Re: information extraction from stock market data

beliavsky_at_aol.com
Date: 02/17/05 

Date: 17 Feb 2005 05:53:21 -0800

lucy wrote:
> Hi all,
>
> Suppose I am working on curves of two stock market series: Microsoft
and
> Cocokola...

> I want to deduce when does the IBM outperform Cocokola(in terms of
growth
> rate, say daily, or weekly, etc.)... and when does Cocokola
outperforms
> IBM...
>
> And I want to deduce some models over these information: for example,
one
> famous phynomenon is the Monday-Friday effect, etc... I try to do
some
> information extraction or "data-mining" to find such models...
>
> What should be my tools to make such discoveries? It seems that there
are so
> many dependent and independent factors impacting the stock market
> simultaneously...

I doubt that there are significant inefficencies in the prices of
large, actively traded stocks such as Coke and Microsoft. You could try
fitting an ARMA model to the log returns of each stock, a vector
autoregression (VAR) to the bivariate time series, or a
cointegrated-VAR to the price levels. There are books on "financial
time series analysis" covering such models. I recommend trying
interpretable linear models such as the ones I mentioned before trying
"data mining" approaches.

Quantcast