Neural network trading | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Neural Network Trading, Serious people only! 127 replies

Neural Network Trading with Simbrain? 0 replies

Neural Network trading using Matlab and Metatrader 8 replies

Interested in Neural Network Assisted Trading 4 replies

neural network based systems in forex 22 replies

Attachments: Neural network trading
Exit Attachments

Neural network trading

  • #1
  • First Post: Nov 22, 2021 12:13pm Nov 22, 2021 12:13pm
Did some neural network analysis on a major forex pair - purely technical - and I have pretty interesting results. Anyone used neural net to analyze forex market data or for trading?

And here's a million dollar question technical analysis by the way - can a variable (like price) predict itself consistently in a completely chaotic system?
  • #2
  • Edited 2:14pm Nov 22, 2021 1:37pm | Edited 2:14pm
Oh how interesting it. I'm sure that neural networks and machine learning is the future. Next Summer I'm planning to hire some coders to start that kind of project. I have some developments in this area end even some good stats but unfortunately that project was frozen by me due to financial issues.

BTW I think it is not chaotic. It is like bird murmuring, looks like chaos but in fact there are a lot of strict rules involved.

I can't post in another thread, so I can answer here.

Quote
Precisely what neural network analysis can show - that a chaotic system (which is by definition unpredictable) consistently in the long-run cannot be predicted. And major markets are pretty chaotic.
So statistically maybe you can get an edge just above 50% (aren't casino winning edges pretty small, too?) if we are speaking strictly technical here, which will be profitable but not too practical for most traders. (Although that could be immune to the fundamental factors, and as such shows profit even in pandemics...in fact, any unpredictable fundamental pressures cuold be the upside)
And ALL technical indicators that are based solely on technical variables - price and any whichever derivative of that price, volatility, etc - are subject to the unpredictability of chaotic systems over the long-run.
Again, I'm writing from a purely technical standpoint.

