Will VAR studies become like yesterday’s newspapers?

When you read old economic journals you come across lots of empirical studies of things that no longer interest us.  In the interwar period there are lots of studies of the world gold market; estimates of newly-mined gold, industrial use, dishoarding from the Indian subcontinent, etc.  I also seem to recall lots of studies of money demand being published in the 1980s; money demand in Turkey, money demand in South Korea, etc.  My impression is that people are no longer interested in those studies.  They are reread about as often as yesterday’s newspapers.  I wonder whether the same will be true of recent macro studies using techniques such as vector autoregression (VAR.)

[By the way, yesterday’s newspapers can actually be quite interesting.  I spent a decade reading every New York Times from the 1930s.  History looks very different in the “first draft.”]

Let me say upfront that I am going way out on a limb here, as I don’t really keep up with this literature, and hence my comments may already be out of date.  I hope you guys will correct me where I am wrong.  But I do think that I may at least have a few novel reasons for expecting macro to soon move on.

If I am not mistaken, VAR models try to find inter-temporal correlations between interesting macro variables, often including policy shocks.  The first time I heard about these studies there was mention of a “price puzzle,” which meant that monetary shocks seemed to have a “perverse” effect on the future price level.  Easy money led to lower inflation, and vice versa.  Let’s stop right there.

Suppose you want to test a model.  You collect the data that you think is most appropriate, and then you set up the statistical model that best tests your theory.  And suppose the test fails to confirm your theory.  Also assume it isn’t a borderline failure, but rather an abject failure.  What should you do?  If the data doesn’t suggest a model that directly contradicts yours first model, I think you have an ethical obligation to just drop it and move on to something else.  Of course I do know that many people are stubborn, or are selfish, and play around with different variables and different model specifications until they get the “right” result.  What is the right result?  It’s the one that you believe.  (See my earlier post on how people assume the universe was constructed to confirm their biases.)  Or if you are very cynical, then you might just want publications and tenure (although that begs the question of why you went into a low paid field like economic research.)

Let’s go back to the VAR studies.  When those first studies showed a persistent price puzzle the whole enterprise should have been dropped.  As I said I am not an expert here, but my guess is that the entire VAR approach is deeply flawed, at least in the area that I am interested in—monetary policy.  My hunch is that the price puzzle comes from a failure to correctly specify monetary shocks, aka the identification problem.  To anyone who has followed this blog, I don’t think this hunch will come as a big surprise.  I have strongly argued that there was a very contractionary monetary shock in the second half of 2008.  I know very few economists (other than Earl Thompson) who look at things this way.  The two most common ways of identifying monetary shocks (interest rates and the monetary base) both gave a highly expansionary reading in late 2008 (and 1932 as well.)

But of course economists did not discard the VAR approach when the initial studies came back with highly questionable findings.  Rather they kept trying new specifications (model mining?) until the tortured data finally yielded the “right” answer.  Did any of these models provide any useful advice to the Fed in the recent crisis?

Today we don’t care much about interwar gold studies because we no longer have a gold standard.  We don’t much care about money demand studies because fewer and fewer economists favor targeting the monetary aggregates.  I am suggesting that if we move away from backward-looking Taylor rules, and I think we will, then no one will much care about VAR studies of monetary shocks.

I have argued that the proper way of thinking about monetary policy is to assume that expansionary shocks occur when the market forecast of the goal variable (such as inflation or NGDP) exceeds the target, and contractionary shocks occur when forecasts of the goal variable fall short of the target.  From this perspective, something like the TIPS spread might become the monetary policy indicator, replacing conventional measures such as interest rates and the monetary aggregates.

I think when other economists hear about my proposal they have a sort of visceral reaction that by equating the policy indicator and goal, I am simply assuming away the problem of how to get from here to there.  The standard way of thinking about monetary policy is where the central bank sets the controls, and then they sit back for 6 months and pray that the economy moves in the desired direction.  In my proposal, there is no “wait and see,” as you immediately know if your setting is optimal.  Or to be more precise, it is optimal if we assume that the market forecast is optimal.

Under that kind of policy what sort of variables should we look at in a VAR study?  It seems to me that the interesting questions will all revolve around the relationship between observable nominal aggregate forecasts and the unobservable social welfare function.  If we have policy futures markets then all the transmission mechanism questions become unimportant, the important questions are about social welfare—are we better off with a stable inflation rate or a stable NGDP growth rate, level targeting or growth rate targeting, one year futures targeting or two years futures targeting, stable prices or 2% inflation?  And I have no idea whether VAR studies can answer those sorts of questions.

