tag:blogger.com,1999:blog-2198942534740642384.post293605345998325338..comments2023-10-24T03:16:41.009-07:00Comments on Econometrics Beat: Dave Giles' Blog: Integrated & Cointegrated DataDave Gileshttp://www.blogger.com/profile/05389606956062019445noreply@blogger.comBlogger80125tag:blogger.com,1999:blog-2198942534740642384.post-56547332666802741602019-08-30T06:11:16.617-07:002019-08-30T06:11:16.617-07:00Often, that will be sufficient, depending on how &...Often, that will be sufficient, depending on how "complicated" your break is.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-78998560139276013802019-08-29T21:21:39.951-07:002019-08-29T21:21:39.951-07:00Prof Giles, to incorporate structural breaks in co...Prof Giles, to incorporate structural breaks in cointegration and VECM, is it enough to add a dummy variables for a specific year in the exogenous box on VAR, Cointegration, and VECM?Amir https://www.blogger.com/profile/08355544087004838174noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-6415443898725327772019-07-11T08:16:47.745-07:002019-07-11T08:16:47.745-07:00No, that would be the short-run m.p.c. To get the ...No, that would be the short-run m.p.c. To get the long-run mp.c. you need to take into account the lags of the dependent variable, in the usual way. And of course the l.r.m.p.c. exceeds the s.r.m.p.c., because the model is dynamically stable (if you look at the coefficient values).Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-21259534058117152872019-07-11T08:07:03.088-07:002019-07-11T08:07:03.088-07:00The very definition of cointegration is as follows...The very definition of cointegration is as follows. If we have 2 or more series that are all integrated of order d (that is, they are all I(d)), and there exist one or more linear combinations of the series that are integrated of order (d-k), where k>0, then the I(d) series are "cointegrated". The most common case is where the series are all I(1), but a linear combination of them is I(0) (and hence stationary), then the series are cointegrated. However, if all of the series are I(0) to begin with, then any linear combination of them will also be I(0). This isn't cointegration - it's actually just the standard situation that we actually assume to be the case when we first learn about fitting a regression. We assume/pretend that we're in a stationary world. If we're not, then it makes no sense to estimate the coefficients, as they can't be constant.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-72198653456693512352019-07-09T02:14:48.127-07:002019-07-09T02:14:48.127-07:00Prof. Dave, thank you so much for the excellent di...Prof. Dave, thank you so much for the excellent discussion. However, I have a quick question. You said "The ADF test indicates that both series are stationary, so they can't be cointegrated." What does that exactly mean? Does it mean that consumption and disposable income are not in a long-run relationship? If the answer is "yes", then how is the regression in level meaningful?Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-12291365340301113132019-07-09T02:03:53.603-07:002019-07-09T02:03:53.603-07:00You said "Now the residuals are serially inde...You said "Now the residuals are serially independent, and the long-run mpc works out to be 0.870." Is 0.870 a typo? Would not it be 0.356?Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-72887035842830306692018-10-11T07:14:58.380-07:002018-10-11T07:14:58.380-07:00You should test for the order of integration of ea...You should test for the order of integration of each series using a formal test (such as ADF or KPSS). If the series are integrated, of the same order, then you should test for cointegration. The ACF and PACF plots just provide informal visual information.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-21331596803956431722018-10-08T00:49:47.944-07:002018-10-08T00:49:47.944-07:00Dear profesore Giles,
I have seen that you use co...Dear profesore Giles,<br /><br />I have seen that you use correlogram of variables to acees their level of integration. However, I was wondering is it necessary to analise the ACF and PACF of variables before enbarking on cointegration tests. Namely, recently I've seen in the literature a statement like "since 1st lag PACF of both variable does not exceed 0,75 presence of cointegration is not tested". I am not sure I understand why it is not tested. Can you please share some insight? Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-55752456786214102442016-05-29T03:27:35.264-07:002016-05-29T03:27:35.264-07:00Thank you very much.
