M2 and US Inflation, 1960–2026: A Three-Regime Empirical Test of the Monetarist Lag

Chart: M2 Growth (15-Month Lag) and CPI Inflation, 1960–2026

M2 Growth and CPI Inflation Across Three Regimes

The two series track closely in the 1970s and during 2021–2022, and run independently for three decades in between.

Key Takeaway

The simple bivariate relationship between M2 growth (lagged 15 months) and CPI inflation appears strong during two windows — the 1970s and the 2021–2022 surge — and absent for the three decades between them. Whether the M2–CPI relationship is “real” depends entirely on which sub-period is examined.

Sources: Federal Reserve H.6 (M2SL), BLS Consumer Price Index (CPIAUCSL). Chart: Eco3min Research.

PNG
SVG
PDF
Free to use with attribution · Cite

How to Read This Chart

The solid line shows the year-over-year change in M2, shifted forward by 15 months. The dashed line shows the year-over-year change in the CPI at the calendar date on the x-axis. Reading the chart at any vertical line tells you: if you knew M2 growth 15 months ago, what was CPI inflation today? The three shaded regions correspond to the regime classification used throughout this study: monetarist (1960–1989), broken (1990–2019), and post-COVID (2020–2026). For the underlying levels, see our M2 Money Supply and CPI Inflation History datasets.

Two windows visually align: 1973–1981 and 2021–2023. Outside those windows, the lines run independently. The narrative impression of an “M2 predicts inflation” relationship comes from the alignment in those windows, not from a stable underlying mechanism across the full sample.

The 15-Month Lag Was Always Smaller Than Advertised

The conventional summary of monetarism, repeated in financial commentary and macroeconomic textbooks, runs as follows: changes in M2 money supply precede changes in the price level with a lag of approximately 12 to 18 months. The figure is associated with Milton Friedman and forms part of the empirical foundation cited for the monetarist policy framework that influenced US monetary policy from the late 1970s through the early 1990s.

The bivariate test of that claim is straightforward: regress CPI year-over-year inflation on M2 year-over-year growth 15 months earlier. Across 345 monthly observations of the monetarist era (April 1960 through December 1989), the R² of that regression is 4.4%. Across 360 monthly observations of the broken era (January 1990 through December 2019), R² is 7.4%, with the slope of the relationship having inverted (from +0.270 to −0.153). The relationship the rule of thumb describes — strong, stable, lagged by 15 months — is not present in the data in either era.

Regime Decomposition: Monetarist, Broken, COVID

If the bivariate relationship at the textbook 15-month lag is weak, an obvious diagnostic is to ask whether the relationship exists at a different lag. The cross-correlation function between M2 YoY and CPI YoY, computed over each regime separately, identifies the lag at which the bivariate Pearson correlation is maximized.

Cross-correlation peak by regime

Two distinct patterns emerge. First, the optimal lag is itself unstable across regimes — it ranges from 3 months in the broken era to 30 months in the monetarist era. A relationship in which the optimal lead time varies by an order of magnitude across sub-periods is not a stable empirical regularity. Second, even at the optimal lag, the monetarist-era R² is 27.1%, far below what financial commentary typically implies. The relationship existed and was statistically significant, but the bivariate fit was always moderate, not dominant.

The Structural Break: a Chow test

The Chow test for a structural break in the M2(lag-15) → CPI regression, computed across candidate break dates from 1985 through 2000 on the pre-COVID sample, identifies January 1992 as the maximum F-statistic (F = 93.4, p < 0.001). The slope coefficient flips sign across the break: +0.246 before January 1992, −0.135 after. A positive M2 growth shock 15 months earlier is associated with higher CPI in the pre-1992 sample and with lower CPI in the post-1992 sample. The relationship did not merely weaken — its sign reversed.

The break date is not pinpoint-identifiable from this test alone — F-statistics in the 80–93 range occur for any candidate date between 1985 and 2000. What the data identifies is that a break exists somewhere in this window. The historical context aligns with this finding: the Federal Reserve formally downgraded M2 from its targeted aggregates list during Greenspan's July 1993 Humphrey-Hawkins testimony, and academic literature on Goodhart effects from sweep accounts identifies the late 1980s as the period during which the operational definition of “money” began to blur. The empirical break is consistent with the institutional and definitional changes that occurred over the same window.

A legitimate analytical qualification is that the post-1992 negative slope coefficient does not have a strong economic interpretation in its own right. It is a statistical artifact of the broken-era pattern in which M2 growth was elevated during the 1990s and 2000s while inflation remained low. The interpretation is not that M2 growth causes lower inflation post-1992, but that the simple bivariate specification no longer captures the inflation process in this period. The relationship was not merely weak — the specification was inapplicable.

The Velocity Steelman: Why M×V Restores the Fit

The Quantity Theory of Money in its identity form states that nominal output equals money supply times velocity (MV = PY). Friedman's monetarist claim was not merely that M2 growth predicts inflation — it was that, holding velocity stable, M2 growth predicts the price level. The simple M2-only regression tests the relationship under the assumption that velocity is constant. If velocity is itself moving systematically, the M2-only test will fail even if the underlying Quantity Theory mechanism is intact.

M2 velocity (M2V), computed by the Federal Reserve as the ratio of quarterly nominal GDP to M2, peaked at 2.192 in Q3 1997 and fell to a trough of 1.126 in Q2 2020. The decline of 48.6% over 23 years is the largest peak-to-trough velocity decline in the post-war US monetary record.

M2 grew. Velocity fell. The product grew much less.

Between Q3 1997 (the velocity peak) and Q2 2020 (the velocity trough), the M2 money stock expanded from $3.92 trillion to $17.07 trillion — a 335% increase. Over the same period, the consumer price level rose by only 59.6%. The simple monetarist intuition — that quadrupling the money supply should produce some commensurate increase in prices — appears to fail. But it fails because velocity moved in the opposite direction: M×V, the product that the Quantity Theory says should match nominal output, grew only 123% over the same period. The M2 expansion was largely absorbed by the velocity decline.

The fit improves substantially when velocity is included

The bivariate regression CPI YoY ~ M2 YoY (lag 15) yields R² of 4.4% in the monetarist era and 7.4% in the broken era. Replacing M2 YoY with the year-over-year growth of M×V — that is, using the combined nominal aggregate rather than M2 alone — yields the following results:

In the monetarist era, including velocity multiplies the R² roughly nine-fold. The Quantity Theory identity retains its empirical content even when the M2-only version fails. The popular form of monetarism — “the money supply predicts inflation” — is therefore not so much wrong as incomplete. It is wrong when it elides the velocity term that the underlying theory says must be present, and right when that term is included.

M2 velocity by decade

Velocity was approximately stable through the 1960s, 1970s, and 1980s. It rose during the 1990s, then declined sharply through the 2000s and 2010s. The collapse from a 1997 peak to the 2020 trough coincides with the period during which the popular M2-only specification produced its weakest correlations with inflation. When velocity is moving, the M2-only specification is misspecified by construction.

What Happened Next? CPI 15 Months Forward by M2 Growth Bucket

The forward-distribution question reverses the regression: given an observed value of M2 YoY today, what was CPI YoY 15 months later? The table below computes this across 778 observations from 1960 through October 2024 (the latest date for which a 15-month forward CPI observation is available), grouping by discrete M2 growth bands.

At the extremes of the M2 distribution, the relationship reasserts itself: when M2 YoY has exceeded 10%, the forward CPI 15 months later has had a median above 6% in every band tested. When M2 YoY has exceeded 15% (n = 13, all observations either pre-1976 or during 2020–2021), the median forward CPI was 7.56% and 100% of cases produced inflation above 2%. The middle of the M2 distribution — values between 4% and 10% YoY, which describes roughly 70% of all observations — produces a much wider forward CPI distribution that overlaps almost entirely with the low-M2 distribution.

