 0.96%. For context on how credit conditions contributed to the demand-side collapse, see our high-yield credit spread study.

May 2022 – September 2022 — The Already-Elevated Episode

The 2022 episode is structurally different from both prior cases. Core CPI began the episode at 6.03% in May 2022 — already the highest core reading since the early 1990s, driven by post-pandemic goods inflation, shelter components, and recovering services. Headline was at 8.54% — a gap of 2.51 pp. The episode reflected the additional impulse from Russia’s invasion of Ukraine (which triggered new energy and food shocks) on top of a core that was already running well above target. The 3-month average gap peaked at 2.72 pp in July 2022. Core CPI rose modestly from there — reaching a peak of 6.62% in September 2022, a rise of only 0.6 pp from the May 2022 starting point — and then declined: 5.34% twelve months out, 3.39% twenty-four months out. The 2022 episode is therefore not a story of transmission from the gap; it is a story of two already-correlated impulses (the prior demand pressure already in core, plus the new energy/food shock in headline) running together for a short period before both moderated.

March 2026 — Current Observation

As of March 2026, the gap stands at +0.65 percentage points, at the 78th percentile of the historical distribution. This is in the “mild positive divergence” regime — neither aligned nor at the threshold for a new episode. The current configuration is consistent with the post-1985 normal range: most months show small positive or negative gaps without sustained divergence. The disinflation that began in mid-2022 has now produced 18+ consecutive months of headline CPI between 2.5% and 4.0%, and core CPI between 2.5% and 4.0%. The transmission risk from supply-side shocks at current core levels (~2.7%) is structurally different from the 2022 starting position (~6.0%) — but as the 1974 episode demonstrates, sustained transmission has occurred in modern US history from lower starting points when shocks have been large and persistent enough.

Methodology

This dataset combines two BLS Consumer Price Index series — headline CPI (CPIAUCSL) and core CPI (CPILFESL) — into a single monthly panel covering January 1957 to March 2026. All derived variables are computed from the two source index levels.

Episode Selection Criteria

The threshold of +2.0 percentage points and minimum duration of 3 consecutive months were chosen based on the sensitivity analysis above. At this specification, three episodes are detected: May 1973 – Aug 1974, Jul – Oct 2008, and May – Sep 2022. Sensitivity to threshold choice: changing the threshold to +1.5 pp increases the episode count to 6 (additional candidates in 1979–80, 2000–01, 2005, and 2011); changing it to +3.0 pp reduces the count to 1 (only the 1974 episode). Changing the minimum duration from 3 to 6 months reduces the count to 1 (only the 1974 episode) at the 2.0 pp threshold. The 2.0 pp × 3 months specification balances informativeness against false-positive risk.

The asymmetric threshold (positive gaps only) is deliberate: this dataset focuses on the question “what happens when headline runs above core” — i.e., supply-side shocks transmitting (or not) to underlying inflation. The opposite case (core above headline) typically corresponds to demand collapses where headline falls faster than core; the 2008–09 oil-price collapse is the canonical example, with a gap minimum of −3.12 pp in August 2009. Researchers interested in the symmetric case can apply the same algorithm with the inequality reversed using the provided CSV.

Literature Anchor

The three episodes identified by the algorithm correspond to the three commonly-recognized US supply-shock inflation events in the post-WWII literature: the OPEC-driven shock of 1973–74 (Blinder, 1979; Hamilton, 1983); the 2008 commodity-and-credit collision (Hamilton, 2009; Reinhart & Rogoff, 2009); and the 2022 pandemic-and-Ukraine episode (Reis, 2022; Bernanke & Blanchard, 2023). The algorithmic detection is consistent with conventional periodization without requiring it as an input.

Dataset Design

Python Reproduction Code

# Reproduce this dataset from FRED
import pandas as pd

# Load both series from FRED
head = pd.read_csv('https://fred.stlouisfed.org/graph/fredgraph.csv?id=CPIAUCSL')
core = pd.read_csv('https://fred.stlouisfed.org/graph/fredgraph.csv?id=CPILFESL')
head.columns = ['date', 'cpiaucsl']
core.columns = ['date', 'cpilfesl']

# Merge and compute YoY changes
df = pd.merge(head, core, on='date', how='inner')
df['date'] = pd.to_datetime(df['date'])
df = df.sort_values('date').reset_index(drop=True)
df['headline_yoy'] = df['cpiaucsl'].pct_change(12) * 100
df['core_yoy'] = df['cpilfesl'].pct_change(12) * 100
df['gap_yoy'] = df['headline_yoy'] - df['core_yoy']
df['gap_3mavg'] = df['gap_yoy'].rolling(3).mean()

# Episode detection (gap_3mavg > 2pp for >= 3 months)
df['in_divergence'] = (df['gap_3mavg'] > 2.0).astype(int)
df['run'] = (df['in_divergence'] != df['in_divergence'].shift()).cumsum()
runs = df[df['in_divergence']==1].groupby('run').size()
episodes = runs[runs >= 3]
print(f"Episodes detected: {len(episodes)}")
  

Dataset Download & Reproducibility

830 monthly observations · 17 columns · Jan 1957 – Mar 2026 · Licensed under CC BY 4.0. The XLSX file includes a “Documentation” sheet describing every column.

