tag:blogger.com,1999:blog-2198942534740642384.comments2018-02-23T10:29:45.641-08:00Econometrics Beat: Dave Giles' BlogDave Gileshttp://www.blogger.com/profile/05389606956062019445noreply@blogger.comBlogger3883125tag:blogger.com,1999:blog-2198942534740642384.post-79598602794827443132018-02-23T09:59:24.802-08:002018-02-23T09:59:24.802-08:00Hello.This article was extremely motivating, parti...Hello.This article was extremely motivating, particularly since I was browsing <br />for thoughts on this matter last week.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-52053881985342349572018-02-23T08:47:27.703-08:002018-02-23T08:47:27.703-08:00Coming from a background in statistics rather than...Coming from a background in statistics rather than econ, I would like to share some of my thoughts on this discussion.<br /><br />For this discussion you use a latent variable specification of the probit model. If a latent variable is to be assumed, a normal random variable seems like a pretty natural choice. This latent variable specification allows you to introduce heteroscedasticity into the model as you discussed. <br />The probit models tends to not be seen much outside of economics; elsewhere everyone tends to default to logit models. Of course the logit model can be given an almost identical latent variable specification: Y* = XB + e, where e is instead assumed to be logistic distributed.<br /><br />For the logit model, however, this specification is quite uncommon. Logistic distributions are a somewhat exotic distribution, they are unlikely to come about naturally in the same way as normal distribution (central limit theorem and maximal entropy considerations). Instead the logit model is usually specified as LOG-ODDS = XB. While the two specifications are mathematically identical, with the alternative specification we think of the observed responses as bernoulli random variables with varying propensities to success. No latent variable is introduced. This is a (to me at least) a much more natural specification as the logistic distribution is such an unnatural distribution. This alternative specification has an important consequence though, to introduce the same kind of heteroskedasticity as in the probit model, the link function would have to vary for different individuals. This would break the log-odds interpretation of the logit model, and so this kind of heteroscedasticity doesn't make much sense in logit models. What to make then of the results of a (in the context of the probit model) heteroscedasticity test for the logit model? In this case it doesn't indicate heteroscedasticity, but rather indicates some kind of non-linearity in the effects or other misspecification. It seems to me then that the discussion you presented here heteroscedasticity actually is a more general discussion on model misspecification. To me, the choice of model (logit/probit), and even the specification of the model (link-function/latent-variable) comes in to play in understanding how to interpret the results of model misspecification test.Kroutonerhttps://www.blogger.com/profile/05758537158353883675noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-51704364283882719662018-02-20T05:09:08.555-08:002018-02-20T05:09:08.555-08:00Thank you Prof Thank you Prof GOPI KRISHNAN K.K.VIJAYARAGHAVANhttps://www.blogger.com/profile/02981320029069586508noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-52224977343841866382018-02-19T15:35:16.357-08:002018-02-19T15:35:16.357-08:00Professor Giles-This is a great blog! You have pos...Professor Giles-This is a great blog! You have posted lots of interesting things on the ARDL Bounds Testing approach to cointegration. One question I had, and you may have touched on this earlier, is the requirement for one of the variables to be weakly exogenous for the ARDL method to be valid. I can't seem to understand this reading the actual 2001 paper, but in Walter Enders "Applied Econometric Time Series", 3rd edition, on page 411 the author states that to use the error correction test of the ARDL it is necessary to assume that one of the variables is weakly exogenous. If so would this mean that if all variables in a system react to the error correction term, that the ARDL method is not valid? Thanks for any insight, <br />Bill MilesWilliam Milesnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-68477904606358165972018-02-19T07:02:01.264-08:002018-02-19T07:02:01.264-08:00No - you'd go back to the version of the model...No - you'd go back to the version of the model you had BEFORE you added the extra lag(s) for TY testing purposes. Those extra lags are ONLY for ensuring that the Wald statistic has its usual asymptotic distribution - nothing else.