The same buckets, split by monetarist vs broken era

Within the broken era, the median forward CPI is approximately 2% regardless of the M2 bucket — the conditional distribution becomes nearly flat. The same M2 readings that produced 6.4% median forward inflation in the monetarist era produced 2.3% median forward inflation in the broken era. The conditional information content of M2 growth, at the 15-month horizon, fell sharply across the regime break.

Methodological note: forward returns are computed from current-date M2 YoY to CPI YoY 15 months forward, with non-overlapping windows where possible. Buckets with n < 10 are reported but should be interpreted with caution. The extreme-M2 bucket post-1990 contains only 6 observations (all from the 2020–2021 COVID episode), so its median should be read as descriptive of that single episode rather than as a general property of the broken era.

Past distributions are not predictive of future outcomes. Regime-conditional statistics describe historical patterns, not expected inflation.

Three Regimes, Three Scatter Clouds

M2 Growth (Lag 15m) vs CPI Inflation, Colored by Regime

Each regime produces a distinct cloud; the relationship’s strength varies as much as the relationship itself.

Key Takeaway

The three regimes produce visually distinct scatter clouds. Monetarist-era points spread vertically across most M2 levels. Broken-era points cluster near 2% CPI regardless of M2 growth. COVID-era points form a tight diagonal line. A single regression on the pooled sample averages out information that is genuinely different across sub-periods.

Sources: Federal Reserve H.6 (M2SL), BLS CPIAUCSL. Chart: Eco3min Research.

PNG
SVG
PDF
Free to use with attribution · Cite

Regime classification: when does the relationship hold?

Historical Turning Points

October 1973 — Setup for the 1974 Inflation Peak

Monthly M2 YoY in October 1973 stood at 7.08%; M2 YoY in August 1973 (15 months before the November 1974 CPI peak) was 9.12%. The CPI YoY peak of the first oil shock arrived in November 1974 at 12.20%. The M2 signal preceded the CPI peak, but the M2 reading of 7–9% was lower than what the subsequent CPI suggested — the relationship was directionally correct but quantitatively weak. The same M2 reading later in the dataset would not produce the same CPI outcome.

March 1980 — Volcker-era peak CPI YoY

CPI YoY peaked at 14.59% in March 1980, the highest reading in the dataset. M2 YoY 15 months earlier (December 1978) was 7.53%. The 1970s M2 YoY peaked at 13.81% in February 1976 — 49 months before the CPI peak, not the 15 months that the popular rule of thumb specifies. The lag between the monetary impulse and the price-level response in the Great Inflation was closer to four years than to one. See our CPI Inflation History for the full decomposition.

July 1993 — Greenspan's Humphrey-Hawkins testimony

In his July 1993 monetary policy report to Congress, Chairman Greenspan formally announced that the M2 aggregate would be deemphasized as a policy target, citing the breakdown of the previously-stable M2 velocity relationship. The decision is contemporaneous with the Chow-test optimum identified by the regression: F-statistic at the 1993-01 candidate break is 92.4, near the maximum of 93.4 at 1992-01. The institutional and statistical break dates align.

Q3 1997 — M2 Velocity peak

M2V reached 2.192 in the third quarter of 1997, the highest quarterly reading in the post-war record. From this peak, velocity declined for 23 years to a trough of 1.126 in Q2 2020. Over the same window, M2 grew from $3.92 trillion to $17.07 trillion — a 335% expansion that produced only a 59.6% increase in CPI, because nominal M×V grew by only 123%.

February 2021 — Peak M2 YoY in COVID episode

M2 YoY reached 26.77% in February 2021, the highest reading in the entire dataset. Pre-COVID, the maximum M2 YoY was 13.81% in February 1976. The 2021 reading was approximately double the previous historical maximum. The peak CPI YoY of 8.98% followed 15.9 months later in June 2022, a lag remarkably close to the conventionally cited “12–18 month” figure that fails to describe the relationship in any other sample.

