Why does money matter?

Everyone but a few crazy post-Keynesians acknowledge it matters for NGDP. Everyone except a few crazy PKs and real business cycle types agrees it matters for RGDP in the short run.  The question is why?  Here are some theories:

1.  Tight money raises interest rates–which leads to less output.

2.  Tight money reduces the amount of the medium of exchange, which leads to less exchange and less output.

3.  Tight money raises the ratio of hourly wages to per capita NGDP, which reduces employment and output.

Nick Rowe thinks the second mechanism is especially important, whereas I focus on the third.  Most people like the first, but I’m not even going to bother with that one.   Here’s Nick:

The unemployed hairdresser wants her nails done. The unemployed manicurist wants a massage. The unemployed masseuse wants a haircut. If a 3-way barter deal were easy to arrange, they would do it, and would not be unemployed. There is a mutually advantageous exchange that is not happening. Keynesian unemployment assumes a short-run equilibrium with haircuts, massages, and manicures lying on the sidewalk going to waste. Why don’t they pick them up? It’s not that the unemployed don’t know where to buy what they want to buy.

If barter were easy, this couldn’t happen. All three would agree to the mutually-improving 3-way barter deal. Even sticky prices couldn’t stop this happening. If all three women have set their prices 10% too high, their relative prices are still exactly right for the barter deal. Each sells her overpriced services in exchange for the other’s overpriced services….

The unemployed hairdresser is more than willing to give up her labour in exchange for a manicure, at the set prices, but is not willing to give up her money in exchange for a manicure. Same for the other two unemployed women. That’s why they are unemployed. They won’t spend their money.

Alex Tabarrok has a very intriguing post that tries to find evidence for the “medium of exchange” theory of money’s importance.  He finds that barter increased sharply during the Great Depression:

Rowe’s explanation put me in mind of a test. Barter is a solution to Keynesian unemployment but not to “RBC unemployment” which, since it is based on real factors, would also occur in a barter economy. So does barter increase during recessions?

There was a huge increase in barter and exchange associations during the Great Depression with hundreds of spontaneously formed groups across the country such as California’s Unemployed Exchange Association (U.X.A.). These barter groups covered perhaps as many as a million workers at their peak.

That evidence seems to support Nick’s view.  Is there any way it could be consistent with my sticky-wage model?  Perhaps in the Great Depression many workers became so poor that they could not afford certain items unless they were first able to earn some “money” (actually wealth) by selling something they could produce.  Barter was a way of affording something that you otherwise could not afford.  (And maybe saving a bit on taxes.)

My theory is more ad hoc than Nick’s, but does have one implication.  In a modern recession you’d expect to see much less barter, as workers are not nearly as poor.  Most people in American have enough “money” (i.e. media of exchange) to buy something if they want it.  What usually holds them back in recessions is not the inability to find some media of exchange, but a lack of income or wealth.  Because we usually have enough media of exchange to make purchases, if wealth is much higher than the 1930s you’d expect barter to be much rarer.  And that is what Alex finds:

What about today? Unfortunately, the IRS doesn’t keep statistics on barter (although barterers are supposed to report the value of barter exchanges).  Google Trends shows an increase in searches for barter in 2008-2009 but the increase is small. Some reports say that barter is up but these are isolated, I don’t see the systematic increase we saw during the Great Depression. I find this somewhat surprising as the internet and barter algorithms have made barter easier.

So I think the evidence cuts both ways; and we are still left with two plausible hypotheses for why money matters.  (I’m calling the first one listed above implausible.)  It’s likely that both sticky wages/prices and a monetary economy are necessary conditions for money non-neutrality; with barter you can’t have the sort of wage-stickiness that would cause unemployment.  So it’s hard to tell which is the most important.

Perhaps I should say a few words about why I think sticky wages matter, and why other economists don’t agree.  The sticky-wage theory is usually assumed to imply that real wages should be countercyclical, at least during demand shocks.  I think they are somewhat countercyclical during demand shocks (Sumner and Silver, JPE, 1989), but the evidence is unclear because we lack good data for either hourly nominal wages or the price level.

The biggest drop in the monetary base in the past 100 years was in 1920-21.  That was also the biggest drop in the WPI in the past 100 years.  And that was also the biggest increase in hourly manufacturing wages divided by the WPI in the past 100 years.  Coincidence?  Maybe, but when the data is awful I look for shocks so enormous that even with awful data the underlying relationships should be visible.  For me that’s 1920-21, and also 1929-1939.  And both periods suggest that sticky wages are a problem.

