Is this a Good Result for BackTest and Forward Test ? - page 3

New comment
 
Show a backtest on real tiks, pls.
 
tumcog #:
Show a backtest on real tiks, pls.
why ?
 

Sir/Mam,
Should I add spread, commission, slippage and swap when calculating the total net profit of a self-optimizing EA in MQL5 ?

 
Vikram J U #:

Sir/Mam,
Should I add spread, commission, slippage and swap when calculating the total net profit of a self-optimizing EA in MQL5 ?

When testing on real ticks, the historic spreads are used. Enter your commission fees if your broker-dealer charges them. The best that you can do for slippage is set Delays to whatever your broker-dealer's latency is, plus a bit of extra just in case.
 
Ryan L Johnson #:
When testing on real ticks, the historic spreads are used. Enter your commission fees if your broker-dealer charges them. The best that you can do for slippage is set Delays to whatever your broker-dealer's latency is, plus a bit of extra just in case.

Thank you sir. I understand.

 
Hi Vikram,

I understand your concern. Everything that looks like a good strategy from the chart eventually falls apart.
Your optimization covered three months. On M5, that may sound like a lot of data. But your EA may only make a few hundred trades during those three months, and you have probably optimized the strategy over that same period. That is curve fitting, and it cannot be fixed simply by extending the period.
Your second question about the forward test is more interesting. A forward test that beats the backtest is exactly what you would expect if the result is noise.

With noise, every outcome is equally likely. So the fact that the forward test performs better is not necessarily reassuring. It can actually be an indication that there is no stable relationship behind the results.
An important question you should ask yourself is: how much of that $24,089 would a coin flip have produced at exactly the same moments?
If the result is roughly the same, then you are not looking at a strategy with a real edge.

 
Rather than giving my opinion in terms of answering your questions one by one, when it comes to forward optimisation you get results based on predictive analysis. Personally it is almost impossible to determine future trends and I would personally run a demo account OR place a live account with a small balance and let it run. Always remember "past performance is not indicative of future results" 
 
Three months is too short for backtesting, even profitable system will have ups and downs. Consider to do the test for a few years.
For forward testing, too bad we can't speed up the time. But since the time is going forward anyway, try to do forward test for some times and later do a real tick backest on your previous forward test time range. 
Both results have to identical, a slight differences is ok of course. That's how you can judge if your backtest is trusted or not.
 

I would not compare only the two profit numbers. Three months can easily contain one very favorable gold regime, especially on M5. Freeze the parameters, test several separate periods and check if profit factor, drawdown and expectancy stay reasonably stable.

For final verification use real ticks, commission and realistic delay. Also test nearby parameter values, because if only one exact combination works, it is probably fitted too closely.

123
New comment