From theory to practice - page 419
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
Here's what I'm thinking.
If the distribution of a sample of, say, 1,000,000 ticks is unstable (and I still can't get that volume on my exponential time) and changes variance over time, then it turns out that in my case neither the arithmetic mean nor the weighted average can be used as a measure of central tendency.
This leaves me with the median.
Channels must be plotted in relation to the median. Is that it?
In my opinion, the point is not just the sample size and the choice of its distribution centre characteristic (mean, median, mode, quartile mean or whatever). Your sample of increments must belong to the same trend - then it can be considered equally distributed. And here you will find that the results will depend on how exactly the concept of trend is formalised.
I dedicate this post to all those desperate in this topic and in the theory of diffusion processes in the market.
Peasants!
I have finally figured out the notorious constant C in the formula for calculating process dispersion (see my earlier posts).
And it's not a constant! It is the rate in the current observation time window.
For the AUDCHF pair in the last 2 days the process looks as follows:
Having lost all the readers of this thread with the most shameful EURUSD trade, I will keep it to myself. As a diary.
Looking at the distribution of deviations from the moving expectation, I become more and more convinced that it is a Laplace distribution given the huge sample size.
In calculating the variance and, consequently, the standard deviation, it would seem that everything is taken into account, both the speed of returnees and their average value and time.
But, so far, it is not possible to reduce the process to a stationary one, no matter how hard I try. Most likely it will never succeed.
Meanwhile, the quantile always = const. And the form of distribution, due to non-stationarity, changes...
It turns out that the quantile - which covers 99% of distribution values - is also a variable, not a constant. It also needs to be calculated at each step... Is that so? It's crazy...
Percentile calculator:
https://keisan.casio.com/menu/system/000000000540
Now I understand why we need a cartoon about the dynamic form of the probability distribution - at least to see if it is single-mode or multi-mode relative to the expectation.
In the first case one should use Petunin-Vysokovsky inequality, in the second case Chebyshev inequality.
Yes, at quantile=const, the solution of the problem is imprecise, it should also be dynamic, but it is not possible for me personally.
Still, I would venture to argue that both the incremental and moving expectation deviation distributions belong to the same class of unimodal distributions, with increasing sample size --> to the Laplace distribution.
This means that we should use a quantile of about =3, corresponding to the 95% confidence level according to Petunin-Vysokovsky.
If one chooses quantile =4.47 for the 95% Chebyshev distribution, then, unequivocally, a lot of cool trade entries will be lost, which is actually what I have. Very rare trades upset my senile soul.
Maybe you should also work on calculating the optimum stoploss, and if you can, you might be able to increase the number of trades.
Maybe you should calculate an optimal stop loss, and if you can, you might be able to increase the number of trades.
Can you give me a link to such calculations and research on the optimal stop loss level? I failed to do it in that shameful trade, and it was because of its absence that I got burned.
Can you give me a link to such calculations and research on the optimal stop-loss level? After all, in that infamous trade, it was the lack of it that got me burned.
No, it wasn't.
if you remove the trend from kotier (i.e. limit yourself to the analysis of increments over x time period), then this very trend takes you into minus.
and vice versa: If you remove the flat from the kotyr, then this very flat...
Your program is blind, unfortunately
It's just like any other channel, because in the channel we analyze the price for only 1-2 ticks or for the time interval.
)
No, it's not because of that
if you remove the trend from the kotir (i.e. restrict it to the analysis of increments over the x-period), it is this trend that takes you into minus
and vice versa: If we remove the flat from the kotir, this very flat will occur...
Your program is blind.
)
Is there a universal forum that divides data into trend / flat? (in real time without delays, not on history)
With such a formula, making money is like sending two bytes.