Wednesday, December 21, 2011

Delusions of helplessness, monetarist edition


I'm very sorry to do this, but today I must hit Scott Sumner with the Bat Boy pic. Bat Boy is deployed whenever an econ blogger makes a claim that is truly batty. Which, to be honest, we do fairly frequently, since we have no editors and we often write while under the influence of various (legal) drugs such as iced tea, Ambien, and YouTube.

But anyway, via Brad DeLong, I find Scott Sumner making this batty claim:
Keynesian economists have never been able to accept my assertion that the fiscal multiplier is roughly zero because the Fed steers the (nominal) economy.
Translated roughly from Monetarese into English, this means: "If Congress tries to boost output by spending money, the Fed will counteract this effort by enacting tighter monetary policy. Thus, stimulus can never work."

Why is this a batty claim? Well, to see why, we must first identify the assumptions that would have to be true for the claim to hold. These are:

Assumption 1: The Fed can control the path of nominal spending (NGDP).

Assumption 2: The Fed does choose to control the path of nominal spending in a way that will cancel out any stimulus.

I admit to being dubious of the first assumption. I think that the Fed probably observes the factors that affect future nominal spending only with a lag, and that there is also a substantial unpredictable component of the effect of the Fed's actions, making it difficult for the Fed to steer nominal spending with precision. But this is not why I think that Sumner's claim is batty. Assumption 1 - that NGDP targeting could work - is something that is not obviously false. It's also something that Scott Sumner and many other smart bloggers say all the time.

The batty part is Assumption 2. This is an assumption about the Fed's "reaction function" - the way that the Fed actually does respond to changes in the economy. For stimulus to be ineffective, the reaction function has to cancel out any and all output changes that result from fiscal policy changes.

What kind of Fed policy would do this? Well, the Fed could target the growth rate of NGDP. Suppose that the Fed decides that NGDP should grow at 4% a year. Then a fiscal stimulus that tried to push NGDP growth up to 6% a year in the wake of a recession would cause the Fed to tighten (i.e. print less money), frustrating Congress and holding NGDP growth at 4%.

Alternatively, the Fed might target the level of NGDP. In this scenario, the Fed might not counteract fiscal stimulus, because stimulus after a recession might work to bring NGDP back to where the Fed wants it to be. However, in this world, the stimulus turns out to be completely unnecessary, because it's only doing what the Fed would do anyway; in the absence of stimulus, the Fed would print money and buy stuff until NGDP went back to pre-recession levels.

But now notice that there is one huge huge huge problem with either of these stories: If the Fed controls either the level or the growth rate of NGDP, where the heck did the recession come from in the first place?! If the Fed both can and does counteract shocks to the NGDP path, then recessions should never happen. In other words, if the positive demand shock of a stimulus must be canceled out by the Fed, then the negative demand shock of a recession should be canceled out as well!

But it isn't. Via DeLong, here is a graph of the recent path of nominal GDP:


As you can see, both the level and the growth rate experienced a massive swing in 2008. That swing was the Great Recession. It was not counteracted by the Fed.

So to believe Scott Sumner's claim about the powerlessness of fiscal stimulus, you must believe that the Fed can and does counteract stimulus, but either can't or chooses not to counteract recessions. Essentially, you must either believe that the Fed's powers of stabilization are severely limited (which Scott Sumner probably does not believe, given everything he writes), or that the Fed wants to torpedo the U.S. economy.

(Update/Aside: This last item deserves more explanation. It is theoretically possible that the Fed targets the NGDP growth rate, but occasionally makes mistakes, and never tries to correct its past mistakes. So when there are big recessions, the Fed simply lets them happen, and then actively prevents recoveries from returning us to our previous NGDP trend line. Since recessions are more abrupt than booms, this means the Fed is actively out to torpedo the U.S. economy. Now maybe this is true - if the Fed has a bizarre nonlinear hard-money bias that manifests more in recessions than booms, it could be true! - but it would mean that past recessions would have manifested as unit-root drops in output...in other words, a bunch of L-shaped recessions in the past. Most people think that that isn't what we've seen; that after past recessions, output has returned to trend, and that drops in output were not "frozen in" by a deranged Fed. Anyway, now back to your regularly scheduled ranting...)