I think that may be it is better to use neural network to find similarity, like the description - to recognize underlying relationships in a set of data. For example I see every day the same pattern on Gold. It looks like the same algo creating the same traps. But there are some fake entries, and they can be filtered out by neural network too.
Observer effect
  • #3
  • Nov 22, 2021 2:07pm Nov 22, 2021 2:07pm
Quoting ryuryu
I'm sure that neural networks and machine learning is the future.
I'm pretty sure institutional players and some others have been utilizing neural nets in trading for years to various extent and with varying success/utility.
Quoting ryuryu
Next Summer I'm planning to hire some coders to start that kind of project. I have some developments in this area end even some good stats but unfortunately that project was frozen by me due to financial issues.
Why? It's free...the software is, that is.
Quoting ryuryu
BTW I think it is not chaotic. It is like bird murmuring, looks like chaos but in fact there are a lot of strict rules involved.
Neural net easily extracts any patterns and can show it's not 100% chaos but it mostly is in terms of overall features. Yes, there may be local patterns, pockets of self-organization but a) not consistent over long-term, and b) small? Again, strictly in the technical sense. It's very important to dissect the technical from the fundamental.
The exploits and regularities from the fundamentals are different...you can't even quantify them well - it's not like exact numbers such as price, volume.
  • #4
  • Nov 22, 2021 2:22pm Nov 22, 2021 2:22pm
I have the answer to that million dollar question. Interesting thread. Time factor is also should be included to sample or quantify the patterns (IMHO).
  • #5
  • Nov 22, 2021 2:43pm Nov 22, 2021 2:43pm
Quoting ryuryu
Oh how interesting it. I'm sure that neural networks and machine learning is the future. Next Summer I'm planning to hire some coders to start that kind of project. I have some developments in this area end even some good stats but unfortunately that project was frozen by me due to financial issues. BTW I think it is not chaotic. It is like bird murmuring, looks like chaos but in fact there are a lot of strict rules involved. I can't post in another thread, so I can answer here. {quote} I think that may be it is better to use neural network to find...
Are you going to relate it to volume again?
  • #6
  • Nov 22, 2021 2:44pm Nov 22, 2021 2:44pm
Quoting Ihlas
I have the answer to that million dollar question.
But in the end it's still complex and not as clearcut.
Quoting Ihlas
Time factor is also should be included to sample or quantify the patterns (IMHO).
Absolutely. Data is key - and just getting good data and reliably can be a headache for average traders. And sampling is at the core for statistics - larger timeframes can filter out some noise or features...but it depends on what exactly you are analyzing too.
Life of beginning traders and even average traders is difficult - those Expert Advisors and trading environments are so constricting, information is kind of scarce and learning opportunities are meager. It's dangerous to be smart even! But everyone should switch to Linux asap and learn it because any real analysis is hard, expensive, deficient in Windows (even with their attempts to include some Linux lately). But we can't even get good forex tick data to play with so there are all sorts of barriers.
  • #7
  • Nov 22, 2021 2:54pm Nov 22, 2021 2:54pm
Quoting costos
{quote} But in the end it's still complex and not as clearcut. {quote} Absolutely. Data is key - and just getting good data and reliably can be a headache for average traders. And sampling is at the core for statistics - larger timeframes can filter out some noise or features...but it depends on what exactly you are analyzing too. Life of beginning traders and even average traders is difficult - those Expert Advisors and trading environments are so constricting, information is kind of scarce and learning opportunities are meager. It's dangerous...
The only data which is important for me is opening price of the day. Also it varies from broker to brokers.
It shouldn't be complex. But it's hidden in the obvious. I can easily see huge pips that last 3-5 weeks direction. It seems to me that daily price movement is just luring or trap.
  • #8
  • Nov 22, 2021 2:55pm Nov 22, 2021 2:55pm
Who has a fast GPU card (or a thousound of them)? Who wants to learn how to use neural network on forex data? It is doable on one older card but not all graphics cards are good for this and it takes longer.
  • #9
  • Nov 22, 2021 3:03pm Nov 22, 2021 3:03pm
Quoting Ihlas
{quote} The only data which is important for me is opening price of the day.
For neural networks it's different - you want to feed it as much data as you can - if you can - and then ideally only feed it what it needs.
Quoting Ihlas
{quote} Also it varies from broker to brokers.
If you are analyzing older data they may switch timezones, this and that, the data is misformatted, missing data points, it's unbelievable. And theeen on top of it you are completely at the mercy of your data source - and if you don't pay for your data then things can go awry when you start trading using that unreliable data...I have seen data sources even change their candles closing by a few seconds and that affected analysis. (Interestingly. There may be information passed on in how middlemen brokers close their candles).
  • #10
  • Nov 22, 2021 3:10pm Nov 22, 2021 3:10pm
Quoting costos
{quote}For neural networks it's different - you want to feed it as much data as you can - if you can - and then ideally only feed it what it needs. {quote}If you are analyzing older data they may switch timezones, this and that, the data is misformatted, missing data points, it's unbelievable. And theeen on top of it you are completely at the mercy of your data source - and if you don't pay for your data then things can go awry when you start trading using that unreliable data...I have seen data sources even change their candles closing by a few...
Where can we get reliable forex data? Opening, close, etc. I started doubt which source is reliable.
  • #11
  • Nov 22, 2021 8:09pm Nov 22, 2021 8:09pm
These ideas seem to be far beyond simple MT5 level backtests and such. I thought MT5 tick data was enough but now I'm not so sure. Seems to be the best accurate fit for testing EAs but once again this project seems beyond that.
You gotta boop the snoot
Learning to scalp NAS100 Part 2 All Time Return: 0.5%
  • #12
  • Nov 23, 2021 1:19am Nov 23, 2021 1:19am
Quoting Ihlas
{quote} Are you going to relate it to volume again?
I think there are several measurements I can exploit:

1. Time
2. Speed (volume)
3. Size (volume)
4. Imbalance (delta)
5. Distance (angle)
6. Price itself (like a starting point of part of calculations)

So yes, volume is a must have data anyway

Quoting Ihlas
{quote} Where can we get reliable forex data? Opening, close, etc. I started doubt which source is reliable.
Forex open/close is reliable. But forex has no volume data, so you have to take it from another source. I'm taking it from CME.

