Machine learning in trading: theory, models, practice and algo-trading - page 3688
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
A couple more SABJ tests. For some reason the person got banned.
On the orange line (the original filtered graph) you can see some semblance of a carrier cycle (actually one of the Fourier cycles, but filtered by the range of values)
Sometimes you can pick up a pretty good "carrier cycle" working on the OOS.
Was re-reading the thread, over the last 2.5 years. I found that some of my posts look harsh, there is no context for it (or moderators have erased it). I apologise to anyone I may have offended. Apparently this is my flaw - one of many.
When is forgiveness Sunday over there? )
High-frequency harmonic (16th out of 20) and shorter OOS
low-frequency ripples are traced on the yellow chart (because the bot opens trades in certain phases of this cycle)
A couple more SABJ tests. Somebody got banned for something.
On the orange line (original filtered graph) you can see some semblance of a carrier cycle (actually one of the Fourier cycles, but filtered by the range of values)
Sometimes you can pick up a pretty good "carrier cycle" working on the OOS.
Sounds like something unreliable.
A short OOS, it seems to me.
Recently there was hope for another test with 100% returns from January to June 2024. All OOS on Walking Forward with retraining once a week.
Checked from 2020 to 2025: also 100%, but already for 5 years and with drawdowns in half a year and with good half-years.
This was on MQ Demo data. And on real DC the swap is an order of magnitude higher - in the end it turned out to be a plummer.
Sounds like something unreliable.
Short OOS, it seems to me.
Recently was hoping for another test with 100% profit from Jan to June 2024. All OOS on Walking Forward with retraining once a week.
Checked from 2020 to 2025: also 100%, but already for 5 years and with drawdowns in half a year and with good half-years.
This was on MQ Demo data. And on real DC the swap is an order of magnitude higher - in the end it turned out to be a plummer.
I tried different methods of generation, of them only GMM pleased me, so the article is only with its participation.
Rating of generators (from memory):
- GMM
- copulas
- autoencoders, including conditional ones (already bad), or lacked understanding of how to use them better.
- GAN (also bad + slow)
- GAN on transformers and other complex models. (tGAN, tsGAN) Very slow and very bad.
- Other very bad ways...
There are already different packages for generating synthetic tabular data and time series, I don't follow the development.I'm experimenting with generators a bit. So far GMM is still better.
I have an interesting idea to train several regression models of different complexity and do augmentation through their predictions. I guess that would help diversify the data, while keeping the overall structure.
Pretty simple and logical way to do it (at first glance)
Reverse engineering a market strategy using this approach
I took a look at what and approximately how it is traded, made small changes to the logic. It turned out to be similar.
Reverse engineer a marketing strategy using this approach
I took a look at what and approximately how it is traded, made small changes to the logic. It turned out to be similar.
What's the comparison? Where's the original?
What to compare it to? Where's the original?
Top to bottom. Just exported the history of the signal. But it was trained not on signal trades, but through trying to pick a TS through known features. For the sake of curiosity, how well the method can find something, given the given constraints :) And it turned out that I found almost the same as the author's. The features are arbitrary, so there are differences.
In general, for some reason nobody is still not engaged in fitting on examples of existing signals, although the idea was voiced a long time ago. It turns out that there is already a large database for NS training. Probably it is lazy to write code.
Imho, it is much easier to copy a ready-made one than to invent it from scratch