Machine learning in trading: theory, models, practice and algo-trading - page 727
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
Today all night I was conducting tests and working out an algorithm for model preparation, starting from preprocessing and ending with optimization and evaluation of the resulting model. I managed to get the result of the work for a month. Profitability is not great, but Profitability is more than three and Equity looks quite promising ...
In order to finally obtain such a model, it takes an average of three hours or more, which is quite acceptable, taking into account the time of model's work...
Again, the yield is not great, but steadily upwards. And this is just one model for buying, for sales did not do, because particularly there was no need. One model was enough to test my approach to trading.
. Now I have made new models, let's see how this month will work out....
No, I personally will not predict unsteady VR. And all my attempts to bring it at least to quasi-stationarity ended in failure.
Finita la comedy.
But! Realizing that the case is rubbish, some person, long enough engaged in NS, simply must put the results here, that people have found some other approaches, instead of stomping on a place.
People's gratitude will be his reward, because he will never find money in the NS.
And where is the money???? The answer is in the "tails" of the distributions. In the "memory" of the non-Markovian process. ALL.And they told you right off the bat it was a dead issue!!!!!
Take your blinders off and take a broader look at the market, not as an unsteady rad quote. Broaden your horizons as a trader. You may find the right approach there.....
Today all night I was conducting tests and working out an algorithm for model preparation, starting from preprocessing and ending with optimization and evaluation of the resulting model. I managed to get the result of the work for a month. Profitability is not great, but Profitability is more than three and Equity looks quite promising ...
In order to finally obtain such a model, it takes an average of three hours or more, which is quite acceptable, taking into account the time of model's work...
Again, the yield is not great, but steadily upwards. And this is just one model for buying, for sales did not do, because particularly there was no need. One model was enough to test my approach to trading.
. Now I have made new models, let's see how this month they will work out....
Curvafitting
Here's the model I wrote down for sale. On tests MUI from the release of these models was not significant, although the quality of training was high, so this model did not come out of the yield above 1.38, OOS from 01.31.2018
Well as a whole both models looked even not bad during the month and by the number of trades and by the PF indicator above 2.
Eh... I would like to see it on the account......
Mihail Marchukajtes:
Eh... I wish I could see it on the account......
So what prevents it?
So, what prevents it?
I don't know. Now I charged it, let's see what's going on....
Today all night I was conducting tests and working out an algorithm for model preparation, starting from preprocessing and ending with optimization and evaluation of the resulting model. I managed to get the result of the work for a month. Profitability is not great, but Profitability is more than three and Equity looks quite promising ...
In order to finally obtain such a model, it takes an average of three hours or more, which is quite acceptable, taking into account the time of model's work...
Again, the yield is not great, but steadily upwards. And this is just one model for buying, for sales did not do, because particularly there was no need. One model was enough to test my approach to trading.
. Now I have made new models, let's see how this month will work out....
i do not understand - what timeframe, symbol?
The lower model looks more stable
//The upper one looks very much like a fitting...
it's not clear - what time period, the symbol?
The bottom model looks more stable
//The top one looks very much like a fitting...
About c 31.01.2018
The topmost picture post above is one model for signals "Buy", the next is the one in the middle, it is for signals "Sell" the third is them both work. In a month about 70 trades that significantly exceeds the period of training almost twice.... Which surprises me...
Analyzing the obtained models, I found a way to finally verify the quality of the obtained model, as well as whether it is inverted or not.....
After deciding on the sample, I build several models on it, usually four. Then I run the results of polynomials (their actual value) through R, where I calculate the mutual information of each polynomial in relation to the sample. Taking into account that one model is two polynomials, then as a rule their VI are different, but here is what interesting fact I found out. It turns out that if you add up values of VI for two polynomials, it should be more than output entropy. Maybe it is so for me, with my sampling interval, obtained models, etc. But there is something about it. So with 40 entries and equal number of zeros and ones in the output. That is 20 of each, the output entropy is equal to about 0.693147 and it never changes, no matter how many ones or ones are between the zeros.
So in the first picture, where the model "Buy" total VI to the output, in the training area was approximately 0.86, in the middle for the sale 0.65, which was slightly below the output entropy of 0.69, indicating that this model has gained by chance..... may not, but the topmost ... there's.... I'm shocked, to be honest .......