Cognitive indicators and neural-networks | Forex Factory

Options

Permalinks
Print Thread

Similar Threads

Any success with neural networks in trading 29 replies

Neural Networks: Myths And Reality 19 replies

Extreme scalping using neural networks - EA/Manual 646 replies

Neural Networks in Trading (Java, Algodeal, Technical Analysis) 3 replies

Price Analysis with Neural Networks 41 replies

Cognitive indicators and neural-networks

  • #1
  • First Post: Jun 12, 2015 4:57am Jun 12, 2015 4:57am
Hi everyone.

​I've joined this community in 2009, though I hardly post on here. Just been a lurker for the past couple of years while agonizingly pursuing a PhD in cognitive sciences in the flat land of the Netherlands

​I wanted to start a thread concerning non-traditional indicators. By that, I kind of have an idea for two types of indicators.

Cognitive: indicators that do not have assumed interpretation of outcome, but rather offer an certain view of the world for the trader to interpret. I realize that traders often rely on several indicators to determine buy/sell decisions, but what about indicators that offer a certain composite view that provides the trader with a new perspective? Of course you could argue that many existing indicator already fit in that category, but what I'm interested in knowing is whether indicators exist for that sole purpose and not meant to be a part of an EA.

Neural: indicators that rely on a neural network output to provide, just like above, that kind of composite perspective. Neural networks provide forecasts with a varying degree of error, but taking forecasts and errors together, certain market views can be formed.

My aim in this thread is to create a discussion amongst beginner and experienced traders alike, regarding the manner in which they prefer to trade, to see if an indicator of this kind is worth developing. My goal is to develop tools that assist traders in understanding the market on a cognitive level and within a perspective they feel comfortable with. I believe that such indicators can be of very high value when combined with traditional EA's.

I hope this thread will generate the anticipated discussion!
  • #2
  • Jun 12, 2015 9:09am Jun 12, 2015 9:09am
i'll keep an eye on this thread, even tho i tried neural networks in the past, form the most basic to the most complex that runs with Matlab on GPUs.
they all failed.

to have a valid forecast you first need to know what informations matter (to feed your NN), and that's where the problem is.

if you only take into account technical analysis (previous quotes data) no matter how curate they are, you'll be missing a big part of what makes the price move.
same think for fundamentals, you could always input the big news releases, but how to input political speeches, public/trader's emotions,..... the list is long.
i lose money for a living
  • #3
  • Jun 13, 2015 7:03am Jun 13, 2015 7:03am
Thank you for replying Tyler.

When you say failed, can you clarify what you define by that? Was the prediction error too high for example?
I've tried to take the NN output (+error) and make it a part of an EA. Since I (just like you did) only inputed historical price data, of course news and other fundamentals were not accounted for. My idea is to make use of that NN output as an assistive tools to situate the trader in a more informed perspective when combined with fundamentals.

BTW, I'm getting the feeling that this sub-forum isn't the best place for this thread. Admin, can you move this thread to "trading discussion"?
  • #4
  • Jun 14, 2015 4:37pm Jun 14, 2015 4:37pm
Quoting x1810
Thank you for replying Tyler. When you say failed, can you clarify what you define by that? Was the prediction error too high for example? I've tried to take the NN output (+error) and make it a part of an EA. Since I (just like you did) only inputed historical price data, of course news and other fundamentals were not accounted for. My idea is to make use of that NN output as an assistive tools to situate the trader in a more informed perspective when combined with fundamentals. BTW, I'm getting the feeling that this sub-forum isn't the best place...
you're welcome, mate.
and yes, the error was too high.

you can't ignore the fundamentals and expect that historical pricees alone will give you a decent enought prediction, because it won't.
from my experience, fundamentals are way more improtant than technicals, if you do you're training on technicals only, you'll be missing most of the data.


i'm just saying that if you want to have a decent forecast, you need to have a decent input.
the thing is, the amount of data that have an impact on a currency value is insane, i just dropped the ball.
it may work on stock, trying to forecast the value of a corporation, you may get a decent prediction.
i lose money for a living
  • #5
  • Jun 14, 2015 5:11pm Jun 14, 2015 5:11pm
Hi ,

