In article <92n1f0$ji9$[EMAIL PROTECTED]>,
Nilufer Pettersson-Arm <[EMAIL PROTECTED]> wrote:
>I am trying to build an ARIMA model for the movements of the returns of a
>stock. I have differentiated my data series once to make it stationary.
>The autocorrelations and partial autocorrelations do not show any clear
>pattern to indicate a model.
That's because stock movements are very close to being a random walk.
So you would expect that there is nothing left to model after differencing.
Ie, the best model is that it's white noise.
>I have tried all kinds of low-order models,
>but they fit the data VERY poorly. However, if I differentiate it three
>times or more, the fit gets better. But, what does this mean?
It means that when you difference white noise, you create autocorrelations,
which you can then attempt to model, though it's pointless to do so.
>The series
>is stationary after the first differencing and should require no further
>differencing.
Right. Differencing after that point just obscures the truth - which
is that you aren't going to make any money by modeling stock prices
using time-series methods that have been well known for over thirty
years. If there are any subtle patterns to stock prices that haven't
been noticed by the many thousands of people looking for such
patterns, discovering them will take more subtle tools that ARIMA.
Radford Neal
----------------------------------------------------------------------------
Radford M. Neal [EMAIL PROTECTED]
Dept. of Statistics and Dept. of Computer Science [EMAIL PROTECTED]
University of Toronto http://www.cs.utoronto.ca/~radford
----------------------------------------------------------------------------
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================
