From theory to practice - page 17
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Not exactly, it is necessary to get statistics for the period of course (it is important data), but the flux itself changes with filter change, and it is found as a rule after the fact.
The problem of physicists is that all their equations are bound to delta t, which means that the frequency generator must have a clear frequency, but with ticks it floats. The only price we can rely on in terms of time is the bar closing price. In this case we know for sure that the bar will close at that time.
And even in close we have gaps in time on weekends.
The basic statistics (mean, median, dispersion) depends on the timeframe: the smaller is the timeframe, the more accurate it is, i.e. close is closer to the high-low. And tick data has no high-low at all. But tick data will contain a lot of "holes", and not only at the weekend. And these holes should be filled not only with averages, but for example with random forest algorithms. Remember, even M1 is full of holes.
All in all, a lot of hassles and no good solutions.
What for?
It is supposed to be stationary. And if the cotier is stationary, then there is no problem with prediction at all. You take any tool and you have millions in your pocket.
And if the cotier is not stationary, there is an article (I gave you the link), it is all laid out, competent, with the indication of insurmountable limits.
Are we talking about the same thing but in different languages?
Let's be more specific:
1. The movement of Ask or Bid price itself is a non-stationary process whose characteristics change over time, which is described by the Fokker-Planck equation.
2. Returns increments, expressed in pips, of Ask or Bid prices are a quasi-stationary process having a particular t2 Student's distribution with a particularnon-parametric mean = 0 and a scale factor s not equal to the standard deviation.
You think differently?????
So are you going to trade increments or price?
Alexander_K:
And the most important detail.
I've said more than once that VisSim doesn't allow to work with sample volumes greater than 16.384.
I would gladly work with all ticks and probably, like everyone else here, would fight for reception of each quote. But there is a problem - the model shows the best results on timeframes H4 and higher - i.e. I need to take more than 16.384 quotes in the FIFO buffer! So life itself makes me search for ways to optimize the model.
I must have been a bit tongue-in-cheek from a trading point of view initially. A beginner, after all... Please excuse me... But it doesn't change the matter - I' m stationary, full stop. And nobody will convince me otherwise.
In 1998 Long-Term Capital Management (LTCM) went bankrupt. It was run by two Nobel laureates in economics and several economists of lower rank - the main reason, too, was that they were convinced that in general the market was quasi-stationary.
Greetings, Dimitri! Have a look at it. It is for 64-bit. On my 32-bit Vista, the Mean and MedianSmooth blocks just don't work. Besides, the Variance block still has a limit of 16384.
What's vista x32, a laptop or something? You can get x64 on just about any PC... Not to mention the fact that MT support for vista is practically discontinued.
Cena of course, Yuri! I must have been a bit tongue-in-cheek from a trading point of view from the start. A beginner, after all... I'm sorry... But it doesn't change the matter - I' m stationary, full stop. And no one will convince me otherwise.
reeks of bigotry...
Once again:
1. The movement of Ask or Bid price itself is a non-stationary process whose characteristics change over time and is described by the Fokker-Planck equation.
2. Returns increments (for example, return=Ask(t)-Ask(t-1)), expressed in pips, Ask or Bid prices - a quasi-stationary process which has a specific t2 Student's distribution with a specificnonparametric mean = 0 and a scale factor s not equal tostandard deviation.
I appeal to physicists - here, until you understand this, friends, some economists will still laugh at us. That's it!
Let's get back to our business: The question of what we trade - price or increments - is not idle. The matter is that order flow near Bid 1.18000 is slightly different from order flow near Bid 1.18367. In other words, when speaking about memory, there is a difference, as it is obvious that price has memory (a trader must remember exactly where he or she opened and which price he or she has to move away from to gain some profit). And the increment may not. The key point is that we do not know the level of increments, near what we trade?
That is, the statistics on rerurns may be able to find the memory, but the reasons will not be found.
The level is still known.
It is known from Vysotsky-Petunin's inequality that 99% of values of one-tailed distributions are in the range +-6.666*s.
That is, if we know the nonparametric coefficient of scale for a specific currency pair, for example s=2 pips, then 99% will be in the range of +-13-14 pips.
But, these are increments. It may be of interest to some.
I am interested in the fact of stationarity of increments for other reasons.
In the work to which SanSanych gave a link, people recognize that if we accept that ALL processes in Forex are non-stationary, then there is NO WAY to calculate an optimal sample size. It will be floating. If one follows the hypothesis that returns is a quasi-stationary process, then the desired sample size can be calculated easily and freely. And this is confirmed by my experiments.
Alexander, my point is that we should dance from physical processes to mathematics.
If a trader has opened an order, will he close it after 100 bars or 100 points?
The X-axis has a one-dimensional scale (vertical) and the time evolution of the process is recorded on it, but the evolution is of minor importance to the trader as long as the market moved back from the order opening. In other words, the events that take place on the vertical axis are important to the trader. Horizontal events are just a history that is used by a trader for making forecasts.
The level is still known.
It is known from Vysotsky-Petunin's inequality that 99% of values of one-tailed distributions are in the range +-6.666*s.
That is, if we know the nonparametric coefficient of scale for a specific currency pair, for example s=2 pips, then 99% will be in the range of +-13-14 pips.
But, these are increments. It may be of interest to some.
I am interested in the fact of stationarity of increments for other reasons.
In the work linked to by SanSanych they explicitly say that if we accept that ALL processes in Forex are non-stationary, then there is no way to calculate the optimal sample size. It will be floating. But if one follows the hypothesis that it is a quasi-stationary process, then the required sample size can be easily and freely calculated. And this is confirmed by my experiments.
Model fitting.
Moreover, non-stationarity is rejected without proof, and quasi-stationarity is also accepted without proof. And all this at the level of a hypothesis. And then... about the mathematical millstones, I hope you remember?
"and if it does not agree with facts - so much the worse for facts" ;)))
By the way, all pipers should know and love the Vysotchansky-Petunin inequality.
If they know the exact scale factor s for a particular currency pair, good for them.
But we are not looking for easy ways, are we? :)))))) That's why we will solve the Fokker-Planck equation and nothing else!
And you say you don't need to know theory!
What equations, inequalities? You don't even know the basics.