Friday, April 5, 2019

Interest rates and inflation

Apparently Randall Wray has set off the econ twitters with a pull quote captured here:
"There is no empirical evidence to support the belief that raising interest rates fights inflation."
It was pretty funny seeing the flurry of "obviously it does" countered with another flurry of studies of the lack of a connection or just various models (including DSGE models which are bad at forecasting) showing different results. I just found myself in a rare agreement with Wray. My own heterodox take — inflation is always and everywhere a labor force phenomenon — would probably be met with equivalent derision if anyone was paying attention.

My working hypothesis is two-fold:

  1. When labor force participation is increasing faster than equilibrium (when women are entering the workforce in a demographic change or men entering the workforce after a war), we get a surge of inflation. When labor force participation declines, inflation flags.
  2. When recessions happen during the a surge in labor force participation, it modulates that increase in participation causing oscillations — the Phillips curve. I have speculated that it is possible these processes are connected and that you always get "gravity waves" of inflation when labor force participation is increasing faster than equilibrium.
It is possible that the Fed triggers the recessions in the second part by raising rates instead of the oscillations being endogenous (i.e. generated by the surge process itself). In that sense, it is possible the Fed causes the recessions which cut off inflation. This would create a pattern of a series of Fed hikes before a recession which cuts off the inflation (and occurs near, but just after the peak). We'll call this A. However, if the Fed raising rates fights inflation, that should create a different pattern in the data: a surge in inflation should have started before the Fed starts hiking rates, and subsequently turnaround and end after or while the Fed is hiking rates — independent of a recession. This could also manifest as a fall in inflation. We'll call this B. Of course, there is also the possibility that Fed rates have nothing to do with the pattern of inflation (which is closer to my working hypothesis where the Phillips curve is endogenous). We'll call this C.

I will look at this through one of my "economic seismograms" that basically are a visual form of Granger causality (but a bit more conservative). I drew the changes in the time series with increases and decreases shown in red and blue (red is associated with economic growth, blue with decline/recession) and layered the Fed rate changes in yellow (raising rates) and green (lowering rates) as lines on top (click to enlarge):

What do we have? Well, in the period around 1960, we have a B — the Fed raised rates and what followed was a decline in inflation. You could possibly say that the '71, '74, and 80s recessions are A's.

In nearly every case though, the Fed's rate increases begin before the surges in inflation and end randomly relative to the turnaround point (the midpoint). It's almost the neo-Fisher hypothesis that Fed rate increases cause inflation! Recessions come at the tail end of these inflation surges as they're cut off by unemployment spikes, but sometimes we have no inflation surge (through most of the 90s and 00s). The Fed is always dutifully raising and lowering rates in sync with recessions. We even have an anti-B during the period of "lowflation" following the Great Recession — the Fed had lowered rates. That is to say most of the time we have C.

In fact, the single best explanation (we'll call it D) is that inflation and interest rates are unrelated but the Fed thinks it controls inflation and mitigates recessions with them, so it raises rates during economic expansion and lowers them after economic indicators turnaround before an oncoming recession. This has the benefit of the premise being incontrovertibly true — the FOMC does believe it can affect inflation or recessions with its rate choices.

I should always bring up Milton Friedman's thermostat here: if the Fed really did control inflation with policy, then it would look like there's no specific relationship between rates and inflation if they were doing it right. Of course this view is both question begging and unscientific in the sense that it shuts down inquiry. But a really good counter is that Fed policy is obviously correlated with recessions, so Occam's razor is that we should assume D until convincing evidence comes along to change our minds.


  1. It's an interesting hypothesis, but I assume you're not claiming it applies to most countries that've had inflation rates above 10%. Presumably, there're no labor force shocks that can explain hyperinflations, such as we see in Venezuela. Even in your model, money matters in those situations.

    So why wouldn't that be true for inflation rates of less than 10%? At least part of it could be a low signal-to-noise problem.

    It's easy to believe that labor shocks can challenge central bankers, who use poor models in flawed regimes, and have a paucity of data besides.

    1. I still see the inflation π >> 10% regime as more of a political scenario (your economy has effectively collapsed into a single dimension described by a single parameter). But even in those situations, I don't think there are any cases where the central bank just raised interest rates and inflation abated. They're usually accompanied by currency revaluation or even major government change. So I'd imagine interest rates might be entirely unrelated to inflation changes — and might even fall before inflation starts to decline. For example, I could see Venezuela's interest rate falling if Guaidó became the de facto leader even before inflation fell.

