how far back a trading strategy should be profitable to be valid ? - page 3

New comment
 
Matthew88819 #:

Are you factoring in real fluctuations, spreads in live conditions? 

The MetaTrader 5 terminal attempts to do that when set to "Use real ticks" 

Otherwise I haven't read of any guarantees that is being done under the hood (especially if your broker can't send the data over the network for whatever reason), unless you possibly write your own back test library. 
 
metinique: Hi, i got a few strategies developed and coded by me, i test them to be profitable for the last 7 years (2018-2024 included). i now have a strategy in my hands that is decently profitable 2016-2024 but is not profitable 2012-2015, is that a reason to not take it into consideration? In my mind is the fact that back then the trading conditions and the way the market moves were very different . what is your opinion? 

ok, I think one of the best questions in MQL5. There are several fronts, so lets start:

1- Data issues: You have to be sure the quality of the backtest data is good. What we want is stability: If some years are good and some years are bad, the EA is simply not good enough.

2- You need to ask yourself, do you really need such a long period of backtesting? What we want from our EAs is that they can learn and trade with the recent dynamics of the instrument. But dynamics are changing very quickly. Will past repeat and our "learned EAs" trade accordingly*- It is what we expect but most of the case, it is not the case.

What is worse is that the fake sense of the safety by presenting long term simulation results  are effectively used for marketing purposes. Then, in the end what we have is "data reading EAs" those sold $$$ and got 3 stars.

⚠️ Post edited by moderator.

Please be advised that attempts at direct or indirect self-promotion, or advertising is not allowed.

Don't use "Artificial Intelligence” to generate automated answers. They will be removed.

 
metinique:
Hi, i got a few strategies developed and coded by me, i test them to be profitable for the last 7 years (2018-2024 included). i now have a strategy in my hands that is decently profitable 2016-2024 but is not profitable 2012-2015, is that a reason to not take it into consideration? In my mind is the fact that back then the trading conditions and the way the market moves were very different . what is your opinion? 

Thanks!
to my experience, a good strategy is not must be long time profit when back-testing in the case that you know the main trend of the year. 
 
metinique:
Hi, i got a few strategies developed and coded by me, i test them to be profitable for the last 7 years (2018-2024 included). i now have a strategy in my hands that is decently profitable 2016-2024 but is not profitable 2012-2015, is that a reason to not take it into consideration? In my mind is the fact that back then the trading conditions and the way the market moves were very different . what is your opinion? 

Thanks!

very subjective, one size doesn't fit all. Market structure changes and goes through periods where any strategy will be tripped up. Try not to focus so much on the past. 

 
Jordan Lewis Wells #:

very subjective, one size doesn't fit all. Market structure changes and goes through periods where any strategy will be tripped up. Try not to focus so much on the past. 

Well said.

Forum on trading, automated trading systems and testing trading strategies

newbie EA, asking for advice

Ryan L Johnson, 2025.03.08 18:47

Assuming that no undisclosed info is bad, I'd say it looks tradeable. And about 4 or 5 trades per day is certainly active enough.

I always demo test to confirm though.

Forum on trading, automated trading systems and testing trading strategies

newbie EA, asking for advice

Ryan L Johnson, 2025.03.08 19:06

I prefer a minimum of 10 years of historic data in a backtest. The more, the merrier. Having said that, my bots don't trade that actively. I mean if I'm trading 3 times per week, I need a bigger sample.

In your case, I believe that you could do with less due to your bot's trading frequency.

On a side note, I'm more suspect of anything that I've optimized a lot due to overfitting. In my experience, a bot that backtests profitably right out of MetaEditor is more reliable.


123
New comment