M2 and inflation since 1960: what the data shows and what it does not

"M2 growth leads inflation" is a claim that comes with a chart of two lines and, usually, a chosen window. This article computes the relationship over the whole span of the monthly M2 series, splits it into three periods, and lets the lag vary from zero to thirty-six months, so that the strength of the relationship and its instability can both be seen. The data and the script are attached; the correlations are not causal claims and the article does not test causation.

The data

• M2SL: M2, seasonally adjusted, billions of dollars, monthly, from the Federal Reserve's H.6 release, beginning January 1959. The FRED notes describe the May 2020 redefinition, when savings deposits were folded into M1 and the "other liquid deposits" component was created; the M2 total is continuous across that change, but its composition is not.
• CPIAUCSL: the CPI for all urban consumers, all items, seasonally adjusted.

Both retrieved 2026-09-05. The last month with both series is July 2026: M2 at 23,218.0 billion, up 5.41 percent over twelve months, and CPI inflation at 3.30 percent. The CPI series has no October 2025 observation, so the October 2025 and October 2026 twelve-month changes are gaps.

The method

df["m2_yoy"] = 100 * (df["M2SL"] / df["M2SL"].shift(12) - 1)
df["cpi_yoy"] = 100 * (df["CPIAUCSL"] / df["CPIAUCSL"].shift(12) - 1)
for k in range(0, 37):
    cors[k] = p["cpi_yoy"].corr(d["m2_yoy"].shift(k).reindex(p.index))

Both series become twelve-month percent changes. For each period, the script computes the Pearson correlation between CPI inflation in month t and M2 growth in month t minus k, for k from 0 to 36 months, and reports the correlation at lag zero, the lag with the highest correlation, and the correlation at a fixed 24-month lag for comparison across periods. The lag is applied to the full series before slicing, so the 1990 to 2019 window can use M2 growth from 1987 to 1989.

The tables

Output of python code/m2-and-inflation.py on 2026-09-05.

Decade means: 1970s M2 growth 9.5 percent against inflation 7.1; 1980s 7.9 against 5.6; 1990s 4.0 against 3.0; 2000s 6.5 against 2.6; 2010s 5.8 against 1.8; 2020s to date 6.9 against 3.9. M2 growth exceeded 10 percent in 104 months, in the years 1971 to 1973, 1975 to 1977, 1981 to 1983, 2001, 2009, 2012, and 2020 to 2022. It was negative in 15 months, all between December 2022 and February 2024, the only contraction in the series.

What it shows

At the same month, the two series are close to uncorrelated over the whole sample (0.07). The relationship appears with a lag: at 24 months the whole-sample correlation is 0.42, and in the 1960 to 1989 period it rises to 0.54 at a 33-month lag. That is the period in which the textbook version of the claim was written, and the data from that period supports a loose, lagged association.

From 1990 to 2019 the association is absent at every lag tested. The best correlation across 37 lags is -0.08, and at 24 months it is -0.20. The 2000s and 2010s had M2 growth of 6.5 and 5.8 percent a year against inflation of 2.6 and 1.8 percent, and the 2009 and 2012 surges above 10 percent were followed by the lowest-inflation decade in the sample.

The 2020s look different again. With only 78 months, the correlation peaks at 0.85 at an 18-month lag: M2 growth peaked at 26.8 percent in February 2021, and CPI inflation peaked at 9.0 percent in June 2022, sixteen months later. The M2 contraction that began in December 2022 was followed by inflation falling to the 2 to 3 percent range through 2024 and 2025. As of July 2026, M2 growth has recovered to 5.4 percent and inflation is 3.3 percent.

What it does not show

A correlation of 0.85 over 78 months that contain one large up-and-down cycle is close to a single observation. The 2020 to 2022 episode also coincided with fiscal transfers, supply disruptions, and an energy price shock, none of which are in this dataset, and the correlation cannot attribute the inflation among them.

The 1990 to 2019 result is equally hard to interpret. Velocity, the ratio of nominal spending to money, is not constant, and a stable relationship between money growth and inflation requires that it be at least predictable. A thirty-year stretch with no lagged correlation is evidence that it was not, in that period, and nothing more.

Choosing the "best lag" from 37 candidates and reporting its correlation inflates the number by construction; the 24-month column is there so that a fixed, pre-chosen lag can be compared across periods. On that column, the whole-sample and 1960 to 1989 figures (0.42 and 0.44) are moderate, the 1990 to 2019 figure is negative, and the 2020s figure (0.66) is high but rests on the one cycle.

None of this supports a forecast of inflation from the current M2 growth rate. The historical record contains periods where the relationship held loosely, one where it did not hold at all, and one recent episode where it held strongly; the data does not say which of those the next few years will resemble.