Guidance on the timeframe for a backtest needed

New comment
 

Fellow MT4 quants,

I've been backtesting a sell version of a trending EA that I've developed for trading the EUR/USD on 5 minute bars . The timeframes for which I have optimized the EA and show a decent profit are 02-01-09 through 02-17-10 and 09-08-09 through 02-17-09. I was impressed that the EA showed as much profit as it did for the larger of these timeframes considering that it only sells (shorts) and 2009 was largely a bullish year for the EUR/USD. Also, I've been careful to only optimize several parameters at a time in an effort to avoid curve fitting.

I thought I was ready to start forward testing the EA beginning today, but when I tested the EA for the last few months only (in which EUR/USD has largely gone down) it no longer shows a profit. I don't understand this as it should perform well within this time frame.

So here are my questions that I hope a few of you backtesting experts can enlighten me on. For an EA designed to sell on trending trades (looking at 5 minute bars):

  • How long a period should I backtest for and how many different periods should I test against? I've heard that more recent data is more meaning and data too far back is not as good for testing, but I just don't know how far back is good enough.
  • Should I hand pick a period which has approximately as much down movement as up movement (in terms of bars and/or in terms of movement in the rate (price))? In other words, would choosing calendar year 2009 be a bad period to optimize a sell only EA given that it was largely an up year?
  • How comparable will the downloaded historical data from Metaquotes be to actual current broker data? My EA uses price patterns as the primary entry signals, thus the importance that the price action be similar to current market price action.
  • Finally, does that fact that my EA tested well against several periods but not against 2008 and not against recent months mean that my strategies are probably flawed, or just that I incorrectly optimized parameters and my strategies may be sound?

Thanks for any ideas you can give me.

JB

 

My initial EA made 115million from 100 when optimized for 2008 using trader history, but lost everything for 2009 and optimizing with downloaded history, for any other time period was a complete looser.

Not opening loosing trades is just as important as opening winning trades. I'd download history data (don't use the data comes with the trader.) Then optimize from 2006 to mid 2009.

Then test from mid 2009 to present. If it wont make money then, your strategy is flawed.

New comment