Machine learning in trading: theory, models, practice and algo-trading - page 116
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
The market moves against its own statistics, this is a theory that I have confirmed with practice, this is the only theory I know that gives answers to all questions, from why the model does not work on new data to why everyone is losing money in the market...
why is this so hard for you to accept?
does the old knowledge and habits so much suppress the perception of new information?
Why concentrate so much on the model if the difference in performance between models is between 0.5% and 5%?
No model can help here because the essence is in the data itself
It is possible to understand a thought, information, information which must be clothed in common terms and concepts.
But you with your repeated calls demonstrate the dense ignorance, not knowing the most basic things of statistics.
The market moves against statistics, but there are different statistics. The notion of this is basic when the word "statistics" is used. If you don't understand that, then you show graphs that don't say anything.
All that you write as a refutation of statistics in your understanding is a refutation of the part of statistics that is applicable to STATIONARY random processes, while there are no stationary processes on financial markets - only non-stationary ones that are different from stationary ones in that the characteristics obtained on historical data are not applicable to new data. I'm sorry, but this is the basics. You cannot apply the tools of stationary statistics to non-stationary financial time series.
Here we are busy trying to teach the models on historical nonstationary data in such a way that they have approximately the same characteristics in the future. That's the problem we're solving. And I argue that with careful adherence to a number of techniques, this problem can be solved.
I'm sorry, I'm fed up with my ignorance.
The market moves against its own statistics, this is a theory that I have confirmed with practice, it is the only theory I know that gives answers to all questions, from why the model does not work on new data to why everyone is losing money in the market...
why is that so hard for you to accept?
That's right, almost. It doesn't go against it, it just randomly goes from one noise to another, according to the fits to the noise built into the model. That's if that's how you did the modeling. You're the only one responsible for that.
Do another model where the market will walk on the found statistics.
What do we need to accept here? The fact that we make models and they are overtrained? Or the belief that "The market walks against its own statistics." We're not a religious organization here to believe in holy fictions.
But if you write a study in which you try 100 methods on all of the Forex data and conclude that all of them are overtrained and do not give any profit, then we would love to read it. And one micro-study is not clear on what principles made it is not an indicator.
I share an interesting article. It makes sense to me, it seems like a good attempt. There are thousands of models that show some nice trade results in training and testing. There is out-of-sample data. The model selection procedure starts to be checked. The graph in the article is the same as mine. Weak correlation of the test and validation. It means that selecting models by the best test results does not give any out-of-sample superiority.
And then see for yourself what they do there.
https://blog.quantopian.com/using-machine-learning-to-predict-out-of-sample-performance-of-trading-algorithms/
The market moves against its own statistics, this is a theory which I have confirmed by practice
If the market is dominated by a trend or sideways and the main model factors are trying to predict the trend continuation or reversals from the channel boundaries, then it is not surprising if the counter-trend (learned from sideways) model is "reversed" in a region where the trend prevails and it will profit.
But the point is that your "theory" is not always confirmed in practice, but only when trendiness changes into counter-trendiness and vice versa.
This happens most often if a large part of the predictors are insignificant - garbage - and some one is the most significant, such as momentum. In that case, the most significant predictor will get a fat weighting factor during "learning", and the others will only be slightly noisy with small weights.
If the market is dominated by a trend or sideways and the main model factors are trying to predict the continuation of the trend or reversals from the channel boundaries, then it is not surprising if the counter-trend (trained on sideways) model is "reversed" in the area where the trend prevails and it will make a profit there.
But the point is that your "theory" is not always confirmed in practice, but only when trendiness changes into counter-trendiness and vice versa.
It happens more often if a considerable part of predictors are insignificant - garbage, and one of them is the most significant, for example momentum. In this case the most significant predictor will get a fat weighting coefficient during "training", while the others will only make some noise with small weights.
Such a theory is confirmed in practice ALWAYS.
Because the only stable statistical characteristic of the market is non-stationarity - the longer the trend or flat phase, the easier it is to identify it statistically and the more probable it is to change the trend.
It is the only stable statistical characteristic of a price series.
This theory is confirmed by practice ALWAYS.
Because the only stable statistical characteristic of the market is non-stationarity - the longer the trend or flat phase, the easier it is to identify statistically and the higher the probability of trend change.
This is the only stable statistical characteristic of a price series.
1) Is this a graph with the data on which the training itself took place, or is this just a test on new data?
2) If it were that simple, then it would be enough to train any model and just invert its predictions. Unfortunately it doesn't work.
3) The problem is not that the models give opposite results, but that on some bars the result will be correct, on others - wrong, and all this is random and without the ability to filter out only the correct results.
4) And why do you take away prediction S from prediction B? Maybe you should do the opposite, S-B? Then all of a sudden the correlation will be correct too.
1) test on new data, didn't even look at training data, for some reason
2) yes! It doesn't work, because the obtained blue chart has no predictive ability, it reverses at the same time as the regular price chart, or even a candle later, but it's rare, so it's impossible to make profit with this chart, but it gives you insight into the process I wanted to show you
3) Well you see, there is no randomness on the chart, the chart is completely inversely correlated with the price
4) If we express buy and sell signals quantitatively, for example buy = 10 points, sell has 5 points
buy - sell = 5 ( more bays all right)
and if
buy - buy = -5 ( and here it is something absurd)
1) Right, almost. It doesn't go against it, it just randomly goes from one noise to another, according to the model's built-in fits to the noise. That's if that's how you did the modeling. You alone are responsible for that.
2) Do another model where the market will walk on the found statistics.
3) What needs to be accepted here? The fact that we make models and they are overtrained? Or the belief that "The market walks against its own statistics." We're not a religious organization here to believe in holy fictions.
1) Do you see randomness in the blue chart in relation to price?
2) It's not as easy as it seems
3) try to refute it ;) what you say is a fantasy - it's also a fantasy, but it's just your own, dear ... do you agree?
But the point is that your "theory" is not always confirmed in practice, but only when trendiness changes into counter-trendiness and vice versa.
The resulting blue chart moves against both small movements and large ones. If you look at a 200 candlestick chart will go against the price, and if you look at 20 000 candlesticks the picture will be the same
Sorry, I'm sick of your ignorance.