That is why the claim is a bit batty. If the Fed is truly omnipotent, then we shouldn't see any recessions, or any calls for fiscal stimulus in the first place. The fact that we're even having this debate makes the "helplessness" position untenable.

Incidentally, this is not the first time I've seen Sumner make this claim. In a post on trade policy back in October, he wrote:
If the Fed follows its announced policy of inflation targeting, the contractionary effect of China’s [decision to float its currency] on world output will not be offset by any expansionary effects on the US.  
This is exactly the same idea. If the Fed both can and does control output, no (demand-side) policy other than a change in the Fed's policy rule can ever have an effect on output. It doesn't square with the very real fact of recessions.

But here's something else I've noticed - this "helplessness" perspective kind of clashes with Scott Sumner's usual line. Sumner spends a lot of time arguing that the economy would be OK if the Fed would just target NGDP. But if the Fed already effectively steers the path of nominal output - so effectively that stimulus and trade policy are ineffectual - then what is Sumner complaining about?

So anyway, I'm sorry Bat Boy had to be trotted out. I am sure it will be me making a batty claim sometime soon, and I hope someone out there cares enough to point it out. Speaking of which, I need some more iced tea...


Update: I wrote this in a comment and thought I should move it up to the main post:

Really what it comes down to is this: If you believe that the Fed can move around the NGDP path at will, it is possible to postulate a reaction function such that the Fed will always cancel out any fiscal stimulus. But to assert that such a reaction function does exist is a claim that has little or no evidence to back it up. And to assert that such a reaction function logically must exist, simply because it can, is nonsense. Hence, Bat Boy.

I'm not sure if this last, strongest, and most preposterous claim is the claim that Scott Sumner is making...it sounds like it might be. But even if it's simply an empirical claim that in practice the Fed does try to counteract stimulus, then I want to see the evidence.

Update: I may have overinterpreted Sumner's claim. If so, the Bat Boy is hereby revoked. This is important, since Bat Boy is only used when theoretical claims are made that are clearly false, and hence is not to be summoned lightly...

27 comments:

  1. What matters is not actual output but expected output. Monetary policy affects output with a lag. Therefore, if everyone (including the Fed) overestimates future output AND the Fed acts to fully offset the effects of fiscal policy on EXPECTED output, there will still be a recession AND the fiscal multiplier will still be zero. There is no contradiction between the fact of recessions and the premise that the Fed fully offsets fiscal policy.

    In fact the recent recession is quite consistent with a Fed that is targeting the growth rate of NGDP and made one big mistake, which it subsequently (in accordance with the policy of growth rate targeting) refused to correct. Presumably if the Bush administration had (presciently but far out of consensus) pushed through a fiscal stimulus the size of ARRA in mid-2008, the Fed would have fully offset it, and we would still have had the Great Recession.

    (Mind you, I don't think it's very plausible that the Fed is actually targeting the NGDP growth rate, but it isn't contradicted by the data.)

    ReplyDelete
  2. So if you believe that A) the Fed targets the NGDP growth rate, B) the Fed occasionally makes big mistakes and refuses to correct them, and C) the Fed acts to counteract the effect of any stimulus that would act in opposition to the Fed's past mistake, then yes, Sumner's argument is not a logical fallacy. And in fact, I kind of got the sense that this was the argument he was making. But to me, that seems like the idea that the Fed is out to screw the American economy. And it seems inconsistent to call for the Fed to adopt an NGDP level target (which, presumably, would correct past mistakes of the type being discussed) but refuse to call for the Fed to simply allow fiscal stimulus to have an effect.