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-56544171736474847932018-02-19T06:58:24.895-08:002018-02-19T06:58:24.895-08:00Yes - and you use TWO "extra" lags of al...Yes - and you use TWO "extra" lags of all variables in the specification of the VAR (but in the Wald test), as "2" is the maximum order of integration within the set of variables under consideration.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-31917554471693224312018-02-18T21:53:41.240-08:002018-02-18T21:53:41.240-08:00Hello Professor. If The series are I(1), I(1), I(1...Hello Professor. If The series are I(1), I(1), I(1) and I(2). Can we use Toda & Yamamoto test in this situation?Manzoor Ahmadhttps://www.blogger.com/profile/06350334120124783265noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-10671396339924813432018-02-18T19:51:28.557-08:002018-02-18T19:51:28.557-08:00Dear Prof Giles,
I have done causality test using ...Dear Prof Giles,<br />I have done causality test using TY approach. Can i use the same model to interpret impulse responsive results? GOPI KRISHNAN K.K.VIJAYARAGHAVANhttps://www.blogger.com/profile/02981320029069586508noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-46174920019714091212018-02-15T13:14:26.091-08:002018-02-15T13:14:26.091-08:00Thank you Dr. Giles!Thank you Dr. Giles!Puneethttps://www.blogger.com/profile/00752994732971657100noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-25742210667889297022018-02-14T09:56:10.688-08:002018-02-14T09:56:10.688-08:00You might want to take a look at these 2 books:
h...You might want to take a look at these 2 books:<br /><br />https://www.amazon.com/Introduction-Modern-Bayesian-Econometrics-Lancaster/dp/1405117206/ref=pd_sim_14_5?_encoding=UTF8&pd_rd_i=1405117206&pd_rd_r=F5BZ6SJA3K00FHKJHGCM&pd_rd_w=g0F68&pd_rd_wg=n1fbp&psc=1&refRID=F5BZ6SJA3K00FHKJHGCM<br /><br /><br />https://www.amazon.com/Introduction-Bayesian-Econometrics-Edward-Greenberg/dp/110743677X/ref=pd_cp_14_1?_encoding=UTF8&pd_rd_i=110743677X&pd_rd_r=BDH7P24FMTVFPGZS1Q4N&pd_rd_w=mcLTl&pd_rd_wg=QYfli&psc=1&refRID=BDH7P24FMTVFPGZS1Q4N<br /><br /><br />Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-37199356908727189212018-02-13T20:17:38.543-08:002018-02-13T20:17:38.543-08:00Thank you Dr. Giles for sharing your materials. Ju...Thank you Dr. Giles for sharing your materials. Just curious to know: Which book you would recommend to a practitioner/applied researcher i.e. someone who wants an intuitive understanding but wants to be able to implement the procedures in R or SAS. Some of your earlier posts cite Zellner's (1971) book but it seems to be a theoretical book. Thank you.Puneethttps://www.blogger.com/profile/00752994732971657100noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-57838155722415794172018-02-10T20:25:48.360-08:002018-02-10T20:25:48.360-08:00Hi Dave: This is extremely relevant in intraday mo...Hi Dave: This is extremely relevant in intraday modelling in finance because, in that case, one has tons ( multiple thousands ) of observations, so the chances of accepting any null hypothesis ( such as say beta = 1 in a simple regression ) is essentially zero. ( because the t-test wasn't designed for 1000's of observations ). But, there are bootstrapping approaches that one can use to check whether the estimated coefficient really does improve forecasts compared to coefficient under the null. Clark and McFadden ( Federal Reserve of St. Louis, IIRC ) have a ton of papers regarding the approach. Great post. Thanks.mark leedshttps://www.blogger.com/profile/13213841692738932471noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-9223639655531221322018-02-07T09:10:14.623-08:002018-02-07T09:10:14.623-08:00Roland - yes, there is definitely a finite-sample ...Roland - yes, there is definitely a finite-sample bias in the parameter estimates. In principle, this could be reduced by bootstrapping.