Backtesting question

Salamander · Post by **Salamander** » Tue May 24, 2005 11:27 am

Could someone please explain to me the following phenomenon I've noticed when testing systems: If I settle on a given set of parameters, etc. for a system and backtest it from, for example, 1990-2005, I get one set of results. Then, if I keep all parameters, etc. exactly the same and change only the date to, for example, 2000-2005, the results for the overlapping years (in this case 2000-2005) change a lot. Might this be due to differing equity levels affecting the choosing of trades, or what? Thanks.

damian · Post by **damian** » Tue May 24, 2005 11:49 am

In your first test you possibly have open trades at the start of 2000 (very likely if you have a LTTF system). In your second test you will not have any open trade at the start of the test period (2000). The impact of this will depend on how your testing platform/your entry code deals with this.

The equity level will also have a huge impact if you are using position sizing such as fixed fractional. Try with single contracts. If there is still a difference then investigate the 'open trade' issue.

TrendMonkey · Post by **TrendMonkey** » Tue May 24, 2005 7:39 pm

I have noticed this too and I think if your smaller date range was 5 years, open trades is likely to be less relevant. Equity is more likely to be the key factor, especially if you had a particularly 'good year' early in the longer date range.

One thing you can do is make sure that (for example) your results for 1990 match whether run for 1/1990-12/1990 or 1/1990-12/2004. I have found this to be the case (there may be a very small difference due to the open-trade effect as VT seems to assume all trades get closed at the end of the period, which is of course logical). All this really does is confirm that VT is reliable.

damian · Post by **damian** » Wed May 25, 2005 4:30 am

Indeed, I was wrong.