Quoting costos
Who has a fast GPU card (or a thousound of them)? Who wants to learn how to use neural network on forex data? It is doable on one older card but not all graphics cards are good for this and it takes longer.
I have 1080ti card but only one
Observer effect
  • #13
  • Nov 23, 2021 2:06am Nov 23, 2021 2:06am
Quoting Ihlas
{quote} The only data which is important for me is opening price of the day. Also it varies from broker to brokers. It shouldn't be complex. But it's hidden in the obvious. I can easily see huge pips that last 3-5 weeks direction. It seems to me that daily price movement is just luring or trap.
Absolutely. Agree 100%.
  • #14
  • Nov 23, 2021 2:26am Nov 23, 2021 2:26am
Quoting HimmerJimmer
I thought MT5 tick data was enough but now I'm not so sure.
That's really one issue. You gotta be sure about the data you are using, you can't assume anything. That tick data might be fine but you should always double check.
Otherwise - garbage in, garbage out.
  • #15
  • Nov 23, 2021 2:31am Nov 23, 2021 2:31am
Quoting ryuryu
{quote}I have 1080ti card but only one
That's a good card. You can do lots of machine learning with GPU computing on forex with that.
  • #16
  • Nov 23, 2021 3:08am Nov 23, 2021 3:08am
Quoting ryuryu
{quote} I think there are several measurements I can exploit: 1. Time 2. Speed (volume) 3. Size (volume) 4. Imbalance (delta) 5. Distance (angle) 6. Price itself (like a starting point of part of calculations) So yes, volume is a must have data anyway {quote} Forex open/close is reliable. But forex has no volume data, so you have to take it from another source. I'm taking it from CME. {quote} I have 1080ti card but only one
But different brokers have different open/close. So which one should be reliable. At least there are 1 pip difference in the daily opening price.
Volume is the important part, but we can't get the all the forex volume data that could be exploited to profit big. CME gives the piece of the game, I don't think it can be enough for longer term predictions. Short term directions, I see it can be exploited with the volume from CME. Good to see you're doing great.
I'm ready to learn anything useful, because sometime in the future everything can stop working if we don't understand how it works in reality. On alert just in case
  • #17
  • Nov 23, 2021 3:14am Nov 23, 2021 3:14am
Quoting acetrader
{quote} Absolutely. Agree 100%.
Have you found something from Crucial Point's thread?
  • #18
  • Nov 23, 2021 3:21am Nov 23, 2021 3:21am
Quoting Ihlas
{quote} But different brokers have different open/close. So which one should be reliable. At least there are 1 pip difference in the daily opening price. Volume is the important part, but we can't get the all the forex volume data that could be exploited to profit big. CME gives the piece of the game, I don't think it can be enough for longer term predictions. Short term directions, I see it can be exploited with the volume from CME. Good to see you're doing great. I'm ready to learn anything useful, because sometime in the future everything can...
It is not so important. For example I was trading on one exchange using data from another for years. And now I'm using CME data but trade on forex.
Also I have found that daily open price is not good. Asian middle is much more accurate.
Regarding to long-term it is better to ask some who trade it. I'm trading only short-term and use volume/delta. With long-term on higher tf it is impossible to use big lot sizes. The DD will be dangerous. May be on higher tfs the imbalance also can be found, I just don't know (but why not, you can check it).
Observer effect
  • #19
  • Nov 23, 2021 3:27am Nov 23, 2021 3:27am
https://www.kaggle.com/c/g-research-...cussion/287528
  • #20
  • Nov 23, 2021 3:41am Nov 23, 2021 3:41am
Quoting ryuryu
{quote} It is not so important. For example I was trading on one exchange using data from another for years. And now I'm using CME data but trade on forex. Also I have found that daily open price is not good. Asian middle is much more accurate. Regarding to long-term it is better to ask some who trade it. I'm trading only short-term and use volume/delta. With long-term on higher tf it is impossible to use big lot sizes. The DD will be dangerous. May be on higher tfs the imbalance also can be found, I just don't know (but why not, you can check it)....
Yeah, I see your point. To me 300 pips targets are more reliable than 30 pips target. Good luck.
0 traders viewing now