tag:blogger.com,1999:blog-2198942534740642384.post1131173220824434905..comments2023-10-24T03:16:41.009-07:00Comments on Econometrics Beat: Dave Giles' Blog: More on Orthogonal RegressionDave Gileshttp://www.blogger.com/profile/05389606956062019445noreply@blogger.comBlogger3125tag:blogger.com,1999:blog-2198942534740642384.post-31569439866780503562016-12-28T16:16:59.505-08:002016-12-28T16:16:59.505-08:00You're right - what I had written was far from...You're right - what I had written was far from accurate, and I have amended the post accordingly. Thanks for pointing this out.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-33277582395483995972016-12-28T09:51:06.304-08:002016-12-28T09:51:06.304-08:00Perhaps I read the post too quickly, but I cannot ...Perhaps I read the post too quickly, but I cannot quite make the ends meet. Total least squares (TLS) is one thing, principal components regression (PCR) is another. TLS is related to the principal components of all variables (p independent ones and the dependent one). PCR is related to the principal components of the p independent variables alone (_excluding_ the dependent variable). So in TLS the principal components are obtained from a system of p+1 variables while in PCR principal components are obtained from a system of p variables. Thus PCR is not a multivairate extension of TLS, they are two different beasts. (An interesting related method is partial least squares which is in some ways superior to and more intuitive than PCR.)Daumantasnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-73388400421642768782016-12-28T02:54:57.949-08:002016-12-28T02:54:57.949-08:00Thank you so much for this blog post. I really enj...Thank you so much for this blog post. I really enjoyed reading it and found the topic so useful in our work: we use orthogonal multiple regression for tax revenue forecasting in the Institute of Fiscal Studies in Spain. Here we have and example of how we use macroeconomic partial indicators and Principal Component Analysis to obtain orthogonal regressors for a transfer function.<br /> It is the working paper " Combining the predictive ability of factorial analysis and transfer functions for VAT revenue forecasting"<br /> http://www.ief.es/documentos/recursos/publicaciones/papeles_trabajo/2016_05.pdf<br />We use SAS software instead of R software. <br />Thank you for taking the time to show all this material and this helpful discussions of this topic. <br />Anonymoushttps://www.blogger.com/profile/00631545215357047871noreply@blogger.com