Sunday, March 04, 2012

Why macro is hard (Taylor/Krugman edition)


In the latest volley in the Stimulus Wars (giggle), John Taylor explains why he doesn't think the ARRA, commonly called "the stimulus" or "Obama's stimulus", increased GDP:
For the parts of the packages which include temporary tax rebates or temporary tax cuts I find no significant consumption effect using regression analysis and controlling for other factors that affect consumption. If you look at a chart of the tax rebates in 2008, for example, the evidence is striking: There was a big increase in personal disposable income at the time of the rebate, but no similar change in consumption... 
In the case of the 2009 stimulus package, there was also an attempt to increase significantly government purchases of goods and services. But the evidence is that this attempt largely failed. A special satellite account produced by the Bureau of Economic Analysis shows that federal infrastructure investment—at the peak quarter—increased by only .05 percent of GDP as a result of the stimulus and federal government consumption by only .14 percent... 
While state and local governments received substantial grants under the 2009 stimulus, a statistical analysis by John Cogan and me shows that they did not use these grants to increase their purchases of goods and services as many had predicted. Instead they reduced net borrowing and increased transfer payments.
It's important to realize, as Mark Thoma explains, that Taylor is not saying that stimulus can't work theoretically, but that the specific "stimulus" enacted in the ARRA didn't work.

Tayor makes three separate arguments here. The first is that the tax rebates in the ARRA failed to increase household consumption. This doesn't seem like an incredibly controversial argument, since A) there exist decent ways to forecast what consumption would have been in the absence of the ARRA, and B) Keynesian theories predict that transfer payments make for much worse stimulus than government expenditure. But note that one of those things that forecasts consumption is wealth; when the ARRA was passed, household wealth was plummeting due to the collapse in housing prices, and thus it is possible that the ARRA stopped consumption from collapsing.

Taylor's second argument is that the ARRA failed to increase federal government purchases - i.e., "true" federal stimulus - by much. This definitely matches the evidence gathered by Paul Krugman in a recent series of posts (see here, here, here, and here) showing that total U.S. government expenditure fell during the recent recession. So, not much argument here. 

Finally, Taylor claims that the ARRA did not do much to change government purchases at the state level. This is very hard to prove, since we don't have a good model of state government behavior. Paul Krugman claims that without the ARRA, state expenditure would have fallen even more, because states would have run up against borrowing constraints. Taylor obviously doesn't think that would have happened.

Who's right? It's hard to know, and this really illustrates the main difficulty in doing macroeconomics: lack of a good "counterfactual." History only happens once; it's just damn hard to tell how things would have changed if people had made different decisions, just by looking at what ended up actually happening.

The only reliable way to get a "counterfactual" - and, thus, the only way to tell if a policy worked - is to have a good theoretical model of how policies work. But one of the things required for a model to be "good" is for it to be tested against data. And all too often, the only data we have to test our macro theories is...you guessed it...one single run of history. And that run of history is pretty short - less than a century of quarterly aggregate data on a small handful of variables.

Now, if we could believe in cross-country comparisons, then we could basically have several simultaneous "runs" of history to compare, and this would help us be more certain about our models. But cross-country comparisons are notoriously hard to do, and countries' economies are also dependent on each other to some degree.

This is why, in my opinion, the only way to build really believable and reliable models is to use microfoundations (or, if the macroeconomy is big enough to display emergent properties, to use agent-based modeling). Without those techniques, macro is doomed to be a lot more like history than science. 

Which is not to say macro is doomed! History is not a useless way of understanding the world. Observing and recording events and then making reasonable speculations as to their causes - e.g., saying "I think states would have run up against borrowing constraints without ARRA" - is not a useless endeavor by any means. For thousands of years of human civilization - up until Francis Bacon & co. - that was the only way we had of understanding the world around us, and it did lead to a slow steady increase in human knowledge. It's just that history is a less powerful way of understanding the world than science. If we could make macro a science, that would be awesome.

In the meantime, regarding the current debate, you may ask: Who do I believe, Taylor or Krugman? I guess I mostly side with Taylor here. Krugman is probably right that if there had been no ARRA, states would have spent even less as they bumped up against borrowing constraints. But the states still would have saved a portion of the ARRA money (or passed it on to households as tax rebates or transfers), substantially reducing the ARRA's fiscal multiplier. I believe that direct federal government purchases - increased infrastructure investment - makes for by far the best stimulus. The fact that there was almost none of this in the ARRA was a major, major policy failure. I think that that failure overshadows the small benefit that the ARRA might have had in supporting state purchases. And I think that that is a point on which Krugman and other stimulus advocates would broadly agree.

12 comments:

  1. California save money? You need to get out more... Here's a candidate topic for you (useful to you, too, since you'll be teaching soon, Congrats!): a survey of macro textbooks. Thanks!

    ReplyDelete
  2. "Which is not to say macro is doomed! History is not a useless way of understanding the world."

    Reassuring to see that, while you're out hunting the snark, potting pheasants is not beneath you.

    ReplyDelete
  3. Historically, going to the general from the specific has usually been a terrible idea.

    According to the manufacturing "theory of constraints" there are a few issues that drive the profitability of a specific firm. To manage those driving constraint(s) they must subordinate all other issues to that constraint -and by definition, reduce their efficiency on those issues.

    As someone who does firm specific financial models for a living, I've learned you cannot assume parity in terminology between GAAP audited financial statements between firms. If I'm not able to accurately apply standard models between firms in the same industry, it becomes very difficult for me to imagine models that hold true across industries, much less varying types of non-industrial actors.

