tag:blogger.com,1999:blog-17232051.post84824873603279213..comments2023-02-04T12:49:36.805-05:00Comments on Noahpinion: The death of theory?Noah Smithhttp://www.blogger.com/profile/09093917601641588575noreply@blogger.comBlogger96125tag:blogger.com,1999:blog-17232051.post-89800509692500724422015-05-11T18:24:41.566-04:002015-05-11T18:24:41.566-04:00I don't fit any of Noah's id choices!
The...I don't fit any of Noah's id choices!<br /><br />There's one other aspect of this that I think is extremely important: the advances in causal analysis of observational data. Recently, the proliferation of techniques and available software for examining causal relationships in large-scale datasets rigorously has been both surprising and gratifying. It makes the entire business of the social sciences nearer and nearer (largely because the physical and life sciences are doing the same thing) to the usual ideal elsewhere: "Why think? Do the experiment!" Combine that with massive datasets that are available to test assertions and you get a real sea change in how people do business.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-17232051.post-12613408068247328812013-08-25T14:15:43.934-04:002013-08-25T14:15:43.934-04:00The length of a typical paper has significantly in...The length of a typical paper has significantly increased. Maybe there are now more ideas per paper? Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-17232051.post-31371281299198535752013-08-19T12:19:05.605-04:002013-08-19T12:19:05.605-04:00The quantitative and the qualitative are merging a...The quantitative and the qualitative are merging and the complexity of things like the soul conjecture, so befuddle most laymen, that they neglect to be aware of its occurrence. Qualitative data has never been as available in size and scope as it is today and we've never had computers and software so able to query, poke and prod it as we are able today.<br /><br />The quantitative and the qualitative are merging such that art is becoming science and science is being realized as art.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-17232051.post-31685282524627262462013-08-19T12:13:54.873-04:002013-08-19T12:13:54.873-04:00I am completely puzzled by the attempt to contrast...I am completely puzzled by the attempt to contrast between "deductive proofs" and "proofs by computer". Computer assisted proofs are deductive. And arguably more rigorous than human written proofs. Using computers to generate data in support of conjectures or to draw conjectures or to discover structure is an entirely different matter from computer assisted proofs. Mathematicians have made tables of calculations to test their hypotheses for ages, but no one ever called those tables "proof" and no one does now.Sashohttps://www.blogger.com/profile/09380390882603977159noreply@blogger.comtag:blogger.com,1999:blog-17232051.post-23505944827369587962013-08-08T22:10:28.422-04:002013-08-08T22:10:28.422-04:00I was waiting for someone else to point this out, ...I was waiting for someone else to point this out, but no one has. So even though the comments on this blog post are getting stale, let me point it out: <br /><br />Behavioral Econ was WAY theoretical. In fact, in many ways, Behavioral Econ was the last hurrah of theoretical economics. Have you ever had the pleasure of being made to prove the existence of Von Neumann expected utility function from the Savage axioms? Well, prospect theory is like that except worse, with even more esoteric math (that's where measure theory and those Borel sets come in - I see Barkley did mention Aumann above). You want things more general, and funky, and you want it precise and formal... well, the math/theory is actually gonna get strange.<br /><br />So if you're gonna go looking for Colonel Mustard and the candlestick, it wasn't Kahneman and it wasn't "irrationality". (It was Levitt and instrumental variables (joke))YouNotSneaky!https://www.blogger.com/profile/06378267534638281151noreply@blogger.comtag:blogger.com,1999:blog-17232051.post-31581989326503795482013-08-08T01:02:01.939-04:002013-08-08T01:02:01.939-04:00Nathanael,
I take your comment seriously. What i...Nathanael,<br /><br />I take your comment seriously. What is going on here is a transdicslipinarity, Certainly psych is a major player, but a lot more is going on here. For example, defenders of neoclassical orthodoxy simply say "add social preferences" or whatever to utility functions,and, hey, back to normal,although at a minimum once one does that one moves from impersonal GE to game theory, at a minimum.rosserjb@jmu.eduhttps://www.blogger.com/profile/09300046915843554101noreply@blogger.comtag:blogger.com,1999:blog-17232051.post-72733253569060273212013-08-07T22:18:23.287-04:002013-08-07T22:18:23.287-04:00There are vast realms of opportunity for fundament...There are vast realms of opportunity for fundamental work in econ, but it's going to be of a radically different sort. We have to actually start understanding mass psychology before we can get decent fundamental discoveries in econ.