If we knew exactly how the price was moving... - page 5

 
Neutron писал(а) >>

I'm talking about the distribution of incremental bribes - it's asymmetric in my example and introducing TR doesn't change anything, and that doesn't agree with your statement above.

and I didn't say it was about distributing incremental bribes or deals. It's about the distribution of price increments. Of course, on any series any system with sl<>tp will give an asymmetric distribution of results. And of course this does not mean that it can be reduced to mo+.

P.S. And of course it makes no sense to consider the distribution of increments in isolation from the choice of momentum. The average temperature across the hospital and there is no way to use it. imha. But if we consider it selectively - i.e. the distribution of increments when a certain signal or situation detection appears which can become a condition for entering a trade. Then the asymmetry of this selective distribution of increments may indicate the presence of effective output levels. But as a rule the good outputs are dynamic and they won't show themselves this way.

 
avtomat >> :

is wrong in principle here!

0<p<1 is probability

tp, sl are "kilos"

you cannot put them in the same key

Avals is right in this case. The price BP itself cannot be considered from a theorist perspective. Because it has no events and therefore no meaning. It is possible to consider only frequencies of some events (frequencies, not probabilities, because probabilities in non-stationarity are also unacceptable). And the frequencies of all considered events must be equal to 1 in total. Which at Avals is strictly observed.


Roughly speaking the event in the price series is a tick up, against the event a tick down (which is not quite true, since the price series is discontinuous due to gaps, but that's the model in a simplified way).


Since the up tick and the down tick are increments, deltas, only price differences can be considered from a theorist point of view. Correspondingly, it is also possible to consider events N pips up vs M pips down. As for any TS (market model, but not the price series), take and loss events should be considered.


And trying to calculate something using the price series instead of the series of increments (differences) is a purely botanical approach that has no applied meaning. It will result in an empty empty set of numbers without any relation to trading.


-------------------------------------------------------------------------------------------------------------------


If the price series is random and the probability of an up tick is equal to the probability of a down tick - 0.5, spreads, slippages, swaps and commissions are not taken into account, a movement can be only 1 pip up or 1 pip down, then by the theorem the probability of taking a long position is equal to p(tp) = sl / (tp + sl), where tp and sl are take and loss values in pips. While duration of a trade, i.e. average time from opening a position till the take or stop trigger will equal t = tp * sl in ticks.


We can calculate that according to Avals' expected payoff formula we will have a martingale, i.e. MO = 0

 
Reshetov >> :

Avals is right in this case. The price BP itself cannot be considered from a theorist point of view. Because it has no events and therefore no meaning. Only frequencies of some events can be considered (frequencies, not probabilities, because the probabilities under non-stationarity conditions are also unacceptable). And the frequencies of all considered events must be equal to 1 in total. Which at Avals is strictly observed.


Roughly speaking, the event in the price series is a tick up, against the event a tick down (which is not quite true, because the price series is gapped due to gaps, but that's a simplified model).


As the up tick and the down tick are increments, deltas, only price differences can be considered from a theorist perspective. Accordingly, it is also possible to consider events N pips up, versus M pips down. Or in terms of any TS (market model, not the price series itself), take and loss events respectively.


And trying to calculate something using the price series itself, instead of the series of increments (differences) - is a purely botanical approach, which has no practical value. It will result in an empty empty set of numbers without any relation to trading.

Perhaps we speak different languages.


/*

Mo=p*tp-(1-p)*sl-spread=(2p-1)*tp-spread>0, where p- probability of winning

p>spread/2tp+0.5

if for example sl=tp=10p and spread is 2p then p>0.6

and if for example sl=tp=100p then it is enough p>0.51

*/


???

 
avtomat >> :

Perhaps we speak different languages.


/*

mo=p*tp-(1-p)*sl-spread=(2p-1)*tp-spread>0, where p is the probability of winning

p>spread/2tp+0.5

if for example sl=tp=10p and spread is 2p then p>0.6

and if for example sl=tp=100p then it is enough p>0.51

*/


???