    What it is, is assuming a set reaction function when discussing policies you (i.e. Sumner) don't like, but then arguing for changing the reaction function. This seems to embed an implicit anti-stimulus bias that is (so far) unexplained.

    ReplyDelete
  3. Andy, I added a lengthy aside, just for you! :)

    ReplyDelete
  4. "it would mean that past recessions would have manifested as unit-root drops in output"

    Not necessarily. Nobody suggests that the Fed is targeting the RGDP growth rate. Recessions over the past 30 years have been associated with declines in the inflation rate, and after each such decline, the Fed allowed an increase in the RGDP growth rate to offset the decline in the inflation rate.

    The policy of targeting the NGDP growth rate, while I agree that it is a bad idea, is not far out of the mainstream: if I'm not mistaken, it is what Bennett McCallum still advocates. The Fed may actually (though I doubt it) have been following such a policy over the past 30 years, and until 2008 it worked out pretty well, specifically because the Fed always ultimately preferred a lower inflation rate than the one it had initially faced. Unfortunately, it managed to get the inflation rate down to its ultimate target and then apparently continued following the same policy (again, I doubt it was actually doing so intentionally, but it's consistent with the data).

    Contrary to Occam's razor, I do tend to believe the explanation that the Fed's earlier behavior was the result of an opportunistic disinflation policy, while its recent behavior has been the result of an aversion to unconventional policy measures. However, the simpler explanation would be that it has been targeting the NGDP growth rate all along.

    I would also note that, last I heard, Greg Mankiw was still making the case that there is a unit root in RGDP.

    ReplyDelete
  5. Re: Mankiw - Yes, I know he thinks there is a unit root in output, which is why I said "most"... ;)

    Re: past recessions - Have people concluded that there is a unit root in the level of detrended NGDP??

    Really what it comes down to is this: If you believe that the Fed can move around the NGDP path at will, it is possible to postulate a reaction function such that the Fed will always cancel out any fiscal stimulus. But to assert that such a reaction function does exist is a claim that has little or no evidence to back it up. And to assert that such a reaction function logically must exist, simply because it can, is nonsense. Hence, Bat Boy.

    ReplyDelete
  6. "Have people concluded that there is a unit root in the level of detrended NGDP??"

    I don't know if people have concluded that, but based on a quick inspection, it seems to fit the post-1980 data pretty well. It certainly fits better than a constant, mean-reverting trend in NGDP. So the parsimonious interpretation is that the Fed is targeting NGDP growth. The alternative interpretation is that the there is a mean-reverting trend but the rate of trend growth keeps declining. If you want to make that interpretation, you need to offer specific evidence, or you are arguing against Occam's Razor. The prima facie evidence supports the view that the Fed is targeting NGDP growth, which, if it is true, implies that the Fed will offset fiscal policy.

    ReplyDelete
  7. I'm actually starting to convince myself that Scott Sumner is right about this.

    ReplyDelete
  8. So you think all the time Sumner spends plumping for NGDP targeting is wasted effort, since we already do it? ;)

    ReplyDelete
  9. @Andy: You're talking about this data series, right?
    http://research.stlouisfed.org/fred2/graph/?s[1][id]=GDP

    Did you do some unit-root tests on it?

    ReplyDelete
  10. Scott Sumner favors NGDP level targeting, which is a much more sensible policy than growth rate targeting.

    I haven't done any explicit unit root tests, but plot the past 30 years on a log scale and look closely at the recessions. They look like one-time shifts between lines that are approximately parallel.

    ReplyDelete
  11. Like I said, Andy, if you believe the Fed has infinite traction, you can ascribe any fluctuations in nominal GDP to the Fed's unobservable reaction function. I want some good statistical analysis showing that the Fed acts to counter the NGDP impact of fiscal policy (but NOT recessions) before I will seriously entertain the notion... ;)

    ReplyDelete
  12. Noah,

    For some empirical evidence on what the Fed did during the Great Moderation see this paper by Josh Hendrickson: http://mpra.ub.uni-muenchen.de/20346/

    And here I am siding with Sumner on the size of the fiscal multiplier: http://macromarketmusings.blogspot.com/2011/09/how-big-is-fiscal-multiplier.html

    ReplyDelete
  13. Thanks, David, I'll check those out!