The logic behind policy futures markets is so powerful that I believe it is only a matter of time before the older approaches to monetary policy fade away.  Twentieth century monetary studies won’t simply be unread, that’s an easy prediction, but they will almost seem unreadable.  They will seem almost superstitious, like finance models before the EMH was discovered.

Here’s an example.  It is only an accident of history that we don’t already have an NGDP futures market.  There was a CPI futures market as far back as 1985.  If there had been an NGDP futures market last September, I think it would have showed very low expected NGDP growth, probably below 3% and falling fast.  That would have put great pressure on the Fed to move aggressively at their meeting on the 16th.  Instead they did nothing.  They can’t keep making these errors without more and more people noticing.  Not everyone accepts my view of the low TIPS spread last fall, but you can be sure that if it happens again, people are going to be saying “last time it that NGDP growth fell off a cliff. . .”  The human brain is very good at noticing patterns.

In a recent diablog, Robert Wright asked Tyler Cowen what he thinks about God.  Tyler said something to the effect that “the issue isn’t what I think of God, it’s what God thinks of me.”  (I hope that’s not too far off.)  The major central banks still may not think much of the forecasts embedded in financial markets.  But it would be in their interest to start paying attention.  In the long run what matters won’t be what the Fed thinks of futures markets, it will be what the futures markets think of the Fed.

Off topic:  Any film buffs out there?  After his previous fiasco, I almost didn’t see Coppola’s latest film, Tetro.  I’d be interested in the reactions of others.  On the one hand it seems like a must-see film.  But it also seems too derivative; too many scenes reminded me of other films, or other actors.  Did Coppola try to do too much?  Did he lack a powerful overarching vision to tie it together?  How can a film that is so well-crafted in so many ways leave such a blah impression?  Am I so jaded that all I care about is originality?


Tags:

 
 
 

12 Responses to “Will VAR studies become like yesterday’s newspapers?”

  1. Gravatar of Joshua Joshua
    21. June 2009 at 12:45

    It looks like some spammer has hijacked your RSS feed, since the feed for this post does nothing about babble about spammers’ favorite medicine…

  2. Gravatar of Angelo Angelo
    21. June 2009 at 12:55

    Hi Scott,

    Long time reader, but first time commenting here.

    About your questions on the VAR literature, I think they will take a route like Christiano, Eichenbaum and Evans’ (2005, JPE) paper, serving more as an empirical counterpart to validate theoretical models. There is a long way to go in order to measure, for instance, the effects of fiscal policy shocks, and a VAR is one possible way to do that.

    In this sense, it does not make a lot of sense measuring the effects of alternative policies in a VAR: if it’s only possible recover past shocks in a VAR, the experiment of welfare effects of alternative policies must be made on the theoretical model, and not on its empirical counterpart.

    Hope I made myself clear.

    Best!

  3. Gravatar of ssumner ssumner
    21. June 2009 at 13:55

    Joshua, Yes, I’ve had problems for several days. I will try to get it fixed on Monday. Viagra should be paying me for all the free advertising I am giving them.

    Angelo, Thanks for the comment. Fiscal shocks can take many forms. In my view fiscal policy will no longer be viewed as a stabilization tool once monetary policy starts targeting the forecast. In that case VARs won’t be able to estimate the multiplier, because there will be no multiplier (or it will be zero.) But fiscal policy can affect all sorts of other variables, such as real interest rates and exchange rates. In some ways it might be easier to estimate those effects if monetary policy has already stabilized expected NGDP growth. I.e., if you take monetary shocks off the table, it might be easier to estimate the impact of fiscal policy on variables other than NGDP.

    I don’t know enough to have an intelligent opinion on your second paragraph, but my sense is that theory will always be important when we address questions of social welfare in macro. Or at least it will be important for a long, long time.

    One issue I didn’t mention is that policy is always improving on past mistakes. If the term “shock” in some sense denotes “mistakes,” then past data may not be very helpful, as you suggest. Thus one reason money supply shocks don’t seem to explain much anymore is that the Fed has seen the folly of allowing exogenous monetary shocks to impact AD.

