Machine learning in trading: theory, models, practice and algo-trading - page 1809
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
Well almost every other article there ends with a test on new data
Vladimir, could you please tell me how in R you can teach AMO not to classify or regress, but something more vague...
I don't know what it should look like and what values it should take, and it is not important for me, I can only describe the anticipation and let AMO maximize the anticipation criterion in the anticipation function, which it created itself
How it's done? Or it's a pure optimization problem and has nothing to do with AMO?
ok, I just thought, that there's a summary of all this in a specific product with obvious results (not necessarily good)
Well it's articles, it's a summary into a product, an information product :)
Can we get closer to the ground after all?
A thought on real market implementation:
The best sobering thing is a very simple test to start with.
There's a famous binary option (google it) with a great API and MetaTrader5. You can download their real native quotes and train the neuronet on them, and then test them on a demo account. Everything is transparent and understandable, there cannot be any nonsense about their binary trading platforms, because everything is in MT5.
With the neural network we solve a simple task "above/below" and look at the results. If you don't go through this stage it is not serious to speculate about targets and stops that are controlled by a neuronet. And moreover beautiful graphs of backtests are also not meaningful.
Any neuronet will give at least ten signals per day for short-term forecasts, i.e. it is not difficult to gather 200-300 results a week. Considerations about testing for months and years are meaningful only for "human" strategies. If a neural network has been trained for a long period, it will work adequately in any market, only the number of signals will change. When the market becomes inadequate a neuronet simply stops recognizing legal patterns and does not give signals.
If anyone has real experience I suggest following this way and it does not matter how good the result will be, the main thing is that this is a real, understandable, posorachny result that does not require much time for testing and is not technically complicated.
What to start winning on binary should be only 56% of successful trades, can't we all neural network academicians here get such a result on the REAL MARKET?
Can we get closer to the ground?
A thought on real market implementation:
The best sobering thing is a very simple test to start with.
There is a famous binary option (google it) with a great API and MetaTrader5. You can download their real native quotes and train the neuronet on them, and then test them on a demo account. Everything is transparent and clear, there cannot be any nonsense about their binary trading platforms because everything is in MT5.
I think this is not the best way to get the trades, I would never do that.
We solve a simple task "above/below" using the neuronet and see the results. If you don't pass this stage, it is not serious to discuss takeoffs and stops that are controlled by the neuronet. The more so, beautiful graphs of backtests are also not meaningful.
Because of the constant changes in the market characteristics, you can't work with fixed parameters, your "above/below" is the same for some period of time, say 10 candles. It's the same story there, the same problem, only multiplied by the number of signs - those 100+... So your "above/below" becomes inadequate to the market very quickly together with the signs... And everything will fall apart almost immediately!
Of course you can try to solve the fractality problem by training hundreds of AMO on data with different periods, as you do, or to train one model on hundreds of data with different periods as I do, but it's all crutches and nonsense, it's not an affective solution...
I want the network to look for the optimum by itself, let it determine what is the optimal target at the moment, what is the optimal attribute at the moment, it will be much more effective than our crutches that we have now.
When the market becomes inadequate the neural network simply stops recognizing legal patterns and does not signal.
The market is always adequate, the model is inadequate, for the above-mentioned reasons !!!!
Why waste time testing on the demo? Can't you simulate trading in your own code? It's not optimal, I would never do that.
I do not want to be under any illusions. Binary options are a platform that will not play in your favor.
Now the whole topic of neural networks and trading already looks like masturbation, maybe it's time to try it with a real woman?
Maybe it's time to try it with a real woman.
I've tried it... I'd rather not))
But seriously, you want to race a week/month on a demo instead of writing 5 lines of code and check it on the history?
And if my robot doesn't work, I have to use the demo for a month to figure it out?
And if I want to test it in 5 years, I will have to play it in the demo for 5 years?
Some kind of dead-end talk unfortunately...
Case in point...
On my backtests I had a neuro result of 95% in the "above/below" test. Then I found a bug and it was 67%. Then I ran it on binary and the result was 55%. After this stress, I have found another, more fundamental error and binary has got 66%.
Conclusion: You need an independent judge, otherwise you will be floundering in your own illusions.
Had to... I wish I hadn't tried it ))
But seriously, are you suggesting to run a week/month robot on a demo instead of writing 5 lines of code and testing it on the story?
YES! That's the way and no other way!
mytarmailS:
And if the robot does not work, then I need a month to run it on the demo to understand it?
Usually 3 days and everything becomes clear
It is better to test without emotions))) Trade all the more)))))))
To be honest, the theme of predictors is not disclosed. As well as the logic of models, which ones should be applied when, and what is the criterion for their selection.
Recommendations on how to prepare data have nothing to do with the result. Although without it you can't start)))))