s-news
[Top] [All Lists]

Re: prediction function for AR processes (was No subject)

To: "Leeds, Mark" <mleeds@mlp.com>
Subject: Re: prediction function for AR processes (was No subject)
From: Prof Brian Ripley <ripley@stats.ox.ac.uk>
Date: Wed, 25 Jun 2003 22:17:56 +0100 (BST)
Cc: s-news@wubios.wustl.edu
In-reply-to: <C497480353EF904FBF7A8B0851E5422CA9C7B5@MAIL002.AD.MLP.COM>
Please use a meaningful subject.

What you want is often called the eventual forecast function.  For an
AR(1) it is beta^r for r steps ahead.  The distribution of Y(t+tau) | Y(t)
= y* is in all good time-series books, certainly in those by Brockwell &
Davis. You can also get it from the arima.forecast function in S-PLUS.

On Wed, 25 Jun 2003, Leeds, Mark wrote:

> can anyone answer the following or know of a reference for
> the following type of question ?
>  
> suppose i have an AR(1) model where the coefficient
> is Beta. 
>  
> so, the equation is y_t = mu + beta*y_t-1 + epsilon
>  
> if the model is estimated with  a daily frequency,
> so that is t is daily, are there formulas
> that exist for how many days it will take for
> y_t to "return" to its long run average
> when it at say y* at time t ?
>  
> i imagine, if there is a formula, then it's a function
> of beta and the volatility of epsilon  but i haven't seen one in any
> books 
> that i'm familar with ? thanks.
>  
>                                                 mark
> 

-- 
Brian D. Ripley,                  ripley@stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595


<Prev in Thread] Current Thread [Next in Thread>
  • [no subject], Leeds, Mark
    • Re: prediction function for AR processes (was No subject), Prof Brian Ripley <=