CSI -UA and market correlation

Discussions about the testing and simulation of mechanical trading systems using historical data and other methods. Trading Blox Customers should post Trading Blox specific questions in the Customer Support forum.
Post Reply
Chris67
Roundtable Knight
Roundtable Knight
Posts: 1046
Joined: Tue Dec 16, 2003 2:12 pm
Location: London

CSI -UA and market correlation

Post by Chris67 » Wed May 23, 2007 2:25 am

On csi data you have the maulti-market analyser which enables you to do correlation studies on markets.
My question / thoughts are that you get the choice of whether to set the data up as daily or weekly / perpetual or back adjusted. I am a bit confused on this so I read the CSI help notes on the issue but it doesnt make it too much clearer - My instinct is to analyse correlation based on daily back adjusted data as this is what I trade.
Does anyone have any thoughts on the best way to test market correlations over 20 years ? CSI recommends weekly perpetual which is in fact the exact opposite to what I do.

Any thoughts are, as ever, warmly appreciated.
Chris

stamo
Roundtable Knight
Roundtable Knight
Posts: 522
Joined: Tue Mar 13, 2007 5:27 pm

corr

Post by stamo » Wed May 23, 2007 8:13 am

the short answer: match your lookback to cover a few cycles of your time horizon and then periodically update your correlations.

in equity quant work, the 'standard' has been 60 months of monthly data, probably first popularized by BARRA.

if your holding period is days to weeks then 1-2 years of daily data might be better.

i'd only look back 20 years if i were a long-term investor with at least an 12-18-24 month target holding period.

if you have the time (no pun intended) then do a series of lookback correlation analyses and watch the correlations change.

for example do these four 5 year periods: 15-20, 10-15, 5-10 & 0-5 years ago and see how things change over time.

or look at the differences between 60 months of monthly, 2 years of weekly and 1 year of daily data.

it can be a real eye opener for folks who think they can look at a corr matrix once and then forget about it.

Post Reply