tag:blogger.com,1999:blog-2198942534740642384.comments2019-04-22T07:48:10.488-07:00Econometrics Beat: Dave Giles' BlogDave Gileshttp://www.blogger.com/profile/05389606956062019445noreply@blogger.comBlogger4099125tag:blogger.com,1999:blog-2198942534740642384.post-80238292359694996012019-04-22T07:09:02.172-07:002019-04-22T07:09:02.172-07:00Dear Dave,
I have a couple of queries about using...Dear Dave,<br /><br />I have a couple of queries about using the first principal component (of two variables) in a panel VAR. My understanding is that the first principal component is proportional to the average of its components. So I would like to shock this variable to see the effect of lowering the average of its component on other variables in the VAR.<br /><br />However, I have a few issues. <br />1) I am not sure whether this can be done in the context of a VAR.<br /><br />2) I am not sure what the size of the shock should be. The standard deviation of the first PCA is approximately 5 (zero mean). So a one standard deviation shock will cause the component to rise by about 500% which seem enormous. <br /><br />Many thanks in advance for your help. Seyihttps://www.blogger.com/profile/01366180882138855538noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-79041677371351944082019-04-19T19:56:36.408-07:002019-04-19T19:56:36.408-07:00thank u for thisthank u for this <a href="https://www.learnrscit.com/" rel="nofollow"> </a><br /><a href="https://www.ccconlinetyari.com/" rel="nofollow"> </a><br /><a href="https://www.computermcq.com/" rel="nofollow"> </a><br />Gurvinder sirhttps://www.blogger.com/profile/15703313365023361565noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-24963775363325895302019-04-18T05:34:35.653-07:002019-04-18T05:34:35.653-07:00Sorry Jack - not sure on this one.Sorry Jack - not sure on this one.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-10637285479078257242019-04-18T02:00:59.077-07:002019-04-18T02:00:59.077-07:00Hi Lei Xu,
I am not the Professor, but another s...Hi Lei Xu, <br /><br />I am not the Professor, but another student of econometrics and willing to help out if I can. <br /><br />Are you asking about a confidence interval for your estimated coefficients. <br /><br />If so let us know and I can help explain.<br /><br />Best wishes, <br /><br /><br />JackJack Salahnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-10536827533308308502019-04-18T01:58:05.724-07:002019-04-18T01:58:05.724-07:00Dear Unknwon,
First of all, think of ARIMA model...Dear Unknwon, <br /><br />First of all, think of ARIMA modelling as a slight extension of ARMA modelling. So if you're familar with ARMA modelling then this will help. I would recommend a good what I describe as a beginners graduate level text, Verbeek 'A guide to modern econometrics'. On page. 290. they give a good introduction to ARIMA modelling and even provide an empirical example. <br /><br />If the process that you're trying to modelling as a unit root. Then you will have to first difference it before completing ARMA modelling. Then it would be the change in the series that is modelled by the ARMA process. <br /><br />A series that becomes stationariy after first differencing, is said to be integrated of order 1. <br /><br />To use an ARIMA model for forecasting something. <br /><br />1) Decide if it has a unit root or not (or multiple unit roots for that matter). <br /><br />2) If it has (got a unit root or more), differecing it will alow you use ARMA modelling as normal. <br /><br />3) If it hasn't, go back to modelling with an ARMA model.<br /><br />For good forecasts, it is generally accepted to use the BOX-JEKINS method of modelling. <br /><br />This will help you select the appropriate model (i.e. AR and MA lags) and from here forecasting is straightforward. Especially in eviews. <br /><br />Have a look at the eviews manual for unit root tests as this is the best place to start. <br /><br />I hope this message reaches you well.<br /><br />Best wishes, <br /><br /><br />JackJack Salahnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-14777984259030388502019-04-18T00:26:40.710-07:002019-04-18T00:26:40.710-07:00Thank you very much Professor. Sorry to bother you...Thank you very much Professor. Sorry to bother you again but are you perhaps familiar with research by Vilasuso (2001), "Causality tests and conditional heteroskedasticity: Monte Varlo evidence". Jounral of Econometrics 101 (2001) 25-35. I am actually trying to replicate his results. <br /><br />He sets up a VAR(1) model and finds that when using least-sqaures, to test for causality in mean AND there is a causality in variance relationship (that is the conditional variances of the disturbance terms exhibit conditional heteroskedasticity which is related) least sqaures exhibits severe size distortion. <br /><br />I am just wondering how least squares could even pick up that there was a causality in variance relationship. As you say, for a VAR model with the same lags it should be equivalent use OLS line by line.<br /><br />Any tips or pointers would be greatley appreciated and I would like to thank you once again for a excellent blog, website and publications. You have certainly helped out me and tens of thousands of students worldwide.Jack Salahnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-65705648548810193992019-04-17T13:39:28.969-07:002019-04-17T13:39:28.969-07:00Jack - if every equation in the VAR has exactly th...Jack - if every equation in the VAR has exactly the same regressors (i.e., the same number of lags for each of the variables), then OLS is identical to SURE estimation. The second error has no impact on the estimates of the parameters in the first equation, and vice versa. You'll find this result explained and proven in most grad.-level econometrics textbooks. On the other hand, if you have different lag lengths in the two equations, and hence different regressors, the OLS estimates are no longer efficient - they differ from the SURE (system-GLS) estimates. In such cases, in EViews you would create a system object, spell out which lags you want in each equation, and then choose SURE as the estimation method. I hope this helps.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-8740639690786775812019-04-16T23:54:19.508-07:002019-04-16T23:54:19.508-07:00Dear Professor,
I too am a big fan of Eviews.
