tag:blogger.com,1999:blog-2198942534740642384.comments2018-08-17T23:22:03.719-07:00Econometrics Beat: Dave Giles' BlogDave Gileshttp://www.blogger.com/profile/05389606956062019445noreply@blogger.comBlogger3968125tag:blogger.com,1999:blog-2198942534740642384.post-24643321784038385522018-08-17T14:13:30.205-07:002018-08-17T14:13:30.205-07:00I'm a bit confused - you say you chose the lag...I'm a bit confused - you say you chose the lag length by MAXIMIZING SIC - shouldn't you be MINIMIZING it? Also, if you have evidence of structural breaks, did you allow for this when applying the Johansen cointegration tests? If not, this may be the source of your difficulty.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-63223042706856663532018-08-17T14:06:53.533-07:002018-08-17T14:06:53.533-07:00Dear Prof Giles,
I am doing thesis report on Gran...Dear Prof Giles,<br /><br />I am doing thesis report on Granger causality of export on GDP. My variables are GDP, Capital, Productivity (GDP/Labour), Import, and Export. All variables are expressed in natural logarithm (LN) and have 45 annual data each. All variables are I(1) using ADF, KPSS, and PP. Also there is evidence of structural breaks using Zivot Andrews and Multiple Breakpoint Test using OLS where Constant was only independent variable.<br /><br />I found lag 5 to be best for VAR model through trial-and-error basis considering SIC. At this lag, the VAR has highest SIC and no serial auto-correlation using LM test.<br /><br />Using Johansen Cointegration test at lag 5, there is existence of 4 cointegrating equations. But using both T-Y VAR and VECM, I cannot establish any causality between Export and GDP. I also used dummy (amateurish attempt) (1973-1999=0; 2000-2017=1) which still reinforces no causality between GDP and Export. <br /><br />Can you suggest what you would recommend to do...???<br />Can I attribute my ordeal to the problem that Johansen Cointegration Test has regarding lag lengths?? Also, if I use lag 2 and 3, I have 3 cointegration equations.<br />Fahim Hassannoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-23703675305643556212018-08-05T03:41:47.550-07:002018-08-05T03:41:47.550-07:00Yes that is fine. ANY of the variables in the mode...Yes that is fine. ANY of the variables in the model can be I(0) or I(1), but NOT I(2). And you don't have to know in advance which ones are I(1) and which ones are I(0) - that's the whole advantage of bounds testing in the context of an ARDL model.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-15876029804269982292018-08-04T19:00:48.541-07:002018-08-04T19:00:48.541-07:00sir will it ok if the dependent variable is statio...sir will it ok if the dependent variable is stationary but independent variables are mixed I(0),I(1) for ARDL bond testing? please reply<br />Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-27287839353346139742018-08-04T14:57:33.155-07:002018-08-04T14:57:33.155-07:00Wow!
Very educatingWow! <br />Very educatingAkuso Jonahhttps://www.blogger.com/profile/03428937816491462345noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-46975541679852013232018-07-31T13:59:32.786-07:002018-07-31T13:59:32.786-07:00See my earlier post at https://davegiles.blogspot....See my earlier post at https://davegiles.blogspot.com/2016/05/forecasting-from-error-correction-model.html<br /><br />That's a VERY long forecast horizon, no matter what sort of model you used. It doesn't sound like short-run dynamics to me.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-67854450055589693012018-07-31T05:32:04.600-07:002018-07-31T05:32:04.600-07:00Thank you very much for the clarification. Thank y...Thank you very much for the clarification. Thank you for your time!Nada Ben Mariemnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-30699750747714994712018-07-31T05:15:27.726-07:002018-07-31T05:15:27.726-07:00This happens all the time! Keep in mind that each ...This happens all the time! Keep in mind that each of these tests have different power properties, and the KPSS test has the reverse null & alternative hypotheses from the other two tests. Especially when the sample size is relatively small, or there is some structural break that you haven't been able to detect, conflicting results can arise. Ideally, try and report the conflicting results (if only in a footnote). However, generally you still have to come to a choice! If there is any chance that the series is I(1) rather than I(0), then treat it as being non-stationary. The reason is simply that the (statistical) "costs" you incur when you wrongly treat a series as being stationary are usually MORE than if you wrongly treat it as being non-stationary. For instance, suppose it is I(1) but you treat it as I(0) then this can be a big problem. On the other hand, if it is really I(0) but you treat it as being I(1), and decide to difference it, then the resulting series will still be stationary (even though it is not I(0) after being differences. A final comment - this inability to detect I(1) and I(0) series with perfection is precisely what makes the use of an ARDL model so appealing in many situations. With an ARDL model you don;t have to know whish series are I(0) and which ones are I(1). You just have to be confident that none of them are I(2).Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-87799911592233837622018-07-31T03:46:42.216-07:002018-07-31T03:46:42.216-07:00Dear Prof.
