From theory to practice - page 337
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
Still, the real time intervals arehttps://www.mql5.com/ru/forum/221552/page337#comment_7230221,
In fact, 0 means that the time stamp, in seconds, has not changed and the quote has. (Obviously, this is rounding to 0 the time of a quote received in faster than 1 second).
And as I'm working with a discreteness = 1 sec, those 0's were transferred to 1's and a discrete logarithmic distribution was obtained, which is not correct.
PS
And again, what - I'm incredibly lucky with the quote's flow and I unintentionally, thinking I'm doing a good deed, distorted it so badly that I have no deals at all? Is that it?
I keep wondering what you will do on such short intervals??? Suppose you get a satisfying result, but it uses ticks within a second. Then what? Think about how it can be applied in real trading with 1-3 trades a day.... Why do we need so much tick data????
It's always interesting to watch the "statists" about exactly!!!! Now I will call you "STATISTICS" These are people who see a quote as a non-stationary time series and nothing more. Who use words such as "distribution", "Dispersion" or "Markov's Law" or "Not Markov's Law" as the experts in this field say.
Now you are STATISTICS!!!!! :-)
Misha! You're our dear man! All this rubbish with Erlang flows is written for you! Convert initial BP to 150-th order Erlang flow - get M15 analog of prices. Take their returns and feed them into a neural network. ALL.
So let's work out what is a price analogue??? How would they be different from the normal 15 minutes. And explain what returns are, because everyone is talking about it but I don't understand the meaning of the word...... to be honest...????
Misha! You're our dear man! All this rubbish with Erlang flows is written for you! Convert initial BP to 150-th order Erlang flow - get M15 analog of prices. Take their returns and feed them into a neural network. ALL.
I have, in fact, the same question as with Michael - why all this? What to do with all this?
Ticks are actually necessary, but in the last few tens of seconds before entering the trade. During the rest time, even for 5-10 trades per day, they do not contribute anything to the existing information on the 1F. And even if it increases accuracy, its real use gives nothing - you will obtain +/- fractions of a percent of profit in a trade, and nothing more.
By the way, in your recently posted chart it is already visible that one can do without ticks, and without Erlangs, and without many other tricks - and it will not affect anything. Don't attract unnecessary entities (c).
The difference between the current price and the previous price. This should produce a stationary series in which any neural network, in general any with a single input, will predict the next following current value with a very high probability.
Well, the returnee is now taken care of. Although I'm not sure that the series will be stationary. It will be normalized, but not stationary and if the next one is predicted with high accuracy, then you should at least work on hourly timeframes to make profit and have enough time to build a model. Why do you need ticks then????
And don't flatter yourself about returneys. It's unlikely they will be as good as you say they will be. Too easy. By now everyone would be a millionaire....
I actually have the same question as Mikhail - what is the point of all this? What is there to do with it all?
The ticks are actually necessary, but in the last tens of seconds before entering the trade. During the rest time, even for 5-10 trades per day, they do not contribute anything to the existing information on the 1F. And even if it increases accuracy, its real use gives nothing - you will obtain +/- fractions of a percent of profit in a trade, and nothing more.
By the way, in your recently published chart it is already visible that one can do without both ticks and Erlangs. Don't attract unnecessary entities (c).
Agreed. There has to be a balance in the market!!!! Between simplicity and complexity in building the TS as a whole. Here is an example with models: Too simple model will not work long and will be accepted as random. Too complex will work longer but worse. Below the required threshold of profitability. You should choose such models that are not too large and not too small. This conclusion I've drawn from my experience ....
While training one and the same training file I obtain several models. And no matter how many of them I choose, it's always that one which has the average number of inputs and polynomial size among all obtained models, that wins. As an example:
1. 4 inputs, polynomial size small
2. 5 inputs polynomial size medium
3. 8 inputs polynomial size large.
Comparison of polynomial size naturally occurs between models. So, as a rule, model number 2 wins, which has 5 inputs and medium polynomial size. No matter how many I tried to get models with more inputs (from the theory that the more parametric the model, the more intelligent it is) as a rule they all merged into feedbacks.
Once again - I made one gross mistake - I distorted my flow with an exponent and then started collecting the right flows. It will be corrected.
there you go...
Long time ago I said that the scales came out.
The only people who need to use tics are those who sit on the broker's side and write HFT arbitrage.
You can't win them, believe me.
The difference between the current price and the previous price. This should produce a stationary series in which any neural network, generally any with a single input, will predict the next highest probability of the current value.
Agreed. There has to be a balance in the market!!!! Between simplicity and complexity in the construction of the TS as a whole. Let me give you an example with models: Too simple model will not work long and will be accepted as random. Too complex will work longer but worse. Below the required threshold of profitability. You should choose such models that are not too large and not too small. This conclusion I've drawn from my experience ....
While training one and the same training file I obtain several models. And no matter how many of them I choose, it's always that one which has the average number of inputs and polynomial size between obtained models wins. As an example:
1. 4 inputs, polynomial size small
2. 5 inputs polynomial size medium
3. 8 inputs polynomial size large.
Comparison of polynomial size naturally occurs between models. So, as a rule, model number 2 wins, which has 5 inputs and medium polynomial size. I've tried to get models with more inputs (from theory more parametric model is more clever) as a rule they all got merged into feedbacks.
Yes, something like that should be the case. Once I was at a seminar in the Keldysh institute, where mathematical models of complex systems were discussed. Briefly:
Simple model - describes poorly.
Medium complexity - much better,
High complexity - becomes unstable or gives no significant gain.
I.e., for process models there is some optimal complexity and a limit of complexity beyond which one should not go.
Back to the facts - why is Doc in his tick junk with time intervals = Gamma+Koshi getting 0 and my crystal clear 2nd order flow unrestrainedly happy ? After all, this is the key, gentlemen!!! And you are trying to throw it in the mud.
Want me to guess ?
In theoretical musings and searches the physics of the process and say "nuances" are forgotten. To make reliable analysis of ticks with DDE you should at first answer questions what are ticks, how and what do they consist of (minimal process model), what do we count, why, what is DDE, in what mode it works, is there data loss, how we choose time intervals and why. And there's a lot more. How to check intermediate results is out of the question for some reason...
your "crystal clear stream" is a close analogue of s2 s3 s15. what do you want to end up with ? some kind of fractal indicator or something ? how mutually self-similar the different frames are.
For a rather curious thread to get good manners, you need a clear introduction - what we're doing, based on what, why and for what. So far, the most useful things are criticism and rare references.
So far, the greatest benefit is only criticism that is inadequately responded to and the occasional reference.
Criticism here is pointless. They're insane.)
I think I'll shut up now. Tired of it, I confess.