<br /><br />I tend to think that math is of relatively little value in economics (as contrasted to physics, for example), and I say this as someone trained in mathematics by one of your students.Nathanaelnoreply@blogger.comtag:blogger.com,1999:blog-17232051.post-6597680876339972062013-08-07T22:15:42.317-04:002013-08-07T22:15:42.317-04:00Empirical data was better-researched in the *19th ...Empirical data was better-researched in the *19th century* than it is today. In the 19th century, economists paid attention to actual production levels of specific individual goods; railcar loadings; factors of production for each major item in the economy; etc. I don't see a lot of that stuff. Most macroeconomists are playing with vague aggregates, and most microeconomists are looking at overly narrow situations without enough cross-comparison.<br /><br />Excessive corporate secrecy is one part of this: it may be hard to collect the data. :-PNathanaelnoreply@blogger.comtag:blogger.com,1999:blog-17232051.post-67031681445116904232013-08-07T22:12:44.255-04:002013-08-07T22:12:44.255-04:00The price does not converge to the "supply-de...The price does not converge to the "supply-demand equilibrium", ever. This is empirically proven. So, yeah, rational expectations belongs in the trash heap.<br /><br />Supply and demand analysis is still useful. What isn't useful is *equilbrium analysis*. Supply and demand analysis is best used in the form of analyzing the directional effect of shocks, for which it is highly effective.Nathanaelnoreply@blogger.comtag:blogger.com,1999:blog-17232051.post-89438316666867736242013-08-07T22:09:59.702-04:002013-08-07T22:09:59.702-04:00"Psychology has been cranking out the empiric..."Psychology has been cranking out the empirical results for decades, and still doesn't have that many working theories to show for it..."<br /><br />It actually kind of does. It has lots. They're all just very *limited in application*.<br /><br />This is a bit like physics back when optics and mechanics were two totally unrelated subjects -- and the optics theories weren't even consistent with the mechanics theories. Or perhaps it's like an even earlier, even more "small model" stage of physics.<br /><br />It's a good developmental stage to be at. After 100 years, there will be enough "small models" of local application to start actually constructing some Big Theories.Nathanaelnoreply@blogger.comtag:blogger.com,1999:blog-17232051.post-75281080239088095772013-08-07T22:05:05.163-04:002013-08-07T22:05:05.163-04:00Do you want to model expectations realistically? ...Do you want to model expectations realistically? Go over to the Marketing department and you'll find out how expectations are created.Nathanaelnoreply@blogger.comtag:blogger.com,1999:blog-17232051.post-6537078936205528782013-08-07T22:03:47.399-04:002013-08-07T22:03:47.399-04:00I wonder why it took so long in economics. In oth...I wonder why it took so long in economics. In other sciences, that particular philsophical change was definitely done by the end of the 19th century.Nathanaelnoreply@blogger.comtag:blogger.com,1999:blog-17232051.post-42819281275443938442013-08-07T22:01:44.606-04:002013-08-07T22:01:44.606-04:00"Or... the "top journals" make up a..."Or... the "top journals" make up a less-important means of publication than they used to, so papers are migrating to other channels."<br /><br />This is going on. The good theories -- we're talking models for specific areas, not the Grand Unified Bullshit which was so popular for decades -- are often still blacklisted from the "top journals", but used in other journals.Nathanaelnoreply@blogger.comtag:blogger.com,1999:blog-17232051.post-51843650895302099932013-08-07T21:59:20.191-04:002013-08-07T21:59:20.191-04:00Testing that you haven't accidentally generate...Testing that you haven't accidentally generated a contradictory model (which will prove anything) is actually very important. That's one of the reasons you have to check for existence of the solution.<br /><br />Uniqueness? Well, think: does it matter whether this is a unique solution or not, in your application?Nathanaelnoreply@blogger.comtag:blogger.com,1999:blog-17232051.post-9169078476775665092013-08-07T19:30:58.798-04:002013-08-07T19:30:58.798-04:00http://www.edge.org/3rd_culture/anderson08/anderso...http://www.edge.org/3rd_culture/anderson08/anderson08_index.htmlAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-17232051.post-67963314252540109252013-08-07T17:12:40.611-04:002013-08-07T17:12:40.611-04:00Let me elaborate on the RBC stuff (can't help ...Let me elaborate on the RBC stuff (can't help myself)...<br /><br />In an RBC world, the real rate is always at the RE equilibrium. That leaves two possible consequences for the nominal rate and the inflation rate. Either the inflation rate is endogenous and therefore the CB can't set the nominal rate. It has to be at r+pi. Or, the CB sets the nominal rate and therefore inflation just has to follow at pi = i-r. The second possibility seems to me to be totally destroyed by Howitt '92: when the CB raises the nominal rate inflation does *not* go up (Kocherlakota doesn't even believe it any more.) The first possibility, that the CB *can't* set the nominal rate, is just sort of ridiculous. Does anybody think that the Fed can't set IOR at 10%, and thus floor the FF rate at 10%? It's just silly, no? (I only raise the possibility because it's actually sometimes claimed, or implied, by RBC types, that the CB can't set rates.)