The Lag in Effect of Monetary Policy
📄 Summarized from the full manuscript · Human-reviewed for faithfulness before publication
In brief
How long after the money supply changes does the economy respond? Replying to a critic, Friedman defends three claims: money exerts an important independent influence, the delay is long relative to the business cycle, and it varies a great deal. Comparing turning points in money growth with business cycle peaks and troughs across eighteen non-war cycles since 1870, he finds money turns about sixteen months before peaks and twelve before troughs, with a spread of six or seven months that he argues measurement error inflates. The evidence is timing comparison, not formal estimation. It matters because a long, variable lag means active policy may add instability rather than offset it.
What this paper finds — and why it matters
This 1961 Journal of Political Economy paper is Milton Friedman’s reply to J. M. Culbertson’s criticism of an earlier finding, developed jointly with Anna J. Schwartz, that monetary actions affect economic conditions “only after a lag that is both long and variable,” and it defends three separable parts of that claim: that changes in the behavior of the money stock exert an important independent influence on subsequent events, that the average lag is large relative to the length of the business cycle, and that the lag varies substantially across episodes. Friedman’s evidence is not a formal econometric identification but a set of timing comparisons: matching turning points in the percentage rate of change of the money stock against National Bureau of Economic Research reference-cycle peaks and troughs over eighteen non-war cycles since 1870, and, separately, cross-correlations between money-stock changes and income/consumption drawn from the Meiselman-Friedman investment-multiplier study. On the average of the eighteen cycles, money’s rate-of-change peaks lead reference-cycle peaks by sixteen months and rate-of-change troughs lead reference troughs by twelve months (five and four months respectively using step-dates, an alternative dating method); the quarterly 1948-1958 cross-correlation exercise finds money leading consumption and income by three to four quarters, with peak correlations of .52 and .58 (both significant at the .001 level), implying a nine-to-twelve-month lead that Friedman treats as consistent with the longer-sample estimate given ordinary sampling variation. The standard deviation of these timing intervals is about six or seven months, which Friedman argues overstates the “true” variability of the lag because measurement error inflates the variance without similarly biasing the mean; on this basis he rejects Culbertson’s implicit claim that the lag’s standard deviation is under 0.9 months. Friedman explains the length of the lag through a balance-sheet transmission mechanism in which an open-market purchase leaves the non-bank public holding temporarily excess cash that is spent down gradually across a widening set of assets and expenditures, with interest rates and asset prices possibly serving only as a conduit rather than a necessary channel. He concludes that because the lag is long and variable, discretionary countercyclical monetary policy is likely to act as an additional, highly serially correlated disturbance rather than an offsetting one, strengthening the case for policy rules over discretion — while noting the fuller evidentiary presentation was still forthcoming in the Friedman-Schwartz Monetary History.
Summary of a classic paper, AI-assisted and human-reviewed. See the linked original for the authoritative claims and full conditions.
Questions & answers
Q1. What is Friedman responding to, and what exactly is the empirical claim he is defending?
Friedman is replying to J. M. Culbertson’s (1961, same JPE issue) criticism of the earlier Friedman-Schwartz finding that monetary actions affect economic conditions “only after a lag that is both long and variable,” and he separates that finding into three distinct sub-claims that must be evaluated individually (pp. 447-448): (A) changes in the behavior of the money stock exert an important autonomous influence on subsequent economic events; (B) the average lag between a monetary change and its effects is large relative to the typical length of a business cycle; and (C) the length of the lag varies substantially across episodes, again relative to typical cycle length. Culbertson’s critique targets the timing comparisons used to support (B) and (C); Friedman’s reply defends the comparisons and argues the policy implications (rules over discretion) follow even more strongly once the criticisms are addressed.
Q2. What method does Friedman use to measure the lag, and why does he prefer the rate of change of money over its level?