I appreciate your answer!Thank you very much.<br />I appreciate your answer!anonymoushttps://www.blogger.com/profile/10297632631843398427noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-35856488305886279592016-05-28T18:25:39.927-07:002016-05-28T18:25:39.927-07:00Whichever package you're using it's very s...Whichever package you're using it's very straightforward. I've prepared a post for you, here: http://davegiles.blogspot.ca/2016/05/forecasting-from-error-correction-model.htmlDave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-60553879634524617152016-05-28T13:00:40.993-07:002016-05-28T13:00:40.993-07:00Prof. Giles!
Thank you for an informative post!
...Prof. Giles!<br /><br />Thank you for an informative post!<br /><br />I would like to forecast consumption using disposable income as an exogenous variable in R. I have found a cointegrated relationship between the two variables and have estimated an Error Correction Model. However I have difficulties to find a function that produces forecasts for ECM. I've heard that you must create your own function in order to do this forecast, but this is a little bit to hard for me since Im relatively new to R. Do you have any suggestions on how I can create this function, or does it exist a function in R that will produce these forecasts?anonymoushttps://www.blogger.com/profile/10297632631843398427noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-19279137974876191022015-11-20T09:26:01.810-08:002015-11-20T09:26:01.810-08:00This could be happening for a number of reasons.
...This could be happening for a number of reasons. <br />Keep in mind that unit root tests typically have pretty low power, so depending on the tests you're using, you may be getting a "false positive" that one or more of your series are I(1).<br />Structural breaks tend to lead unit root tests to "discover" unit roots that aren't there, so once again, it may be the case that not all of your series are really I(1).<br />If you are using seasonally unadjusted quarterly or monthly data, the unit roots that you are detecting at the zero frequency may only be part of the story. There may also be unit roots at the seasonal frequencies - these can be detected using the HEGY tests. This will mess up your cointegration testing.<br />In short, there could be lots of reasons why your unit root/cointegration tests may be giving false signals. There may not be cointegration at all.<br /><br />In addition, if you're using the Johansen methodology to test for cointegration, you need to be sure that the VAR model that is used as the basis for this testing is properly specified with respect to lag lengths, serial independence of the errors, and normality of the errors.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-88348822373362864732015-11-19T21:47:07.298-08:002015-11-19T21:47:07.298-08:00Dear Prof. Dave
I am facing a peculiar problem. Al...Dear Prof. Dave<br />I am facing a peculiar problem. All my tests for cointegration between non stationary time series are suggesting the presence of cointegration, but my error correction term is turning out to be either positive or insignificant or an absolute value of more than 1. How should I interpret this result?Anonymoushttps://www.blogger.com/profile/10557169555440504758noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-29030370291388939912015-10-09T01:16:55.114-07:002015-10-09T01:16:55.114-07:00thats is great blog i love this
thats is great blog i love this<br />Meo School Of researchhttps://www.blogger.com/profile/12442042136056227524noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-11808834159619129402015-09-17T08:24:53.113-07:002015-09-17T08:24:53.113-07:00It doesn't matter what sign it is.It doesn't matter what sign it is.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-62853561681589713752015-09-17T07:22:31.200-07:002015-09-17T07:22:31.200-07:00what if the intercept (constant coefficient) is ne...what if the intercept (constant coefficient) is negative? then what to do? what will be the explanation? Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-50549092673494809212015-07-02T14:13:56.957-07:002015-07-02T14:13:56.957-07:00Kerry - my apologies for being so slow in respondi...Kerry - my apologies for being so slow in responding. The use of AIC/SIC for lag length selection should take place after you have dealt with any autocorrelation in the residuals. The reason is this - the AIC and SIC are based on the log-likelihood function, which in turn assumes independence of the observations. This assumption is violated if you have autocorrelation. I don't have a general resource to refer you to. Sorry!Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-76601783635555874172015-05-13T07:43:08.805-07:002015-05-13T07:43:08.805-07:00can you please tell me about the step by step proc...can you please tell me about the step by step process of johanson co integration and its basic requirements?Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-55328354444874735772015-05-05T08:18:18.632-07:002015-05-05T08:18:18.632-07:00It's not possible for them to be cointegrated....It's not possible for them to be cointegrated.<br />You can't use an ARDL mode, if any series is I(2).<br />Logically, you could estimate a VAR with the levels of the first 2 variables and the second difference of the third variable. I'm not saying that the results will make any economic sense, though.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-6714297554405093512015-05-05T00:53:36.107-07:002015-05-05T00:53:36.107-07:00Respected Sir,
I have 3 time series.Two of them ar...Respected Sir,<br />I have 3 time series.Two of them are stationary at level and third one is stationary at second difference.These3 are not cointegrated at level but cointegrated at first difference.I am trying to model their relationship.Should I go for VAR or VECM or ARDL? While modelling should I use data at level or at difference?<br />Regards,<br />RohitAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-39573576302580546992015-05-02T11:04:04.640-07:002015-05-02T11:04:04.640-07:00Good day Prof. Giles
Thanks for a very informativ...Good day Prof. Giles<br /><br />Thanks for a very informative and helpful post. I always find myself referring to your blog first before looking else where as your posts have helped me resolve a number of issues in the past.<br /><br />My question builds on what has been discussed here, as well as your comments earlier about dealing with auto-correlation if concerned about the long-run (and not just co-integration).<br /><br />I too have this 'problem' of my AIC and SIC moving in opposite directions as I try to determine optimal lag length on the VECM. Given your comments and my readings on the topic your advise on preferring SIC makes sense.<br /><br />I would like to confirm that the process of lag length selection would take place after correcting for the auto-correlation in the residuals?<br /><br />That is, after running the VECM at the first lag length I pause to check the residuals characteristics and find auto-correlation. Adding more lags helps a little but some AC still obvious (even using the LM test over the Portmanteau Test), but I start to worry about the d.f. issues of having too many lags (and a number of lags which don't make economic sense).<br /><br />I imagine that once I note this AC issue, I should then bring the Residual Term into the VECM and restart my lag length selection, but I would like to confirm this with you.<br /><br />Any resources/extra advice you could link me on this matter of lag length and auto-correlation in the VECM would be most appreciated.<br /><br />Kind Regards<br /><br />KerryKerryhttps://www.blogger.com/profile/09313679634890191752noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-7580287677821731382015-04-28T12:21:29.202-07:002015-04-28T12:21:29.202-07:00I prefer SIC - see my other posts on Information C...I prefer SIC - see my other posts on Information Criteria. Remember that you're trying to minimize it's value across competing models. You need to do a broad search.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-12607048156468368452015-04-28T12:19:10.437-07:002015-04-28T12:19:10.437-07:00Thank you for the swift reply. Which of the two in...Thank you for the swift reply. Which of the two information criteria is preferential? I have a case where increasing lags on certain variables improves the AIC but decreases the SIC. Lastly, in terms of the whole process, is determining the number of lags essentially a case of trial and error, wherein we must stumble across the ECM with the best AIC/SIC? Apologies for all the questions, I've found myself quite stuck!<br /><br />Many thanks,<br />Basty TonksAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-88788355260564453522015-04-28T09:54:31.432-07:002015-04-28T09:54:31.432-07:00Use SIC or AIC, and make sure that you enough lags...Use SIC or AIC, and make sure that you enough lags for the errors to be serially independent.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-25071970923073793532015-04-28T09:47:45.564-07:002015-04-28T09:47:45.564-07:00Hi Prof,
I must estimate an ECM with one dependent...Hi Prof,<br />I must estimate an ECM with one dependent and two independent variables. There is a cointegrating relationship between those three but we do not know where it lies. How do you determine the number of lags used on both the dependent and independent (both of them) variables in an error correction model?Anonymousnoreply@blogger.com