[EMAIL PROTECTED] (Vj) wrote in message news:<[EMAIL PROTECTED]>... > Hi, > > Can we use multivariate regression for forecasting? > > -VJ
If you have one Endogenous Variable and the rest Exogenous then........ Yes you can if you take care to test for and remedy any evidented viloations culminating in a Transfer Function which can be re-stated as a multiple regression equation. The regression assumptions are http://www.autobox.com/blp013.html Please see http://www.autobox.com/conditions2.html for a technical discussion if these things. For another view please see http://www.autobox.com/t1c6.html and even more at http://www.autobox.com/t1c8.html In summary one need to identify the relationship between the series of interest and the candidate X's in a way that is robust and not flawed by the autocorrelation within the series. After doing that normal augmentation ( step-up strategies ) along with step-down strategies should be followed. Additionally one needs to possibly augment a potentially deficient model with an ARIMA structure TO PROXY OMITTED STOCHASTIC SERIES and/or INTERVENTION CARIABLES such as Pulses, Step , Seasonal Pulses and Local Time Trends to proxy omitted Deterministic Structure. You can get more info by perusing http://www.autobox.com/teach.html and reviewing FreeFore at http://download.com.com/3120-20-0.html?qt=freefore&tg=dl-2001 If you have more than one endogenous then http://www.autobox.com/t1c10.html http://www.autobox.com/mts.html Hope this helps .... Dave Reilly Automatic Forecasting Systems 215-675-0652 http://www.autobox.com P.S. If I can help further please call or email me. . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
