Thanks Patrick, Let me try to paraphrase. You (1) recognise the theoretical assumptions don't hold in applied econometrics; (2) argue that this doesn't render the results meaningless but means we must use a range of methods. I actually argued that 'strictly speaking' the results are meaningless. A curious turn of phrase I know. The point I'm trying to make is that it is quite bizaar that theoretical econometrics should go to such lengths at formal rigour when, in application, the most basic of assumptions are blatently broken. By the 'rigorous' standards of theoretical econometric this indeed renders applied results totally meaningless. I accept that often the results must be telling us something because I don't abide by the 'rigour' of theoretical econometrics. But, in general, I cannot really work out what it is they are telling us because it involves working out meaning without the theoretical assumptions which render that meaning obvious (i.e. 'meaningful' according to the theoretical standards upon which the analysis is constructed). Moreover, neither theoretical nor applied econometric literature etc. are of much use in addressing this basic problem (which is really quite astonishing since any science worth its salt should be extremely concerned to establish the meaning of its statements!!) The state of econometrics is thereby in a horrible mess. Your 'hope' that some good comes out of all this appears no more than a 'hope'; it is not justified by any stated argument in econometrics texts (or elsewhere). Maybe your hope will turn out to be justified - my problem is that there does not appear to be any established basis to argue for or against this hope. Or, to put it more starkly, the only well known basis we do have is theoretical econometrics but this would tell us that applied econometrics is, indeed, hopeless, as discussed above. Best wishes, Andy On 20 Nov 2001, at 10:59, Patrick L. Mason wrote: > Andrew: > > I do not disagree with your assessment of data mining. I'll go one step > further. Journals will often reject papers that "get the wrong result." I > might add however that editors also tend to reject theoretical papers that > challenge the prevailing orthodoxy (of the journal). I even agreement with > your argument that the art of econometrics carries a high degree of > subjectivity. > > But, I do disagree with the argument that empirical results are meaningless > unless all econometric theory assumptions are perfectly satisfied. This is > an impossible constraint. It will never happen. Econometric equations > assume that one has a "well" or "fully" specified model, i.e., no missing > variables. Well, there are always missing variables. You'll note in my > previous postings I have consistently argued that economic theory decides > causality (and perhaps also substantive significance). > > Given an accepted theory, we can then use econometrics (with all of its > imperfections and artistic subjectivities), simulations, experimental data, > natural experiments, calibrations, input-output analysis, etc. to assess > the strength of relationships implied by the accepted theory. Rarely, will > any one study be sufficient to decide theoretical debates, but the hope is > that a collection of studies employing different approaches and assumptions > will be sufficient to provide the desired level of quantitative insight. > > peace, patrick > > > At 10:33 AM 11/20/01 +0000, Andrew Brown wrote: > >Allin, > > > >I doubt very much whether there has ever been one refereed journal > >article which 'did econometrics right', even on your own criteria > >(which I think are a small subset of the full required criteria). This is > >the case even for say the Hendry type procedures which ostensibly > >are all about testing for the required properties of the error term. All > >practicing econometricians I know end up bending their own rules > >(the rules you describe) to get the 'right' results. Data mining is > >simply endemic and inescapable in a world of 'publish or die'. More > >interesting is to actually ask someone doing econometrics with > >some data set what they *really* think it tells them - and this may > >give some scientifically interesting answers - but what ever the > >answers are they always involve making difficult judgements in the > >face of some or other necessary theoretical assumption being > >broken. This 'art' of econometrics - making judgements as to the > >quantitative significance of breakdowns of theoretical assumptions > >seems an entrirely 'tacit' art with no rules written down in > >textbooks. For myself, it is an art that I struggle with - I simply > >cannot work out what the output tells me without the assumptions > >that would make it unambiguously meaningful. It's not a question of > >being an imperfect, imprecise buisines as Patrick seems to > >suggest. It's more a question of a 'perfect' formal theory (theoretical > >econometrics) whose application outside of such perfection (in the > >real world where some necessary assumption or other does not > >unambiguously hold) is, strictly speaking, *utter nonsense*!! > > > >Re Lawson: I think you must disagree with Lawson's > >characterisation of neoclassical economics somewhere along the > >line since, whether one agrees with his critique or not, his > >characterisation firmly ties the growth of neoclassical economics to > >the growth of econometrics - something you appear to want to > >deny. I'm yet to read your article on Lawson (I recently did a web > >search and it was one of the few articles that I could not actually > >download)- would be interested in your reflections on the exchange. > > > >Thanks, > > > >Andy > > > > > > > > > >On 19 Nov 2001, at 14:47, Allin Cottrell wrote: > > > > > On Mon, 19 Nov 2001, Andrew Brown wrote: > > > > > > > 1) Whether the only game or not, the trouble I find with typical > > > > econometric methods (typically based on multiple regression), is > > > > that the theoretical assumptions are *always* broken... > > > > > > It's difficult to do econometrics right -- but doing it right involves > > > rigorous testing for violation of the required statistical properties > > > of the error term, and the use of alternative estimators where needed. > > > > > > > 2) Does 'calibration' count as another game so suggesting there is > > > > another game in town after all? > > > > > > I think you're right, calibration is an alternative -- for researchers > > > who are certain in advance that their theory is right and who just > > > want to quantify parameters rather than test hypotheses. > > > > > > > 3) Why do you (Allin) reject Tony Lawson's characterisation of > > > > neoclassical economics as deductivist? As you know, Lawson's > > > > arguments tie econometrics closely to neoclassical economics, if > > > > they are accepted. > > > > > > I don't think I disagree with Tony Lawson's characterization of > > > neoclassical economics as such. I'm less persuaded by his critique of > > > econometrics. This is a rather complex subject; I wrote about it in > > > "Realism, Regularities and Prediction" (Review of Social Economy, > > > LVI/3, Fall 1998). > > > > > > Allin Cottrell. > > > >
This archive was generated by hypermail 2b30 : Sun Dec 02 2001 - 00:00:06 EST