The FapTurbo Evolution Backtesting Blues – Track 1
A new version of FapTurbo Evolution has been released! This was extremely welcome news for us here at the Trading Gurus because backtesting the robot was proving to be something of a nightmare. According to the email we received announcing the new version it has:
Optimized Java code, making FapTurbo Evolution 20 times faster!!!
Whilst the speed increase is certainly welcome, it does make one wonder what was wrong with the Java code in the initial release of the robot. Helpfully the FapTurbo team provide a list of the other problems they've fixed:
- Fixed load time to reduce/eliminate occurences of ERROR 101
- Fixed GMT inconsistency bug
- Fixed rare bug "Label not unique"
- Fixed bug that prevented opening new orders at the same direction after Rejects
- Added auto-reconnect if there is no ticks. New parameter "Auto-reconnect Time period" added.
- Bug related to No Tick Error fixed
- Added multiple instances check (to prevent multiple copies of FAPT Evolution running on the same account)
- Fixed TradeMonday parameter to mean Monday-Tuesday night trading instead of Sunday-Monday night.
Since backtests that cover a few months no longer take all night we've managed to do quite a few now, and we've discovered some interesting things! Let's start with the EUR/GBP pair. Here's the equity curve we obtained from Dukascopy's JForex historical tester for the period from January 1st 2009 onwards.
Looks OK doesn't it? A nice profit of $10424.18, less Dukascopy's commission of course, which amounts to $1692.33 in this instance. However after a little bit of tweaking we managed to produce an even more impressive graph. Here it is:
The profit has increased to an amazing $28805.68, less commision of $2120.55 of course.
Now you may be wondering how we managed to achieve such a huge improvement. Possibly we've had a huge grid of computers slaving away for the last week optimising the large number of input settings you can twiddle with in FapTurbo? Actually nothing so complicated was involved. Both curves were produced using the default settings that come with the new release of the robot! The first chart was produced by feeding the robot tick data. The second was produced by feeding it one minute bars. That's the only difference.
Now consider this. The MetaTrader strategy tester doesn't allow you to use tick data when you're backtesting expert advisors. The best it has to offer in the way of historical prices are one minute bars. Maybe that has some bearing on the fact that whenever you set a MetaTrader robot loose on a live account, and feed it ticks rather than bars, the bottom line of your account never seems to shoot skywards in quite the same way as in all those backtests that seem to be plastered all over the web these days? Our FapTurbo MetaTrader live account is certainly looking a bit sorry for itself after just over a week with the robot in charge of trading the EUR/GBP pair: