Distribution of price increments - page 12
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
Matlab has a similarSimulink package. The convenience lies in the link with Matlab itself.
Alexander, a methodological question: why take the ticks? There is a lot of noise there. It's worth taking close prices, imho. Or more precisely the returns from close prices.In terms of my analysis, only the so-called "gap" can be considered as "noise". I consider the rest of the data to be reliable and well within a tolerance (confidence) interval for the t2-distribution Student's t-test.
Matlab has a similarSimulink package. The convenience lies in the link to Matlab itself.
Can we calculate the median(20000), i.e. the median of a sample of 20000 values, in Matlab?
Yes. In fact, there are only hardware limitations...
Here, for example, is a sample of pseudorandom numbers of type double with size 2e6, where values are from 1 to 10k. The last element of the sample is 9439. The median value is 5003.
Yes. In fact, there are only hardware limitations...
Here, for example, is a sample of pseudorandom numbers of type double with size 2e6, where values are from 1 to 10k. The last element of the sample is 9439. The median value is 5003.
VisSim, unfortunately, has a sample size limit = 16384. But it works great with dynamic data obtained both from DDE and other data sources. And the mathematical power of the functions can't be compared with that in MQL.
This is not an advertisement! But I haven't seen a better system for statistical analysis, especially in dynamics.
Matlab has a similarSimulink package. The convenience lies in linking with Matlab itself.
Alexander, methodological question: why take ticks? There is a lot of noise there. It's better to take close prices, imho. Or more precisely, returns from close prices.Denis, please forgive me for saying earlier that all ticks should be taken into account. I was in a hurry. I've taken quite a difficult pair CHFJPY. I cannot use the Tudent distribution for its increments but that's it.
I decided to take the average value between two incoming ticks - and here it is (see attached files)
Conclusion: apparently DCs even with NDD/ECN accounts manage to provide not all data, or distort it in some way, and yes - we need to apply simple filters for data processing (further sample increase for averaging had no effect, so just take the average between 2 incoming values).
2. The probability distribution of price increments (returns) in the flow of tick quotes is a discrete, asymptotically described by the Student's distribution with 2 degrees of freedom and the quantile functionQ(p) = 2*s*(p-1/2)*sqrt(2/a), where a=4*p*(1-p), s is the nonparametric standard deviation.
The Student distribution with 2 degrees of freedom has an infinite variance.
In some institutions the trader gets immediately scolded by the risk managers for trading with risks higher than the limit, while under the assumptions of your hypothesis the risks are infinite. Hence a logical question arises - why do such institutions need traders if they are not allowed to trade?
The amount of margin required for futures is also tied to risks and has finite values.
Alas, your hypothesis is untrue.
The Student distribution with 2 degrees of freedom has an infinite variance.
In some institutions a trader will be immediately sacked by the risk managers for trading above the risk limit, while under the assumptions of your hypothesis the risks are infinite. Hence a logical question arises - why do such institutions need traders if they are not allowed to trade?
The amount of margin required for futures is also tied to risks and has finite values.
Alas, your hypothesis is not true.
Yes it is. You just have to apply non-parametric measures of variance, expectation and asymmetry.
Corresponds. You just have to apply non-parametric measures of variance, expectation and asymmetry.
Please follow your theory to the end or accept that it is incorrect. Since you claim that the process obeys Student's distribution with 2 degrees of freedom - then even theoretically there is no finite variance, and with non-parametric methods on such a process you will get rubbish, which has no relation to reality.
Please follow your theory to the end or accept that it is incorrect. Since you claim that the process obeys a Student's distribution with 2 degrees of freedom - then even theoretically there is no finite variance, and with non-parametric methods on such a process you will get rubbish, which has no relation to reality.
Please read the literature http://www.mathnet.ru/php/archive.phtml?wshow=paper&jrnid=znsl&paperid=1692&option_lang=rus and further on the literature used, including the English language literature.
We are talking about the t2-distribution as a representative of the family of distributions with a scale factor instead of the standard deviation. This coefficient of scale is a nonparametric parameter and is calculated not even as half of the interquartile range, but a bit more complicated.
But, I say again - it is important for me now to get practical results, not a direct proof of my hypotheses.
I am working very hard at the moment - I will soon present the results of modelling the process with entry/exit points on historical data on the forum. This will be followed by testing the model on a demo account and then on a real account.
Then why am I writing intermediate results? Just for people to read, especially young people - the topic is quite interesting :)))
So, I formulate basic hypotheses about the processes in the Forex market, which can be considered proven empirically and experimentally (in fact, the person who proves these hypotheses in analytical form, can easily go to the Nobel Committee for a prize :))))
1. The process of Ask and Bid price formation is non-Markovian.
In practice - all Expert Advisors, indicators and advisors that do not consider analysis of historical data (such as Bollinger Bands, Fast Fourier Transform, etc.) can not be considered from the word "at all".
2. The probability distribution of price increments (returns) in the flow of tick quotes is a discrete, asymptotically described by the Student's distribution with 2 degrees of freedom and the quantile functionQ(p) = 2*s*(p-1/2)*sqrt(2/a), where a=4*p*(1-p), s - nonparametric standard deviation.
In practice -all EAs, indicators and advisors that use Gaussian normal distribution in their calculations (as well as other classical distributions), "3 sigmas" rule etc. can also be ignored.
3. Theprobability distribution of the Ask or Bid price is a superposition of the Student's distribution with 2 degrees of freedom.
In practice, it is a cool task to extract specific distributions from the superposition.
Actually, based on analysis of historical tick data, or simply by averaging statistical parameters at certain sample sizes, a conclusion about the current price value exceeding certain historical boundary conditions is made. Only after that, the current distribution parameters - dispersion, skewness, skewness ratios, etc. are analyzed in order to find out whether a new Student's distribution has begun or has already finished. In the first case - the deal is made following the trend, in the second - against the trend.
Sincerely,
Alexander_K
I again suggest you to comment on the tick increments in the market review window (USDJPY) and in the trade opening window (EURUSD) shown in the picture of ticks just taken. Now from the perspective of the three hypotheses quoted above. The account is real.
Don't you want to immediately analyse groups of consecutive changes of one point back and forth? What is their quantile function?