Friedman’s method is a timing comparison — dating turning points in a money-stock series and measuring how many months they lead or lag the corresponding turning points in the National Bureau of Economic Research (NBER) reference-cycle chronology — rather than any formal econometric identification (Section II.A, pp. 451-455). He argues the percentage rate of change of the money stock (its first log-difference) is the dimensionally appropriate and statistically trend-free series for this comparison, since the level of money is dominated by a strong secular trend that would swamp cyclical timing information. Two variants of the timing comparison are used: (a) turning points in the rate of change of money against reference-cycle peaks/troughs, drawn from eighteen non-war cycles since 1870; and (b) cross-correlations at various leads and lags between money-stock transformations and income/consumption, a by-product of the Meiselman-Friedman investment-multiplier and velocity study.
Q3. What do the eighteen non-war cycles since 1870 show about the average length of the lag?
Averaged across eighteen non-war cycles since 1870 using the NBER reference chronology, rate-of-change peaks in the money stock precede reference-cycle peaks by sixteen months and rate-of-change troughs precede reference-cycle troughs by twelve months (Section II.C, p. 457). An alternative dating method — “step-dates,” which mark the shift from a “high” to a “low” rate of change rather than the turning point of the (choppy) rate-of-change series itself — gives shorter leads of five months (peaks) and four months (troughs) (p. 458). Friedman reports that the two dating methods “run nearly a dead heat” in stability, and he chooses the rate-of-change comparison on economic grounds as the better estimator of the mean interval between a monetary action and its effects.
Q4. Does the independent Meiselman-Friedman cross-correlation evidence corroborate the eighteen-cycle timing result?
Yes: in quarterly U.S. data for 1948-1958, cross-correlations between quarter-to-quarter percentage changes in the money stock and percentage deviations of income and consumption from trend peak when money leads consumption and income by three to four quarters, with correlations of .52 (consumption) and .58 (income), both significant at the .001 level (Section II.C, pp. 459-460) — implying a lead of roughly nine to twelve months. Friedman calls this “somewhat shorter than the lead of twelve to sixteen months found” in the eighteen-cycle comparison but judges the difference “almost surely within the range to be expected from sampling fluctuations” (p. 459). A supplementary exercise using first differences of money against first differences of income/consumption finds the highest correlation when money leads consumption by one quarter (correlation .34, significant at the .02 level) — a result Friedman describes as “less clear cut” (p. 460).
Q5. How variable is the lag, and how much confidence does Friedman place in the variability estimate?
The standard deviation of the timing intervals — computed either from rate-of-change turning points or step-dates against reference-cycle turning points — is about six or seven months for both comparisons (Section III, p. 463), which Friedman uses to reject Culbertson’s implicit claim that the standard deviation of the lag is under 0.9 months, calling that figure “hardly credible.” He is explicit, however, that the variability evidence is “less satisfactory” than the average-length evidence: because measurement error inflates variance but largely cancels out of the mean, the observed 6-7 month standard deviation likely overestimates the true variability of the lag (p. 463) — a caveat that qualifies rather than undermines claim (C).
Q6. Why does Friedman think the lag is long, mechanically?
Friedman attributes the length of the lag to a balance-sheet transmission channel: an open-market operation leaves the non-banking public holding a stock of cash that is temporarily high relative to its other assets, and holders spend down this excess cash gradually, first bidding up prices of government securities or commercial paper and then spreading demand to equities, housing, durables, and producer goods (Section II.D, pp. 461-463). The expenditure effects can occur “without any change in interest rates at all,” with interest rates and asset prices potentially serving only as a conduit rather than an independent channel of transmission. Two features make the lag long: balance-sheet adjustment itself is sluggish and spread over time, and the resulting effects on expenditure are themselves spread out, so the relevant lag for policy purposes is the weighted average interval between the monetary action and the (distributed) effects on expenditure, not the moment effects first appear (p. 463).
Q7. What robustness does Friedman offer for the timing results?