    ReplyDelete
  14. Anonymous6:12 AM

    If a third of the corporate profits were from the financial sector and half of this sector did not produce anything useful in hindshight, does it really make sense to blow up the money supply to correct for this decline?

    This just reinforces the wrong incentives. And I do not think that another bubble really helps right now.

    ReplyDelete
  15. Here's a statistical test for you: take NGDP from 1981Q3 to 2011Q2 (30 years of data). Regress ln(NGDP) on a time trend and one lag of itself. The coefficient on the time trend is precisely zero (SE=.0002). The coefficient on the lagged dependent variable is precisely one (SE=.014). I don't recall what the critical values for this test are, but it seems like a pretty strong indication. If I add 3 additional lags, the sum of the lag coefficients is still one, and the trend coefficient is still zero. Also run with annual data, one lag and three lags, with similar (though obviously less precise) conclusions. Also tried ending the quarterly series in 2007Q4 (to make sure the recent episode wasn't critical to the conclusion). Results were broadly similar, though less extreme: the coefficient on the time trend is 0.00047 with a SE of 0.00026, and the coefficient on the lagged DV is 0.963 with a SE of 0.018. Again, I don't recall what the critical values are, but it still looks like a unit root to me.

    ReplyDelete
  16. OK, I went into EViews so I could get critical values and a standardized default unit root test. The default Augmented Dickey Fuller Test on ln(NGDP) 1981q3 to 2007q4, gives a test statistic (for the null hypothesis of a unit root) equal to 1.65, compared to a 10% critical level of 2.58. This is generally robust to changes in the sample endpoints, number of lags, and inclusion of a time trend for the first differences, although I can get it to reject if I stack the deck by starting near a business cycle trough and ending near a peak. (Compare my test to the ADF test on ln(RGDP) for the same sample, which rejects at the 10% level.) Conclusion: yes, Virginia, there is a unit root.

    ReplyDelete
  17. The majority of economists, I think, agree that the fiscal multiplier is near zero when the zero lower bound is not an issue: the Fed can easily offset fiscal policy under these circumstances, and there is no reason it would not choose to do so. The question is whether this is still true at the zero lower bound. If ARRA had produced a dramatic recovery that had broken the pattern of the unit root in NGDP, this would have been evidence of a positive fiscal multiplier. But that dog didn't bark. What would have happened without ARRA? Surely the Fed would have pursued a more aggressive QE1 than it did. How much more aggressive? I don't know. My offhand guess is that it would not have been aggressive enough to compensate for the absence of ARRA. But Scott Sumner makes a different guess, and frankly, the statistical evidence does not support my guess.

    ReplyDelete
  18. Here is anecdotal evidence that the Fed "has a bizarre nonlinear hard-money bias that manifests more in recessions than booms:"

    http://www.federalreserve.gov/newsevents/press/monetary/20111213a.htm

    Paraphrased: "We expect slow growth, high unemployment, and low inflation, but we're not going to do anything about it."

    ReplyDelete
  19. Thanks for doing that, Andy. Very interesting. I actually intend to follow up on this, research-wise, now. If I can convince myself that the Fed is actually nuts enough to allow every recession and then suppress every recovery (while suppressing booms, since booms happen more slowly than recessions), I will revoke the Bat Boy from Scott Sumner and award 87,000 Bat Boys to the Fed itself...

    ReplyDelete
  20. That said, note that in the Great Recession NGDP fell for several consecutive quarters. If the Fed was willing and able to return NGDP to some target growth rate after a surprise, it would only allow a one-quarter fall in NGDP before it enforced a return to the target growth rate. Unless the fact that negative shocks to NGDP were correlated was itself the surprise, and kept surprising the Fed for 3 quarters in a row.

