September 27, 2013
-
The dawn of cognitive macroeconomics
JDN 2456563 PDT 15:10.
A review of Animal Spirits by George A. Akerlof and Robert J. Shiller.
When I first came to CSULB about a month and a half ago, we had an orientation for graduate students. One of the faculty members there (Seiji Steimetz, for whom I am now a graduate assistant, and whom I have come to adore) asked us all a question: “What kind of research do you want to be involved in?” Most of the students didn’t have an answer. I had an answer I didn’t quite know how to explain, so I basically coined a new term: “Cognitive macroeconomics. Basically, what happens to our understanding of the macroeconomy when we stop assuming people are rational—because they’re not?” He replied, “Like prospect theory to explain inflation?” That wasn’t quite right, but it was a more accurate response than I’d expected. Actually that would count as cognitive macroeconomics I think; it’s just not in particular what I had in mind. “Yeah, something like that,” I said.
Animal Spirits is, in all but name, an introductory textbook on cognitive macroeconomics. It is written in a very readable style, and uses hardly any math; but it marks a paradigm shift in macroeconomic theory. Instead of assuming that workers and capitalists are rational, let’s study how they actually think and behave.
Daniel Kahneman and Daniel Ariely (collectively I shall call them “the Dans”) basically founded cognitive economics, but they are really cognitive microeconomists. They talk about issues at the level of individual firms and consumers. I find neoclassical microeconomics mind-numbingly boring; cognitive microeconomics is more interesting—and more valid—but it still lacks the glamor of large-scale impact that macroeconomics promises. If we want to live by Keynes’s “the world is ruled by little else”, it is in macroeconomics that we will do so.
Indeed, it could be argued that Keynes himself was a cognitive macroeconomist; after all, he was the one who coined the term “animal spirits” from which the book draws its title. But the paradigm shift didn’t happen then, because Hicks distorted Keynes’s vision into something quasi-neoclassical, making what could have been a fundamental advance into a incremental improvement. It is as if we shoehorned Newtonian physics onto Ptolemy’s epicycles, or told Darwin that his theory was useful for other animals, but God still made human beings in his own image. (Come to think of it, a lot of people still think that, don’t they?) I doubt Keynes would have recognized what we know call “Keynesian”.
I actually know George Akerlof’s older brother, Carl Akerlof; he’s a physicist at Michigan (whom I interviewed for the Physics Historical Project). Hopefully someday I’ll get to work with George as well; his work sounds like almost exactly what I want to be doing.
Part One explains the five “animal spirits” Akerlof and Stiller think are most important: Confidence, fairness, corruption, money illusion, and stories. The first three are relatively self-explanatory; the fourth is familiar to economists ever since (you guessed it) Keynes, though it has fallen out of favor.
The fifth I think is worth exploring further, since it may actually be the most important. Cognitive scientists have basically established that human short-term memory comes in two basic data formats, image and audio. The latter is literally audio, basically a two-second ring buffer: You can remember longer sentences if you say them faster. The former is not a bitmap image, but more like vector graphics; you can scale and stretch the image in your mind, but it has limited detail. To say that the human brain stores in SVG and WAV is only a slight exaggeration.
But long-term memory takes a fundamentally different format; I think the best way to describe it is to say that the native data format for long-term memory is narrative. We use stories to organize our own past, our culture, our ideas for the future—even our scientific theories. This is why epic poems were so successful in ancient times, and novels are so successful today; they link into this fundamental data format. (Epic also makes use our two-second audio buffer through techniques like rhyme and meter.)
The universe, by contrast, does not appear to be made of stories. Sometimes things happen randomly. Often the cause is unavailable to us. Most historical events are driven by slow pressure from millions of sources, with no clear “Great Man” to be the hero or the villain. Reality is unrealistic, and sadly, the good guys don’t always win.
Animal Spirits uses this to explain the observation that Nassim Nicholas Taleb made in The Black Swan: Why do pundits always come up with a story to explain any random fluctuation? Because human beings like to organize their world in terms of stories.
The second part of the book tries to apply these “animal spirits” to real-world problems in macroeconomics; this is where the book comes up a little short. Akerlof and Shiller sketch out a plausible qualitative account of what happened in the 2008 crash and the Second Depression, and offer some basic suggestions on how we might fix the problem… but much of what they say is vague, and none of it offers sharp, quantitative predictions.
This is the one criticism neoclassicists make of cognitive economics that I do take seriously: It’s easy to show flaws in the current models—but do you have a better one? Neoclassical economics succeeds as a science in one sense: It is wrong. It has gone beyond not even wrong and actually made it to wrong—it makes precise predictions that are incorrect. Cognitive economics isn’t quite at that point yet; we have the potential to make predictions that are correct, which is of course the goal; but right now we aren’t making a whole lot of predictions at all, and the ones we are making aren’t very precise. Akerlof and Stiller think they can explain all the recessions of the past (or rather specifically the non-oil-related peacetime recessions) by their model; and that would be useful, to be sure. But can you predict or prevent the next recession? With enough free parameters you can fit an elephant; it’s easy to make a model that fits the past. The trick is making a model that fits the future. (This is also why “maximum likelihood” is sort of a perversion of Bayesian methods. The maximum likelihood is basically just what you actually found. What you want is the maximum probability, and for that you need a prior distribution.)
They have precise models that give the wrong answer. We have imprecise models that give answers somewhere near the right answer. So we’re not quite there yet. We need to make precise models that give the right answer.
That is why this is only the dawn.