I have some similar projects in my mind.Generally traditional indicators are either noisy or smoothed but they lag.I want to combine predictive indicators with lagged indicators. I created a static indicator algorithm with high prediction rate with machine learning algo.What I want to do is feed the self trainning neural network algo(preferably recurrent nn type) and random forest with this formula.Then combine it with the lagged indicator.If you are interested to cooperate in private I am ready.I have created good trading systems with laser-beam accurate entry exits with machine learning algos.I have very limited coding skills but I am a good trader
  • #6
  • Jun 14, 2015 5:13pm Jun 14, 2015 5:13pm
join to learn
we don't predict, we only ride
  • #7
  • Jun 15, 2015 3:43am Jun 15, 2015 3:43am
Quoting olsen-yersen
Hi , I have some similar projects in my mind.Generally traditional indicators are either noisy or smoothed but they lag.I want to combine predictive indicators with lagged indicators. I created a static indicator algorithm with high prediction rate with machine learning algo.What I want to do is feed the self trainning neural network algo(preferably recurrent nn type) and random forest with this formula.Then combine it with the lagged indicator.If you are interested to cooperate in private I am ready.I have created good trading systems with laser-beam...
Hi Olsen-Yersen,

Yes the general idea of this thread is to create real chances of collaboration. Can you tell me a little more about the predictive indicator? As far as I know, no real technical indicator is non-lagging, since they all use price data. But you say you used machine learning? It would be great if you can explain a little bit what you mean by that, without spilling out any secret recipes if there were any

We can communicate in private, but since this thread is at its beginning, it would be nice to see what the rest of the community has to offer. We can potentially widen the collaboration that way.

Admins, another quick request to move this thread to "trading discussion". I think it would fit better over there-- I cannot move it myself.
  • #8
  • Jun 15, 2015 7:42am Jun 15, 2015 7:42am
Quote
Cognitive: indicators that do not have assumed interpretation of outcome, but rather offer an certain view of the world for the trader
to interpret. [....] a certain composite view
Could you please elaborate on this a bit more? Do you mean an indicator which gives a vector of N characteristics of the price action instead of N indicators each in charge of estimating one of them?
No greed. No fear. Just maths.
  • #9
  • Jun 15, 2015 11:46am Jun 15, 2015 11:46am
I have no idea what you mean by saying cognitive indicator, but I can give you a few references to the previous works and attempts to create indicators and EAs based on Neural Networks with poor performance (if we do not want to say with no sucess). The previous attemps show that the best designed NN for trading cannot perform better than trading based on a simple moving average. Nevertheless, I am looking forward to having more detail about your approach on both the cognitive and NN based indicators for trading.
Trading is the hardest way to make easy money...
  • #10
  • Jun 15, 2015 2:17pm Jun 15, 2015 2:17pm
Quoting MathTrader7
The previous attemps show that the best designed NN for trading cannot perform better than trading based on a simple moving average.
MathTrader7,

I disagree with you on that.
  • #11
  • Jun 15, 2015 2:38pm Jun 15, 2015 2:38pm
Quoting x1810
{quote} Hi Olsen-Yersen, Yes the general idea of this thread is to create real chances of collaboration. Can you tell me a little more about the predictive indicator? As far as I know, no real technical indicator is non-lagging, since they all use price data.
Hi x1810,

I tried to predict if not hundreds of yet, tens of filters like different types of kalman filter ,Holt-Winters filter ,Hull, FIR,jurik,polynomial fitted moving avrage etc.. and indicators, only couple o them worked very well.
  • #12
  • Jun 15, 2015 3:00pm Jun 15, 2015 3:00pm
Quoting olsen-yersen
{quote} MathTrader7, I disagree with you on that.
The important words are ....based on.
I use a single ma, not more. I also trying to get benefits from NN, not better entries or exits.


Makes simple things complicated.

NN
  • #13
  • Jun 16, 2015 1:50am Jun 16, 2015 1:50am
Quoting PipMeUp
Could you please elaborate on this a bit more? Do you mean an indicator which gives a vector of N characteristics of the price action instead of N indicators each in charge of estimating one of them?
Quoting MathTrader7
I have no idea what you mean by saying cognitive indicator
By cognitive, I mean an indicator that would give you a "larger" view of the market, or a certain pair. A "helicopter view", if you prefer that term. It could be a vector of several characteristics, but the key idea is how it would be presented to the trader.

It is an idea, and I don't know what exactly could exactly fit in there, that is why I started this thread. If you are a trader, and you wanted a one certain visual indicator on your chart (or on a side bar) that will allow you to interpret the market your own way while you're about to make a buy/sell decision, what would you have in such an indicator?

