From theory to practice - page 247

 
Alexander_K2:

Yesterday I started reading the thread myself from the beginning and realised that, yes, it is almost impossible to understand what we are talking about here.

I am too lazy to write an article - it would require a lot of work, rigorous formulas and proofs.

I offered to arrange for the model in VisSim to be sent to those who wanted it on request. Very few people contacted me. Either they are shy, or they consider it beneath their dignity to contact someone... I don't know. I can't put the model right here on the forum. It turns out that it will get both those who literally fought the market like lions, but something did not work out, and those who did nothing at all. This is not fair. I will still think about what to do.

For example, I just don't have time to deal with VisSim, after all, mostly everyone uses mql language here... or at least pseudocode :)

 
Maxim Dmitrievsky:

For example, I just do not have time to deal with VisSim, most of us use mql language, or at least pseudo code :)

Now we are figuring out how to open PAMM account. In a week or two we will organize everything and I will implicitly promote this data through my relatives. In my opinion, it is the best solution of all problems. If the trader sees that the trading is going well, why not subscribe, right? If it's obvious that the case is rubbish and that A_K2 has messed up something here, then there's nothing to talk about.

 
Alexander_K2:

Now we are figuring out how to open this signal and a PAMM account. In a week or two we will organise everything, and I will promote this data implicitly through my relatives. In my opinion, this is the best solution to all the problems. If the trader sees that the trading is going well, why not subscribe, right? If they see that the deal is shit and A_K2 has messed up something here, then there's nothing to talk about.

it's an ideal scenario, but we will have to wait a few months for the statistics all the same.

 
Maxim Dmitrievsky:

it's generally an ideal option, but you have to wait a few months for the statistics anyway

Not a problem. We'll wait. And let this topic stay. Maybe someone will get seriously interested, double-check, write an article. Or successfully combine two topics - diffusion equations and neural networks. It will be a masterpiece.

 
Alexander_K2:

Nah, Doc. Again, as this is a conceptual point.

1. It is necessary to work with ticks, if the analysis of a non-Markovian process in its entirety is supposed. The depth of analysis for decision making in this case - the more the better, up to the whole tick archive. I.e. we work on a logarithmic time scale, which is formed by ticks ("time spiral").

2. I, on the other hand, generate exponential time intervals and read the data for each pair separately, regardless of whether it was a real tick or not. I get pseudo-markov time series (I now have 12 in total - by the number of pairs), to which I apply mathematics to solve the diffusion equations. In this case, it is enough to analyze some observation window of the process. The time scale is exponential.

Cool, isn't it?

This kind of "black box" I definitely don't understand, but if I may, let's clarify the interfaces to it.

You take the tick quotes unloaded from the terminal into the base (file)? Right?

A tick quote for each pair in question is simply a tick time and the price of that pair at that tick. If you have an " exponential time scale", then I understand, that you do not pull all ticks from the base, but with some step (sampling, recalculation, whatever you want to call it), and the time is recalculated in your exp-scale, but the price in ticks remains unchanged? So we have a stream of tick prices, for each pair, which is sucked into VisSim. Right? Or is it not?

The black box in VisSim somehow "solves the diffusion equations", for the above described flow of prices (unchanged), and time (recalculated). Right?

What is the output from VisSim? Model parameters? Forecast of the next tick? Prediction of market movement for the hour/day/exp time ahead? A buy/sell command? What exactly comes out of the black box in VisSim?

Next, apparently, what VisSim has calculated you shove into your trading Expert Advisor (bot). What does the bot do? Does it have a working logic?

 
Serge:

A tick quote for each pair in question is simply a tick time and the price of that pair at that tick. If you have an "exponential time scale", then I understand that you do not pull all ticks from the base in a row, but with some step (selection, recalculation, whatever you want to call it), while you recalculate time to your exp-scale, but price in tick remains unchanged? So we have a stream of tick prices, for each pair, which is sucked into VisSim. Right? Or is it not?

I set the time intervals for reading quotes forcibly with the help of the NA generator with exponential distribution.

The Bid and Ask prices at a given time step are considered to be the received quote. And the obtained time series will consist of two types of data: 1. real tick quotes with real time stamp 2. pseudo-quotes, i.e. actually the value of the last really accepted tick.

The ratio of the total amount of such data (time window size) to the real ticks with the time stamp is the trading rate (or intensity). This value is involved in the calculation of the diffusion coefficient of the process.

This is one of the easiest and most reliable ways to replace a non-Markovian process with a pseudo-Markovian one. Now we can use mathematics for Markovian processes, i.e. consider diffusion equations, e.g. Fokker-Planck equation.

