Long and variable nonsense

I recently heard a NPR discussion of the “long and variable lags” in monetary policy. Not surprisingly, it made me cringe.

They discussed the fact that Milton Friedman believed that monetary policy affected the economy with a lag as long as 18 months (but variable in length.) Then the host explained that “monetary policy” meant changes in interest rates. Friedman must be rolling over in his grave. He explicitly rejected the notion that interest rates were monetary policy and he did not believe that changes in interest rates affected the economy with a long lag. It would be more accurate to suggest that he believed the economy affected interest rates.

Unfortunately, Friedman’s preferred indicator (M2) is only modestly less bad than interest rates as an indicator of the stance of monetary policy. Consider the fact that M2 is up 34.8% in the past 3 1/2 years, and is down 3.7% in the past year. I occasionally see old-line monetarists claim that the big rise in M2 caused the inflation overshoot, and that the recent fall in M2 shows that money has now become too tight. Maybe, but I don’t see the logic of this claim.

Over the past 3 1/2 years, the PCE price index is up 14.9% and NGDP is up 23.6%. Both of these increases are far less than the rise in M2. So if you took monetarism seriously, you’d conclude that the 34.8% rise in M2 hasn’t yet worked its way though the system. We’re still waiting for those mysterious “long and variable lags”. Velocity was temporarily depressed by Covid and (the theory suggests) once it returns to normal we’ll see a lot more inflation. Instead, monetarists seem focused on the recent drop in M2. Why? Are the money supply figures since the beginning of 2020 meaningful? Or not?

To be clear, monetarists might be correct that money is currently too tight and we’ll soon enter a recession. I just don’t see how one would conclude that from the fact that M2 is up 34.8% over the past 3 1/2 years and NGDP is up only 23.6%.

Here’s how I see things. A change in monetary policy affects future expected NGDP (say one or two years out) almost instantaneously. Current NGDP immediately responds by a very small amount (higher commodity prices), and over a few months the effects become quite large. The scientific way to look at policy lags would be to create a deep and liquid NGDP futures market, and then look at how long it takes changes in future expected NGDP (i.e. “monetary policy”) to affect current and near term NGDP. I suspect the effect mostly occurs quite fast, within a few months.

The whole long and variable lags nonsense is an excuse to explain flaws in existing Keynesian models (which suggest monetary policy is changes in interest rates), and flaws in monetarist models (which suggest monetary policy is changes in M2.) Since both models are wrong, they need to explain their failed predictions in roughly the sort of way an astrologer explains his bad forecasts by being intentionally vague. Lots of mumbo jumbo about long and variable lags.

Future generations will be embarrassed by the pseudoscience that masquerades as “monetary theory” in the 2020s.

PS. Many economists seemed to believe the Fed adopted a “tight money” policy in 2022 and that we’d have a recession in the first half of 2023. How’d that prediction work out?

PPS. The “credit channel” people thought last March’s banking crisis would worsen the recession. How’d that prediction pan out?

PPPS. The Atlanta Fed nailed the RGDP forecast (2.4%).



23 Responses to “Long and variable nonsense”

  1. Gravatar of Kenneth Duda Kenneth Duda
    27. July 2023 at 14:19

    I feel like you’ve been writing posts just like this one for more than 10 years.

    Do you think it makes sense to prepare a short monetary policy fact sheet, with facts that are similar in nature to the “facts” most reporters present (such as, “higher interest rates mean tighter money”). It might include these basic points:

    – sometimes higher interest rates lead to less spending, but sometimes they don’t; it depends on many other factors that are hard to observe, including people’s expectations of the future. For example, people may spend more when interest rates go up if they expect higher interest rates or inflation in the future. So it is simply wrong to say that when the Fed raises interest rates, it’s tightening monetary policy.

    – the reliable way to determine whether monetary policy is getting looser or tighter is to look at market expectations of future NGDP growth. Anything the Fed does that increases those expectations is loosening. Anything the Fed does that decreases those expectations is tightening.

    – there is no long or variable lag from changes in monetary policy to market NGDP growth expectations. The market reacts to Fed news instantly.

    – NGDP growth expectations determine NGDP growth with some lag, but it’s neither long nor variable, a few months or so.

    and then send a copy to every economist and news reporter who reports on this incorrectly, asking them to please use the sentences above instead of nonsense about how interest rates determine monetary policy but with a macro effect that takes a large and unknown amount of time.

    Can you imagine if physicists said that objects fall to earth based on their temperature, but it takes a long and variable lag for their temperature to cause them to fall, which is why it’s so hard to see how temperature is causing the falling?

    I don’t understand why any of this should even be controversial. It’s obvious, for example, that rising interest rates can accompany monetary loosening, if (for example) expectations of future inflation are rising faster than rates.



  2. Gravatar of ssumner ssumner
    27. July 2023 at 14:49

    Ken, I’ve tried to make these points in every way possible. Academic papers, think tank papers, op eds, blog posts, public speaking, and now an entire book on the subject.

    Just not sure anyone pays attention.

  3. Gravatar of Solon of the East Solon of the East
    27. July 2023 at 15:23

    The scientific way to look at policy lags would be to create a deep and liquid NGDP futures market, and then look at how long it takes changes in future expected NGDP (i.e. “monetary policy”) to affect current and near term NGDP.-SS

    I like this idea.

  4. Gravatar of Ahmed Fares Ahmed Fares
    27. July 2023 at 15:27

    The equation of exchange is flawed because it doesn’t account for non-GDP transactions, i.e., GDP transactions are a subset of all transactions. A quote from Richard Werner (he uses the older equation of exchange here):

    Solving the enigma of the ‘velocity decline’

    The problems arising from the implicit assumption that nominal GDP (that is, PY) can be used to represent total transaction values (PT) are obvious. GDP transactions are a subset of all transactions. The mainstream quantity equation that uses income or GDP to represent transactions will thus only be reliable in time periods when the value of non-GDP transactions, such as asset transactions, remains constant (thus dropping out when considering flows). However, when their value rises, this will cause GDP to be an unreliable proxy for the value of all transactions. In those time periods we must expect the traditional quantity equation, MV = PY, to give the appearance of a fall in the velocity V, as money is used for transactions other than nominal GDP (PY). This explains why in many countries with asset price booms economists puzzled over an apparent ‘velocity decline’, a ‘breakdown of the money demand function’ or a ‘mystery of missing money’ – issues that severely hampered the monetarist approach to monetary policy implementation.
    [end quote]

    The historically low interest rates we had previously explains the rise in financial engineering, which is why the money supply grew without inflation.

  5. Gravatar of marcus nunes marcus nunes
    27. July 2023 at 15:41

    I´ve tried in a “tongue in cheek” way to dispel the notion of “long & variable” lags in the effect of monetary policy:

  6. Gravatar of Daniel Carroll Daniel Carroll
    27. July 2023 at 16:44

    I’ve been following your blog sometime around the financial crisis. I’ve been an investor for 30 years. You’ve said this so many times that I’ve been forced to think about it. I see high frequency asset prices reflecting monetary policy immediately. The economy then goes through various adjustments in a process that takes 0-18 months. My thought is that the inflation impulse is immediate, but measured inflation (and by extension measured GDP) is not immediate. The “long and variable lag” theory confuses measured inflation with actual inflation. The two are correlated but are not the same thing.

  7. Gravatar of ssumner ssumner
    27. July 2023 at 17:09

    Daniel, Thanks for following the blog. Here’s one way of thinking about your point. Imagine a sudden monetary shock that raises the future expected price level (and NGDP) by 10%. The spot CPI might rise by 0.5% or 1%, as things like gasoline and other commodities respond immediately. House prices rise fairly quickly, and other consumer prices more slowly. Real output rises within a few months, but in the long run is unaffected. NGDP rises within months.

    Notice I said a monetary shock that raises the future expected price level. That’s not identical to a monetary shock as described by the media, as a cut in rates may merely be following the natural rate lower. I’m talking about an actual change in monetary policy that moves the expected future NGDP higher.

    When you say “actual inflation” I think you mean inflation once prices have responded to the monetary shock. Initially, it shows up in commodities and other asset prices.

    I think we are on the same page, but I wanted to spell my view out more clearly.

  8. Gravatar of ssumner ssumner
    27. July 2023 at 17:10

    Everyone, Once again, comments will be held in moderation until I can fix the problem.

  9. Gravatar of Brent Buckner Brent Buckner
    28. July 2023 at 02:22

    Some of what appears to be “long and variable lags” may be measurement error, specifically the use of lagged prices in the relevant price index.

  10. Gravatar of spencer spencer
    28. July 2023 at 05:49

    The equation of exchange uses transaction’s velocity. It is a mathematical certainty.

    Financial transactions aren’t random. But you obviously omit NYC’s transactions in computations on the G.6 release to remove the preponderant distortions.

    Even so, using rates-of-change, you can compare GDP transactions to total transactions. They move in concert.

    M2 is mud pie. Banks don’t LEND deposits. George Garvey explains that some deposits don’t turn over at the same rates.

    The resilience of the economy is being determined by the changing composition of the money stock, viz., the activation of monetary savings. Ceteris paribus, there is a one-to-one correspondence between interest-bearing time deposits and demand deposits, as time deposits are depleted, demand deposits grow.

    As Powell removed reserve requirements, the percentage of demand deposits is growing unchecked. And DDs turn over more frequently.

    The distributed lag effect of monetary flows, the volume and velocity of money, have been mathematical constants for > 100 years.

    Behavioral economics is the dismal science.

  11. Gravatar of spencer spencer
    28. July 2023 at 06:04

    Targeting nominal GDP growth is the optimum objective of monetarism (which has never been tried). N-gDp is a proxy for aggregate monetary purchasing power in American Yale Professor Irving Fisher’s truistic equation of exchange.

    The transactions concept of money velocity is where (M) equals the volume of means-of-payment money; (Vt), the transactions rate of turnover of this money; (T), the volume of transactions units; & (P), the average price of all transactions units.

    As statistics on P &T are unavailable, it is impossible to calculate P & T in the equation. Nevertheless, the equation is a truism: to sell 100 bushels of wheat (T) at $4 a bushel (P) requires the exchange of $400 (M) once, or $200 (Vt) twice, etc.

    Or a dollar bill which turns over 5 times can do the same “work” as one five-dollar bill that turns over only once.

    Since nominal-gDp has continued to rise, and (M) has remained more or less constant, we can infer that (Vt), has risen (because roc’s in MVt = roc’s in nominal-gDp).

  12. Gravatar of spencer spencer
    28. July 2023 at 06:27

    “Separately, inflation pressures continued surface, with the May 2023 Money Supply reflecting still-extreme flight to liquidity. The most-liquid “Basic M1” (Currency-plus-Demand Deposits) held 119.2% above its Pre-Pandemic Level and was increasing year-to-year with intensifying inflation pressure, versus the Aggregate M2 Money Supply holding up by 34.7%, but declining year-to-year, amidst no signs of an overheating economy.”

  13. Gravatar of spencer spencer
    28. July 2023 at 06:37

    As Friedman said: “The only relevant test of the validity of a hypothesis is comparison of prediction with experience.”

  14. Gravatar of spencer spencer
    28. July 2023 at 07:58

    As Dr. Sumner has said, the reaction to a monetary policy change is immediate. Thus, the monetary transmission mechanism can’t be interest rate manipulation. A change in interest rates, as Powell has said, is indeterminate. Policy must be driven by base money, or reserves.

    Powell didn’t change interest rates when SVB collapsed. Powell interrupted QT. Powell temporarily increased base money. And after the quick injection, Powell resumed QT, draining base money.

    Therefore, base money must be used to control N-gDp.

  15. Gravatar of dlr dlr
    28. July 2023 at 08:02

    this is a useful model, but i think you overlook how deftly it slides the user toward cheating in practice. in november of 2022, you claimed “not only did the Fed not do too much tightening, there’s almost no evidence that the Fed has tightened monetary policy at all, at least to any measurable extent.” at the point you wrote that, the implied sep23 FFR had jumped from 5% from 2.9% in just three months. the real two year rose from flat to 2%. during this period both the S&P and the CRB index had declined and the 5-year breakeven also declined. i agree that rates changes suffer a severe identification problem. but when both fed funds and real interest rate expectations rise markedly amid declining asset prices and nominal results coming in near consensus expectations (unlike 21-mid 22), what you absolutely do not have is “almost no evidence” of tightening.

    if believing in leads and understand the identification problems steers you to looking only at concurrent economic data (you pointed to high levels of job openings and strong weekly payrolls), you are doomed ignore most of the most informative, if difficult, evidence about policy. we were only weeks removed from powell’s obviously (from market price reactions in rate and asset markets) tightening jackson hole speech. even in a mostly variable-leads economy, we are not going to get highly useful indicators of policy in all but the most draconian cases from weekly payrolls and levels of job openings. we can’t run from the fact that some data will in fact lag (or, like payrolls, have multiple paths all potentially consistent with a given nominal policy path given our imperfect understanding of the interaction of real and nominal variables) as an excuse to ignore the language in which the fed unfortunately chooses to speak. it’s more useful to speak interest rates while putting the identification problem front and center, or leadism is just a license to be overconfident about the information in a few concurrent cherry-picked economic data.