    As a result, while I can see going from the general to the specific i.e. going to the hospital (Macro Health System) when you are in a cancer treatment program (Micro/Specific System) it is not clear that the cancer treatment model covers all hospital visits.

    I agree with Krugman - using Micro to check Macro assumptions is useful. I just don't think you can start at Micro and work back to Macro because the number of actors you are required to model is excessive.

    ReplyDelete
  4. physicist6:12 AM

    If you're hoping to derive macroeconomics by starting from the microfoundations -- that is something which in general doesn't even work in physics. It's not a coincidence that thermodynamics was discovered before the statistical mechanics that underlies it. And universality classes and effective theories are the tool of choice in condensed matter physics.

    If you say it's impossible to come up with a decent "effective theory" for macroeconomic behaviour, I think the outlook for economics is bleak.

    ReplyDelete
  5. Anonymous8:15 AM

    Macroeconomics should cease pretending to be an experimental science and focus on becoming a more disciplined form of history -- more like cosmology and less like particle physics.

    ReplyDelete
  6. "if the macroeconomy is big enough to display emergent properties, to use agent-based modeling)"

    1. There is a middle way: dynamic simulation based on *sector*-based behavioral algorithms. Like weather modeling. As far as I can tell, Steve Keen is the only person who is actually trying to do this. It's seems glaringly obvious that the macroeconomy *is* big and complex enough, so you gotta ask: why do all economists stick with their childish comparative statics drivel?

    2. As far as I can tell the only outfit even talking about agent-based macroeconomic modeling are the Santa Fe guys, and they've been talking about it for a long time without showing us anything.

    If the economics discipline were anything like a science, we'd be seeing huge resources poured into both of these. The fact that we're not reveals it be more similar to medieval scholasticism.

    So: "doomed"? It took MS centuries to die.

    ReplyDelete
  7. BTW, lots of work, going back to maybe Goodwin (1951), explores the possibilities of multiple equilibria, complex dynamics emerging in standard macroeconomic models with simple non-linear tweaks.

    Goodwin, I expect, is disappeared from orthodox economics. But not all of this work is from heterodox economists. For example, John Geanakoplos, at Yale and the Cowles Commission, has written about complex dynamics in Overlapping Generations Models. You can read about some complex dynamic macroeconomic models in the Journal of Economic Behavior and Organization.

    I don't know how threatening this work is to orthodox macroeconomists, but my impression is that it is mostly ignored by many groups of macroeconomists.

    ReplyDelete
  8. Noah: "Who's right? It's hard to know, and this really illustrates the main difficulty in doing macroeconomics: lack of a good "counterfactual." History only happens once; it's just damn hard to tell how things would have changed if people had made different decisions, just by looking at what ended up actually happening."


    "This is why, in my opinion, the only way to build really believable and reliable models is to use microfoundations (or, if the macroeconomy is big enough to display emergent properties, to use agent-based modeling). Without those techniques, macro is doomed to be a lot more like history than science. "

    Please note that the objections in the first paragraph apply to the second paragraph, as well. You're skipping over the problem of coming up with good parameters for your agents.

    ReplyDelete
  9. Noah: Enjoying your blog. Some props, from an earlier post:

    http://www.scpr.org/blogs/economy/2012/03/05/4949/economists-are-special-breed-nerd/

    ReplyDelete
  10. Why not look very very closely at the state level to see this in action?

    I would tend to laugh at Taylor, because if you pick an obvious state, such as Arizona, that is fairly small, suffered a huge decline in revenues, had measurable stimulus that specifically backfilled lost revenue, and faced a very hard borrowing and revenue constraint set due to politics. In this state, one sees the immediate effects of stimulus through cancelled layoffs, with those earning continuing to circulate through the local economy.

    So, why don't you start here in a small state of 6 million people with a full microfoundation set of data available and see if really happened the way Taylor postulates.

    Hint: Using my macro spidey sense, let us talk about failures in an epic sense:

    The first is that the tax rebates in the ARRA failed to increase household consumption. This doesn't seem like an incredibly controversial argument, since A) there exist decent ways to forecast what consumption would have been in the absence of the ARRA, and B) Keynesian theories predict that transfer payments make for much worse stimulus than government expenditure.

    Uh, the research has been done, thanks to the first Bush tax money mailout- high end households saved the money, low end spent it. Um, so now that we have established that a significant population of households are not ration enough to practice Ricardean Equivalence over a one year period, can we tell Taylor he is wrong?

    I could go on and on, but that rational expectations Chicago model just refuses to quit telling people the earth is perfectly spherical, when it is actually oblate.

    Now, coming from highly mathematical models of physics, you think you can model a cultural agglomeration function with a few handy dandy tools?

    Your hopes that microfoundations can prevail on an aggregate level are nearly risible, when we can't even do an adequate job of predicting unemployment rates.

    If you want true microfoundations, first adequately model micro, which still thinks in terms of the rational expectations model.

    After you leave that behind, you might have a chance to succeed.

    At the first mention of twice differentiable, or separating hyperplanes, you might as well begin discussing your Erdos number, it has more relevance.

    ReplyDelete
  11. Anonymous12:12 PM

    Amazing blog site. Part econ theory, part political commentary, part entertainment. Keep up the good work Noah.

    ReplyDelete
  12. Surprise!! John Taylor found what he was looking for; proof that Obama policies don't work.

    ReplyDelete