In these equations the 2 components we are interested in are the drift and diffusion parameters, which we calculate and use.

It is easier to look at the model. Do you need it?

Only please - study the model as much as possible on your own, no time now - busy with dreams of unrestrained cash in my purse :)))

 
Alexander_K2:

The answer is that I strongly believe that describing the market by diffusion equations is the only true solution. That's why I don't put a stop loss. I know that everything will return to the average in a well-calculated observation window.

But! Naturally, I am worried about the result. And since I go to work every day, I only look at the result in the evening and my nerves are fine. What would happen if I watched the equity in the course of a deal - I don't know, I can't say.

Also - the system still needs 1 improvement. When going beyond the support/resistance lines determined by the diffusion coefficient, we need another parameter for analysis, let us call it the trend/float coefficient.

At present time I have this parameter - Nonparametric Skew (asymmetry coefficient). It does not work well, but Hurst does not work at all!

I believe that the real additional parameter for analysis is non-entropy, but this has yet to be proven.

When this additional parameter is found - everything will be 100% positive trades, i.e. the Golden Grail. My current grail is not very good, it is wooden... But - the Grail!

If we were hardcore mathematicians, then of course the words "I am deeply convinced" would result in your expulsion from the "academy" =) But we are all just "digging" the market here and if you like doing it with a shovel called "diffusion equations", we are only happy to do it. Andyou don't have to prove anything!

As long as there is no harm to others, everyone is free to spend his time and money on whatever he wants. (c)

"I only watch in the evening and my nerves are fine" - that is until one evening you see a floating drawdown, painful for you by its size - that night you will not forget! And it is better to come up with a plan of action in advance, depending on the floating, but still drawdown (when to close, when not, when partially). It is even better to program this plan into the Expert Advisor code! Otherwise, at the moment of violent emotions, there is such a rush, that you can lose everything, and even more importantly, your health.


Now the most interesting thing, from what I read on these 247 pages:

"When going beyond the support/resistance lines determined by the diffusion coefficient, we need an additional parameter for analysis, let's call it the trend/float coefficient".

What space do you draw the lines in? Time - price? Or exponential_time-price? Can you recalculate into time-price?

Doesn't price going beyond these lines automatically mean there is a trend?

Do you have any ideas on how to improve the Trend/Fly Factor?

 
Serge:

"When the support/resistance lines, defined by the diffusion coefficient, are exceeded, we need an additional parameter for analysis, let's call it the trend/float coefficient".

What space do you draw the lines in? Time - price? Or exponential_time-price? Can you recalculate into time-price?

Doesn't price going beyond these lines automatically mean there is a trend?

Do you have any ideas on how to improve the Trend/Fly Factor?

As a matter of fact, we cannot completely transform a non-Markovian process into a Markovian one no matter how hard we try. But on an exponential time-price scale, most entry points behind support/resistance lines mean a return to the mean.

However, in my case 20% of 100% of crossings of these lines mean, well, let's say, the middle of the trend which is exactly the sign that the process memory cannot be completely "destroyed".

I need an additional parameter to improve the result.

As I said before, I'm using asymmetry coefficient now. But... Not the right one! Not quite right!

Absolutely sure that this parameter is the non-entropy coefficient https://en.wikipedia.org/wiki/Negentropy, i.e. a measure of deviation of a non-random process from a random one.

But, no time to do it - the housemates are shaking like a pear, demanding to stop research and get busy immediately replenishing the cash bags :)))

 
Novaja:

Converting a non-Markovian process into a Markovian one using Erlang flows.

http://www.ngpedia.ru/id346136p1.html

Might help.))

PS. Bas, thanks for the book, read it. I liked it, complex things in an accessible form, it was interesting))

OOO! Thank you!

Simple and accessible and the exponent popped up:

https://lektsii.org/8-40682.html

But the exponent is a 1st order flow, i.e. the simplest, as I understand it.

And if you experiment, you can get a nice regular flow!!!

Thanks so much again!!!

PS

So Sanya, you've been outflanked by Nova.

Потоки Эрланга, их свойства и применение
  • lektsii.org
Потоком Эрланга k-го порядка называется поток Пальма, у которого интервалы времени между событиями распределены по закону Эрланга k-го порядка. Поток Эрланга k-го порядка может быть получен из простейшего с помощью его прореживания. В простейшем потоке сохраняется каждое k-е событие, остальные отбрасываются. Привести схему формирования...
 

Gentlemen!!!!

While I'm away from the forum, if you don't mind, please post links to formulas for calculating additional time series parameters in this thread:

1. Hurst coefficient (verified formula!!!, because there are too many of them...)

2. Non-entropy coefficient

3. anything else you want.

Regards,

Alexander_K2