Friedman points to convergence across independent evidence sources rather than a single robustness check: the eighteen-cycle NBER-chronology timing and the separate Meiselman-Friedman cross-correlation study give consistent lead estimates, two different dating methods (rate-of-change turning points versus step-dates) agree closely on stability even though they differ in lead length, Macesich’s Canadian data show roughly the same timing as the U.S. despite very different financial structures, and the cyclical relation between money and business has held with similar timing and amplitude across nearly a century of very different monetary arrangements (pp. 450, 457-460). No single decisive robustness test is offered; the argument rests on this pattern of corroboration across time periods, countries, and measurement methods.
Q8. What limitations and caveats does Friedman himself acknowledge?
Friedman explicitly denies that “the” lag is a single well-defined number: the effect of an instantaneous monetary change would begin immediately, rise to a “crescendo,” and then decline gradually without fully disappearing for an indefinite time, producing a distributed lag for which “the” lag is only a (still oversimplified) weighted-average summary (Section II.B, p. 455). He also flags: the variability (standard deviation) evidence is less reliable than the average-lag evidence because measurement error inflates it (p. 463); a “so-called identification problem” arises in estimating multi-equation systems from timing evidence (footnote 19, p. 456); timing comparison 1(b) using levels of money against levels of income/consumption is dominated by trend and unreliable, with percentage-deviation-from-trend correlations “extremely low” for the postwar period and first-difference comparisons only “less clear cut” (p. 460); and the policy assessment is described as preliminary pending the fuller Friedman-Schwartz Monetary History (p. 466).
Q9. What policy conclusion does Friedman draw from long and variable lags, and why?
Because the lag between a monetary action and its effects is both long and variable, Friedman argues that discretionary countercyclical monetary actions are likely to be poorly timed relative to the cycle they are meant to offset, so that “actions taken for countercyclical purposes [become] additional and unnecessary random disturbances” (Section IV, pp. 464-465). He adds that because monetary and fiscal authorities act on a scale that is large relative to other economic actors, they can sustain an inappropriate action for an extended period, introducing high serial correlation into the disturbances they create — disturbances of “just the kind that contribute most to the temporal variance of time series” (p. 465). This is the basis for Friedman’s broader case for monetary policy rules over case-by-case discretionary management.
Key terms in this paper
Definitions below follow the paper's own usage.
- "The" lag (weighted average interval)
- Friedman insists there is no single lag — a one-time monetary change produces effects that begin immediately, build to a peak, and decay slowly without fully disappearing, so "the" lag used in his timing comparisons is really the weighted average interval between the action and this whole distributed pattern of effects, an already-simplified summary statistic (Section II.B, p. 455).
- Reference cycle chronology
- the National Bureau of Economic Research's dating of general-business-cycle peaks and troughs (Burns and Mitchell 1946), used throughout the paper as the benchmark against which turning points in the money-stock series are timed; it is a descriptive dating convention, not a model-based decomposition.
- Rate of change of the money stock
- the percentage (first log-difference) change in the money stock, which Friedman treats as the dimensionally correct, trend-free series for cyclical timing comparisons, in contrast to the level of money, which is dominated by secular growth and yields unreliable timing correlations with income/consumption.
- Step dates
- an alternative to dating turning points directly in the (choppy) rate-of-change series; step dates mark the point where the rate of change shifts from a sustained "high" phase to a sustained "low" phase (or vice versa), yielding shorter, more stable estimated leads (five and four months) than the rate-of-change turning points themselves (sixteen and twelve months).
- Long and variable lags (the doctrine)
- Friedman's three-part claim that money exerts an autonomous influence on the economy (A), that this influence operates with a lag that is large relative to cycle length (B), and that the lag's length varies substantially across episodes (C) — the combination of (B) and (C) is what, in Friedman's argument, makes discretionary countercyclical monetary policy likely to destabilize rather than stabilize.