I have one more question, I checked the...Dear Prof.<br />I have one more question, I checked the stationarity with three different tests (ADF, PP and KPSS). I found the same results for both ADF and PP tests (the series is first difference stationary (with no trend and no constant)). But when using the KPSS test, I found a different result: I could not reject the null hypothesis of stationarity in level which means that the series is stationnary in level (under a constant and constant plus trend) . I am a little bit confused about the conclusion that I have to make: Is the series first difference stationary I(1) according to the ADF and PP tests, or is it stationary in level I(0) according to he KPSS test? Or Should I simply put the results as they are and conclude for each test separately? Thanks in advanceNada Ben Mariemnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-33098141934723694022018-07-30T04:33:06.398-07:002018-07-30T04:33:06.398-07:00This is really up to you. Sorry, but I can't a...This is really up to you. Sorry, but I can't advise you in such detail out of context. I hope that you understand.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-57655130284042093872018-07-29T20:30:48.255-07:002018-07-29T20:30:48.255-07:00Dear Prof Dave, I used the Option 2 with more prac...Dear Prof Dave, I used the Option 2 with more practical implications variables and shorter lags, but more CointEq(-1) is smaller than -1 i.e. -1.12, although negative and is significant (p-value 0.0002). I knew smaller than -1 is not good as speed of adjustment seems too fast? Prof, is there any good advise you may give me whether I can adopt this result, or should I drop one variable instead?Saranoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-11264725069646073112018-07-29T08:05:50.195-07:002018-07-29T08:05:50.195-07:00Thank you for your prompt response, I really appre...Thank you for your prompt response, I really appreciate your help.Nada Ben Mariemnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-34695508399772443362018-07-29T06:10:04.700-07:002018-07-29T06:10:04.700-07:00Hi - you can compute the critical values for any s...Hi - you can compute the critical values for any sample size using James MacKinnon's updated table: http://qed.econ.queensu.ca/working_papers/papers/qed_wp_1227.pdf<br /><br />See pages 9 and 10. Note that the sample size is denoted "T", and "N" is the number of variables. So, for the case where N=1, you are computing critical values for the ADF test of a unit root. When N=2 you are computing critical values for the Engle-Granger cointegration test, etc. By way of an example, if T=19, and you want the 5% critical value for the ADF test with a drift(constant) and trend, the number will be c = [-3.4126 - (4.039/19) - (17.83/(19^2))] = -3.625.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-75649961965959944642018-07-29T05:06:28.815-07:002018-07-29T05:06:28.815-07:00Dear Prof,
Thank you for your precious advice
I ...Dear Prof,<br /><br />Thank you for your precious advice<br /><br />I have to check the stationarity of a series of annual stock price index over the period 1997-2017. In this case, I have 21 observations of annual data, but when I run the ADF test (SIC used to select maximum lags with automatic selection=4) the included observations after adjustments become 19 and this appears "Warning: Probabilities and critical values calculated for 20 observations and may not be accurate for a sample size of 19" and this is the case of the following 1/ in Level: with intercept 2/ in Level with Trend and intercept 3/ in First difference with Intercept 4/ in First difference with Trend and Intercept 5/ in First difference with None. Note that the series becomes stationary in first difference with None. My question is the following: Should I ignore the warning and conclude that the series is stationary at first difference I(1) without trend or intercept? What do you suggest?<br /><br />Best regardsNada Ben Mariemnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-21811115650544056722018-07-28T07:21:14.000-07:002018-07-28T07:21:14.000-07:00Prof Dave, a million thanks for your insightful ad...Prof Dave, a million thanks for your insightful advise to select the one with more practical implications by including the variables. I was doing a research related to innovation in developing countries. So was at crossroad when came across this result. Thank you so much again, I will be careful on the issue of autocorrelation.Saranoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-87279280692827025162018-07-28T06:07:58.534-07:002018-07-28T06:07:58.534-07:00That's an interesting question - and it's ...That's an interesting question - and it's one that could be asked of any regression model, really, not just an ARDL model. The over-riding consideration will be to avoid under-specifying the model, because this would have serious consequences for estimator properties, etc. If certain variables are significant, why not include them - even if this means shorter lags? Presumably you think that these extra variables are relevant, on theoretical grounds, in the first place? The lag structure is going to alter the dynamics of the model, and this may be important if you are going to use the model for forecasting etc. On balance, without knowing the exact context here, I'd be inclined to go for Option 2. What do you need to be careful about? Probably the most important thing in a model of this type will be possible autocorrelation in the error term. watch out for that. And in the end it may be that you need extra lags to deal with it -in that case, you may be pushed back towards Option 1.