...Dear Professor, <br /><br />I too am a big fan of Eviews. <br /><br />I have a question about VAR modelling. I understand that the default option in Eviews if OLS. <br /><br />My question is related to the error terms in a VAR system. Lets say if I estimate the VAR by OLS, how can the error term in the second equation (u2), impact the estimation of the first equation of the VAR. <br /><br />Every post is an interesting read. Thank you very much for your efforts. <br /><br />Best wishes, <br /><br /><br />Jack Jack Salahnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-48940525199209576412019-04-16T11:04:10.260-07:002019-04-16T11:04:10.260-07:00Thanks for the kind comment. You can use the T-Y a...Thanks for the kind comment. You can use the T-Y approach with ANY combination of I(0), I(1) and I(2) variables. For instance, If one variable is I(0) and one is I(1), then set m=1, in terms of the notation used in my post. If both are I(1), then again set m=1. If one is I(0) and the other is I(2), set m=2. If one is I(1) and one is I(2), set m=2. And if both are I(2), again set m=2.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-84153057799042555032019-04-16T08:52:36.258-07:002019-04-16T08:52:36.258-07:00Thanks much Dr. DG for the excellent interpretatio...Thanks much Dr. DG for the excellent interpretation of the T-Y approach. I have not seen any other explanation better yours. Hats off to you.<br />I have one question...can we still use T-Y approach even if our all series (i.e. Xt and Yt) are either I(0) or I(1) and what if both are I(2)?Himmy Khanhttps://www.blogger.com/profile/06702422968121938601noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-9211646770726263102019-04-15T08:20:00.694-07:002019-04-15T08:20:00.694-07:00Hi - not sure, off-hand. Why not check on the EVie...Hi - not sure, off-hand. Why not check on the EViews Forum? DGDave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-37281780490843649502019-04-15T08:13:26.946-07:002019-04-15T08:13:26.946-07:00Possibly. But it can also signal that the converge...Possibly. But it can also signal that the convergence to the long-run equilibrium is oscillitary. For example, see this paper: https://www.imf.org/external/pubs/ft/wp/2005/wp05170.pdf . Also, see P. K. Narayan and R. Smyth, 2006. "What Determines Migration Flows from Low-Income to High-Income Countries? An Empirical Investigation of Fiji-U.S. Migration 1972-2001". Contemporary Economic Policy 24(2):332-342.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-54768821423553621322019-04-15T07:01:48.589-07:002019-04-15T07:01:48.589-07:00Yes, it's OKYes, it's OKDave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-64274259137744427142019-04-14T17:36:44.803-07:002019-04-14T17:36:44.803-07:00Dear Professor,
I have two series of daily futures...Dear Professor,<br />I have two series of daily futures' returns data for about 10 years making it about 2500 data points. But both the series are stationary in nature. Is it okay to run an ARDL model to measure the short term and long term relationship between the returns ? Ipsita Singhhttps://www.blogger.com/profile/06343374486184937486noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-64315501654859651042019-04-14T11:57:56.607-07:002019-04-14T11:57:56.607-07:00Hi Dave, I thought you might be able to help with ...Hi Dave, I thought you might be able to help with this - can you advise me on how to run a vecm in eviews where the data in the short run equation is entered as YoY change? As you know, the vecm in eviews automatically converts the data for the short run equation to one period changes, but in alot of cases we want to use YoY change. <br /><br />RegardsAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-78576841005289738562019-04-14T07:19:23.973-07:002019-04-14T07:19:23.973-07:00On the face of it, you should be O.K.On the face of it, you should be O.K.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-1043490070167755482019-04-13T17:42:18.245-07:002019-04-13T17:42:18.245-07:00If the coefficient of the error correction term tu...If the coefficient of the error correction term turns out to be negative and significant but between -1 and -2, does this signal a problem with model specification?Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-70391535361378454722019-04-13T11:13:03.923-07:002019-04-13T11:13:03.923-07:00Thank you very much Professor. And one more questi...Thank you very much Professor. And one more question, I have 60 observations (Quarterly) , and after running ARDL with 4 independent variables , using Schwarz criteria, ARDL(1,0,1,0,0,3) was chosen. Bound test shows co-integration and there is no serial correlation in residuals, so can I trust this results? , considering that some of my independetn variables are I(0). Thank you again.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-38236351541150144782019-04-13T06:20:50.807-07:002019-04-13T06:20:50.807-07:00ARDL models are estimated by OLS (& not just i...ARDL models are estimated by OLS (& not just in EViews). OLS will be biased (for small samples) in any model that has lagged values of the dependent variable as regressors, so that includes ARDL models. However, it is a consistent estimator (as long as the errors are independent), so the bias vanishes for large samples. You shouldn't use an ARDL model with a very small sample. And keep in mind that the long-run relationship is, again, just that - it needs plenty of observations in the sample to be meaningful. Finally, if there is a long-run cointegrating relationship, then OLS is a really good choice for estimating its parameters. The reason is this. Under cointegration, OLS is "super-consistent". The estimates converge to the true parameter values at the square of the usual rate as the sample size (n) grows. In standard models this rate of convergence is SQRT(n), but the rate is the same as "n" itself under cointegration.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-41250925741040368962019-04-13T05:26:47.361-07:002019-04-13T05:26:47.361-07:00Hello Professor Giles, I want to know, If OLS esti...Hello Professor Giles, I want to know, If OLS estimation of ARDL model gives Biased results then how can we rely on the Long-run or short run coefficients? given by the ARDL model? . When runing ARDL model, does Eviews use OLS to estimate it? Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-6898078300452296912019-04-11T04:14:42.512-07:002019-04-11T04:14:42.512-07:00coolcoolUnknownhttps://www.blogger.com/profile/10913983684470179044noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-65894583150985849392019-04-10T08:16:22.546-07:002019-04-10T08:16:22.546-07:00Hi - no. What we need is a suitable estimate of th...Hi - no. What we need is a suitable estimate of the variance of the error term - not the variance of the forecast.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-55232897796278571222019-04-09T20:02:58.494-07:002019-04-09T20:02:58.494-07:00Lovely post Prof. I will take time to look at the ...Lovely post Prof. I will take time to look at the permutation test for split-plot design model.David I. Johnhttps://www.blogger.com/profile/04354378602008692834noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-17867915410671605902019-04-09T14:52:09.341-07:002019-04-09T14:52:09.341-07:00Hi. My first post went into a cyber black hole (!)...Hi. My first post went into a cyber black hole (!). I have a model as described above, ln(y)=XB. I understand that when I want to make a prediction of linear y I need to "de-bias" the prediction by including the s^2/2 term. My question is when we estimate the standard error of a forecast (stdf), stdf=s^2*x(X'X)invx' + s^2), where little x is a vector of independent variables, and big X is our data matrix, shouldn't we use the stdf to de-bias rather than the rmse from the regression? That is, yhat(linear)=exp(yhatln+stdf^2/2)?<br /><br />Ref on stdf: https://www.stata.com/manuals13/p_predict.pdfcjabohttps://www.blogger.com/profile/14984084897323486985noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-41602674272780326742019-04-09T14:27:45.985-07:002019-04-09T14:27:45.985-07:00Hi - for the life of me, I don't understand wh...Hi - for the life of me, I don't understand what you're referring to here. By all means elaborate, and I'll be happy to respond.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.com