Including the spread, if the probability of a tick up or down is 0.5


p(tp) = (tp - spread) / (sl + tp)

p(sl) = (sl + spread) / (sl + tp)

p(tp) + p(sl) = 1


If the probability of a tick approaching a take is not equal to the probability of a tick approaching a stop, the formula is more complicated. But, according to the theorem, it is known that the increase of price will be approximately equal to dprice = dt * (p(tick up) - p (tick down)), where t is time in ticks. It is a perfect Bernoulli scheme, where the increase per tick can be only 1 pips up or down. If p(*) is frequencies and not probabilities, then the price increment formula is accurate.

 
Avals писал(а) >>

But for example, let the distribution be

incremental probability...

Z.U. And of course this distribution is not like a tick distribution. It is different from HP as I recall. At least it won't be the same probability at zero. But it will also be symmetrical. Of course, we can analyze it in a real series and "mow" it down to remain equal to zero, but I'm too lazy.

Yes, everything is correct, with such a distribution is possible. We could think of an even simpler example: -5 is probability 0.2; +10 is probability 0.1; 0 is probability 0.7; the rest is 0. However, what does this have to do with reality ? An example is only indicative if it explores at least to some extent meaningful conditions.

Besides, the reasoning was about bars, and your example is more about ticks. And you started talking (suddenly) about ticks. Why is that?

In the market the asymmetry is expressed in a small divergence of values of the right and left branches of the distribution. The absolute value of this divergence is only a fraction of the value of the PDF. And the PDF itself is more or less a smooth function. Under such conditions, all the benefits of asymmetry are instantly eliminated by the spread.

Not to mention that this asymmetry is even more non-stationary than the distribution itself.

 

/*

p(tp) = (tp - spread) / (sl + tp)

p(sl) = (sl + spead) / (sl + tp)

p(tp) + p(sl) = 1

*/


this is a normalisation operation by one, but it is in no way a probability, neither tp nor sl

 
avtomat >> :

/*

p(tp) = (tp - spread) / (sl + tp)

p(sl) = (sl + spead) / (sl + tp)

p(tp) + p(sl) = 1

*/


this is a normalisation operation by one, but it is by no means a probability, neither tp nor sl

You can say what you like, but these are theorized formulas from the player's ruin problem. The only thing that's been added is spread accounting. But the essence does not change.

 
Reshetov >> :

P&%%d anything you want, but these are theorist formulas from the player's ruin problem.

is the whole argument...

 
Yurixx писал(а) >>

Yes, that's right, with this distribution it is possible. You could think of an even simpler example: -5 is probability 0.2; +10 is probability 0.1; 0 is probability 0.7; the rest is 0. However, what does this have to do with reality ? An example is only indicative if it explores at least to some extent meaningful conditions.

Besides, the reasoning was about bars, and your example is more about ticks. And you started talking (suddenly) about ticks. Why is that?

In the market the asymmetry is expressed in a small divergence of values of the right and left branches of the distribution. The absolute value of this divergence is only a fraction of the value of the PDF. And the PDF itself is more or less a smooth function. Under such conditions, all the benefits of asymmetry are instantly eliminated by the spread.

Not to mention that this asymmetry is even more non-stationary than the distribution itself.

I agree with you. You can't outplay the spread if you consider the asymmetry across the entire series. It would be significant. I wrote about this in the post at the top of this page. It can only be used as a benchmark when analyzing sample distributions of increments. It can mean for example, that there is a significant technical level in the vicinity, which can be used. But this is rather an exception. Although it is possible to check the strong level definition rule for significance in this way. Actually, the question was originally a theoretical one.

 

avtomat писал(а) >>

Reshetov wrote (a) >>

... are theorver formulas from the player ruin problem.

that's all the argument...

Is it not enough? Or do you have something more reasonable other than siftings?