Neural network trading | Page 2 | 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

  • #21
  • Nov 23, 2021 9:58am Nov 23, 2021 9:58am
Quoting nathaniel
https://www.kaggle.com/c/g-research-...cussion/287528
As you can see: a) kind of porous research, nothing too meaty publicly, and b) nobody reporting great prediction results...just improvements blah blah blah.
It's like with true Artificial Intelligence. It is very easy to create but it's never gonna happen in this current state of global affairs because true AI would instantaneously expose a tonload of really bad things, point at a ton of bad people for egregious wrongdoing against humanity on the planetary scale and of course solve all problems and empower people who actually care and have morals and ethics maybe. True AI is death to many people in control. Same with forex and trading. Can't expose major economic shams. Can't have people too smart in trading or finances or they will figure out the really bad stuff about this and other stuff, too.
Here's the neural network analysis/strategy: https://www.forexfactory.com/thread/...twork-strategy
  • #22
  • Nov 23, 2021 10:04am Nov 23, 2021 10:04am
Quoting costos
{quote} 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.
Well the data is imported directly from the broker themselves, so for EA testing it's much more accurate because it simulates the exact spreads and it's changes throughout the day, swaps, commissions, and even slippage. I trusted it enough to go live with similar results so far.
You gotta boop the snoot
Learning to scalp NAS100 Part 2 All Time Return: 0.5%
  • #23
  • Nov 23, 2021 10:07am Nov 23, 2021 10:07am
Quoting HimmerJimmer
{quote} Well the data is imported directly from the broker themselves, so for EA testing it's much more accurate because it simulates the exact spreads and it's changes throughout the day, swaps, commissions, and even slippage. I trusted it enough to go live with similar results so far.
Where do you think brokers get their data? That's what I was referring to.
  • #24
  • Nov 23, 2021 10:50am Nov 23, 2021 10:50am
Quoting costos
{quote} Where do you think brokers get their data? That's what I was referring to.
Ah. From what I know, they get it from their respective liquidity providers and provide the best bid and ask prices from a group of providers.. correct?
You gotta boop the snoot
Learning to scalp NAS100 Part 2 All Time Return: 0.5%
  • #25
  • Nov 23, 2021 11:09am Nov 23, 2021 11:09am
Quoting HimmerJimmer
{quote} Ah. From what I know, they get it from their respective liquidity providers and provide the best bid and ask prices from a group of providers.. correct?
Well, the fact that you don't know for sure and transparently in detail is what part of my point is.
  • #26
  • Nov 23, 2021 11:19am Nov 23, 2021 11:19am
Quoting costos
{quote} Well, the fact that you don't know for sure and transparently in detail is what part of my point is.
Fair enough. I see what you're saying.
You gotta boop the snoot
Learning to scalp NAS100 Part 2 All Time Return: 0.5%
  • #27
  • Edited 3:11pm Nov 23, 2021 1:47pm | Edited 3:11pm
Not a fan of neural nets. They are very expensive to train (I already see GPUs mentioned in this thread), very picky how you feed it data (must be normalized, no redundant inputs, etc.) and impossible to understand afterwards (just a huge bunch of numbers).

I have moderate success using Random Forests. The advantage is that RFs will just select whatever signal it finds most useful. That means I just throw thousands of signals at it, and it will figure out which onces have predictive value and which don't. This is a huge advantage. I just try any indicator I can get my hands on (thanks @mladen ;-)). After training, RFs can also report back the significance of each signal which helps me get rid of useless indicators or signals.

RFs are "quick" to train (compared to neural nets). I am using a 6 year old Macbook, no servers, no GPUs. Below is an example of something I am working on. EURUSD gives 36.61% return per year (when risk is scaled using Monte Carlo to 5% chance on max. 25% drawdown). Note that I don't use the last year on purpose (see here for my approach on backtesting and why I save the last year for validation just before going live).

Attached Image (click to enlarge)
Click to Enlarge Name: Screenshot 2021-11-23 at 22.07.52.png Size: 60 KB
  • #28
  • Nov 23, 2021 3:36pm Nov 23, 2021 3:36pm
Quoting Ihlas
{quote} Have you found something from Crucial Point's thread?
Ihlas,

I did find something & YOU did too.....now what to do with it is the question. The previous close is important. Read everything Crucial Point shared as I did & there is something else there. I printed everything CP shared that I could find & after many readings put something together. What I found CP never really mentioned, but
his postings put me on to something else.