<br /><br />The way I see it, the only thing that can rescue most of RBC is the assumption that we are everywhere and always at the RE equilibrium. And intrinsic structural parameter uncertainty just kills that possibility.<br />Khttps://www.blogger.com/profile/09226058602565040485noreply@blogger.comtag:blogger.com,1999:blog-17232051.post-50973125965048310092013-08-07T17:05:23.731-04:002013-08-07T17:05:23.731-04:00Just my two-bit opinion, but maybe there's mor...Just my two-bit opinion, but maybe there's more money to be made in empirical work than theory? I don't see too many grants given to scholars who want to work out a deductive theory. Firms, think tanks, government, etc. probably prefer to have stats to back up whatever issue they support rather than complex theory which is difficult, if not impossible, to empirically prove. But you're going to get more support with a statistic than a theory. <br /><br />How many half-assed statistical studies are out there, all in an effort to further an agenda? Lies, damn lies and statistics - but hey, it works. <br /><br />But again, just a theory. Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-17232051.post-55066725656033784982013-08-07T14:42:28.470-04:002013-08-07T14:42:28.470-04:00ivansml,
Evans and Honkepoja is exactly the kind ...ivansml,<br /><br />Evans and Honkepoja is exactly the kind of deviation from RE that CA was assuming above. I.e. errors and frictions. I agree that this is not the achilles heel of RE. The achilles heel is that "agents in those models usually ignore parameter uncertainty" as you say, which leads to the fundamental, inescapable and large deviations from RE. <br /><br />Regarding Howitt '92, the implications for RBC are slightly more subtle than I implied. Howitt shows, that under very general conditions, if r is below r* that results in rising i, and vice versa. Once you permit that the real rate may differ from the RE equilibrium (the "natural rate"), and that any difference of r relative to r* results in a rising, not falling, difference, then you have a model in which the central bank can directly control the real rate via changes in the nominal rate and therefore the path of nominal rate has a direct impact on output, in direct contradiction of standard RBC. It also requires you to have a theory of the dynamics of prices and output *away* from the RE equilibrium, though it may be that you can do it without sticky prices (I'll admit). Anyways, it's a long and complex debate, and not really central to my point, which I'll let rest on the asset pricing literature. <br />Khttps://www.blogger.com/profile/09226058602565040485noreply@blogger.comtag:blogger.com,1999:blog-17232051.post-85207581357576632862013-08-07T13:38:39.790-04:002013-08-07T13:38:39.790-04:00K: Sampling given interval at increasing frequency...K: Sampling given interval at increasing frequency is definitely not what people have in mind when they discuss convergence to rational expectations. Anyway, my point is that there's a difference between saying "agents cannot learn REE at all" and "agents will learn REE asymptotically, but given typical timescales asymptotic behavior is misleading".<br /><br />Howitt shows that learning process is unstable in a model where central bank pegs interest rate at constant value. Since most DSGE models these days assume that CB instead follows some kind of Taylor rule (with coefficients chosen to yield determinacy), this is not very applicable result.<br /><br />On the other hand, literature on adaptive learning (Evans & Honkapohja,...) shows that RE equilibria in NK and RBC models are typically learnable (though agents in those models usually ignore parameter uncertainty, which is a bit strange). And if drifts change over time, you can construct a model where agents learn and update their beliefs about hidden state by solving a filtering problem (e.g. Veronesi 1999 in asset pricing, though surely there must be some papers in macro too). I'm not really an expert on this, but suffice it to say, I think it's a bit more complicated than you say.ivansmlhttps://www.blogger.com/profile/00955626621561436702noreply@blogger.comtag:blogger.com,1999:blog-17232051.post-6780691299435950272013-08-07T13:06:25.218-04:002013-08-07T13:06:25.218-04:00Barkley, if you see above, I noted that experiment...Barkley, if you see above, I noted that experimental work was clearly a small share of the decrease in theory, given the chart. You can see that theory as a share drops from 57.6 to 32.4 from 1983 to 1993, but experimental work only goes up from 0.8 to 3.7. It's clearly not true that behavioral work's rising caused the "death of theory" that Noah describes.<br /><br />I realize the "behavioral non-theory works" may have been interpreted differently from how I meant, but you're clearly being unnecessarily obtuse.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-17232051.post-27557354943946355732013-08-07T12:21:58.490-04:002013-08-07T12:21:58.490-04:00ivansml,
"Are you sure about this?"