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-11345769415094016472018-07-28T04:17:32.348-07:002018-07-28T04:17:32.348-07:00Dear Prof Dave, may I seek your kind advise on ARD...Dear Prof Dave, may I seek your kind advise on ARDL which is a better model in your view when the overall model for both options are statistically significant (pvalue<0.05):<br />Opt 1 - less number of regressors, with longer lags<br />Opt 2 - more regressors, with shorter lags<br /><br />Is there anything I need to be caution of? Many thanks.<br />Saranoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-24355449992161114932018-07-27T11:48:08.789-07:002018-07-27T11:48:08.789-07:00Thanks you so much!Thanks you so much!Sri Thiruvadanthaihttps://www.blogger.com/profile/12443401299603289168noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-73371532060789825932018-07-27T08:43:21.086-07:002018-07-27T08:43:21.086-07:00Yes, you can.Yes, you can.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-70944259634936463562018-07-27T08:07:31.205-07:002018-07-27T08:07:31.205-07:00Dave,
One question. If two series are I(1) but on...Dave,<br /><br />One question. If two series are I(1) but only one of them has a structural break, can we still use this methodology? Thanks<br /><br />Sri Thiruvadanthaihttps://www.blogger.com/profile/12443401299603289168noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-45064031303986312842018-07-27T02:37:26.996-07:002018-07-27T02:37:26.996-07:00Thank You, Professor Giles and apologies I should ...Thank You, Professor Giles and apologies I should have been more clear. X is a measure of demand and Y,Z...are explanatory variables like income etc. They are all non-stationary and I(1)<br /><br />Interesting, so an ECM is for short run dynamics I see. So if these coefficients were to be used for forecasting (yearly out to 2050) would you recommend an ECM?Unknownhttps://www.blogger.com/profile/03432530496138818529noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-85127330184564992382018-07-26T10:03:07.200-07:002018-07-26T10:03:07.200-07:00It's impossible to answer your first question ...It's impossible to answer your first question without knowing what X, Y, and Z are, and what are their orders of integration.<br />Regarding your second point: If all of the series are I(1) AND they are cointegrated, then regular OLS using the LEVELS (not differences) of the data will be super-consistent. That is, the estimator converges to the true parameter values at the rate "n" (the sample size), and not the usual SQRT(n) rate. This is well known, and is unaffected by adding lags of the levels to the regression. Of course, such a regression will only provide information about the long-run equilibrating relationship among the variables. It will tell you nothing about the short-run dynamics - and the latter is precisely what any sort of ECM is all about.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-56034510986481304252018-07-26T07:33:09.302-07:002018-07-26T07:33:09.302-07:00Professor Giles, I was sure I asked this question ...Professor Giles, I was sure I asked this question already but can't see my post? Maybe I'm looking at the wrong post. I'm working on a project which was last done 8 years ago and I have a couple of questions. <br /><br />1) The last time it was done they used an Unrestricted Error Correction model (UECM) of the form:<br /><br />d.X = d.Y + d.Z + L.X + L.Y + L.Z<br /><br />This seems to be the similar to Pasaran and Shins ARDL approach? With a lagged dependent variable on the right hand side with the differenced and lagged independent variables. It's very hard to find documentation of this UECM elsewhere. Also in Pesaran and SHin they have not differenced the dependent variable. Can you tell me which is the most appropriate method? I am dealing with a small sample (24 yearly obs last time and no more than 30 this time)<br /><br />Also I am only interested in the coefficient estimates to be used in another model. According to Eric Sims time-series notes he recommends just regressing cointegrated series in levels as the estimates will be "superconsistent"? This seems a lot simpler and he even says including lags of the dependent variable protect from potentially not cointegrated independent variables? Do you agree?Unknownhttps://www.blogger.com/profile/03432530496138818529noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-75035944844021589422018-07-24T03:21:35.464-07:002018-07-24T03:21:35.464-07:00Seriously??? You start up EViews, you click on the...Seriously??? You start up EViews, you click on the "HELP" button, and then you type in "BDS". :-)Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-54114249516326348532018-07-24T03:12:33.706-07:002018-07-24T03:12:33.706-07:00Just click on the numbers (1, 2, ....) - they are ...Just click on the numbers (1, 2, ....) - they are highlighted!Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.com