  4. Gravatar of David Landry David Landry
    21. June 2009 at 14:22

    Scott,

    You use the word “embedded” in the last paragraph not about Tetro to describe forecasts produced by markets. I take this to mean “objective” indicators like TIPS spreads or the yield curve. However, I’m curious to know what you think of private forecasts made by private forecasting professionals. In a widely cited paper by Romer and Romer (2000), it was found that the Fed’s forecasts, particularly for inflation, were vastly superior to those of these professional private forecasters. So what I’m wondering is this: If “embedded” market forecasts are public information, why did both the Fed and professsional private forecasters not utilize them in this crisis. This is of course assuming that professioanl private forecasters did not predict this crisis (feel free to educate me on whether this was the case), but if they did, is it really necessarily negligent of the Fed to ignore them if the Fed has consistently been more right than them over the past couple decades? I’d love to see you write a post sometime fitting together the implications of forecasts embedded in markets, forecasts by the Fed, forecasts by private professionals, and policy transparency. Particularly, what added information can forecasts provide if all relevant information is embedded in market prices? It would seem that if weak market efficiency holds, only holders of private information can provide useful additional information in forecasts.

    David Landry

  5. Gravatar of JimP JimP
    21. June 2009 at 17:08

    Well here at least is one hard money guy accusing the Fed of making the same mistake they made in the 1930s – aiming at deflation. Would that someone at the Fed were listening. But they aren’t.

    http://www.telegraph.co.uk/finance/comment/ambroseevans_pritchard/5586043/Dont-believe-the-hyperinflation-hype—dare-to-make-cuts.html

    to quote from the article:

    We know that the Fed’s balance sheet has exploded (to $2.07 trillion), but that is only half the story. Data from the St Louis Fed shows that the “monetary multiplier” has collapsed from a decade-average of 1.6 to the depths of 0.893. The “velocity” of money has slowed to a crawl.

    snip

    Tim Congdon – a hard-money Friedmanite from International Monetary Research – says the Fed is still not easing enough, perhaps because it is spooked by so much criticism or faces a mutiny by its own hawks. “If Ben Bernanke and his officials are listening to this sort of stuff and taking it seriously, they are making the same mistake as the Fed in the early 1930s,” he said. The US “output gap” is near 7pc. That is a powerful lid on inflation.

  6. Gravatar of JimP JimP
    21. June 2009 at 17:15

    The Fed is listening to that utter moron Larry Kudlow – and Laffer (a laugh, that guy) – and Jimmy Rodgers – who prefers the wonders of China – and to Jim Grant – for whom the one right thing always is to buy gold. Economic moralists – all of whom figure that we deserve what we are getting and we should get a whole lot more of it.

  7. Gravatar of Josh Josh
    21. June 2009 at 18:25

    Scott,

    The two puzzles were as follows:

    — Liquidity puzzle: the interest rate immediately rose following a monetary shock (think of an instantaneous Fisher effect).

    — Price puzzle: the price level would initially decline before rising following a monetary shock.

    These problems were solved rather easily. First, the liquidity puzzle could be solved by using base money or the federal funds rate as one’s monetary target. Second, the price puzzle was solved by placing cointegration restrictions on the appropriate variables.

    I do agree that it is questionable why we would give so much credence to models that predict something that we don’t really believe. However, I don’t think that VAR studies as a group will be left in dustbin of history as they do provide a useful guide for testing the implications of the theoretical models (as another commenter pointed out). Nevertheless, we must differentiate between estimated models that are derived from economic theory and those that are not. I think that the latter of these are destined for the dustbin.

  8. Gravatar of tom s. tom s.
    21. June 2009 at 18:56

    “…the Fed has seen the folly of allowing exogenous monetary shocks to impact AD.” In other posts you have spoken of the monetary base as endogenous. Can you elaborate on this exo/endo distinction? I’ll bet there are other readers like me who aren’t clear on this concept. Thanks.

  9. Gravatar of Clipping matinal « De Gustibus Non Est Disputandum Clipping matinal « De Gustibus Non Est Disputandum
    22. June 2009 at 00:58

    […] O fim dos vetores auto-regressivos? […]

  10. Gravatar of ssumner ssumner
    22. June 2009 at 04:39

    David Landry,

    Good question. I am not surprised that the private consensus forecast does rather poorly. In the last couple years it has looked to me like monthly changes in the consensus forecast are serially correlated. If there are any experts out there I would be interested in knowing if I am right. Serial correlation might occur if some of the individual forecasters do not take a fresh look at the economics outlook each month. I would expect market forecasts to be much better, and they probably are. The Fed devotes a lot of resources to forecasting, so they also may be more efficient than the private consensus.