  16. Gravatar of ssumner ssumner
    28. July 2023 at 12:47

    Brent, Yes, especially for housing. But they claim long lags even for RGDP.

    dlr, I think subsequent events confirm my call last November. NGDP growth has remained far too high over the past 8 months.

    Also, it’s important to not confuse day to day changes in the policy stance with an overall stance of tightening. On any given day, a sharp rise in the fed funds target is usually a tighter policy than the day before, but it doesn’t indicate that policy is tighter than 6 months before.

    I agree that current data doesn’t always reveal the current stance of policy, but I continue to believe that interest rates are virtually useless as a policy indicator. Once you know the market’s expected NGDP growth path, interest rates add nothing useful.

  17. Gravatar of Brent Buckner Brent Buckner
    29. July 2023 at 04:36

    @Dr. Sumner, you wrote:
    “But they claim long lags even for RGDP.”

    Sure, but the measures of RGDP supporting those claims rely upon measures of inflation, so the same measurement error is in play.

  18. Gravatar of dlr dlr
    29. July 2023 at 05:35

    dlr, I think subsequent events confirm my call last November. NGDP growth has remained far too high over the past 8 months.

    qoq ngdp growth went 7.7%, 6.6%, 6.1%, 4.7%.

    in december you also wrote:

    At a minimum, the Fed should have tried to get NGDP growth to level off at no more than roughly 4% or so. Instead, it rose by 9% over the past 4 quarters, and 2022:Q4 looks like another hot one.

    To prevent further overshoots, the Fed needed to quickly get interest rates up to a neutral level. Instead, they raised rates at a very slow pace, which meant that policy remained effectively expansionary through most of 2022. The problem is getting worse as each month goes by. We aren’t making progress, just the opposite.

    but by spring you were saying:

    Yes, I don’t see much evidence that the current stance of monetary policy is too easy or too tight…

    i don’t see how you can say the subsequent marked decline in nominal growth confirms your claim in late 22 that the fed didn’t tighten and the problem was in fact worsening. it’s not like you came out in february or whenever and said “finally the tightening has arrrived!”. it just seems like you are using leadism and the identification problem to ignore the facts. but instead of phrasing like a gotcha, let me ask you: the fed brought nominal growth down from double digits (even after we had crossed the precovid trend in q121) to 4.7%, and extremely substantial tightening. when do you think they did this?

  19. Gravatar of ssumner ssumner
    29. July 2023 at 08:46

    dlr, I focus more on NGDP relative to trend. It’s been getting further and further above trend. The NGDP growth rates of 2022 were clearly way too high, so I think I’m justified in calling 2022 monetary policy “too expansionary.” Recent policy is more debatable.

    I will concede that Fed policy is something of a mess, because it’s never clearly explained its policy target. What does “average inflation targeting” actually mean? Until they tell us, people can disagree as to whether the current stance of policy is appropriate. But if the Fed truly is looking for inflation to average 2% over a period of 5 or 10 years, then policy has clearly been way too expansionary.

    This is why we need level targeting—it would make it crystal clear whether policy is off course.

  20. Gravatar of ssumner ssumner
    30. July 2023 at 09:52

    Brent, But the measurement errors go in the opposite direction, whereas monetary policy affects P and Y in the same direction.

  21. Gravatar of Brent Buckner Brent Buckner
    30. July 2023 at 10:28

    @Dr. Sumner:

    Stipulated – thanks!

  22. Gravatar of viennacapitalist viennacapitalist
    30. July 2023 at 23:54

    NGDP futures as a monetary policy signal you say?
    (In our discussion a few weeks back, you denied that futures would play the role of policy signal – you said it is enough if the fed just offers to buy or sell at a fixed price without the need for adjusting the price, i.e. no signal)
    Anyway, how do you envision the market design of this market?
    After all, it is very difficult to create a futures market without the possibility for market makers of hedging their book by trading the underlying.
    Multiple, independent, market makers are needed for the price to fulfill its signalling function – that is standard market microstructure theory.

  23. Gravatar of ssumner ssumner
    31. July 2023 at 12:02

    The signal comes from volume, not price. Check out chapter 5:


Leave a Reply