Artificial neural networks. - page 4

New comment
 
alsu: But it is not certain that it is convenient to output the information in such form, maybe it would be easier to give us not more roughly 1.5 bits (buy-sell-stop), but for example 10 bits of information?

I was also thinking about it, in principle for indicator-based TS it would be enough to use NS as a predictor not of further price movement, but of a series of losses, i.e. to predict when indicators are trading at a profit and when they are at a loss

 
IgorM:

I was also thinking about it, in principle for indicator-based TS it would be enough to use NS as a predictor not of further price movement, but of a series of losses, i.e. to predict when indicators are trading at a profit and when they are at a loss

by the way yes, it's a good idea for research
 

I would like to enliven the discussion and ask everyone involved to answer the following question: What do you think neural networks model in the market:

1. Hidden non-linear relationship between prices, or

2. The work of the trader's brain in making trading decisions based on incoming data.

 

gpwr:

Hidden non-linear relationship between prices, or

Is it so non-linear? At times of strong moves that literally hit the liquidity limit, yes, I would probably agree. At other times, it is probably very linear, so it seems to me.
 
I can suggest another direction for meshes - the selection of parameters for a dynamic process model. You can even do it in real time.
 
NS do not use any formulas, so they find all dependencies both linear and non-linear!
 
gpwr:

I would like to enliven the discussion and ask everyone involved to answer the following question: What do you think neural networks model in the market:

1. Hidden non-linear relationship between prices, or

2. The work of the trader's brain to make a trading decision based on incoming data.

The first is more likely. NS operates with numbers. And the trader with: (1) form, (2) real-time dynamics (speed of movement, jerks, crawls), (3) memory and associations, (4) emotions, moods, ideas, (5) "counterext". And this is already called mindfulness.
 
I understand that neural networks can be used with any standard strategy from indicators to martin. I am wondering what the results will be when combining neuronics with pair trading (spread trading) or volatility trading. Does anyone have any experience on this subject?
 

What matters is what we feed into the neural network inputs. These should be significant factors that influence the price.

Well, if we make Mars weather as inputs and take EUR/USD quotes as outputs and train the neural network on them, then with a certain number of neurons and layers the neural network will find dependence.

But it will have nothing to do with reality.

As I was taught (a long time ago =) ) there are 3 types of inputs:

1. Inputs of which we know they exist, but cannot obtain or measure them in any way. In our case it is noise (interference).

2. Inputs that we can receive (including delayed) and measure.

3. The outputs of the system at previous moments in time. (If the system is dynamic, it depends on its previous states).

Imho:

1. the first type of inputs - cataclysms, terrorist acts, Central Bank interventions, actions of big players. We do not find out about them at all or we find out after the fact. We are not interested in them.

The second type - macroeconomic indicators (GDP, refinancing rate, etc.).

3. The price at time t depends on the prices at time t-1, t-2, ... t-n.

Amateurs of thechanalysis consider only type 3 inputs. Trend-followers believe that if the price moved in one direction, it will still move in the same direction. And how significant are inputs of this type? I think the price hardly depends on them by more than 5%.

P.S. If anyone here has tried to inputs of the 2nd type - macroeconomic indicators, please share the results. =)

 
I too have come to the conclusion that it's worth delving into neural networks, although it seems a bit complicated.
1234567891011...14
New comment
Reason: