Machine learning in trading: theory, models, practice and algo-trading - page 133
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
We will consider the LSTM later.
For now, my colleague and I have arrived at R^2 0.2 on the test. A few convolutional filters and a few neurons in a fully connected layer. The idea is that recurrence is not needed there. What is needed is proper extraction of features.
I'm working on the following concept: from the current pattern "B" in the history we look for a pattern analogous to "A". With the dtw algorithm we look for similarity...
The sad thing is that we don't know what size "B" and "A" may end up in the search and that's a lot of headache,
The sad thing is that we don't know what size "A" and "A" might end up being, so it's a lot of headache, so besides the search we need to dynamically shrink/expand these patterns...
If anyone has any ideas how to make such a search as effective as possible with interest I'll listen...
The standard indicator "Bill Williams' Fractals" in MT is also a search for a certain pattern, and without dtw, just by bars. It worked quite well once, until it failed because of its popularity (on some symbols on D1 you can still use it, the profit is minimal, though).
But trading strategy with this indicator is more complicated than "buy/sell on 1 bar". It uses pending, tp and sl, so in addition to looking for patterns you need to look for trading strategies to which they are applicable.
The task of the trading system can be formulated as follows:
on each bar where recalculation takes place, perform two actions:
1) choice between "up" and "down"
2) selection of the trade volume(from 0 to a certain maximum)
It looks like this:
up 0.1
up 0
up 0.13
up 0
down 0.23
down 0.3
...
This is an extremely difficult task for optimization.
To begin with it is possible to make an "ideal" alternation of these simple actions, leading to the maximum profit with the minimum drawdown, by means of rather heavy calculations on the whole history.
In the second step, by selecting training features, you can try to train the machine on the previously selected actions. Example:
ficha1, ficha2,... itch10: 0
ficha1, ficha2,... ...thing10: 0.4
...
fich1, fich2,... fich10: -0.25
Regression in classical form.
Even the first point is not easy to implement. Any thoughts, fellow mathematicians?
The task of the trading system can be formulated as follows:
On each bar where recalculation takes place, perform two actions:
1) choosing between "up" and "down"
2) Choice of volume of transaction(from 0 to a certain maximum)
It looks like this:
up 0.1
up 0
up 0.13
up 0
down 0.23
down 0.3
...
This is a very difficult task for optimization.
I used an indicator that searches for price patterns by Euclidean distance and draws continuation into the future with confidence intervals. I used an indicator that searches for patterns of price with Euclidean distance and draws the continuation to the future with confidence intervals.
Difficult in what way?
In computing. Imagine you have 2,000,000 5-minute trades. For each you need to optimize the value from -1 to 1 in increments of 0.01. The goal is to get a perfect, perfect trade.
For example, it is not necessary to open purchases on each minute on the increasing trend - it is a loss of spread. It is necessary at the very beginning of the trend to buy a lot at once. And it will be like this:
up 1
up 0
up 0
...
up 0
... down 1
instead of inefficient scheme
up 0.02
up 0.02
...
up 0.02
..
down 1
I did it too, I spent about a year on it, and we had a conversation about it, like 50 pages ago, but apparently you have a short memory and therefore no interest in it
That's not a very logical suggestion.
I remember, but there is no interest. But not because it's a crazy idea or because I don't respect your ideas. It's because I'm busy with another idea for now.
In computing. Imagine you have 2,000,000 5-minute trades. For each you need to optimize the value from -1 to 1 in increments of 0.01. The goal is to get a perfect, perfect trade.
For example, it is not necessary to open purchases on each minute on the increasing trend - it is a loss of spread. It is necessary at the very beginning of the trend to buy a lot at once. And it will be so:
You need to find the points of ideal ZZ, am I right?
Exactly ideal. That is, without compromises. If it were possible to optimize actions for every bar ideally, it would be your ideal zigzag.