Menu
Sister Sites

Options

Bookmark Thread

First Page First Unread Last Page Last Post

Print Thread

Similar Threads

Neural Network Trading, Serious people only! 127 replies

Neural Network trading using Matlab and Metatrader 8 replies

neural network based systems in forex 22 replies

Neural Network anyone? 12 replies

What trains the neural network? 2 replies

Attachments: The "Better" neural network
Exit Attachments

The "Better" neural network

  • Post #1
  • Quote
  • First Post: Nov 27, 2007 5:22am Nov 27, 2007 5:22am
Well I can't really make an opinion about neural networks w/ regard to forex, but these results are pretty amazing:

http://championship.mql4.com/2007/users/Better/reports

900% in 2 months. Sure, it was high risk. Sure, it was luck. But how much of it was luck???

The equity curve is very smooth..... and he lost 35% or so of his trades...... so is NN the dominant predictor available to us?


The problem is, you can't just write a neural network overnight..... it requires knowledge and specialization, which few of us with ever have the time or interest to pursue. But it makes one wonder, doesn't it?
Okay, I downloaded some free NN software, and am training it with a CSV created by MQ4.

Any takers on working on a NN project?
Hi tdion, could you possibly direct has in a direction where we can read up on this a bit. It sure sounds interesting
Sure....

This morning I Googled "Neural Networks How Stuff Works" and got a lot of good informative pages.
Here are three books to get you going...

Computational Intelligence; An Introduction (this thing is long and full of history but not terrible useful from a design standpoint)

Artificial Intelligence; A Modern Approach (this was okay, lays the ground work for several types of systems, NN GP EC etc)

Practical Nueral Network Recipies in C++ (I'm not a coder but worked with a fellow a little on one of these recipies... it was considerably slower than the cheapest application I had bought and tested... which is not why we went to the trouble of building our own system )


*****************************************

Almost all of the White Papers you find will basically tell you that long term success with NN as a forecasting tool in investmets doesn't work over the long haul. I've never tried to duplicate their research or hit it from a different angel so I can't say for certain. What I did notice with the testing that I did was, the applications I used were simply not powerful enough to produce a meaning ful forecast in a reasonable period of time. Forecasting got better with addintional candadite predictors but computational overhead began to exceed my computers ability... big words for the file was just to damn big.

Don't let that discourage you from your work though; you may find a mix of inputs that was better than mine and you may have an improved nn alogorithm that works better than mine did...

Good luck, I'll be interested in knowing how it turns out.

Oh, and on the 900% return; the trader probably hasn't let this ride through a major market shift yet... Algorithmic trading is basically extremely complex curve fitting and it will fall apart during market transitions; if proper money managements isn't used the draw down will be extremely deep. Those returns are very likely cherry picked for a very good period or they just aren't long enough to properly show the applications weakness.

If you build a NN based system; get something together that produces resonable results and keep your MM in check... there are no shortcuts to Long Term success in the markets.

Good Luck!!!!
Trade Small. Be Patient. Learn to Survive.
Matlab is a great tool for designing systems. You can use it for NN systems as there is a NN toolbox. I haven't done any NN work with it though...
I deconstructed the trade statement for the "Better" NN model to give you an idea where its signaling in relationship with the underlying price action. Like most AI programs it tends to display a bias towards mean regression; you'll see a lot of the trades tend to be counter-trend, although I'm using trend lightly here. When the price pops or pushes out (relative to recent volatility) the system appears to be fading those moves... Being on the right side of a couple of pretty large moves has really really propped returns. A system like this, with good money management has a low correlation to trend/break-out models and can add some diversification to a traders portfolio...

Red Diamonds were shorts
Blue Triangles were Longs
If the trade was profitable it got a "+"
If it was a loss it got a "-"

I think I've wasted enough of my time today...
Attached Image
Trade Small. Be Patient. Learn to Survive.
I've said it before and I will say it again...

