I tried to pick Dr. Athreya's brain today. Here was my email to him:
"Hello Sir. I know your busy but I wonder if I could ask you a quick question-I admit I feel like being able to pick your brain now and again is a kind of resource. This is kind of a question statistics."
"Btw--Noah Smith reviewed my book, and interestingly, argued that it would be inaccessible to most readers without background in economics. Your experience, and others I've heard directly from, belie this--and I hope he's not putting off too many non-economist readers."
"If you're one of those few people who wants to understand macro by poring through dense literary tomes (Keynes, Hayek, Minsky), but who flees at the sight of a single equation, then I guess Big Ideas could be just the book for you. But to be frank, I don't think you'll really understand modern macro by reading this book. Maybe it's possible to write a book that teaches modern macro without equations - macro's answer to the Feynman Lectures? - but Athreya has not written it here."
"Hello Sir. I know your busy but I wonder if I could ask you a quick question-I admit I feel like being able to pick your brain now and again is a kind of resource. This is kind of a question statistics."
"If a salesman consistently gets 15-20 sales a week selling drain cleaner to maintenance guys for 8 months and then has a month where he's getting just 10 sales a week on average how likely is this to be an aberration or a new lower trend?"
"Yes this could be autobiographical!"
Basically I wanted to know if he has anything in his economist's bag of tricks that might tell me if I'm just in a slump or if maybe now I have now regrettably peaked as a salesman. I guess there is no Royal Way to Truth eve in in economics. I always dream that maybe the technical analysts in the stock market are right but apparently to no avail. Here was his response.
"Tough to tell, actually. You'd need to compare with the more routine volatility in sales you experience. If you *always* get sales of 15-20, then the drop seems more significant. But, one month is not everything. Good luck!"
He did do one very clever trick on second thought-he turned my concern into a discussion about his concern-Noah's review. Nevertheless, I have no problem with it and am still starstruck enough to be flattered. After the treatment I've received by Sumner it's nice that some economists are honest enough to admit my worthiness!
I must disagree with Noah on the question of accessibility. Whether Dr. Athreya is all right or all wrong or some place in the middle I must say that I found his book very accessible and I'm certainly a noneconomist-just ask Sumner. So I left Noah this note.
"I will say this in his defense. I read his book-and I come in as one of those laypeople that folks like Sumner stick his nose up about-who are disposed not to believe the Neoclassical economists-and I found him highly readable. I mean certainly at least compared to what has been offered by establishment econ before, his work is very accessible."
"Traditionally establishment economists take Sumner's attitude-laypeople are too stupid to understand anyway and there's no time for displaying pearls for the swine."
"So I have to disagree with you at least that it was inaccessible. I feel that I came away knowing a lot that I didn't before reading him. Also because he's been very nice to me and has had ongoing email correspondences with me while you have always seemed to treat me like I'm kind of like the economic blog equivalent of that chick who used to break into David Letterman's house. You seem like you take pains not to encourage me."
"So on a personal level-and hey I'm not a professional economist so I will capriciously go with my intuitions!-I like him a lot more than you whether I agree with him or not!"
"Here is more on my conversations with him.
http://diaryofarepublicanhater.blogspot.com/2014/04/i-exchange-emails-with-katrik-athreya.html
http://diaryofarepublicanhater.blogspot.com/2014/06/kartik-athreya-on-adaptive-expectations.html
"Traditionally establishment economists take Sumner's attitude-laypeople are too stupid to understand anyway and there's no time for displaying pearls for the swine."
"So I have to disagree with you at least that it was inaccessible. I feel that I came away knowing a lot that I didn't before reading him. Also because he's been very nice to me and has had ongoing email correspondences with me while you have always seemed to treat me like I'm kind of like the economic blog equivalent of that chick who used to break into David Letterman's house. You seem like you take pains not to encourage me."
"So on a personal level-and hey I'm not a professional economist so I will capriciously go with my intuitions!-I like him a lot more than you whether I agree with him or not!"
"Here is more on my conversations with him.
http://diaryofarepublicanhater.blogspot.com/2014/04/i-exchange-emails-with-katrik-athreya.html
http://diaryofarepublicanhater.blogspot.com/2014/06/kartik-athreya-on-adaptive-expectations.html
http://noahpinionblog.blogspot.com/2014/06/kartik-athreya-responds-to-my-review.html#comment-form
I do feel that Noah doesn't like me. He seems quite stuck up though perhaps it's just that I bring that out in him-in other words he just doesn't like me. However, I have plenty of other economists who have embraced me-or at least have been very friendly-Nick Rowe, Brad Delong, others, and now Kartik. At least Dr. Athreya's heart is in the right place-he clearly wants to educate the public-rather than sneer at them as Sumner and perhaps Noah prefer to do.