      That is all to say it's not clear that when "money matters" for π >> 10% that money matters in a mechanical sense with scientific and empirically valid relationships between interest rates, money stock, inflation, growth, etc. It almost assuredly "matters" in a political/social sense but that will change from crisis to crisis and country to country. If people feel like the situation is being controlled (signalled by the central bank or political actors), inflation will abate. But there is no reason that will have a 1:1 correspondence with the magnitudes of the changes being made.


      As a side note: insufficient signal to noise is the same thing as not having a direct relationship (i.e. regressions fail for the same reason that low SNR foils a detector).

    2. This is a more interesting reply than expected. This is a unique blend of heterodox economics and sociology.

      This would seem to indicate you think liquid asset markets are quite inefficient, assuming you agree with common interpretations of market reactions to central bank policy announcements and other statements.

    3. It depends on what you mean by efficiency! In the traditional econ sense, sure this means markets are inefficient. But if markets circumscribe their state space narrowly while exploring all of it, they're efficient in an information equilibrium/MaxEnt sense.

      If I can choose anything, but I only choose from inside a small box that only I adhere to, that's "inefficient" from an info theory standpoint. But if we all decide that small box is the state space, then if we all stay in it that's "efficient". Even if they're invented constraints — if they apply to all agents, they're still constraints.

    4. Yes, so you would say markets are roughly as relatively efficient as conventional econ suggests, but less absolutely efficient. "Relatively" in this case refers to large, representative samples of the market or the market population versus less representative subsets.

  2. What about sectoral inflation? For instance, 3 sectors have experienced the highest rates relative to the economy. College, Medical Care, and Housing/Shelter.

    What do these 3 sectors have in common? Massive government intervention in the form of direct subsidies or in kind transfers (subsidized loans).

    Although, i do agree with you that general price level inflation is better explained by changes in the labor force I'm uncertain about the exact mechanism. This is however a feature of the data. Even during the great 'price revolutions' of the 15th century. This experience is often framed as a massive monetary expansion (gold silver increase due to colonization of America), but the data points to massive surges in population right around this point as well.

    Anyways, would like to get your thoughts on the above. Given the sectoral inflation and the laborforce story i suspect the mechanism is the traditional aggregate demand story. That is, govt intervention is creating demand for specific services and supply isn't elastic enough. The laborforce story would be the same, AD rising too quick relative to AS and maybe enhanced by an expectations channel.

    Lastly, I'll ask if you find any role for inflation targeting. The Fed has succeeded in anchoring expectations at 2%. This is evident in the amazing stability of InFEx but also in realized inflation. Do you think this 'anchoring' reduces the sensitivity of inflation to respond to employment/ad shocks?

    1. Regarding your last question first — to some degree you can predict Canada's and New Zealand's "about 2% inflation" from the 80s:

      There are two issues that always come up: 1) what do we mean by "2% inflation" (the countries that have set expectations all have around 1.6 to 1.8% inflation in the price indices that are targeted so that means up to 50 basis points error which is a massive 25% slop we're allowing for), and 2) inflation was by this standard "about 2%" before the 70s demographic surge.

      More to come ...

    2. Regarding the first part on inflation components, I wrote a post on this a couple years ago:

      But the general idea is that there is a distribution of price changes for the entire economy, and that distribution is set based on macroeconomic constraints (which can and do include government-based forces). However, individual prices within that distribution can change — sometimes by large amounts — while the properties of the macro distribution (e.g. average and variance of inflation) remain fairly constant.

      Like a gas at a given temperature in a container, some molecules are moving very fast, some slow, but the typical energy is ~ kT based on the constraint that the overall kinetic energy of the system isn't changing.

      (Maybe we could think of individual prices as kinetic energy of individual molecules E = 1/2 m v^2, but macro price level as a temperature which although related to kinetic energy E ~ kT is a different thing. With government involved, we would think of that as separate rooms in a house with some warmer than others but with an overall "equilibrium" distribution that's a sum of different local temperatures.)


Comments are welcome. Please see the Moderation and comment policy.

Also, try to avoid the use of dollar signs as they interfere with my setup of mathjax. I left it set up that way because I think this is funny for an economics blog. You can use € or £ instead.

Note: Only a member of this blog may post a comment.