Expert Advisor Robot - stochastic + artificial intelligence
The market is constantly changing, and old ideas periodically outlive themselves. But if the problem cannot be solved by the traditional method, an unconventional approach, such as neural network analysis, can help. This is the approach used by the Expert Advisor with the simplest name Robot, published in our forumwhich combines traditional technical analysis with the beginnings of artificial intelligence.
Characteristics of the Expert Advisor Robot
Since it is a neural network robot, it has the ability to work on any currency pairs and markets. The teachability of the principles of the dynamics of a given asset does not limit it in terms of timeframe and working time. So you can use it in long-term trading as well as in scalping. The only difference will be the training data.
- Timeframe: any
- Currency pairs: any
- Trading hours: 24 hours a day
Description of the principle of operation of the Expert Advisor Robot
The Expert Advisor Robot uses 4 most traditional stochastics with different shift values as input data. The input data is processed by the usual linear classifier, which is the simplest model of a nerve cell. The learning process of such an algorithm is the selection of optimal values of weights, which, with the right approach to the learning process, fixes some regularity (pattern), if it really exists.
And although the Expert Advisor does not possess a full-fledged artificial intelligence, the use of such an algorithm gives several advantages over the traditional (manual) selection of logical conditions.
- First, there is the flexibility of the conditions themselves. All you set in the beginning: the type and the number of input parameters, but, by default, the Expert Advisor does not yet know what to do with these parameters. Only in the process of optimization (learning) the Expert Advisor acquires some picture of the market and can further predict its movement, based on its past experience.
- Second, it is the possibility of quick retraining. If the market has changed and old patterns have lost their relevance, all you need to do is re-optimize values of weights, and then (in theory) the robot will work profitably again.
Operating parameters of the robot
In this case, all input parameters for the single-layer perceptron are shown in the picture. The first parameter "shag" defines the shift value for each of the stochastics. That is, the first indicator has no shift, the second one has a shift shag, the third one has a shift shag * 2 and the last one has a shift shag * 3. We will select the shift value from 1 to 6. In order not to make the optimization last for a week, we select a not too small step for the selection of weights - in this case, 20 iterations for each of the multipliers. This should be quite enough to determine the approximate type of the pattern.
Testing and optimizing the Expert Advisor Robot
Optimization parameters:
- Currency pair: EUR/USD
- Initial deposit: 10 000
- Tick generation
- Time frame: H1
- Testing period: April 1 - May 1, 2016
- Spread: 10 pips
As a result of optimization, we got the best option (taking into account profit/depreciation) with these parameters:
- Step - 5
- x1 - 0
- x2 - 70
- x3 - 150
- x4 - 160
According to the results forward test with the selected parameters we can judge about the good teachability of the strategy. Profitability and general dynamics almost completely correspond to the results of back-testing.
However, the relative drawdown of the Expert Advisor is almost equal to the amount of the initial deposit, which is an unfavorable condition for using it on a real account and is quite possible to lead to a rapid loss of the deposit on a real account. But this problem can also be solved by using less aggressive money management.
Conclusion
We can say that Robot is practically a universal Expert Advisor, which, among other things, can be easily adapted to any market situation. The parameter mult is responsible for multiplying the lot in the sequence, which allows you to apply both aggressive and more conservative approaches to trading (for a fixed lot set the value of 1.0).
Nevertheless, it is not at all a Grail, as it may seem at first sight - the Robot requires constant attention, because once "having learned" to work on one variant of the market movement, the Expert Advisor will simply follow the market paradigm familiar to it. When the situation changes, it can start to lose money unexpectedly for the owner. The efficiency of this approach is generally similar to manual periodic selection of conditions for Expert Advisors, but its main advantage is the significant saving of time for the trader, and the result can even be superior to more "advanced" strategies.
Therefore, in order to keep the results predictable and satisfactory, periodically re-adjust the weights, then the robot will always deal with the current market picture.
The conclusion: there is no smell of artificial intelligence here, a common robot on stochastics (I admit that I did not do optimization). what do you do? if you test it, it learns and then after a while it loses you, then its name is a financial diverter. but the idea is good, but very crude.
I tested, good up to 20 points. and optimized and still turned out to be shitty and ill-conceived. all the same plum. and in general, what they came up with is the intelligence of the creators (apparently amoebas).
Does it close positions by itself or not? Because he opened about 30 positions for me now
It closes. But you should first look at the work in the tester to understand the meaning.
In the tester everything is always good, I just opened a separate account with a small deposit (100$), let's see.
And what parameters should be prescribed for stochastics?