Wednesday, February 3, 2010

Scott Sumner on macroeconomics and blogging

Tyler Cowen writes: “Read the whole thing. It is one of the best statements of how blogging can make a difference” and answer the question “In what way is blogging science?”

Scott Sumner (TheMoneyIllusion) is often not easy to read and his arguments about macro and monetary policy are quite subtle. He makes an important point in his recent post "Seeing the world in a different way (one year later)" about the “Official Method” in macroeconomics. He doesn’t think much of it. Excerpts:

“You devise a model. You go out and get some data. And then you try to refute the model with some sort of regression analysis. If you can’t refute it, then the model is assumed to be supported by the data, although papers usually end by noting “further research is necessary,” as models can never really be proved, only refuted.

My problem with this view is that it doesn’t reflect the way macro and finance actually work. Instead the models are often data-driven. Journals want to publish positive results, not negative. So thousands of macroeconomists keep running tests until they find a “statistically significant” VAR model, or a statistically significant “anomaly” in the EMH. Unfortunately, because the statistical testing is often used to generate the models, and determine which get published, the tests of statistical significance are meaningless.

I’m not trying to be a nihilist here, or a Luddite who wants to go back to the era before computers. I do regressions in my research, and find them very useful. But I don’t consider the results of a statistical regression to be a test of a model, rather they represent a piece of descriptive statistics, like a graph, which may or may not be usefully supplement of a more complex argument that relies on many different methods, not a single “Official Method.””

One could add too that in so many published papers, mostly in fields other than macroeconomics, the story goes like this: first a rather simple, or basic economic statement is made about a plausible causal link between two variables. It usually relies on very simple economics, although it is a speculative hypothesis not justified by deeper argument or any other evidence. Then follow twenty pages of abstruse and irrelevant mathematics, preferably with a score of triple integrals. And then only, some statistical tests of the sort Scott Sumner criticizes are presented. Here the mathematical apparatus, not the statistical testing, is what determines which paper gets published, wasting a lot of publication pages.

Deirdre McCloskey has been suggesting for a long time, as Scott Sumner does in a way, that economics is a persuasion instrument. In this perspective one has to wonder what can be the value of mathematical developments and “proofs” derived from rather arbitrary formal models designed mostly to justify, in an apparently “scientific” way, the statistical presentations that follow. A more useful discussion should be centered on the choice of the main building blocks of potential models, and a comparison with alternative choices, in the process of constructing the model.

This is congruent with the author’s experience with journal publishing that he reports later in his post:

“In 2008 and 2009 I sent papers on the economic crisis out to journals like the JMCB and the JPE, journals that I have published at in the past. But now for the first time in my life the articles come back without even being sent out to a referee. “It’s not the sort of thing we publish” they’d say. I gather this means they don’t see enough equations. I hope it doesn’t mean “because it addresses the most important macroeconomic problem since the 1930s.” "

There are many gems in the post. I especially liked the beautiful summary of what the efficient market hypothesis (much criticized – wrongly - for being disproved by the crash) really is : it “merely tells us that the market forecast is the optimal forecast, not perfect foresight.”

The second part of the post, “Don’t ask me to become a blogger”, is about the experience of publishing a blog. It is very well worth reading and invites one to think about the respective influences and impact of the blogosphere and the established institutions of universities and journal publishing.

Hat tip: Tyler Cowen (Marginal Revolution).

No comments: