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
By MA I meant smoothing in general, i.e. any low pass filter. In theory, tails for WMA on Wiener should be longer than for simple MA or EMA.
Well, we just established ourselves that some of these tails are produced by the methodology itself. And Wiener itself is self-similar.
For some reason, I thought I was halfway there. Fuck it.
once even wrote an article that any indicator based on smoothing ideas is irrelevant to a quote because the approximation error has a variable variance and can reach arbitrary values multiple of three sigmas. This is on the surface. People trading know it very well on their own depo.
Moreover.
If we take some smoothing, even the most abstruse one, and build a regression for it, the parameters of such a regression will not always be significant.
Therefore any idea of smoothing must be used very carefully in trading.
I'm incredibly glad that SanSanych, with his experience, has just joined the discussion! And with what you wrote I absolutely agree. Simple thoughtless application of any MA and variance around them (as Citizen Bollinger does) is a road to nowhere.
The problem is that a Wiener process with drift is only a first approximation of this process. We have Steudent instead of Gauss, and such a serious one, almost like Cauchy and NEMARCLE - don't forget that! We will describe this "memory" more than once or twice - and I can imagine what a debate it will be here! I feel a little uneasy - and I wonder whether I should write about it at all...
We know it's not Gauss, it's something else. It's been known for about 10 years. There is no objection to Student's tails - he describes them well).
I am now more interested in tails generated by the methodology itself, and it is best to watch them on Wiener. By the way, it takes you 5 minutes to look through the tails on WMA Wiener.
By the way, do you know WMA coefficients? I would like to look at them.
In general, it's better to use Bessel instead of WMA, imho. And there are less calculations).
It might be better. We'll have to check it out. Are the coefficients the weights w? This is one of the keys to this problem!
Once again - weightsw are determined from the formula of probability density of increments. I.e. at the current step the increment for the pair AUDCAD = 1. We know that the nonparametric coefficient of scale s for this pair = 1.95. This coefficient is tabular and does not change over time. Substitute it into the formula: w=s^2/[2*sqrt((s^2+x^2)^3)]by Mikhail Dovbakh. We obtain the weight of the current price value.
Where can I find s table coefficients for currency pairs? The answer is nowhere. I should calculate them myself (it's not the standard deviation) which is what I'm doing now.
Yes, weights. So far I'm interested in specific coefficients of WMA (or one of WMA) applied by you without regard to currency pairs or even parameters of the distribution itself. If not a big-big secret, of course).
Anyway, I took two brokers, real accounts. I compared bid ticks on EURUSD. Took several weeks: 1 week in October and 2 in November.
I got, by the tests, that the samples are different. And the second broker has more ticks, sometimes by 2 times. Initial impression: the second broker plays with the spread, creating the appearance of an active market, when in fact there is no activity. That's how it is. I will still test with others. If the tests will show the same results, then I will have to change the method, and may read the ticks with a certain periodicity. Question.
For example, for EURJPY the coefficient is s=2.35. The increment expressed in pips and this coefficient is substituted inw=s^2/[2*sqrt((s^2+x^2)^3)] and you get the price weight at each tick receipt to calculate the moving weighted average. (sorry, maybe by WMA in MQL you mean something else? - I work in VisSim)
I don't mean MQL.) By WMA we mean the polynomial Yi=sum(Aj*X(i-j) or Y(z)=sum(Aj*z^(-j), where j is from 0 to N. In general, I take it as a polynomial of degree N WMA doesn't exist, as the coefficients change as the play progresses. Well, all right then, let's leave it at that.
Anyway, it's clear that from such smoothing in the frequency domain something is going on.
If the tests show the same results, then we need to change the method, maybe really read the tics at some intervals. Question.
The system is not crude. I have already checked and rechecked it and theoretically justified it. Now I'm preparing to run it on a real account for 18 pairs at once. This is an important moment. That's why I am asking some technical questions from professionals.
What is important is not the differences in specific quotes values - with a large sample size they won't have a significant impact on WMA - this expected value is stable enough. What is important is the NUMBER of quotes. For if I take 1000 ticks in 1 hour, and you take 10 000 ticks, then how can I or you recommend me these or those sample sizes?
Yes, here's more: Alexander_K:
"Even if you have your own view of the market and do not want to understand my developments, I assure you - these tools are very useful, you just need to know how to apply them. And in MQL, in my opinion, there is no moving average and therefore there is no nonparametric skew. How the algo-traders work without it - I don't know. :)))"
The algotraders don't care if there is a moving median in MQL4, and in MQL5 as well (though one of the threads created by you has already created a very fast indicator), or not. If they need it, they will write it in the environment where the core of their TS with decisive rules is implemented. Executives to collect ticks and transmit trade orders to the server should not do the analysis. Their minimal functions and the same middleware interface make them easy to implement in a variety of trading platforms, not just in the two versions MT4 and MT5. And, of course, VisSim should not be required either - it is a completely non-specialized alien tool where even the maximum sliding sample size cannot be changed. Everything must be in your own hands, you only have to outsource what you cannot do yourself - communication with the server, above all.
Oh, these darkies, physicists, lyricists and other graduate conservatoire graduates.
I have attached a text, which explains everything without all the crap about ticks and different sliders. maybe someone with experience will understand it and post the result.