Using neural networks in trading - page 6
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
Standard indicators are a reliable basis for successful trading. Recommended.
I am not asking how to use them, give examples.
I'm not asking how to use them, give examples.
For example, any of the oscillators.
Which one do you prefer? Take a closer look at it. Look at it on different TFs. In the vast majority of cases they give an unambiguous answer to your question "What's next?".
You see, it means that successful trading depends not only on the indicators themselves, but also on their correct application... And it means that if I can find dependencies, it means that a neural network will find them as well (maybe even some patterns of its own). Right?
For example, any of the oscillators.
Which one do you prefer? Take a closer look at it. Look at it on different TFs. In the vast majority of cases, they will give an unambiguous answer to your question "What's next?
I prefer the macd.
Yes, but first you have to figure it out for yourself and then teach the network. Just as you teach your son the knowledge you have accumulated, rather than leave him in the woods with the confidence that he will figure out what to do and how to do it on his own.
That's understandable. But that's not what I mean. What matters is not the "exchange of experience" but the result. Today I have launched simultaneous training on 28 currency pairs. 4 hours later I have made 218779 points of profit (including spreads per trade), while the average ratio of profitable trades to loss-making ones was 2.12461. Every hour the sample changes, a fresh bar is added and the oldest one is deleted; the depth of the whole sample is 1 year. So, though on historical data, the network has found dependencies, even though the inputs are only prices.
That's understandable. But that's not what I mean. The main thing is not the "exchange of experience", but the result. Today I have started a simultaneous training on 28 currency pairs. In 4 hours their total profit has reached 218779 points (taking into account spreads per trade), while the average ratio of profitable trades to loss-making ones is 2.12461. Every hour the sample changes, a fresh bar is added and the oldest one is deleted; the depth of the whole sample is 1 year. So, though on historical data, the network has found dependencies, even though only prices are entered.
So although on historical data, the network has found dependencies, with only prices as input.
Fit. Watching the backtest is as meaningless as watching a stranger's uncle's trading stats. Just the forward. Why fool yourself?
So far, yes, everything is nice and rosy. But I'll wait a little longer, the learning curve is too slow. And when it comes to acceptable results, I'll put on a demo, without any forwards. I do have memorization problem, but I haven't solved it yet. That is, there are 1010 weights per pair. Lagging pairs in training inherit weights of the most profitable pair once an hour. After inheriting weights profitability naturally decreases but tends to parental pair's values. For now I have stopped on such algorithm.