Data Sources & References

• Primary U.S. Bureau of Labor Statistics (BLS) — Consumer Price Index for All Urban Consumers (CPI-U), series CPIAUCSL. Monthly, seasonally adjusted, 1947–present. Retrieved May 2026.
• Primary U.S. Bureau of Labor Statistics (BLS) — Consumer Price Index for All Urban Consumers: All Items Less Food and Energy, series CPILFESL. Monthly, seasonally adjusted, 1957–present. Retrieved May 2026.
• Primary Federal Reserve Bank of St. Louis (FRED) — Programmatic data retrieval for both BLS series.
• Research Blinder, A. S. (1979) — “Economic Policy and the Great Stagflation,” Academic Press. Analysis of the 1973–74 supply shock and its transmission to underlying inflation.
• Research Hamilton, J. D. (1983) — “Oil and the Macroeconomy Since World War II,” Journal of Political Economy, 91(2): 228–248.
• Research Hamilton, J. D. (2009) — “Causes and Consequences of the Oil Shock of 2007–08,” Brookings Papers on Economic Activity, Spring 2009.
• Research Bernanke, B. and Blanchard, O. (2023) — “What Caused the U.S. Pandemic-Era Inflation?” Hutchins Center Working Paper.
• Research Reis, R. (2022) — “The Burst of High Inflation in 2021–22: How and Why Did We Get Here?” CEPR Discussion Paper.
• Reference Federal Open Market Committee — Statement on Longer-Run Goals and Monetary Policy Strategy, originally adopted January 2012. Specifies PCE (not CPI) as the Fed’s preferred inflation target measure.

Methodological Limitations

• Small episode count. The primary specification identifies only three episodes across 69 years. Any claim about “what the gap predicts” based on n=3 is necessarily descriptive rather than inferential. The forward-distribution analysis (n=28 for gap > +2pp) addresses this in part by using the full distribution of high-gap months rather than just the three episodes, but adjacent observations within the same episode are not independent.
• Asymmetric threshold. The primary algorithm filters for positive divergence only (headline above core). The opposite case is documented in the data (gap minimum −3.12 pp in August 2009) but is not the focus of this dataset’s narrative. Researchers interested in symmetric analysis can apply the threshold logic to both tails of the gap distribution using the provided CSV.
• CPI is not the Fed’s target. The Federal Reserve’s stated inflation objective since 2012 is the year-over-year change in the PCE (Personal Consumption Expenditures) price index, not the CPI. The PCE and CPI are constructed from different baskets and using different weights, and they have diverged by an average of approximately 30–40 basis points historically. Conclusions about the BLS’s CPI series do not directly translate into conclusions about the Fed’s preferred measure.
• Composition changes over time. The CPI basket of goods and services has changed substantially since 1957 — both because of consumer behavior shifts and because of BLS methodological revisions (most notably hedonic adjustment introduced in 1998 and the gradual shift to chained measures). The headline-core gap in 1973 is technically not measured the same way as the headline-core gap in 2022. The BLS considers the series comparable; users should be aware of this caveat.
• Seasonally adjusted series. Both CPIAUCSL and CPILFESL are seasonally adjusted by the BLS. Year-over-year calculations partially eliminate seasonal effects but do not fully remove them when seasonal patterns change over time. The not-seasonally-adjusted versions (CPIAUCNS, CPILFENS) are available on FRED for users who prefer to apply their own seasonal adjustment.
• Three monetary regimes, one dataset. The 1957–2026 sample spans the post-Bretton Woods transition (1971), the high-inflation regime of the 1970s, the Volcker disinflation (1979–82), and the modern inflation-anchoring regime (post-1995 informal, post-2012 formal). Inflation dynamics differ structurally across these sub-periods. The pre-1985 versus post-1985 comparison in the Executive Summary partially addresses this but does not fully resolve the non-stationarity concern.

Frequently Asked Questions

Source

Related Eco3min Research

• The Inflation Tax — 113 Years of US CPI Data (1913–2026) — full purchasing-power decomposition with four major episodes
• US Core CPI Inflation Dataset — Raw Data Page (CPILFESL)
• US PCE Inflation Dataset — The Fed’s Preferred Measure
• Federal Funds Rate History (1954–Present) — Fed reaction function and policy response
• US Real Interest Rates History (1962–Present) — real-rate context for each inflation episode
• 10-Year Breakeven Inflation Rate (T10YIE) — TIPS-implied long-term inflation expectations