January 2026 — Current Observation

M2 stands at $22.43 trillion. M2 YoY is 4.09%, in the “low” band of the historical distribution. CPI YoY is 2.39%, near the Federal Reserve's 2% target. M2 velocity is 1.411, having recovered 25% from the 2020 trough but remaining 36% below the 1997 peak. M2 YoY 15 months earlier (October 2024) was 2.89%, in the “very low” bucket where the historical median forward CPI has been 2.86% — close to the actual current reading.

Methodology

This dataset combines three FRED series at monthly frequency, computes year-over-year growth rates and lag-shifted versions, classifies observations into three regime periods, and tests the bivariate relationship between lagged M2 growth and CPI inflation at fixed and optimal lags.

Core formula

Filter Definitions

All conditional statistics in this study use the following canonical filters (per the project's formal-definition rule):

Regime classification rationale and sensitivity

The boundary between “monetarist” and “broken” regimes is set at January 1990 as a conservative pre-test threshold, before the Chow-identified maximum at January 1992. This choice is robust: across candidate break dates from 1985 through 1997, the post-break R² remains in a narrow band (3.5% to 7.6%), and the pre-break R² remains 4.4% to 9.7%. The qualitative finding — a sharp reduction in the M2(lag-15) → CPI relationship across the late-1980s-to-early-1990s window — does not depend on the exact choice of break date within that window.

Lag selection

The conventionally cited “12–18 month” lag is tested directly at 15 months. The optimal lag for each regime is identified empirically by computing the Pearson correlation between M2 YoY[t − k] and CPI YoY[t] for k from 0 to 30 months and selecting the k that maximizes |r|. This avoids the look-ahead bias of choosing a lag based on post-sample knowledge of the relationship. Cross-correlation values at all tested lags are reported in the open-source companion notebook.

Chow test

The Chow F-statistic is computed across candidate break dates from January 1985 through January 2000, on the pre-COVID sub-sample (truncating at December 2019 to exclude the 2020 episode from break identification). For each candidate date c, the procedure (i) fits a pooled OLS regression on the full pre-COVID sample, (ii) fits separate OLS regressions on the pre-c and post-c sub-samples, and (iii) computes the F-statistic comparing the sum of squared residuals under the restriction (single regression) versus the unrestricted (separate regressions) specification. Standard Chow distributional assumptions are noted as a methodological caveat in Limitations.

Dataset Design

Python Reproduction Code

# Reproduce this dataset from primary sources
import pandas as pd
import numpy as np
from scipy import stats as sst

# Load three FRED series (M2SL monthly, CPIAUCSL monthly, M2V quarterly)
m2  = pd.read_csv("M2SL.csv", parse_dates=["date"])
cpi = pd.read_csv("CPIAUCSL.csv", parse_dates=["date"])
m2v = pd.read_csv("M2V.csv", parse_dates=["date"])

# Merge to monthly grid, forward-fill velocity within quarter
df = pd.DataFrame({"date": pd.date_range("1959-01-01", "2026-01-01", freq="MS")})
df = df.merge(m2, on="date", how="left").merge(cpi, on="date", how="left").merge(m2v, on="date", how="left")
df["m2v"] = df["m2v"].ffill()

# Compute YoY rates and lagged M2 series
df["m2_yoy"]       = df["m2sl"].pct_change(12) * 100
df["cpi_yoy"]      = df["cpi"].pct_change(12) * 100
df["m2_yoy_lag15"] = df["m2_yoy"].shift(15)
df["m_x_v"]        = df["m2sl"] * df["m2v"]
df["mv_yoy"]       = df["m_x_v"].pct_change(12) * 100

# Regime regression — monetarist era
mono = df[(df.date >= "1960-04-01") & (df.date < "1990-01-01")]
mono = mono.dropna(subset=["m2_yoy_lag15", "cpi_yoy"])
slope, intercept, r, p, se = sst.linregress(mono.m2_yoy_lag15, mono.cpi_yoy)
print(f"Monetarist R²: {r**2*100:.1f}%, slope = {slope:.3f}")

Dataset Download & Reproducibility

805 monthly observations · 21 columns · Jan 1959 – Jan 2026 · Licensed under CC BY 4.0.