The modern data is much less clear.  In my view the right way to define real wages is nominal wages over per capita NGDP.  Since nominal wages are sluggish, and NGDP is highly cyclical, that definition would make “real” (actually relative) wages look countercyclical.  Unfortunately, other economists would argue that this pattern doesn’t prove much, as it would be expected under almost any business cycle theory.  So we are back to square one.  I think sticky wages are the key to money non-neutrality, but can’t prove it to those who use more conventional models.

In the comments someone recently argued that I don’t rely on formal models.  Yes I do, but like Milton Friedman I like ad hoc partial equilibrium models, not the DSGE models that are so popular today.  One model for output and employment, another for monetary policy and NGDP, another for interest rates, another for exchange rates, etc.  Yes, you can try to put them all together in a general equilibrium model, but I don’t think macroeconomists know enough for those large abstract models to be useful.  The world is too complicated.  Better to stick with what we do know.

BTW, don’t confuse the issue of real wage stickiness and the natural rate of unemployment, with nominal wage stickiness and cyclical unemployment.  For instance this post Tyler Cowen discusses evidence that has a bearing on the real wage/structural unemployment issue, but no bearing on nominal stickiness and cyclical unemployment.  (That’s not to say real and nominal stickiness cannot interact to make the problem worse, they almost certainly do.)

Monetarism >>>> New Keynesianism (and Paul Krugman tells us why)

My approach to macro has always been “nominal shocks have real effects.”  As a result, I’ve never cared for the Keynesian view that inflation was caused by an overheating economy, and disinflation was caused by economic “slack.”  If that were true, then it would be the case that “real shocks have nominal effects.”  I.e. changes in real GDP affect inflation.  And that just didn’t seem right.

Milton Friedman (and Irving Fisher) interpreted the Phillips curve in the same way I do—deflation causes unemployment, not the other way around.  As I got older I modified this view slightly, now I see nominal shocks as changes in NGDP growth.  But I still believe nominal shocks have real effects.  Only now an unexpected change in NGDP would simultaneously change both prices and output.  So how does this differ from the Keynesian view?

Mankiw and Reis (2006) identify “three key facts” of modern business cycle theory:

  1. The Acceleration Phenomenon . . . inflation tends to rise when the economy is booming and falls when economic activity is depressed.”
  2. The Smoothness of Real Wages . . . real wages do not fluctuate as much as labor productivity.”
  3. Gradual Response of Real Variables . . . The full impact of shocks is usually felt only after several quarters.”

None of these facts even comes close to fitting the economy of 1933, and the first is especially strongly rejected.  Inflation soared in 1933, when unemployment was at the highest level in history.  The monetarist model, however, can easily explain 1933.  After all, a powerful nominal shock (dollar devaluation) sharply raised output.  But I get tired of talking about 1933, unless there are other good examples then the distinction I’m trying to make might not be that important.  Fortunately, Paul Krugman has found some recent examples:

Via Mark Thoma, the Dallas Fed notes some acceleration even in its core inflation measures, although these remain below target. What should we make of this?  .  .  .

What this suggests to me, anyway, is that there’s a rate-of-change effect as well as a level effect: when the economy is growing, even from a low base, some firms gain pricing power, and some firms raise their wage offers a bit. So the acceleration of US growth and the fall in unemployment since last summer has produced an inflation uptick; if past experience is a good guide, however, this will be only temporary unless the economy continues to accelerate.

I’m sure if you add enough epicycles to the Keynesian model you can make it all work out.  For my part I’ll stick with Fisher and Friedman—nominal shocks have real effects.

PS.  Check out Krugman’s short post.  It’ll make more sense if you look at his graphs.  I don’t like to copy entire posts, if I can avoid it.

PPS.  Krugman refers to an “acceleration of US growth and the fall in unemployment since last summer”.  Did QE cause that acceleration?  Martin Feldstein says yes.  Wasn’t he one of those economists who said the Fed was out of ammo back in 2008-09?  Any help from commenters would be appreciated.

Crises can’t be forecast

Here’s a graph from a recent David Beckworth post:

Spot commodity prices can’t be forecast (EMH.)  Spot commodity prices are highly correlated with the business cycle.  Ergo, the business cycle cannot be forecast.  This is in response to a recent Tyler Cowen post:

Raghu Rajan nails it:

“I would argue that three factors largely explain our collective failure: specialization, the difficulty of forecasting, and the disengagement of much of the profession from the real world.”

Rajan’s second point is exactly right.  The other two are completely irrelevant.  Someone please tell the Queen.