Machine learning in trading: theory, models, practice and algo-trading - page 17
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
My experience as a stock speculator began with Borovoy's checks. Before that, I spent another 20 years investing in the real sector.
And you, by the checks, were already born?
You could call any granny in the market a practitioner. Do you have any practice of profitable TS? Stat. significant profitable TS - thousands of non-intersecting in time deals?
Call George Soros a practitioner of algorithmic profitable TS! He is a complete zero in this field. You are much more competent than he is - no sarcasm. But this does not make you not a theorist.
Anton Zverev
Let's not talk that way, people who learn and share their experiences here are willing to help each other, while you take the position of saying you are stupid here and I know everything) You'd better help me understand what you think and experience is right.
You have an inappropriate tone between the lines, not in the text (read it). With developers and moderators can be, a lot harsher conversations, getting a daily / weekly bans for insulting the self-esteem)) Do not worry, in short. I am good!
Dr.Trader knew it right away. He told it like it is. So, respect and admiration to him.
After it I realized a kind of indicator, I took the cumulative sum of all buy prices and also the sum for profit, built their difference and got some index, when I compared it with the price it appeared to be opposite to the price, the correlation was -0.7 to -0.9, simple - the market goes against its own statistics, that's something to think about and reconsider
There's nothing interesting there, it's the conclusions themselves that are interesting...
looks like http://prntscr.com/aqg96r at best...
And to reproduce it, you have to write code to search for patterns, then run it for a couple of days to process several years of history
Hello!
Who worked with depmixS4 package ? or in general with hidden Markov models in R, there is an interesting idea and there are some questions
Hello!
Has anyone worked with depmixS4 package ? or in general with hidden Markov models in R, I have an interesting idea and have some questions
There's nothing interesting there, it's the conclusions themselves that are interesting...
looks like http://prntscr.com/aqg96r at best...
To reproduce it, you need to write a code to search for patterns, then run it for a couple of days to process several years of history
The point of any machine learning algorithm is to look for patterns. I gave an example with trees above. You can print them out and see what patterns were found. For 100 predictors with 18000 bars it takes a few minutes.
Not (but I will read your ideas with interest.
Yesterday I was inspired by this article or blog https://forum.mql4.com/ru/26460 no matter, the idea is to divide the graph into frequencies, impose a trading system on them and identify only the frequencies (parts of the graph) in which the system makes money, and trade by this system only these frequencies
I kept thinking how it could be done easier and faster (it took the author 16 hours to calculate one frequency, and his frequencies were 500 and something like that)
I remembered that I dabbled, albeit very superficially, with SMM (hidden Markov models). SMM are used for probabilistic forecasting of non-stationary processes, voice recognition, I even read somewhere that they tried to forecast sunspots...
I tried to apply them to the market in their pure form, like a network or RF, like a target and forward... I didn't get good results, although there are people who got something out of it (for example http://www.quantalgos.ru/?p=1759).
The idea of SMM to divide the object into n states and assess the probability of transition from one state to another. I propose to divide the market into a bunch of states, let's assume 10, cut out from the graph all sections that correspond to well say, state №5 and glue them together, as a result (in theory) we get a stationary series that will be stable (in theory)its attributes, having estimated it even visually it is possible to make a trading system on it, optimize it and when the same market condition will occur again it can be traded and it should make some profit (in theory) because the new series will have the same attributes as the previous one
To begin with all that is needed is just to cut out sections of one state and glue them together, and just look visually and assess whether it is stationary, then if everything is "even") then you need to take and look at the quality of recognition of new states, that is whether the predicted state number 5 corresponds to the found old state number 5, if both tests say "yes" then there is a sense to develop the idea.
I'm sure I didn't say something and something is not clear, ask, I'll answer if I know the answer.)
Yesterday I was inspired by this article or blog https://forum.mql4.com/ru/26460 no matter, the idea is to divide the graph into frequencies, impose a trading system on them and identify only the frequencies (parts of the graph) in which the system makes money, and trade by this system only these frequencies
I kept thinking how it could be done easier and faster (it took the author 16 hours to calculate one frequency, and his frequencies were 500 and something like that)
I remembered that I dabbled, albeit very superficially, with SMM (hidden Markov models). SMM are used for probabilistic forecasting of non-stationary processes, voice recognition, I even read somewhere that they tried to forecast sunspots...
I tried to apply them to the market in their pure form, like a network or RF, like a target and forward... I didn't get good results, although there are people who got something out of it (for example http://www.quantalgos.ru/?p=1759).
The idea of SMM to divide the object into n states and assess the probability of transition from one state to another. I propose to divide the market into a bunch of states, let's assume 10, cut out from the graph all sections that correspond to well say, state №5 and glue them together, as a result (in theory) we get a stationary series that will be stable (in theory)its attributes, having estimated it even visually it is possible to make a trading system on it, optimize it and when the same market condition will occur again it can be traded and it should make some profit (in theory) because the new series will have the same attributes as the previous one
To begin with all that is needed is just to cut out sections of one state and glue them together, and just look visually and assess whether it is stationary, then if everything is "even") then you need to take and look at the quality of recognition of new states, that is whether the predicted state number 5 corresponds to the found old state number 5, if both tests say "yes" then there is a sense to develop the idea.
I'm sure I did not say something and that something is not clear, ask, I'll answer if I know the answer myself)
You can divide the series into parts (quantize) either by clustering or, for example, by convolution with SKH (Kohonen). And then it is a pure experiment.
Cluster is a little different, well, say the market now is the cluster number 5, the next candle will be the cluster number 18, it will not give us anything because we do not have time to trade cluster number 5, and in SMM is the concept of state, the state can last a certain time
Or maybe I don't understand your thought?