    Although I think the market forecast is likely to be superior to the Fed’s forecast, the real advantage lies in it’s conditional forecasts. I don’t think the Fed is very good as estimating the effect of different instrument settings on forecasts during periods of great instability. Example; last October NGDP growth forecasts had probably fallen to about 0%, or maybe even negative. Suppose you had then asked Bernanke “what instrument setting right now would lead to 5% expected NGDP growth over the next 12 months?” My hunch is that Bernanke wouldn’t even know how to answer the question. Because the Fed lacked credibility, it is a very difficult question to answer. Bernanke would have had to guess what the market thought the future policy trajectory would be. (I.e. he’d have had to guess whether the policy would be perceived as credible.) The market would not have that problem, because the same people would be forecasting the effect of policy, and forecasting the credibility of the policy.
    Without NGDP futures targeting things are more difficult, as there is a circularity problem if the Fed starts targeting private sector forecasts. But that doesn’t give the Fed a pass, as they should at least be targeting their internal forecasts. But they didn’t even do that after last October. And even if they are targeting internal forecasts they can still look at private forecasts to see if they are far off course, even if they cannot perfectly target those forecasts.
    Your question reminds me that my original post was slightly misleading. I gave the impression the the Fed really needs to believe TIPS spreads. But what I really meant was that the Fed really needed to OBEY TIPS spreads, that is, to target the forecast. Or at least target their own forecasts. Last September the market forecast was clearly far superior to the Fed forecast. After that I am less sure, although my hunch it continued to be superior in the fall.

    JimP, Thanks for that article. I also found one from Bob McTeer, who used to head the Dallas Fed, who complained about the interest on reserves policy. I was interested in the comment that M2 has slowed. I hadn’t watched that indicator, but it would be another similarity with Japan if true, and would also expose the silliness of the hyperinflation fears.

    I had not thought of Laffer as a tight money guy, as he has argued for looking at market indicators of inflation in the past, so I was a bit surprised to see him take that view today.

    Josh, You may be right but that is not at all the impression that I had. What was the indicator of monetary shocks used in the early VAR studies that found a price puzzle? I thought it was short term rates?

    Even recently I have seen VAR studies which bragged that they found some way of solving the price puzzle. Why would they emphasize this claim if it had been easily disposed of 15 years ago. I am not saying you are wrong, as you know more about this than me, but am just a bit puzzled.

    I feel a bit more confident in disagreeing with your view of the future perception of these studies. You could also argue that money demand studies are necessary in order to understand the implications of theoretical models, but people lost interest once we stopped targeting the monetary aggregates. I believe that in the future no one will pay much attention to short term interest rates as a policy indicator. Theoretical models now view interest rate changes as monetary shocks. In the future economists will start viewing changes in the expected path of the goal variable as monetary shocks. And then studies will focus how those changes in expectations will impact the macroeconomy. Interest rates and the monetary base will be ignored. If so, then VAR studies that used interest rates and the base won’t be seen as being useful.

    BTW, It would be interesting to plug the ff rate or the monetary base into one of those VAR models in the fall of 2008, and do a “horse race” with TIPS spreads. I wonder which one would best predict the future path of the economy? (Yes, that’s a rhetorical question.)

    tom s. You are probably not clear because I am not clear myself. I should not have used the term “exogenous.” What I meant was some sort of shock that comes from outside the monetary policy system, such as the banking panics of 1930-33 which reduced M1 and M2. The Fed now knows that if something like that occurs again they need to offset that change with some sort of monetary injection to keep M2 from falling.

    Elsewhere I said that monetary policy was endogenous under the gold standard, or that the money supply was endogenous once the Fed decided to target interest rates. Endogenous means from inside the system, or determined by the system in question. The terms are vague because from a broad enough perspective almost everything is endogenous.

    I meant the Fed should not let the money supply fall for reasons that had nothing to do with the goals of monetary policy

  11. Gravatar of TGGP TGGP
    22. June 2009 at 16:47

    Division of Labour (where Larry White, among others, occasionally blogs) often features stories from the NYT a century ago to the year. At MR Tyler just linked to the blog News from 1930.

  12. Gravatar of ssumner ssumner
    23. June 2009 at 05:09

    TGGP, Thanks, I also saw the Tyler link, and did a post on it.

Leave a Reply