tag:blogger.com,1999:blog-2198942534740642384.comments2016-05-24T21:50:47.129-07:00Econometrics Beat: Dave Giles' BlogDave Gileshttp://www.blogger.com/profile/05389606956062019445noreply@blogger.comBlogger3356125tag:blogger.com,1999:blog-2198942534740642384.post-87362794265032141832016-05-24T15:39:04.196-07:002016-05-24T15:39:04.196-07:00Michael - thanks for the comments. First, regardin...Michael - thanks for the comments. First, regarding Bayesian inference, I totally agree - my PhD in '75 was in Bayesian econometrics (and see various posts on this blog).<br /><br />The point about the pre-testing literature (which goes back to Ted Bancroft's work in the 1940's) is the following:<br /><br />1. It is dealing only with non Bayesian inference, where the sampling distribution is a core concept. Hence the emphasis is on issues such as Bias, MSE, etc.<br /><br />2. It does not condone pre-testing (which only very rarely is an optimal strategy), and is always inadmissible. <br /><br />3. Rather, the emphasis is on pointing out the (usually adverse) consequences of pre-testing.<br /><br />4. (Non-Bayesian) applied researchers invariably pre-test, often in complex ways. Then they interpret their results as if this pre-testing had not taken place. That's where the problem lies. Their final estimators and tests don't have the properties that they ascribe to them.<br /><br />The literature on pre-test issues continues to grow, which I believe is a good thing. A lot of this literature can be quite technical, so some simple illustrations can be helpful. Simulation exercises usually convince students about the associated issues pretty quickly too.Dave Gileshttp://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-73494791788509165782016-05-23T23:30:31.351-07:002016-05-23T23:30:31.351-07:00Short answer - NO. In both cases. I've been me...Short answer - NO. In both cases. I've been meaning for ages to do a post on logs vs. levels when it comes to unit roots. Must do!Dave Gileshttp://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-89664615126844409682016-05-23T23:28:36.553-07:002016-05-23T23:28:36.553-07:00Dear Prof,
Suppose a variable is non-stationary in...Dear Prof,<br />Suppose a variable is non-stationary in level. If we transform it to log, will it become stationary? Or,at least, will the probability of becoming stationary increase?<br /><br />Thank you.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-48327272953956843632016-05-23T02:58:25.057-07:002016-05-23T02:58:25.057-07:00From a superficial look at the links, I gather you...From a superficial look at the links, I gather you have found that pre-testing can be justified by certain loss functions (e.g., mean squared error) under certain circumstances. <br /><br />If I understand correctly, this analysis treats the likelihoods as probability distributions over the parameters, which makes sense only if they are are seen as, in particular, the posterior distributions proceeding from an uninformative prior.<br /><br />Question: is there any case for pre-testing now that doing full Bayesian computation is easy (as of course it was not when you wrote the survey paper)? It seems to me if you can specify a loss function, you might as well report your whole posterior together with expected loss as a function of the parameters. Then the loss-minimizing point-estimate is evident, together with much else; the purpose is clear; and alternative loss functions can be considered without muddling up the statistical starting point. <br />Michael Margolishttp://www.blogger.com/profile/00454854828001494319noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-69329573011185665202016-05-22T13:44:12.673-07:002016-05-22T13:44:12.673-07:00As long as your dependent variable is the same in ...As long as your dependent variable is the same in each case you rank the models using AIC or SIC (see my other posts on this topic). Note - this is not a test, just a ranking procedure.Dave Gileshttp://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-78551820830389558772016-05-22T13:41:07.445-07:002016-05-22T13:41:07.445-07:00Then you model is probably mi-specified - e.g., om...Then you model is probably mi-specified - e.g., omitted regressors; wrong lag structure, etc.Dave Gileshttp://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-264069541318089722016-05-22T13:39:10.057-07:002016-05-22T13:39:10.057-07:00Either would be OK - differencing an I(0) variable...Either would be OK - differencing an I(0) variable still leaves it stationary.Dave Gileshttp://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-57911827594156691512016-05-22T11:29:45.156-07:002016-05-22T11:29:45.156-07:00In step 7: What if the Z(-1) coefficient is positi...In step 7: What if the Z(-1) coefficient is positive and significant? What can infer of that result? No long run cointegration?<br />brauliohttp://www.blogger.com/profile/09142703315035505040noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-41581719227969923662016-05-22T10:28:37.619-07:002016-05-22T10:28:37.619-07:00Dear sir, i want to compare two models on the basi...Dear sir, i want to compare two models on the basis of their best best..in one model i have used cash flow as predictor of dividend along with some control variables and in another model i have used earnings as predictor of dividend along with control variables. i want to test that which one variable among cash flow and earning is the best predictor of dividend. kindly guide me that what is the suitable statistical test should i used. thanksMuhammad ilyashttp://www.blogger.com/profile/09019985905158223452noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-40649181031105599192016-05-19T02:33:39.017-07:002016-05-19T02:33:39.017-07:00I'm not the original commenter, but I just wan...I'm not the original commenter, but I just wanted to say that that was a remarkably restrained and informative response to a quite rude comment. Kudoz to you, Prof Giles. Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-20766220782747256452016-05-17T13:55:04.177-07:002016-05-17T13:55:04.177-07:00Thanks for your reply.Thanks for your reply.Thom Goodwinhttp://www.blogger.com/profile/00676882553395837298noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-87047336485464440642016-05-17T13:50:53.157-07:002016-05-17T13:50:53.157-07:00Thom - I should add that the assessment also inclu...Thom - I should add that the assessment also includes several assignments, and an empirical project worth 20% where the students have to apply some of the techniques learned in the course to a problem of their choice. Dave Gileshttp://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-782356529518050852016-05-17T13:47:53.048-07:002016-05-17T13:47:53.048-07:00Thom - we use Greene as a background text (partly...Thom - we use Greene as a background text (partly because the students have it already from the previous compulsory "intro" course). When I'm teaching this course I have my own detailed slides for the bulk of the material. There are weekly computing labs., where the students get to apply the material to real data. I use both EViews and R throughout the course - mainly EViews for the MLE part (and IV/GMM if that's included), but almost exclusively R for the Bayesian material.Dave Gileshttp://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-67563531705068140132016-05-17T13:41:45.658-07:002016-05-17T13:41:45.658-07:00Are you able list the resources prescribed for thi...Are you able list the resources prescribed for this course?Thom Goodwinhttp://www.blogger.com/profile/00676882553395837298noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-25152747141763371452016-05-17T13:35:37.984-07:002016-05-17T13:35:37.984-07:00Thanks for the feedback. You're entitled to yo...Thanks for the feedback. You're entitled to your opinion, of course, but it would mean more if it was put in context. If you bothered to check our dept. website you'll see that at the grad. level we offer a general (theory/applied) econometrics course for all incoming students. Apart from the theory elective mentioned in this post, we offer three applied econometrics electives - a general one; one on applied time-series econometrics; and one on applied microeconometrics. And BTW, the students DO use the topics from the "Themes" course - in their other courses, and elsewhere. Sorry if that sounds defensive, but our students get a pretty balanced "diet" of econometrics.Dave Gileshttp://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-74918019456610815732016-05-17T11:48:23.397-07:002016-05-17T11:48:23.397-07:00uff, a typical exam with a lot of theoretical topi...uff, a typical exam with a lot of theoretical topics that students will not use ever.<br />Good job|Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-22035861504328174912016-05-13T15:15:44.698-07:002016-05-13T15:15:44.698-07:00There is no way I would attempt to do any serious ...There is no way I would attempt to do any serious ARDL modelling with so few observations.Dave Gileshttp://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-57032640030942994502016-05-13T11:01:21.319-07:002016-05-13T11:01:21.319-07:00Sir I saw some Papers in which no of observations ...Sir I saw some Papers in which no of observations were 14 to 15 and they applied ARDL.me confused I have 14 years time series annual data.can I use this?check this https://www.researchgate.net/publication/281857078_MACROECONOMICS_INDICATORS_AND_BANK_STABILITY_A_CASE_OF_BANKING_IN_INDONESIAloving soulhttp://www.blogger.com/profile/11632612922377893039noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-66558054293449422972016-05-10T03:05:55.402-07:002016-05-10T03:05:55.402-07:00In my undergraduate econometrics class, one studen...In my undergraduate econometrics class, one student asked quite astute question of why these OLS used when far more accurate estimation procedures are available.<br /><br />The lecturer answered that the choice was between paying for the computer time or having money from a research assistant. The choice was obvious.Jim Rosehttp://www.blogger.com/profile/02233668500637892711noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-40738892286919498482016-05-09T11:17:16.714-07:002016-05-09T11:17:16.714-07:00In the early 70s, I worked briefly as a research a...In the early 70s, I worked briefly as a research assistant at Michael Evans' econometrics firm. Perhaps Mike's greatest claim to fame was using PDLs (Almon lags) in estimating investment functions. In any case, a little bit of algebra made estimating the two or three unbound PDL coefficients no worse than the same number of independent variables.<br /><br />Computing of the late 60s / early 70s was expensive enough that econometrics required some smarts about computation. Pre-computed matrix inverses that could be used easily with multiple dependent variables, simplified PDLs and other shortcuts were part of the toolkits.Walt Frenchnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-71921149678248876582016-05-08T18:41:55.450-07:002016-05-08T18:41:55.450-07:00You can use the ADF test or the KPSS test to test ...You can use the ADF test or the KPSS test to test for the stationarity of the data even if the series is non-normal. You should avoid the Johansen procedure for testing for cointegration in this case, but the Engle-Granger 2-step procedure is still valid.Dave Gileshttp://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-68798400121923207762016-05-08T18:20:11.471-07:002016-05-08T18:20:11.471-07:00No - all of the series have to be I(1).No - all of the series have to be I(1).Dave Gileshttp://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-32258000880635054312016-05-08T07:05:00.773-07:002016-05-08T07:05:00.773-07:00Hello! Many thanks for your blog! Could I use Joha...Hello! Many thanks for your blog! Could I use Johansen test with I(1) and I(0) process?<br />ThanksAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-8116769453488310332016-05-08T06:44:10.635-07:002016-05-08T06:44:10.635-07:00Hello! Many thanks for your blog, it is really coo...Hello! Many thanks for your blog, it is really cool and usefull. Is there a way to check the stationarity and cointegration for abnormal series (large kurtosis)? Many thanks for your answer! Mr. Z.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-40397426574436531322016-05-07T04:34:35.954-07:002016-05-07T04:34:35.954-07:00Mr Ozcan maybe you are able to retain the dummy in...Mr Ozcan maybe you are able to retain the dummy in the case that your ARDL model faces stability problems with Cusum tests and the dummy variable ''fixes'' f.e the CUSUM of Squares diagnostic for 5% level. I am not sure if that is legitimate although i have seen it before in papers. What do you think Dr Giles?Anonymousnoreply@blogger.com