Hello, I would like to ask you a question. I have seen that you have been doing backtesting since 2017 or 2018 and my question is when I try to do an optimization for more years. Since 2013, for example, the DD has risen a lot. And as a general rule, good optimizations after 5 or 6 years fail. What do you think, is it necessary so many years?
Thank you
In 2019 we had the covid pandemic, the markets changed a lot. Markets change over time. It is difficult for a strategy to adapt to all the changes that happen in the world. Maybe 2 or 3 years of backtesting.
As you mentioned, I don't think there is a perfect strategy and in the end there is always a year that fails. What I think about a lot is how to change my strategy before something happens that leaves me without money, but I guess this will be the million-dollar question. ;-))
Exactly! If you know when the market will change, you won't need anything else.
It was (and still is in part) also my million-dollar question.
I generally do this: optimize to the best of my ability from 2020 to today and then forward test...into the past! If it passes at least from 2018 (preferably 2016) to now it can go into production.
Luckily for me I have a network of very powerful 6x 5950x computers and can afford to do a ton of backtesting.
I have not yet found the holy grail though but if I do I will let you know :-P
As you mentioned, I don't think there is a perfect strategy and in the end there is always a year that fails. What I think about a lot is how to change my strategy before something happens that leaves me without money, but I guess this will be the million-dollar question. ;-))
In my case, what I am doing is with some Indicators that tell me the DrawDown in real time (PZ Risk Management y pz_balanceGuardianEA) and an audible alarm that warns me when it exceeds a certain value (>20% I'm already worried) and with a DrawDown > =50%, I insert double the capital into the account to reduce it by half and I stop the martingale while I think about any change in strategy. The Any To Any close at $1 helps me reduce martingales.
I'm looking for a utility that will notify Telegram when the Drawdown is >=X, but I can't find it.
It was (and still is in part) also my million-dollar question.
I generally do this: optimize to the best of my ability from 2020 to today and then forward test...into the past! If it passes at least from 2018 (preferably 2016) to now it can go into production.
Luckily for me I have a network of very powerful 6x 5950x computers and can afford to do a ton of backtesting.
I have not yet found the holy grail though but if I do I will let you know :-P
Thanks for the advice, I also optimize at least before Covid-19 and then look at previous years to see if something can be fixed. Sometimes raising the SL or lowering the lots works.
In my case, what I am doing is with some Indicators that tell me the DrawDown in real time (PZ Risk Management y pz_balanceGuardianEA) and an audible alarm that warns me when it exceeds a certain value (>20% I'm already worried) and with a DrawDown > =50%, I insert double the capital into the account to reduce it by half and I stop the martingale while I think about any change in strategy. The Any To Any close at $1 helps me reduce martingales.
Thanks Ulysses, the truth is that I once closed operations manually trying to upload lots and with reverse operations, but quite badly. It is difficult to optimize the strategy when you already have operations underway. What bothers me is that, in the end, there are very few operations that go wrong over time but destroy the entire strategy.
I'm looking for a utility that will notify Telegram when the Drawdown is >=X, but I can't find it.
https://www.mql5.com/en/market/product/13639?source=External#description
send email not to telegram
It depends on a lot of factors, from your strategy to how often you trade.
And there is no right answer here.
Longer period leads to overfitting, but allows to check different market phases.
Shorter period can miss some valuable events.
Google it a bit, it is very interesting topic.