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-49562270949436385492018-02-07T08:51:47.931-08:002018-02-07T08:51:47.931-08:00Hello Prof. Giles, terrific post as always. Could ...Hello Prof. Giles, terrific post as always. Could I kindly ask you a follow up question? You're carrying out step 7 where you estimate a restricted ECM, and you are using Least Squares to estimate an ARDL with lagged dependent variables. Wouldn't using OLS to do so result in a finite sample bias for the coefficient estimates as you have pointed out separately (with consistency maintained)? Does this mean that there is a some amount of bias in your restricted ECM estimates? or should instead I be taking into account the fact that EViews does some kind of adjustment on its own in the background and that consequently there is no bias in the restricted ECM estimates? I am assuming that the ARDL add-in for EViews was not used here... Thank you in advance.Rolandnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-10352358832290881822018-02-04T05:49:11.687-08:002018-02-04T05:49:11.687-08:00Santosh - thanks for pointing this out. Now fixed....Santosh - thanks for pointing this out. Now fixed. DGDave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-31616393480126619712018-02-04T01:45:00.433-08:002018-02-04T01:45:00.433-08:00Thank you posting the slides, Prof. Giles. The lin...Thank you posting the slides, Prof. Giles. The link to # 11 is not working.Santosh Dashhttps://www.blogger.com/profile/02016226999263087762noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-3490797251472571492018-02-03T13:40:31.991-08:002018-02-03T13:40:31.991-08:00Not an ARDL model of this "modern" form ...Not an ARDL model of this "modern" form - there can;t be an error correction term, because you can;'t have cointegration if all of the variables are I(1). However, you would just use an "old fashioned" ARDL model. All variables would be in levels (not differences). The Y variable could be "explained" by lagged values of Y, as well as lagged (and perhaps a current) values of X variables.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-35073081531802252342018-02-03T13:01:00.516-08:002018-02-03T13:01:00.516-08:00can i use ardl with stationary variable at level (...can i use ardl with stationary variable at level (all the variable are i(0))?<br />thanks a lotAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-12845508660344328412018-01-30T20:19:34.209-08:002018-01-30T20:19:34.209-08:00Dear Dave,
we are waiting for your valuable contri...Dear Dave,<br />we are waiting for your valuable contribution about nonlinear ARDL and Quantile ARDL, as it is now emerging as important ways of analysis.<br />Thank you for your informative website. It improved the skills of many of us.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-25498044223686948892018-01-22T16:04:25.699-08:002018-01-22T16:04:25.699-08:00Yes, it can.Yes, it can.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-47633151992939008922018-01-22T09:14:46.801-08:002018-01-22T09:14:46.801-08:00Hello Sir,
Dear Professor
In the TODA AND YAMA...Hello Sir,<br />Dear Professor<br /> In the TODA AND YAMAMOTO procedure can the trend be introduced as an exogenous variable?<br /><br />Best wishesecointelligencyhttps://www.blogger.com/profile/09764228437124629520noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-63989253481655370962018-01-22T05:54:31.553-08:002018-01-22T05:54:31.553-08:00It makes no sense, and if you try to actually do t...It makes no sense, and if you try to actually do this you'll find that everything "falls apart". Try it! :-)Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-37290972916176738722018-01-21T22:59:16.090-08:002018-01-21T22:59:16.090-08:00Dear Prof Giles,
May i ask whether it is okay to ...Dear Prof Giles,<br /><br />May i ask whether it is okay to include two perfectly negatively correlated series (i.e. one series is the negative of another series) in a VAR framework to test for Granger causality?<br /><br />Thanks in advance.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-46897696709715689722018-01-18T00:29:01.354-08:002018-01-18T00:29:01.354-08:00This comment has been removed by the author.Wassim Dboukhttps://www.blogger.com/profile/15229578596461920962noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-46289530424020270962018-01-08T17:17:59.318-08:002018-01-08T17:17:59.318-08:00That's one of the tests R actually does for it...That's one of the tests R actually does for its random number generators -- using Massart's inequality, which is the best bound for the one-sample KS-statistic without assuming continuity.Thomas Lumleyhttps://notstatschat.tumblr.comnoreply@blogger.com