I think Neural Networks can be very very useful. It is all about how it is being used though...

What most people do is feed in the historical price and expect the NN to predict the next tick. That's just impossible as there is too much noise.

The NN will never be "smart" enough to do a more detailed analysis of the situation, market conditions, etc.

So in my mind the correct approach is to feed in parameters coming from indicators. As much as they are lagging they are great for filtering out noise.

Another twist is what is the NN trying to output. Instead of trying to predict the next tick or bar high, low, open, close or whatever. I prefer to simulate a trade and risk / reward output. e.g. "If I enter this trade now with 100 pips TP and 100 pips SL, how close am I going to get to each before the other one closes".

This will give you good risk/reward scenarios recognition.

And finally...

It is easy to make a system that is insanely profitable for a small period of time. What is hard is making sure the system can turn itself off and recognize whether it is not suitable for the current market conditions.


I am currently exploring making an adaptive NN system that only trades when it "knows" what it's doing.
Quoting charlinks
I've said it before and I will say it again...

I think Neural Networks can be very very useful. It is all about how it is being used though...

What most people do is feed in the historical price and expect the NN to predict the next tick. That's just impossible as there is too much noise.

The NN will never be "smart" enough to do a more detailed analysis of the situation, market conditions, etc.

So in my mind the correct approach is to feed in parameters coming from indicators. As much as they are lagging they are great for filtering out noise.

Another twist is what is the NN trying to output. Instead of trying to predict the next tick or bar high, low, open, close or whatever. I prefer to simulate a trade and risk / reward output. e.g. "If I enter this trade now with 100 pips TP and 100 pips SL, how close am I going to get to each before the other one closes".

This will give you good risk/reward scenarios recognition.

And finally...

It is easy to make a system that is insanely profitable for a small period of time. What is hard is making sure the system can turn itself off and recognize whether it is not suitable for the current market conditions.


I am currently exploring making an adaptive NN system that only trades when it "knows" what it's doing.
I don't think anyone who has used the technology for any length of time is trying to get a good fit with just price data; it only takes a couple of runs of that to realize that approach will get you nowhere...

The problem with using price based indicators (and pretty much thats all we have) is that it tends to exagerate the regressive/cyclical bias thats already present in the appraoch... It's hard to find a good set of indicators that will still be effecient when price action begins to trend; I spent two years working on a balanced dataset... I ended up coming to the same conclusion as you; there had to be a way to get the model to recognize when the probability for success was low.

Ultimately I wasn't able to do it; trying to get an algorithm to self police unless you're using a completely different set of data is very tough. The application is already under the assumption that its "right". My answer came in comparing the actual trades against specific measurable price action and pinpointing when the application was profitable and when it wasn't. I became the regulatory body that said "yes you can go out and play now" and "no you can't". You just have to adjust the filter process each time you adjust the algorithm... all of this still won't help in transitional markets (at least that was my experience).

Pretty much it comes down to knowing when to pull out which tool or when to be confident enough to simply sit on the side-lines and wait for the right conditions.
Trade Small. Be Patient. Learn to Survive.
We need more NN traders. Step right up. Calling all Quants. Hey, how about some Nobel Peace prizes like those guys from LTCM.

A little something from Trader/Author Mark Douglas:

"I know it may sound strange to many readers, but there is an inverse relationship between analysis and trading results".

But, carry on.
Quoting SunTrader
We need more NN traders. Step right up. Calling all Quants. Hey, how about some Nobel Peace prizes like those guys from LTCM.

A little something from Trader/Author Mark Douglas:

"I know it may sound strange to many readers, but there is an inverse relationship between analysis and trading results".

But, carry on.
I'm currently speachless... mom always said if you can't say something nice don't say anything; guess I'll just bite my tounge...
Trade Small. Be Patient. Learn to Survive.
Quote
The problem with using price based indicators (and pretty much thats all we have)

What about feeding in the news indicators (front page ForexFactory) somehow?
(expected v.s. real v.s. previous)
Quoting FXAnalytics
speachless... tounge...
A Oral surgeon or an English major might help that. Just kidding.