CP used some type of MA's as shown in some of his postings. A small channel of some sort on the 1M time frame, but not sure. He was a scalper from his postings. Take a look at a 55 EMA High & 55 EMA Low on 1M. It seems similar.

What I use needs no indictors such as MA's, RSI, Stochastic or MACD. It took many readings of CP's post for me to discover this & it is NOT a HOLY GRAIL, but works
enough that I'm satisfied trading it. Works equally as well in the futures market.

I know this is off thread topic, but you ask & not wanting to ignore you I answered. Hope this helps.
  • #29
  • Nov 23, 2021 9:58pm Nov 23, 2021 9:58pm
Quoting acetrader
{quote} Ihlas, I did find something & YOU did too.....now what to do with it is the question. The previous close is important. Read everything Crucial Point shared as I did & there is something else there. I printed everything CP shared that I could find & after many readings put something together. What I found CP never really mentioned, but his postings put me on to something else. CP used some type of MA's as shown in some of his postings. A small channel of some sort on the 1M time frame, but not sure. He was a scalper from his postings....
Thannks Acetrader,
I will have a look again, maybe I can add to my Arsenal something more. Thanks again and good luck.
  • #30
  • Nov 24, 2021 12:28am Nov 24, 2021 12:28am
Quoting yoriz
That means I just throw thousands of signals at it, and it will figure out which onces have predictive value and which don't. {image}
I sense problems with this approach. You will know predictive values historically but you will never know what those values will be one step in the future.
  • #31
  • Edited 4:50am Nov 24, 2021 2:34am | Edited 4:50am
Quoting costos
I sense problems with this approach. You will know predictive values historically but you will never know what those values will be one step in the future.
Yes, I was worried about that too. I am using a walk forward approach and update my model after each H1 bar (inspired by this excellent thread of @algoTraderJo). I have seen that the significance of the provided inputs remains relatively stable. There is the occasional outlier when suddently another input seems very relevant, but the top-x indicators remains stable.

Obviously I don't literally use thousands of indicators simultaneously, but I have experimented with many, had them in my pool of inputs for a few weeks while experimenting, and most often decided to stop using it if it didn't bring much. Over time, I filtered out a nice list of indicators and other signals that appear to have some sort of predictive value.
  • #32
  • Nov 24, 2021 3:17am Nov 24, 2021 3:17am
Quoting acetrader
{quote} Ihlas, I did find something & YOU did too.....now what to do with it is the question. The previous close is important. Read everything Crucial Point shared as I did & there is something else there. I printed everything CP shared that I could find & after many readings put something together. What I found CP never really mentioned, but his postings put me on to something else. CP used some type of MA's as shown in some of his postings. A small channel of some sort on the 1M time frame, but not sure. He was a scalper from his postings....
I searched too, read all CP posts not once, but didn't find anything reliable longterm. I think my 2D thinking is a problem and want a backtestable reliable strategy. On a positive side, with my knowledge, there is no way, i will trade real acc with real money.
Does your system backtestable? Do you have more than 1 exit strategy?
Thanks, IenDzi.
  • #33
  • Edited 6:59am Nov 24, 2021 6:28am | Edited 6:59am
Quoting ryuryu
{quote} Guys I have one question.
What if use NN not to predict next candle close or new day close.
For example I have very good entry point. They are perfect most time. But there are let's say 10% chance that it will be not good.
So the task is: to find best possible approach to average the entry, or reverse it.
Сan this be done using a neural network?
I'm not 100% sure I understand the question exactly. Using neural networks you could show it every good entry point and see if network's prediction/classification for the good entry point is better using whatever data you use to find your very good entry points.
  • #34
  • Nov 24, 2021 10:42am Nov 24, 2021 10:42am
Quoting Ihlas
{quote} Thanks Acetrader, I will have a look again, maybe I can add to my Arsenal something more. Thanks again and good luck.
Your most welcome by friend. Take a look at EMA 200, Median Price HL/2 & EMA 3 Median Price HL/2 on a 1Min & see what you think. Look at the crosses & the retraces attempt to the EMA 200. Seems similar to what CP posted.

Best of trades to you my friend.
  • #35
  • Last Post: Nov 24, 2021 1:46pm Nov 24, 2021 1:46pm
Quoting acetrader
{quote} Your most welcome by friend. Take a look at EMA 200, Median Price HL/2 & EMA 3 Median Price HL/2 on a 1Min & see what you think. Look at the crosses & the retraces attempt to the EMA 200. Seems similar to what CP posted. Best of trades to you my friend.
Sure, I will try it. Thanks again. Wish you health and wealth
0 traders viewing now