tag:blogger.com,1999:blog-17232051.post3926479792760304349..comments2024-03-28T03:16:14.104-04:00Comments on Noahpinion: The "let's lump a bunch of models together" approachNoah Smithhttp://www.blogger.com/profile/09093917601641588575noreply@blogger.comBlogger4125tag:blogger.com,1999:blog-17232051.post-40437044362513916092012-04-11T22:38:12.630-04:002012-04-11T22:38:12.630-04:00You make a good point, TED.You make a good point, TED.Noah Smithhttps://www.blogger.com/profile/09093917601641588575noreply@blogger.comtag:blogger.com,1999:blog-17232051.post-88983032821765165612012-04-11T18:02:41.467-04:002012-04-11T18:02:41.467-04:00Isn't "et al" short for "et ali...Isn't "et al" short for "et alii"? I think "et. al." is one period too many.TEDhttps://www.blogger.com/profile/07765245186357910074noreply@blogger.comtag:blogger.com,1999:blog-17232051.post-91196222207494700012012-04-11T14:32:37.859-04:002012-04-11T14:32:37.859-04:00Thanks for the link. I very much agree with one t...Thanks for the link. I very much agree with one thing you wrote here " (Actually in practice ... ."Thus I very strongly disagree with something else you wrote "But let's put this issue aside for now.)". I am typing on an iPad so I will be brief (no applause or sighs of relief are needed). I see no reason to believe that the so called "structural models" are policy irrelevant. In fact I am absolutely sure that they are vulnerable to the Lucas critique.<br /><br />From the first sentence in parentheses, I am also sure you agree. The fact that a model is internally consistent and includes only rational maximizing agents does not imply that it is policy irrelevant. This would be true if the assumed objective functions and constraints correspond to the actual real world objective functions and constraints, that is, it would be true if the model weren't a model but rather the truth the whole truth and nothing but the truth.<br /><br />A model in which we assume that agents have different desires than their true desires can fit the data OK for one policy regime, but not at all for another. This is obvious. A simple example, I observe the sexes of legally married couples and conclude that gay people aren't interested in marriage. My forecast of the effect of policy changes will be way off. I think this is a perfectly fair comparison vs say estimates of the effect of fiscal stimulus on employment based on models in which the labor market clears and estimates of labor supply elasticities from cross sectional data.<br /><br />Such models are ( finally) out of fashion. However, I think the empirical failures of new Keynesian models are different in degree but not in kind. I confess ignorance as usual and report my guess as to what is in the literature I haven't kept up with.<br /><br />As far as I know, consistent micro founded models can fit moments of the data using roughly one free parameter for every phenomenon to be explained. As far as I know, the interaction of the models and the data is about what one would expect if the models were fundamentally wrong. If they are, then they are not useful for poicy analysis.<br /><br />The reason I care about forecasting performance is not that I aim to forecast, but that it is the only possible way to determine if a model has any scientific value at all. Policy invariance is just one of the desirable features a model might have. I see no reason to seriousely entertain the conjecture that existing micro founded macromodels (very definitely including the one's on which I am currently working) have any desirable features.Roberthttps://www.blogger.com/profile/14455788499385673507noreply@blogger.comtag:blogger.com,1999:blog-17232051.post-17388614364245801432012-04-11T00:48:20.109-04:002012-04-11T00:48:20.109-04:00What about the DSGE-VAR approach?
http://www.ecb.i...What about the DSGE-VAR approach?<br />http://www.ecb.int/events/pdf/conferences/schorfheide.pdf<br />It kind of merges the two approaches in a meaningful way.Anonymousnoreply@blogger.com