How Long to backtest a ea
-
Many say a ea will not last a year, but why do many say you have to backtest over years.
If there ANY who is profitable with there ea, how long do you backtest.
Also what I also wonder about why don’t we hear any big stories about successful bots out there. A lot of manual trading/strategies texts is the only content.I would say backtest should maybe be 3 month to 6 month?
-
Markets can change sentiment, so testing for a short time can give unrealistic results, over years can allow the EA to correct short term results. There are successful bots, but people will not give the information for free.
-
@Fireblade
I think there is no definite answer.Apart from the period duration, I would also consider the number of trades done within the period. I would expect that at least 100 trades. Say, if there are 100 trades in 3 months, I would test at least 3 months.
Sure you would backtest again after some time.
-
3-6 months is way too short. I'm assuming that you mean a bot that places trades. An EA could absolutely last a year. Even two. You can put little effort in an EA and it can be be profitable one year but the next year completely blow up. Testing over years puts the EA in different market structures because the more time that goes by, the more market conditions it goes through. Just like anything else especially any type of engineering, things are more so tested in environments that stresses it so that the limits are known. Unless there's some major machine learning stuff going on you'll be hard pressed to find an EA that is profitable in all market conditions over the course of years. I think the amount of processing power needed would bug the EA down and even if someone is pulling it off, they aren't putting that information out. That's why I don't focus on building EA's that take trades because if I have to constantly adjust it to be profitable one year at the expense of drawing down or blowing up another year, it's not viable. I know this goes against saying that you should test for years but it's much better than a few months. There still needs to be human discretion and knowing when to engage the EA or turn it off. Even the software that market makers, banks, and hedge funds use is managed by people. They don't just set it and forget it.
-
@JayPhillbrooks - i also agree with that, right now im testing for one year, the thing is would it be wise to retest it ever 3 month, for a period of time of a year?. To have a stable bot.
Your bot just tell you, have a look now, maybe there is a trade for you?
-
I never try a bot on real with AT LEAST 10 years reasonable backtext and then an additional year on forward test.
-
@Fireblade Yes my bots give signals. I do have ideas to implement executing trades but as of now I'm not interested in trying to figure it out. I do build general bots to trade and have built countless of them in the past. Going back one year or shorter is a good starting base just to see if it's basically doing what it is told. But after that it just makes sense to see how it would have performed in much earlier years. I believe that a bot's stableness depends entirely on the user. An example that I notice from many people sharing their projects on here, their strategy will have a limit buy but no time expiration for it. That order will then get triggered at a much later time with no relevance to the actual strategy because they didn't think to remove it so it affects their results. It could also work in their favor but either way, there's an order that's being reflected when it shouldn't. Build it to where it makes sense and go back as far as you can. See what market conditions crushed it the most and then consider pluging it. If an 8 week downtrend blows it up. Engage it after you see an 8 week downtrend. I've noticed that when most EA's I build are most unprofitable nearly to the point of blowing up. That's usually a good time to deploy. When they are most profitable is usually a good time to take the profit and turn them off. Maybe that's due to the strategies I use.