    See, this is the problem with trying to estimate a reaction function with the same data set that you use to verify the models that explain how the Fed gets traction in the first place...

    ReplyDelete
  21. There are data lags, policy implementation lags, and policy response lags, which can add up to quite a long lag, and surely, on September 14, 2008, nine months into the Great Recession, the Fed would not have forecast the decline in NGDP that happened over the subsequent nine months. In December, even January, IIRC people were still far underestimating how deep the recession would be. (Remember the famous with/without stimulus chart that the incoming administration produced.)

    ReplyDelete
  22. So, this just introduces more degrees of freedom...if we use the Fed's knowledge AND its reaction function as free parameters, how can these time series ever refute the hypothesis that the Fed intentionally suppresses recoveries? Unless we get more data, this theory is not-even-wrong. What kind of data would allow us to disentangle the reaction function from the Fed's forecasting accuracy? Event studies, I suppose. When I get some time (and a job), I'd be interested in writing a paper exploring the hypothesis of an "evil Fed"...

    ReplyDelete
  23. Steve M.6:05 PM

    Not an economist here, so a bit at sea; but I thought I'd pose a question anyhow. If there is the sort of reaction that Sumner is pressing why has Bernanke used many opportunities over an extended period of time to call for more fiscal action in fighting the slump? Only to negate its potential effect? Is his motive purely political?

    ReplyDelete
  24. Phrases like "intentionally suppresses" and "evil Fed" are misleading. (Admittedly they are consistent with Scott Sumner's own criterion of monetary policy in which any reduction in expected NGDP relative to trend constitutes a tightening, but not with the way most people think about ease and tightness of monetary policy.) There has never been (and is not likely ever to be) an example of a fiscal response so strong that the Fed didn't have to ease (in the conventional sense, not in the Sumner sense) during a recovery. The hypothesis is that the Fed completely passes the buck to fiscal policy when fiscal policy is used. It doesn't (in its own perception or by the conventional definition) deliberately tighten to offset intentional fiscal stimulus; it merely fails to ease as aggressively as it otherwise would, so that fiscal policy is ineffective because it substitutes for monetary policy actions that would otherwise have happened.

    It seems to me that, if one could reject the unit root hypothesis for nominal GDP, it would refute this theory. We could then conclude that the failure of NGDP to recover in the recent episode was a result of the Fed's limitations rather than its intentions. It could also perhaps be refuted by looking at RGDP and inflation separately, or by looking at productivity, employment, and demographics, or some such indicators of potential output. But that gets complicated.

    ReplyDelete
  25. Steve,

    The argument is that the Fed prefers, other things equal, not to have to do a lot of unconventional policy, so it would prefer that fiscal policy be used to accomplish its objectives. But if fiscal policy is not used, the Fed will use its own policies to accomplish those objectives. Thus fiscal policy helps make the Fed more comfortable, but it doesn't change the macroeconomic outcome.

    ReplyDelete
  26. Anonymous6:42 PM

    I think the correct interpretation is that the Fed can control MV (i.e. NDGP) but that for political reasons there is a upper bound (increasing in relationship to the unemployment rate) on how high they are willing to let M go.

    ReplyDelete
  27. Noah,

    I don't agree with your personification of the Fed's actions as sabotaging fiscal stimulus. There is no malice here, just math: any credible commitment to stabilizing inflation is a commitment to a fiscal multiplier of zero.

    The chain of logic here is really quite simple: the driving theory of demand-side stimulus is the paradox of thrift, which is to say that a real variable (employment) is affected by a purely nominal mismatch in the supply and demand for money. Fiscal stimulus and monetary stimulus act to the same end: they drive inflation up and so quench the excess demand for money that's creating unemployment.

    Since the target of both actions is the same, of course one offsets the other, but fiscal policy is implemented very slowly, and therefore, it is always the Fed and Fed's forecast of the impact of fiscal policy which ultimately controls.

    ReplyDelete