Tuesday, November 22, 2011

From Russ "I’m an anti-Keynesian because I want smaller government" Roberts

He writes in the comments of this post: "There is no scientific evidence that this [Keynesian policy] has been successful. The model they used to justify spending an extra $800 billion made an explicit prediction about the impact of that policy. It was off by 25%. That means that we are not in the realm of science."

No, no, no. How many times do we have to go through this?

We have two estimates in "the model they used" - first, a baseline unemployment forecast which was taken out of a forecasting model based on information they had available to them in very early January 2009 (three years ago now, people). Those forecasting models are very different from the models that make predictions about the impact of the policy. Forecasting models offer fancy versions of "what happens tomorrow is going to basically be like what happened yesterday". We do not have crystal balls, and forecasting models are never that great - particularly when everything is crashing down around you and no one is sure what's going on.

Then, in addition to that baseline, they had multiplier estimates adjusted by some kind of Okun's law rule which they basically just took and said "if our baseline forecast was X, then our forecast with stimulus is X times (1-Z)".

So the question is - what was wrong? The models producing the multiplier estimates or the models producing the unemployment forecasts? As I've noted here and to Russ numerous times, the multiplier models are very consistent in their results. We don't have a pinpointed multiplier like we have (or had?) pinpointed the speed of light, of course. But we're dealing with a parameter that changes depending on circumstance. Given what we know about multipliers, the results are very consistent and conform to predictions about how the multiplier should behave in different circumstances. Nevertheless, Russ continues to make these misleading claims that the range of multiplier estimates is enormous. And unfortunately, Russ has a much bigger platform for his assertions than I do.

As for forecasting, Russ might be interested in this from the CBO:

"Forecasters collectively have tended to make the largest errors during periods that included either turning points in the business cycle or significant shifts in the trend rate of growth of labor productivity. Large changes in oil prices—apart from their role in precipitating business cycles—also cause errors in forecasts of inflation. The difficulty of forecasting business-cycle turning points explains why most forecasters overestimated the economy’s future growth rate in the forecasts they made just before the two back-to-back recessions of the early 1980s. That pattern was repeated in the forecasts made just before the recessions that began in 1990, 2001, and late 2007."

Russ is looking at two estimates multiplied by each other - one of the multiplier, and one of forecasted unemployment - and he's acting as if he knows he can attribute the error to the first one. I call B.S. on that. Russ Roberts does not know that. Everything we know about fiscal multipliers from past research says they are high right now. Everything we know about forecasts during periods like this says they are bad. Certainly they are going to be bad in January 2009, when the crisis hadn't spread to the real economy yet and we had no idea precisely how the real economy would react.

In the comment section of a previous post I invited people who want to do guest posts to let me know. I would be especially interested if anyone wants to talk about multipliers and these naive accusations of "scientism" that get tossed around. It looks like things are likely to get worse for the world economy in the near future, and people who think this way are getting big exposure. The White House has pushed out its pro-stimulus economists and the worsening budget situation means that we need to be much smarter about how we do things. Let me know if you're interested.


  1. Urgh...not again indeed. The stimulus did work, but it only stopped further contraction. It did not give the boost we may have hoped for because it did not target investment enough.

    Speaking of scientism, Daniel Kuehn, how do you think Russ Roberts will react to Jean-Philippe Bouchaud, Joseph L. McCauley, and their colleagues in the econophysics project? Will he first accuse the application of statistical mechanics to financial economics to be "acts of scientism"?

  2. Id like to also point out that the one forecasting error the administration economists made is not the same thing as there being "no evidence" the stimulus wasn't successful. There seems to be a growing literature suggesting the stimulus had an effect or that it didn't have an effect because it wasn't big enough.

    Man, I don't know why you engage with these yahoos.

    Also, Dan, you should listen to the econtalk interview with V. Ramey. Just bear in mind you are getting a highly skewed introduction to the VAR multiplier literature.

  3. re: "Id like to also point out that the one forecasting error the administration economists made is not the same thing as there being "no evidence" the stimulus wasn't successful."

    Yes - I hope this was clear implicitly, explicitly, backwards and forwards in my post. I was still working on my first cup of coffee, so in case it wasn't clear - YES.

    re: "Man, I don't know why you engage with these yahoos."

    Very simple: they have a big platform.

  4. Taubes had a number of interesting things to say.

    If you don't like Roberts' assessment of the stimulus just ignore it (I've heard his comments on the subject enough that I can easily ignore them).

  5. "Very simple: they have a big platform."

    Besides you (and Russ) I think I'm the only one who remembers Russ's challenge from two weeks ago he abandoned quietly. But then, I think Russ's real goal is advertising.

    Invisible Backhand

  6. Sure it is.

    Now you're just being mean Gary. Don and Russ have built up a well followed blog, Russ's work is regularly featured on NPR and very popular on Youtube. Don is on Fox fairly regularly. And Russ gets to a lot more people through Econlib. They're consistently ranked high among libertarian blogs. They're also featured regularly in the top economics blogs.

    Don't tell me they don't have a big platform. Don't tell me what I should and shouldn't pay attention to.

  7. Daniel,

    I didn't tell you what you should and should not pay attention to (in fact, I made no reference to you at all). However, I will continue to state that they don't have a big platform.

    Cafe Hayek is just one of a large bunch of similarly sized blogs out their on the internets and lots of those blogs have people who end up doing things in the broader media landscape. I don't think of any of them having big platforms.

  8. This comment has been removed by the author.

  9. "Forecasters collectively have tended to make the largest errors during periods that included either turning points in the business cycle or significant shifts in the trend rate of growth of labor productivity."

    If the prediction is a line, y=mx+b, the quote above, restated, is that the largest errors occur when the value of 'm' changes. So, the forecasts are most accurate when the prediction is an extension of a straight line.

    That does not strike me as a useful forecasting model.


All anonymous comments will be deleted. Consistent pseudonyms are fine.