[EMAIL PROTECTED] (MG) wrote in message news:<[EMAIL PROTECTED]>... > Dear all, > > sorry for this preliminary question. I tried to explore through the > texts and web search. But could not understand. So seek your help. > > In the identification of step of transfer function model (ARIMA,box > jenkins), it is said that the shape of cross-correlation diagram give > some idea about s,r,b (in the notation of box,jenkins book). I could > understand 'b', but find kind of difficulty to understand how to guess > s,r. > Your help in this regard will be highly appreciated. > Regards > MG > p.s. I understand there are some softwares which does the estimation > process automatically (including identification and diagnosis). My > concern here is to understand the process. The process is quite straighforward. Firstly identify the pure delay (b) by finding the dirst statistically significant cross-correlation. Now the issue is to identify r and s where r is the prder of the numerator or input polynomial and s is the order of the denominator or output lag polynomial. If the cross correlations can be characterized as following a pattern then the order of that pattern is a clue as to the order of the denominator or output lag. Sometimes this is obfuscated (clouded) by unuusal values in the "beginning" cross-correlation. These unusual values represent the order of the numerator polynomial. I would suggest using a statistical package like AUTOBOX or SAS or GAUSS or one of your own making to simulate various states of nature and examine the sample cross-correlations in order to pursue this. By simulation you are controlling the process ( no outliers , small input variance ) thus it will be quite clear as to how the sample cross-correlation speaks to the the issue of Transfer Model identification. AUTOBOX deals with the outliers so that they have minimal effect on the r,s,b identification. I would be glad to chat with you and give you some more pointers on this issue ...but I would prefer dialogue rather than monologue. At a minimium please visit the AFS website ( http://www.autobox.com ) Dave Reilly ================================================================= Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =================================================================
