Machine learning in trading: theory, models, practice and algo-trading - page 1597
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
As for H1, everything is much sadder there
for example here without spread and commission
and now take into account the spread and commission
or here's WITHOUT
and now WITH the spread and commission
someone (market-maker) has long figured out this trick and is successfully using it.
etc.
and we cannot ignore spread + commission
If on the daily chart there is less spread and commission of just 2% of candlestick shadows, then on the H4 it is 6+% for the eurusd
So, cut off weak signals over the threshold
And again we come to the question of overtraining
Wouldn't such a cutting off be overtraining?
If we cut it bravely, we can raise profitability up to 40% per annum, but
"I'm afraid, Slavik..." (С)
So filter out the weak signals over the threshold
and we come to the question of overtraining again
Wouldn't such a cutoff be overtraining?
If you boldly "cut off", then the profitability can be raised up to 40% per annum, but
"Slavik, I'm afraid..." (С)
It's not retraining but a selection of signals that give profits. Not all returns deviate significantly from zero. My results are in the article. I don't know where the 40% is taken from, it's a matter of money management.
The method of choice and can be overtraining
way of choosing and could be overtraining
My results are in the article. I don't know where you got the 40% from, it's a matter of money management.
I took the maximum drawdown of the balance, multiplied by 6 (drawdown = 1/6 of the deposit), the average annual revenue divided by the deposit size and voila
I do not understand these definitions. You need to beat the spread and the commission, so you set a higher threshold. The distribution of the other patterns does not change in any way.
Choosing among all the available inputs that you are given by the pattern you have identified, only those that you like, is overtraining
The correlation between the current and lagged returnee does not change, we just select those increments that are larger than the spread
you can see there on the jinnplots. We can make a separate study for this subsample, of course. I do not have time yet, but I already have an idea for a new article.
By the way, it is an excellent idea to first exclude the returns less than the spread and then to look at dependencies only for them
The correlation between the current and lagged returnee does not change, we just select those increments that are larger than the spread
you can see there on the jinnplots. We can make a separate study for this subsample, of course. I do not have time yet, but I already have an idea for a new article.
By the way, it is an excellent idea to first exclude the returns with a smaller spread, and then look at the dependencies only for them
communication is the greatest value!
it has been noticed that some pairs already show a "slide", i.e. the presence of a local maximum after which the balance curve starts to look down
The fate of the remaining equations is similar to that of some other pairs, i.e. they have already started to control the process
in this regard, it is better to look for solutions that do not have a "slide", but do not put them in the public domain