Don't get me wrong, while I understood his Athreya's book, I have done a fair amount of reading up on economics the last few years starting in about 2011. Still, if you're an interested layperson who's willing to work at it a little then I should say you should be able to read him with profit. If you're not willing to put in the work then you don't care about understanding it all that much-maybe a little bit but not if it cuts into your time at the bar.
Let me quote Noah's passage where he levels the inaccessibility charge.
"Who should read this book? If you've taken graduate-level econ courses, Big Ideas will be a review, but it may point you toward a couple of important theory papers that you've overlooked; Athreya is verywell-read. So I'd say skim it for the lit-review value. If you haven't taken grad-level econ, but you want to understand modern macro, then my recommendation is to first read a textbook (e.g. this one), and then maybe skim Big Ideas for interesting paper references. If you're mostly just interested in the policy kind of stuff and informal ideas that you can read about in blogs - in other words, if you follow macro debates just for fun - skip Athreya's book; it'll be too dense, and you'll just get bored and quit."
"If you're one of those few people who wants to understand macro by poring through dense literary tomes (Keynes, Hayek, Minsky), but who flees at the sight of a single equation, then I guess Big Ideas could be just the book for you. But to be frank, I don't think you'll really understand modern macro by reading this book. Maybe it's possible to write a book that teaches modern macro without equations - macro's answer to the Feynman Lectures? - but Athreya has not written it here."
http://noahpinionblog.blogspot.com/2014/06/big-ideas-in-macroeconomics-book-review_8748.html
Aha-this time Noah can't get me with the read a textbook line-like Nick Rowe did before- as I'm just finishing up Greg Mankiw's Principles of Economics. That's the second edition but it was a lot cheaper than the latest version-I also got Krugman's recent textbook which I will read soon and Paul Sameulson's original 1948 The Foundations of Economics.
http://diaryofarepublicanhater.blogspot.com/2013/03/ok-nick-rowe-you-win-ill-read-econ.html
http://diaryofarepublicanhater.blogspot.com/2013/03/ok-nick-rowe-you-win-ill-read-econ.html
http://diaryofarepublicanhater.blogspot.com/2014/04/nick-rowe-drops-by-and-promise-of.html
"If Noah doesn't think you can understand 'Modern Macro'-see I'm still very skeptical-by reading Athreya, where would he recommend we start? Ok-after the textbook which is what I've been doing. As for this idea that if you are interested in 'policy stuff' you can skip it, I don't understand this either. Hasn't the point that Noah, and Sumner, and al these other econ bloggers shown over the last 5 years that to really properly understand the policy stuff you need to have some understanding of theory? If you are interested in policy and want to understand how policy actually gets made, this is not a bad place to start-most policymakers-certainly at the Fed believe the kinds of ideas Athreya talks about-Walrasian Equilbrium, Rational Expectations, etc. You might not care about this stuff and just want to debate policy but your policymaker either believes this stuff or has an economist advising him who does."
None of this necessarily means that Athreya is right about most everything in the book-I'm still even now digesting it and still have a number of questions-on these questions, I agree with Noah on a number of things. However, it was a very accessible book and genuinely tries to reach the interested layman. So here I disagree with Smith.
P.S. One thing that Noah and Dr. Athreya seem to agree on is that Keyne's General Theory is a 'literary tome.' The question I have for Noah is if there is any book without equations that he wouldn't find a silly, literary tome?
I wrote about Athreya's own issue with Keynes' literary masterpiece'-the General Theory.
http://diaryofarepublicanhater.blogspot.com/2014/05/katrik-athreya-modern-macroeconomics.html
I notice that all mainstream economists without exception today find nothing more absurd than spending your time sitting down reading GT while stroking your beard and asking Now what did Keynes really mean? Is this just a clever way of shutting off debate with the Post Keynesians?
This contempt for what Keynes really meant could be the defining characteristic of 'Modern Macro.'
P.S.S. Maybe Noah doesn't like me because I'm literary-I never write any equations. Of course, for an economist to call another economist literary is like one supermodel calling the other one fat.
