How Far Back To Test Equities?

JohnTHill · Post by **JohnTHill** » Thu Mar 12, 2009 1:36 pm

I am testing long term trading strategies and am wondering how far back I really need to test. Is ten years of back testing sufficient?

For instance I would like to know how well I would have done during the crash in the eighties as well as the tech wreck. I would also like to be able to optimize my system for these bad times and see how to use it in that environment in the future. However my data provider charges $25.00 per year for any data over ten years old and I prefer not to pay for that data if it is not really necessary. Thank you.

John

Paul King · Post by **Paul King** » Thu Mar 12, 2009 4:13 pm

Well what's the opportunity cost of waiting 10 years to collect enough data (to add to the "included" 10 years) in order to have a total of 20 years to test on without paying any extra?

How does this solution compare to paying up for the data you actually mention you would like to test on in your actual question?

If you add an extra couple of percent annual return to your trading (or shave off a couple of percent on the maximum dradown) by testing on more data, how much will that increase your return (or save you) over a trading career compared to the cost of the data?

JohnTHill · Post by **JohnTHill** » Thu Mar 12, 2009 4:45 pm

Thanks Paul.

My take is that it is stupid not to spend the cash on the older data. I do not want to trip over a dollar to save a penny!

I guess as a newcomer my real question isâ€¦ For long term trading is ten years historical data enough and if not how many years back should I go? I am more than willing to pay for it but do not yet have the experience to know how much history I need. Thanks.

Paul King · Post by **Paul King** » Thu Mar 12, 2009 6:04 pm

OK, a more methodical approach would be to classify markets into types by trend (up, down, sideways) and volatility (low, normal, high) giving 9 possible "types" of market (3 trendiness measures multiplied by 3 volatility measures).

Then make sure you have a decent sample of trades in your testing from each type of market to be representative. This would then determine how much historical data was needed to ensure adequate/representative testing.

Simple method for classifying a market into types attached.

Hope this helps.

Paul

JohnTHill · Post by **JohnTHill** » Thu Mar 12, 2009 6:18 pm

Thanks Paul, this is great stuff. I will re-read it tonight and work on digesting it. I appreciate your input.
John