Quoting MathTrader7
...but I can give you a few references to the previous works and attempts to create indicators and EAs based on Neural Networks with poor performance (if we do not want to say with no sucess). The previous attemps show that the best designed NN for trading cannot perform better than trading based on a simple moving average. Nevertheless, I am looking forward to having more detail about your approach on both the cognitive and NN based indicators for trading.
This statement is too general for me to agree or disagree with. Even if you have only one SMA in your trading system, the results will still vary according to your TP/SL levels, money management, and some randomness.

The idea that I have is that NNs are not being used right. Instead of creating a robotic trading algorithm based on NN forecasts, why not use these forecasts to create a market-view perspective that complements a trader's other EAs/tools?
  • #14
  • Jun 16, 2015 4:10am Jun 16, 2015 4:10am
It's an interesting thread but I'm struggling with the concept and definition of a NN "forecast". Apologies in advance for the kind of blunt example but in everyday terms, the word "forecast" as applied to let's say, the weather, would be used to detail the possible outcome of a certain set of prevailing circumstances; low pressure over high ground = potential rain.

If one were to then look at the most basic "algorithm" of an umbrella, it would read something like; rain = umbrella up.

The reason I cite these examples is that it seems that the "forecast" and "algorithm" element of both systems are inextricably linked; don't they just make up one system?

I may be jumping the gun here but if one applies this to a currency trade forecasting NN then isn't that NN (or the person programming it) going to need access to a profitable trading method in order that the data it uses to compile its forecast has any meaning?

You've asked the following question:

If you are a trader, and you wanted a one certain visual indicator on your chart (or on a side bar) that will allow you to interpret the market your own way while you're about to make a buy/sell decision, what would you have in such an indicator?

I'd be interested to hear your own answer to this question.

Cheers
  • #15
  • Jun 17, 2015 4:30am Jun 17, 2015 4:30am
Quoting simond2002
It's an interesting thread but I'm struggling with the concept and definition of a NN "forecast". Apologies in advance for the kind of blunt example but in everyday terms, the word "forecast" as applied to let's say, the weather, would be used to detail the possible outcome of a certain set of prevailing circumstances; low pressure over high ground = potential rain. If one were to then look at the most basic "algorithm" of an umbrella, it would read something like; rain = umbrella up. The reason I cite these examples is that it seems that the "forecast"...
I can say (at least for myself) that what I consider by a forecast is when my NN, already trained using historical data, outputs the expected closing price for a certain period (1D, 1W, 4H, etc.). That forecast is usually a part of the trading algorithm or EA that traders use.

I believe that most traders here who in the past used NNs will agree with me. I would be very interested if a trader used them differently (I did so once, but that was purely experimental stuff).

Quoting simond2002
You've asked the following question: If you are a trader, and you wanted a one certain visual indicator on your chart (or on a side bar) that will allow you to interpret the market your own way while you're about to make a buy/sell decision, what would you have in such an indicator? I'd be interested to hear your own answer to this question. Cheers
Well, like I said earlier I'm not sure exactly what to put in there, that's why I started this thread. But if you really want to put me to the test, I'd say a colored visualization of where the price is expected to hover in the following day, week, etc. Sort of like you use a weather forecast to decide when exactly to have a BBQ and when to wrap it up
  • #16
  • Jun 19, 2015 4:11am Jun 19, 2015 4:11am
I'm gonna use my get-out-of jail card and rescue this thread from the dark realms of page 2
Hopefully a few others have more thoughts on the subject?
  • #17
  • Jun 30, 2015 7:22pm Jun 30, 2015 7:22pm
i tried it some time ago, im using matlab and yes im still new and need learning lol. my neural network perform really badly. i agree with tylerbose that input data is most important thing to make neural network, and from my experienced previous data will not enough to make a good neural network (im using previous data candle when no high impact news released) . so the most question is , what input should neural network have? since option is so many i think it will be a long way to know the answer
  • #18
  • Aug 7, 2015 5:49pm Aug 7, 2015 5:49pm
Particle filter is a cognitive indicator ?
  • #19
  • Aug 4, 2016 8:07pm Aug 4, 2016 8:07pm
You're a psychologist ? I can predict if the market will be down or up only watching it, i think, I don't know exactly why
  • #20
  • Last Post: Aug 4, 2016 8:12pm Aug 4, 2016 8:12pm
If you are a professional Psychologist, we can talk, but not here, not matter if is a private message,It is a very complex and difficult thing, talking practically
1 trader viewing now