P.S.S. Maybe Noah doesn't like me because I'm literary-I never write any equations. Of course, for an economist to call another economist literary is like one supermodel calling the other one fat.
I'm still impressed you made it through the book and got something out of it apparently, ... and that you are now tight with Kartik too, right. :D
ReplyDeleteOn that subject, perhaps Kartik would be willing to take up this challenge:
http://informationtransfereconomics.blogspot.com/2014/07/a-challenge-to-macroeconomists.html
He seems like just the sort to know where to find that kind of thing, doesn't he? Does he have an email or blog or something (a means of asking him)?
Well I know I can't give out his email without his permission-I had to apologize to him for quoting from his email in this piece.
ReplyDeleteHe had given me permission to do that previously and I wrongly assumed it was therefore always ok to quote from him. I'll take a look at your link later Tom and maybe I'll ask him about it.
However, I do appreciate his correspondence and input and I need to be careful not to abuse it. Again, this is nothing you've done wrong and you ask a good question-just that I have to be careful not to make him feel I'm overstepping again.
Ok Tom I just read your link. I like that Smith points out that Sumner despite all his relentless declarations of victory has proved nothing quantitative.
ReplyDeleteIt seems that all anybody ever comes up with in defense is that 'all models are false which is ok as long as they give enable us to do important and interesting things with them' and that they need to correspond with empirical reality-this is an argument that Milton Friedman used famously in the 50s and I don't think was even new then-the whole thing of 'positive vs. normative' arguments.
Here I can speak of Kartik as this is in his book-I think that's basically his explanation of the unrealistic assumptions of things like the life time income model and Rational Expectations-ok they are false but they're helpful.
I guess it can be seen as this: the place we start drilling may necessarily be unscientific but we have to start somewhere. A model with unreal expectations is often going to be simpler than one that corresponds more faithfully to reality.
ReplyDeleteI' far from necessarily agreeing, just I think this is what the standard answer to this complaint is.
ReplyDelete"and that they need to correspond with empirical reality-" should read 'and that they don't need to correspond to empirical reality'
ReplyDeleteHi Mike, thanks. I would never ask you to pass on his email w/o his permission, but I thought perhaps he had it on a public web page (I haven't bothered spending more than 10 seconds looking).
ReplyDeleteI left the challenge on a "student's" blog, that he gave a link to in Noah's post (I noticed he made some comments there).
Smith has since found a few more models. One from the NY Fed for inflation (which he can do as well) has 42 parameters vs his 3 parameters. He did some calculations using an "information criterion" (that by chance Noah happened to have linked to today as well in his Cochrane post). I don't understand the results, but they look bad for the Fed model.
Then he found another Fed model called P* which Bernanke talked about. That one does better and has fewer parameters (7 or 8), but still produces some odd results (like predicting... or "now-casting") deflation in 1990 (which was pretty far from the mark). Jason's model doesn't do that... and by fitting his three parameters to the same time period (1955 to 1990) he is able to project forward and get a great fit on up to recent years (something he implies would be tough with P*, for which the data was only available from up to 1990)
http://informationtransfereconomics.blogspot.com/2014/07/notes-from-ben-bernanke-and-p-model.html
BTW, as Noah points out, more parameters is generally bad. Here's Jason's analysis:
Delete"I just realized the NY Fed model has 42 parameters compared to the ITM model that has 3. By the AIC (which incidentally, Noah Smith just referred to a few minuted ago), that means the likelihood function for the Fed model would have to be 4.3 x 10^18 % in order to select it over the ITM if the likelihood of the ITM model was 50%. Or the other way ... if the Fed model had a likelihood of 99%, then the ITM would have to have a likelihood of 1.1 x 10^-15 % to lose to it.
Now I'm not sure all of those parameters are necessary to describe inflation and there are other things described by the Fed model like the capital stock, but as we say in engineering, there's a lot of margin.
http://en.wikipedia.org/wiki/Akaike_information_criterion"
Oh no, Tom don't get me wrong-I wasn't concerned about you it's more me-he already let me know he didn't appreciate me quoting from his last email without permission-which I totally understand.
DeleteIf you find his email listed then you can ask him yourself. I just know that I can't do it.
The point about parameters may be why econ have more realistic models? Because to make them more realistic you need more parameters and to have more parameters is to weaken the models into impotence? Is that a fundamental tension?
I mean why econ has such unrealistic models-because too many parameters which would make them more realistic waters them down into saying something so broad as to be pointless.