Ye...ivansml,<br /><br />"Are you sure about this?"<br /><br />Yes. I said "Make as many observations of that path over a *certain period* of time as you want." Not, "sample the process over an arbitrarily long period."<br /><br />So I think we agree. <br /><br /><br />"I'm not sure why such sampling scheme should be more relevant."<br /><br />I think you answer your own question: "Since the economy has indeed existed for only finite time, this may be of course important if it leads to quantitatively significant deviation from RE results"<br /><br />The deviations are, in fact, huge (not merely significant).<br /><br />"but it's not the same thing as proof that there is "no way for you to know the RE measure""<br /><br />Sure it is. First, you can't know it exactly, and for some theories that's fatal. RBC macro, for example, is not stable away from the RE equilibrium. (Howitt 92). Second, it's not reasonable to assume that past drift rates are stable over sufficiently long periods to estimate them reliably. I.e. it takes many decades to estimate drift rates of macro variables (drift of consumption *and* drift of consumption volatility are critical), and there is no reason to assume that those drift rates will persist over the next many decades over which the asset pricing measure is relevant. For sensible models the fragility of the estimators is a dominant consideration.<br /><br />Certainly, though, there are problems for which RE may be relevant. Early exercise of options, for example, since markets are relatively more complete and empirical drift rates don't matter under the risk neutral measure. Even there, though, there can be quite wide divergences in exercise behaviour, so it's questionable whether RE is a good approximation even under the most favourable circumstances.<br />Khttps://www.blogger.com/profile/09226058602565040485noreply@blogger.comtag:blogger.com,1999:blog-17232051.post-26552733592424352362013-08-07T11:08:31.312-04:002013-08-07T11:08:31.312-04:00K: "In the limit of infinite number of observ...K: "In the limit of infinite number of observations, the posterior for the volatility converges to a point (great!) but the posterior for the drift does not."<br /><br />If I may ask, are you sure about this? Let's say I sample the path at intervals of fixed length, then increments will be just iid samples from normal distribution, and my posterior will converge to true values (of both mean and variance) as sample size goes to infinity.<br /><br />If you instead mean sampling more and more points within a fixed interval (i.e. at higher frequency), then maybe you're right, but I'm not sure why such sampling scheme should be more relevant.<br /><br />The papers you cite are about impact of learning after finite period, when posteriors have not yet converged (so predictive density is fat-tailed due to parameter uncertainty). Since the economy has indeed existed for only finite time, this may be of course important if it leads to quantitatively significant deviation from RE results, but it's not the same thing as proof that there is "no way for you to know the RE measure".ivansmlhttps://www.blogger.com/profile/00955626621561436702noreply@blogger.comtag:blogger.com,1999:blog-17232051.post-42141405581877299302013-08-07T10:18:46.428-04:002013-08-07T10:18:46.428-04:00Stephen: "And the 600-pound man decreed that ...Stephen: "And the 600-pound man decreed that running is dead."<br /><br />Fight! Fight! Fight! Fight!<br />Khttps://www.blogger.com/profile/09226058602565040485noreply@blogger.comtag:blogger.com,1999:blog-17232051.post-7893923966981286112013-08-07T10:06:45.730-04:002013-08-07T10:06:45.730-04:00Fun fact:
Assume you know the volatility but not ...Fun fact:<br /><br />Assume you know the volatility but not the drift of a Brownian motion, and you observe a path over some period. Assume also you have some prior for the drift rate. The posterior drift rate is a function only of the prior and the difference between the beginning and end point of the path. Observing more points in the interior of the path adds no information.<br />Khttps://www.blogger.com/profile/09226058602565040485noreply@blogger.comtag:blogger.com,1999:blog-17232051.post-64703824120514272013-08-07T10:03:48.237-04:002013-08-07T10:03:48.237-04:00And the 600-pound man decreed that running is dead...And the 600-pound man decreed that running is dead.Stephen Williamsonhttps://www.blogger.com/profile/01434465858419028592noreply@blogger.com