Data Sources & References

• Primary Board of Governors of the Federal Reserve System (US), M2 [M2SL], retrieved from FRED, Federal Reserve Bank of St. Louis, May 2026. https://fred.stlouisfed.org/series/M2SL
• Primary U.S. Bureau of Labor Statistics, Consumer Price Index for All Urban Consumers: All Items in U.S. City Average [CPIAUCSL], retrieved from FRED, May 2026. https://fred.stlouisfed.org/series/CPIAUCSL
• Primary Federal Reserve Bank of St. Louis, Velocity of M2 Money Stock [M2V], retrieved from FRED, May 2026. https://fred.stlouisfed.org/series/M2V
• Research Friedman, Milton, and Anna J. Schwartz (1963). A Monetary History of the United States, 1867–1960. Princeton University Press. The foundational monetarist text establishing the long-run M-P relationship.
• Research Reis, Ricardo (2022). “The Burst of High Inflation in 2021–22: How and Why Did We Get Here?” NBER Working Paper 30214. Discusses the breakdown and partial revival of monetary aggregates in inflation forecasting.
• Research Bordo, Michael D., and Mickey D. Levy (2024). “Do Enlarged Fiscal Deficits Cause Inflation? The Historical Record.” Economic Affairs. Examines the role of money growth in postwar inflation episodes.
• Research Goodhart, Charles A. E. (1981). “Problems of Monetary Management: The U.K. Experience.” In Inflation, Depression, and Economic Policy in the West. The original statement of what became known as Goodhart's Law on the instability of money-aggregate targets.
• Reference Greenspan, Alan (July 1993). “Monetary Policy Report to the Congress.” Federal Reserve Board, Humphrey-Hawkins testimony, which formally deemphasized M2 as an FOMC target.
• Reference NBER Recession Dates, Business Cycle Dating Committee, retrieved May 2026.

Methodological Limitations

• The bivariate regression specification used throughout this study is deliberately simple. The Quantity Theory of Money, the New Keynesian Phillips Curve, and the fiscal theory of the price level each imply more complex inflation processes that this study does not test. The R² values reported are therefore lower bounds on what a multivariate specification could achieve; they describe the popular form of the monetarist claim, not the theoretical maximum.
• The Chow test assumes errors are independent and identically distributed under both hypotheses. The CPI YoY series exhibits autocorrelation, which inflates the apparent statistical significance of break tests. The F-statistic of 93.4 should be read as evidence of a structural change, not as a precise quantitative measure of its magnitude.
• The M2 series itself has changed composition over time. The May 2020 redefinition (incorporating savings deposits into M1) is the most recent example; earlier definitional changes occurred in 1980 (Depository Institutions Deregulation Act) and during the 1990s (introduction of sweep accounts). Comparing M2 levels and growth rates across regimes assumes a stable underlying measure that is not strictly accurate.
• Forward-distribution analysis uses overlapping 15-month windows. The effective sample size for forward-CPI statistics is smaller than the raw observation count suggests, and standard errors on bucket-conditional medians are not adjusted for the overlap.
• The post-1990 negative slope coefficient (−0.135) does not have a causal interpretation. It is a statistical artifact of the broken-era pattern, in which M2 growth was elevated and inflation remained low. Readers should not infer that increases in M2 cause decreases in CPI within this period.
• M2 velocity is constructed as nominal GDP / M2, which makes it mathematically dependent on the same variables it is supposed to explain. The velocity-restored fit reported in the steelman section is therefore not an independent test of the Quantity Theory; it is a consistency check on the identity MV = PY.

Frequently Asked Questions

Source

Related Eco3min Research

• US M2 Money Supply (1959–2026) — raw dataset
• The Inflation Tax: 113 Years of US CPI Data — episode decomposition
• The Fed's Track Record: 70 Years of Rate Decisions — Fed policy responses to inflation
• Net Liquidity Index — the modern composite that replaced M2 in market analysis
• Real Interest Rates History — the Fisher-equation framework for thinking about inflation
• M2 / GDP Ratio — the inverse of velocity