DeleteMike, regarding parameters and realism. I'm not expert on this ... I do have an intuition about it though: I think adding parameters to make something more realistic is fine as long as you get a big pay off in terms of eliminating the error between your model and the data... and not just the "training data" but data outside the training set.
DeleteFor example, say some natural process that we measure produces N points on a x-y plane, with N unique x values. We might propose an ad-hoc theory to explain these points *exactly* with an (N-1)st order polynomial (e.g. A(N-1)*x^(N-1) + .... + A2*x^2 + A1*x + A0). This can always be done!... but the model is almost guaranteed to suck. Suppose we get another measurement between x3 and x4.... the polynomial will almost certainly have wild excursions up and down between all of the data it was "trained" with (fitted to). It'll suck both for interpolation and extrapolation.
Now if instead we fit a straight line to the data... not passing through any of the training set of data points precisely, but minimizing the sum of squared errors to all the points, and the underlying process produces a straight line (with some random measurement noise), we might do much much better. It'll still be an ad hoc model, but a much better one.
This simple example is the lens I look at this problem with... as the number of parameters goes up, the danger is that the model designer is training his model to better explain the training data, but perhaps at the same time destroying the model's ability to explain data outside the training set.
That one of the reasons it helps to have a theory to go with your model. Take Newton's theory. For a two body problem (Earth and sun say), you might be able to calculate the orbits pretty precisely. Now you could add more parameters to reduce the errors on the training data... but rather than just do this in an ad hoc way, why not make use of the theory... perhaps make it into a three body problem (add in the moon), or four bodies (Jupiter?), etc. You might only get marginal improvements in accuracy, but the nice thing is the addition of parameters probably helps to explain the data outside the training set. If adding parameters consistent with the theory doesn't help, it might indicate that the theory is wrong. Which is the case actually: Einsteins' theory of gravity is more successful than Newton's.
Jason's approach is very interesting to me. It's analogous to the way that 19th century scientists finally started to develop a good theory of heat and heat flows (thermodynamics) and thus sh*t canned the old "phlogiston" concept.
But the interesting thing is that the early thermodynamicists didn't know much about atoms or molecules... the just knew they were bits of matter and that there were a LOT of them in a human sized object, or container of gas. So they were able to develop an ideal gas law, describing the macro behavior of a huge collection of individual gas molecules, with only a tiny number of parameters. You might go back to this today, and using all our knowledge of atoms and molectules attempt to build a simulation of a whole container of individual gas molecules, with lots of parameters, and keeping track of all the states of the simulated molecules bouncing around, and vibrating, and spinning, etc, and come up with the same ideal gas law.
DeleteAn early attempt to justify macro models such as the ideal gas law was "statistical mechanics." Here they basically just looked at the number of possible modes the gas molecules could be excited in: for example, if the gas was composed of two atoms, the atoms might vibrate wrt one another, or move in the x, y, or z dimension within the container. These micro models of what gas molecules could do was reduced to just one number on the macro scale: the number of possible modes was counted, and this count became part of the exponent in the ideal gas law. There was a theory for why this should be, but the point is not much of the micro state or model survived into the macro law.
What Jason is trying to do is analogous to this. In some sense his theory is based on counting possible states... w/o trying to model all the details. Sometimes this kind of approach can be successful. We'll have to see if the way he's doing it is successful for econ (both micro and macro). Take a look at these three posts... they give you a feel for the what he's doing:
http://informationtransfereconomics.blogspot.com/2013/04/the-philosophical-motivations.html
http://informationtransfereconomics.blogspot.com/2014/06/what-if-money-was-made-of-vinegar.html
http://informationtransfereconomics.blogspot.com/2014/05/causality-in-information-transfer.html
Jason is not big on the use of "expectations" (the bedrock of MM theory?) in macro. He calls it the "phlogiston" of economics. Here's a related post of his:
Deletehttp://informationtransfereconomics.blogspot.com/2014/05/expectations-destroy-information.html
I think his concern about how expectations can be (mis)used in econ, arises from this pattern of circular reasoning, that you've probably seen before:
1. CB OMOs don't work as expected because "expectations" aren't being communicated properly (never mind that the ERs have been on bank BSs for going on 7 years now).
2. What are "expectations" here exactly? They're defined as the thing, which if the CB communicated them properly, then OMOs would word as expected.
3. How do you know that expectations are no being communicated properly? Because if they were, then OMOs would work as expected.
Now go away, you troublesome "people of the concrete steppes!"