As far as the Mark Douglas quote, there is a whole paragraph further to that one line as well a couple of books he has written.

The point being, that in order to be a sucessful trader more analysis is not the critical factor. Patience, discipline, an edge, money management, positive attitude, etc etc. are, as well as little things like eating and sleeping well to be in right mental state of mind.

Time and time again when a trader fails the first reaction is that they need more market analysis. IMO learning by making this same mistake - this is not the answer.
Quoting charlinks
What about feeding in the news indicators (front page ForexFactory) somehow?
(expected v.s. real v.s. previous)

Pip... Err Charlinks;

Thats an interesting idea... I haven't as of yet tried to roll fundamental data into a dataset like that; seems like the collection of it would be a bear. It might be worth trying though.
Trade Small. Be Patient. Learn to Survive.
Are you happy now? I'm glad you regurgitate the same banter all lemmings on Forex Factory repeat OVER AND OVER to each other....

  1. "patience, discipline, edge....."
  2. "Mark Douglas (my hero) said...."
  3. "Jessie Livermore (my idol) said...."

This thread is about neural nets, not about how to be like everybody else on this forum. I've watched your negative posts for years on here (ever since you signed up) and I just had to say something.

PS) Not all people here are "traders" some of use are innovators.... And just so you know, my 100% mechanical system trades for me night and day, without me doing a thing.


Quoting SunTrader
The point being, that in order to be a sucessful trader more analysis is not the critical factor. Patience, discipline, an edge, money management, positive attitude, etc etc. are, as well as little things like eating and sleeping well to be in right mental state of mind.
Quoting SunTrader
A little something from Trader/Author Mark Douglas:

"I know it may sound strange to many readers, but there is an inverse relationship between analysis and trading results".
I don't think we must expect a book on Neural Networks from ol' Mark soon......

Here is a link that might be a nice starting point - http://cortex.snowcron.com/ , was for me anyway.
Quoting tdion
Okay, I downloaded some free NN software, and am training it with a CSV created by MQ4.

Any takers on working on a NN project?
Count me in - which software are you using?
Is there some rule that I am unaware of that we can only respond to the original author of a topic and only what they want to. Say no more. I have unsubscribed to the topic.

in your endeavor.
Quote

Oh, and on the 900% return; the trader probably hasn't let this ride through a major market shift yet... Algorithmic trading is basically extremely complex curve fitting and it will fall apart during market transitions; if proper money managements isn't used the draw down will be extremely deep. Those returns are very likely cherry picked for a very good period or they just aren't long enough to properly show the applications weakness.

So true..I being on the path of market shift, the exact time when my account was growing up and was feeling confident..people will not realize this...but one has to be in the market for couple of years at least to know how hardcore is this issue..

You will feel tectonic plates shifting right beneath your feet, as if all your years of experience is worthless when the market shifts...

I thought something was wrong with me..doing so well and then ending in a dustbin..only after reading the forums i came to realize WHAT THE FOLKS REALLY MEAN WHEN THEY SAY MARKET SHIFTS..


The Better guy has degree in physics and mathematics besides he said he has 3 years trading binary options and being trading FOREX for couple of years PLUS he can write his own code in C++ and mq4 and he is 40+

Man, that would easily put him in 0.01% of Forex traders....I don't think we have seen many independent traders with that kind of resume lately...he is already ahead of the pack with that kind of foundation...and NN just aided what he already possessed.

I think he already won the championship....and he don't even have to trade whole year to make a living and go with pain of marker shifts...

He just have to design for 3 month period and take home 50,000 dollars or BETTER..get his cousins a strip down version of his NN and get the second and third spot and pocket 80 K...man what a life
He just rounded 1000% at the end of this week... Amazing results
0 traders viewing now