Length of Backtest

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
Demon

Length of Backtest

Post by Demon » Fri Mar 05, 2004 2:54 am

It seems widely accepted that when backtesting, the longer the period of the backtest - the better. However, if for example you backtest using a large portfolio of markets over 20 years or so, is this not weighted towards 'older' markets that have been in existance for that whole period at the expense of 'newer' markets that may have only been around for the last 10 to 15 years? Would you then not be left with a system/set of parameters that were optimised more to the older markets than the newer ones because the older markets have contributed more examples to the test? I realise the only other option you would have is to shorten the period of the backtest to conincide with the first trade date of the latest market in your portfolio, but I would like to hear what other members thoughts are here on the pros/cons of either or both angles.

Jwebster
Roundtable Fellow
Roundtable Fellow
Posts: 52
Joined: Fri Jul 04, 2003 10:23 am
Location: London, UK

Post by Jwebster » Fri Mar 05, 2004 8:11 am

Darran,

I think you make an interesting point. I thought about the same thing recently when testing 34 years of data on the hang seng index. Just by looking at the chart of this data and the volatility, clearly the market has changed significantly in the past 10-15 years. I have kind of adopted the process used by William Eckhardt adn that is to focus most on the last 10 years of data as this will incorporate current market forces and the fact that computerised trading has increased enormously. I think if you do this, your system will be able to account for the forces you are most likely to encounter in your actual trading. The trade off is you lose length of data, but if your data set is fortunate enough to covers variety of markets, up donw/sideways, etc I think this becomes far less of an issue.

edward kim
Roundtable Knight
Roundtable Knight
Posts: 344
Joined: Sun Apr 20, 2003 2:42 pm
Location: Silicon Valley / San Jose, CA USA
Contact:

Post by edward kim » Fri Mar 05, 2004 10:53 am

I test with a minimum of 20 years of data. I notice that for all markets, my system performance is fairly independent of the 10-year time slice that is chosen during that time period. So even if markets "change", my system still does its job. That way, when a new market comes along, I know I can start trading it right away. It might not work all the time, but it usually does for me.

Edward

Post Reply