Discussing the article: "Neural Networks in Trading: LSTM Optimization for Multivariate Time Series Forecasting (Final Part)" - page 3
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
|
I ran a 1-month, 5-minute run in Research, which produced a 300MB DACGLSTM.bd file. I launched Study. The error rates are alarming. Or is this normal for the very first run?
After the Study, the .nnw neural network files appeared. I ran another round of data collection, but by the very next month, Research had stopped executing trades...
Hurrah! After a few weeks of running around in circles, the error values in Study have finally returned to normal! Now the neural network is actually learning.
I had to expand the reward and penalty algorithm
and also make significant adjustments to the Trajectory’s input parameters
Hooray! After a few weeks of running around in circles, the error values in Study have finally stabilised! Now the neural network is actually learning.
Now comes the interesting bit – how it will trade on the new data… I’m working with wooden models – they’re easier to understand. But on the new data, the trading is almost random, as there are no significant features.
Now comes the interesting bit – how it will trade on the new data… I’m working with wooden models – they’re easier to understand. But with the new data, trading is almost random, as there are no significant features.
I’ve still got a way to go before I can trade for real.) After a few rounds of Research-Study, trades start closing very quickly. The neural network is learning to survive, not to make a profit. And the shorter the trade, the more it seems to the network that it’s incurring a smaller penalty. The dance with the tambourines continues. I need to set up the reward and penalty algorithm correctly, and that seems to be the hardest bit. For now, I’m just trying to impose strict limits. For example, not allowing stop-losses or take-profits to be set below a threshold value. I’m also trying to limit the minimum trade duration so they don’t close almost immediately after opening. I use either LLM, Grok or ChatGPT 4.1, but they sometimes act so daft it’s beyond belief. Still, little by little, we’re making progress. A lot of time is spent on training cycles. It would be great to join forces with someone else; it’s a very promising area.