s-news
[Top] [All Lists]

Re: non splus/time series question

To: "Leeds, Mark" <mleeds@mlp.com>
Subject: Re: non splus/time series question
From: Prof Brian Ripley <ripley@stats.ox.ac.uk>
Date: Mon, 26 Apr 2004 07:42:03 +0100 (BST)
Cc: s-news@lists.biostat.wustl.edu
In-reply-to: <54668D97C0199943A454516A5FF6481E1F9BC5@EXCHUS001.AD.MLP.COM>
On Sun, 25 Apr 2004, Leeds, Mark wrote:

> if one has built an AR(1) model where the
> autoregressive parameter is less than
> 1, does anyone know of a book that derives
> the time till the level reverts back to it's original

It never does!  The predictions decay geometrically at rate \phi (the
AR(1) parameter), but the level is a discrete-time continuous-value
stochastic process which therefore never hits any givne value.

> level.  I think this often referred to as the half life
> of the model in the literature.

I think the `half life' is the time T for predictions to halve, that 
is \phi^T = 0.5, T = -1/log_2 \phi.  But surely the place to look is `the 
literature' which defines this?  (AFAIK, that is not the time series 
literature.)

-- 
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>