Brain-training tasks related to trading in one way or another. Theorist, game theory, etc. - page 14

New comment
 

If the topic is still "alive", I'd like to think collectively about it:

there are three variables: X, Y, Z

there are ratios of variables X/Y and Z/Y, and the order is different X/Y > Z/Y by ~1000 times, i.e. three orders of magnitude older

there is a step change of these variables it is constant and equals delta = 0.01

i.e. to find how many steps Y has changed, i.e. n*delta from the initial value Y0, other variables are not important, but they also change

Googled approximate methods of calculation, I do not remember the mathematics, now I lean towards the derivative, it is simpler, because by definition the derivative is

f(Y)` = f(Y0+delta)

But then a double question arises: may we look for the derivative from the product of X/Y and Z/Y --> (X/Y) * (Z/Y), we get Y squared

all in all, i would like to know the answer to this

thanks

 

Here's another problem: we have a TS at the intersection of two MAs with periods of 5 and 10, for example.

Let's assume that the MA5 is perfectly predicted 2 bars ahead.

We obtain an attractive result.

Question: how many bars should we forecast the time series itself to obtain the value of MA(x) two bars ahead?

P.S. I do not intend to predict BP, the question is purely speculative.

 
Swetten:

Here's another problem: we have a TS at the intersection of two MAs with periods of 5 and 10, for example.

Let's assume that the MA5 is perfectly predicted 2 bars ahead.

We obtain an attractive result.

Question: how many bars should we forecast the time series itself to obtain the value of MA(x) two bars ahead?

P.S. I do not want to forecast BP, the question is merely speculative.


Any MA correctly gives the first value only after it has passed at least one of its period, i.e. we have МА10 on Н1, it means that in 10 hours the first value of МА10 will appear, for МА5 = 5 hours

and taking into account the fact that there is no final price value on the unclosed bar, it means that we need one more bar

But if we have only fast МА, then the time frame will still be limited by the slow МА, i.e. we still need 11 bars to make a decision, because it is not certain that the fast МА will be in the needed position over/under the slow one after 3 bars

 
Swetten:

Question: how many bars do I need to predict the time series itself in order to get the MA(x) value two bars ahead?

Captain Hindsight states that in order to predict the MA by X bars ahead, you have to predict the price exactly X bars ahead))
 

Igor, your problem lacks conditionality: there are three variables and, sorry, only two equations: { d(X/Y) = delta ; d(Z/Y) = delta }

And differentials don't help - they don't reduce the number of variables. In general, if we don't know at least an approximate value of variable X or Z or some combination of them, the problem has an infinite number of solutions.

 
alsu:

Igor, your problem lacks conditionality: there are three variables and, sorry, only two equations: { d(X/Y) = delta ; d(Z/Y) = delta }

And differentials don't help - they don't reduce the number of variables. In general, if we do not know at least an approximate value of X or Z or some combination of them, the problem has infinitely many solutions.


but the problem is for trading and solution is needed only as a mathematical model for approximate calculation, note only the increment from initial value is of interest

Well, the initial value is a constant and will not change - if we had a system of three unknowns with a system of three equations - no problem

I don't mind introducing constants for the initial relations X/Y=const and Z/Y=const, I wanted to speculate about the variants of searching for Y+Y0, namely, Y0 discrete with delta step

 
IgorM:


but the problem is for trading and the solution is needed only as a mathematical model for approximate calculation, note only the increment from initial value

Well, the initial value is a constant and will not change, there would be a system of three unknowns with a system of three equations - no problem

I don't mind introducing constants for initial ratios X/Y=const and Z/Y=const. I would like to think about search options Y+Y0, namely, Y0 discrete with delta step

You won't get anything. If ( X, Y, Z) is a solution, then for any k<>0, the solution is ( k*X, k*Y, k*Z)
 
Mislaid:
This will not work. If ( X, Y, Z) is a solution, then for any k<>0, the solution is ( k*X, k*Y, k*Z)


Thanks, but what if I use a derivative? Here's the first thing I looked up: http: //ftoe.ru/list9/int65a.html

and genetic algorithms also seem to be able to show with a certain margin of error

And again, you don't need XYZ, but an offset from the start/start value, and the offset is only for one Y-value

 
alsu:
Captain Hindsight argues that to predict the MA exactly X bars ahead, you have to predict the price exactly X bars ahead.)

On the one hand, yes.

On the other -- if the MA(10) is fed 1 bar forward, after all, 1 bar is only 10% of its calculation.

Or am I in the wrong place to be afraid?

 
alsu:
Captain Hindsight says that in order to predict the swing forward by X bars, you have to predict the price exactly X bars forward))


mashka is better predicted because for mashka period i: MA(0)=MA(1)+(X0-X(i+1))/i. And if the best predictor of the time series itself is its last value, then for the mask, the last value plus the difference between the last known value of the series and the i-th value (retired when calculating for the future value of the mask) divided by the period of the mask.

The question is what counts as a prediction.

1...789101112131415161718192021...23
New comment