tag:blogger.com,1999:blog-2198942534740642384.comments2019-08-19T22:31:43.793-07:00Econometrics Beat: Dave Giles' BlogDave Gileshttp://www.blogger.com/profile/05389606956062019445noreply@blogger.comBlogger4181125tag:blogger.com,1999:blog-2198942534740642384.post-42924558882768360752019-08-19T00:59:05.084-07:002019-08-19T00:59:05.084-07:00Thank you, Professor Giles! This page provides ver...Thank you, Professor Giles! This page provides very useful information.In Choihttps://www.blogger.com/profile/15460934918703661775noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-85730465222988953302019-08-14T06:19:14.097-07:002019-08-14T06:19:14.097-07:00Hi. I apologize for this very simple question. I h...Hi. I apologize for this very simple question. I haven't found an answer on this blog, on the internet more generally, or in papers I have read.<br /><br />Suppose you want to evaluate the impact of a recently implemented policy using interrupted time series because you have no comparison group and the policy was implemented all at once on the implementation date.<br /><br />Further suppose:<br />- the government has collected monthly data on the outcome variable for 24 months prior to policy implementation and 24 months after implementation<br />- there is no seasonality in the outcome<br />- unfortunately, after collecting the monthly data, the government has chosen to create a 12-month trailing moving average (current period plus prior 11) and seems unable to provide the raw (pre-smoothing) data<br />- no factors are likely to affect the outcome other than the passage of time (linearly) and the implementation of the policy<br />- you want to do the best you can with these data<br /><br />What would you do to deal with the fact that the outcome variable you have to work with is a 12-month moving average? Obviously:<br />1) It is autocorrelated<br />2) Any policy effects that actually occurred during the period in which the effects are being smoothed into the MA outcome variable will be muted (i.e., in the first 11 months)<br /><br />A few options (in my naive mind) might be:<br /><br />1) Ignore the issue and use a model of the form<br />yma = b0 + b1*time + b2*policy + b3*policy*time + e<br /><br />where yma is the moving average of y (which is unavailable) and<br />policy is 0 in the months before implementation and 1 in all months in which the policy was in effect<br /><br />2) Replace policy in the model above with a policy variable that is:<br />a) 0 in the months before implementation<br />b) a fraction during the first 11 months of implementation calculated as month of implementation / 12 (1/12 in month 1 of implementation, 2/12 in month 2, ..., 11/12 in month 11)<br />c) 1 in all later months<br /><br />3) Use a model that has 3 pieces, one for the pre-implementation period, a second for the 12 months in which the pre-period outcomes are being averaged in with the post-period outcomes, and a third for the period after the pre-period outcomes have disappeared from the moving average, such as:<br /><br />yma = b0 + b1*time + b2*partial + b3*partial*time + b4*full + b4*full*time + e<br /><br />where:<br />a) partial is 1 in the first 11 months of implementation and 0 otherwise<br />b) full is 1 in month 12 of implementation or later, and 0 otherwise<br /><br />4) Same as #3 but make partial a fraction as in #2<br /><br />Something else? Again, the government cannot or will not provide the underlying data before creation of the moving average.<br /><br />Many thanks for advice on this.<br /><br />donboyd5https://www.blogger.com/profile/03271340430759131455noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-85083502402390662482019-08-13T05:41:07.598-07:002019-08-13T05:41:07.598-07:00Take a look at this post..... https://davegiles.b...Take a look at this post..... https://davegiles.blogspot.com/2012/01/cointegration-analysis-with-i2-i1-data.htmlDave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-320855356211247892019-08-12T18:01:45.842-07:002019-08-12T18:01:45.842-07:00Hi Professor, my variables are a mix of I(0), I(1)...Hi Professor, my variables are a mix of I(0), I(1) and I(2), what method can I use to determine the long-run relationship? Johansen cointegration only applicable with I(1) variables and ARDL can't have I(2)... Vickienoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-17132824384113261612019-08-10T06:29:55.368-07:002019-08-10T06:29:55.368-07:00Thanks. This really helped! Thanks. This really helped! Kashmira Zhttps://www.blogger.com/profile/18059047261010268266noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-81753581122685692402019-08-09T08:09:03.770-07:002019-08-09T08:09:03.770-07:00It is good to see that IV is still popular 90 year...It is good to see that IV is still popular 90 years from now.EViews Garethhttps://www.blogger.com/profile/02265937096525975321noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-77488886495716279242019-08-08T07:47:55.067-07:002019-08-08T07:47:55.067-07:00If all of your series are I(1) and not cointegrate...If all of your series are I(1) and not cointegrated, then you ALREADY have a classic spurious regression. Sorry! And adding AR() terms won't solve that problem. You need to test to see if you have cointegration - you almost certainly will. In that case, estimate a regular error-correction model. That will take care of the underlying dynamics and I think you'll fins that's the cause of the autocorrelation.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-8227599122162282832019-08-08T01:32:19.775-07:002019-08-08T01:32:19.775-07:00Hi Professor, I run a OLS regression using EVIEWS....Hi Professor, I run a OLS regression using EVIEWS. The dependent variable and five independent variables (mostly are economic time series data) all are I(1). The estimation result shows very serious autocorrelation issue (DW stat is 0.9 something). The residuals PACF indicate it is an AR(1). As dropping/changing of variables is not an option, can I include a AR(1) term in the equation and run the OLS again? Will I get a spurious result? Hope to hear from you soon. Thank you!Vickienoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-64357785983562735742019-08-06T13:02:36.085-07:002019-08-06T13:02:36.085-07:00Thanks for this. The Amazon link is
https://www.am...Thanks for this. The Amazon link is<br />https://www.amazon.com/Econometricians-Pearson-Hotelling-Haavelmo-Finance-ebook/dp/B01N51YEC6/ref=sr_1_2?keywords=the+econometricians&qid=1565121693&s=gateway&sr=8-2Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-64544675153927446062019-08-06T12:56:47.455-07:002019-08-06T12:56:47.455-07:00Yes I have but I have not received a response yet....Yes I have but I have not received a response yet. Thank you for your prompt response. Dudunoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-49540460064382340112019-08-06T11:53:26.399-07:002019-08-06T11:53:26.399-07:00Colin Read's "The Econometricians" i...Colin Read's "The Econometricians" is also an interesting book on the history of econometricsMohamed Merabtinehttps://www.blogger.com/profile/10395548982662605729noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-49005050832598165712019-08-06T09:12:36.511-07:002019-08-06T09:12:36.511-07:00Hi - I'm not sure - have you taken this up on ...Hi - I'm not sure - have you taken this up on the EViews forum at http://forums.eviews.com/ ?Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-73353410658372436332019-08-06T07:51:54.052-07:002019-08-06T07:51:54.052-07:00Dear Professor Giles
Thank you for this wonderful...Dear Professor Giles<br /><br />Thank you for this wonderful blog, it is very helpful. <br /><br />I would like to please ask a question about SVAR impulse responses. I am doing my PhD on policy analysis and I am using SVARs. I have done my impulse responses on EViews, with the decomposition method as Structural decomposition. My challenge is that these impulse responses give me a response to a one standard deviation shock and I would like to change this to a percent shock. I am really stuck on this one. Please assist me on how I can solve this issue. If I can't do this on EViews, how can I compute these impulse responses for myself?<br />I have tried using the "scaled impulse response function - sirf" add-in on EViews but I am stuck on what should go into the menu interface where there is "Scale factor for IRF" as well as "Multiplication of standard error". Looking forward to your assistance and thank you in advance.Dudunoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-21182164651990577692019-08-01T08:38:30.402-07:002019-08-01T08:38:30.402-07:00Waiting for August reading :-)Waiting for August reading :-)Mohamed Merabtinehttps://www.blogger.com/profile/10395548982662605729noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-65369837959124054722019-07-28T17:17:38.558-07:002019-07-28T17:17:38.558-07:00:-) :-) Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-41013155192606366172019-07-28T11:54:28.150-07:002019-07-28T11:54:28.150-07:00Very interesting, Dave! Your last slide reminds me...Very interesting, Dave! Your last slide reminds me of this: https://xkcd.com/1725/ <br /><br />Cheers,<br />PSpsummershttps://www.blogger.com/profile/16146930482281058979noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-41987464849924090862019-07-26T07:32:10.131-07:002019-07-26T07:32:10.131-07:00I'm not sure off-hand. I suggest that you cont...I'm not sure off-hand. I suggest that you contact the people at Gretl.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-16474146809003008152019-07-25T14:11:12.458-07:002019-07-25T14:11:12.458-07:00Hi, Thanks for the wonderful article.
I am working...Hi, Thanks for the wonderful article.<br />I am working on gretl and using lagReg package and using "pdl" function of the package. Please help me how to specify the Matrix of PDL specifications. Lag order= 12, Degree of polynomial = 2<br />Please help me how to define the matrix. <br /><br />Thanks in AdvanceUnknownhttps://www.blogger.com/profile/10418641795044687292noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-33761035539427474772019-07-15T05:29:58.159-07:002019-07-15T05:29:58.159-07:00Dear Professor Giles,
Consider two I(1) series, x ...Dear Professor Giles,<br />Consider two I(1) series, x and y. The ARDL bounds (and residual) tests applied to x regressed on y, are passed with flying colors, thus suggesting x and y are cointergrated. But the bounds tests applied to y regressed on x indicate absense of cointegration...Should one live with this?<br />Regards, Autodidact, retired (but not inactive). Anonymousnoreply@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-41314515214824614442019-07-11T07:58:48.059-07:002019-07-11T07:58:48.059-07:00The test won't be appropriate in that context ...The test won't be appropriate in that context as there are lags of the dependent variable among the regressors. Note that in the post I say: <br />"Notice that if this null hypothesis is true, then the only regressors in the model will be the columns of X, and these are non-random. So, by the Milliken-Graybill Theorem, the usual F-statistic for testing the restrictions associated with the RESET test will still be exactly F-distributed (under the null), in finite samples." The requirement of non-random regressors is crucial.<br />Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-45458361670512045652019-07-10T16:25:41.178-07:002019-07-10T16:25:41.178-07:00Please sir, how important is the Ramsey reset test...Please sir, how important is the Ramsey reset test for an ARDL model. I ran an ARDL model, and all other post estimation results are very good news including the cusum and cusum of square test, but the Ramsey test shows me that the model is not correctly specified. Majority of the research I reviewed did not conduct the Ramsey test of any test for model specification in their research Unknownhttps://www.blogger.com/profile/03396096923137917219noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-58584582182032528242019-07-09T20:56:35.920-07:002019-07-09T20:56:35.920-07:00Wonderful post. Although I have read several time ...Wonderful post. Although I have read several time series books, I have not seen the conversion to polar coordinates so clearly explained. Would you consider explaining the ARMA frequency test in EViews, relating it to comments? ThanksAnonymousnoreply@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?Unknownhttps://www.blogger.com/profile/12641842081503313293noreply@blogger.com