From theory to practice - page 42

 
Alexander_K2:

That's how I see a fairly clean t2 distribution for returns increments.


Alexander, let me ask you a couple of questions:

  1. What exactly led you to the conclusion about the type of distribution? The appearance of the histogram, calculations or something else? No one has mentioned this type of distribution in financial markets before you.
  2. What is the expected payoff (in 5/3 digits) in a trade you expect (or already have) in the output of your model? The question is dictated by the fact that you analyze tick increments, respectively, it is logical to expect the statistical advantage of a close order. The fact is that even if the model actually gives the OOP within the spread, then you will not make money on it.

 
Petr Doroshenko:

The point is, a substantial liquidity provider (#1) has been filtering-drawing quotes/data for a long time and has experience. This provider (as probably others), most likely (purely imho) is disabled at #2 and #3 at certain times of the day on some condition (perhaps a hidden good at certain times that allows not to use higher timeframes). I.e.:

- There is no definite regularity in the continuous flow of ticks into terminals 2 and 3: a couple of hours ticks go this way, then five hours differently, then some other way, and the sign of the change of the regularity is unknown to a user. It doesn't make sense to analyse ticks in simple aggregators of several liquidity providers.

- the found/adjusted tick pattern for #2 and #3 will not be reliable, because the error-divergence of data at certain moments reaches opposite values as -1/+1 (up/down direction).

- During a minute, DT/broker can draw anything, as long as the final minute corresponds to something (maybe someone will go to fca), and a large DT/broker will not change tick drawing frequently, but will just send a history update (and superindicators/advisors "hang").

Has it occurred to you that ticks may come in batches to one, at equal intervals to another, and at some other intervals to a third?

Then when comparing the quotes all the above theory is nonsense?

The point is that first you have to synchronise the graphs by time, and only then compare them.

But I'm 100% sure that nobody will ever be able to do it with ticks, unless we have the same communication channels, the same receiving and transmitting equipment, and so on.

And it is not for nothing that the reference point for comparison is the graph on TF M1.

But even on М1 the charts may not coincide because the first and the last ticks of the minute may occur in different candlesticks of different brokerage companies and will be different again.

And in general I did not like the fact that some geophysicist from the very beginning gave the formula with a mistake and is still banging on his chest - I am mol....
 
Alexander Sevastyanov:

Alexander, let me ask you a couple of questions:

  1. What exactly led you to the conclusion about the type of distribution? The appearance of the histogram, the calculations or something else? No one has mentioned this type of incremental distribution in financial markets before you.
  2. What is the expected payoff (in 5/3 digits) in a trade you expect (or already have) in the output of your model? The question is dictated by the fact that you analyze tick increments, respectively, it is logical to expect the statistical advantage of a close order. The matter is that even if the model really gives an MOS within the spread, you cannot make profit out of it.

Greetings Alexander!

1. I was also interested in the fact that nobody has so far described this distribution in market models. How so? Some Gauss, Laplace, Cauchy... And further on ALL the models that are available at the moment can not be looked at from the word "at all". I can repeat the following - there is a relationship between two successive quotes, the so-called "memory". We have 2 degrees of freedom. The increments form a non-Markov chain the probability distribution of which must be described by the t2-distribution. It has to be there. Nothing else is there and can't be! Only it is a discrete analogue of the continuous t2-distribution which can be read, for example, hereevery tick. Well, the spread and commissions will just reduce your profit.

But, I thought - am I a physicist or not a physicist? What is more important to me - money or truth? Truth! That's it!

So I decided to move on - to describe a non-stationary process of price itself, which RELATED to the stationary process of increments through this t2-distribution. And that's where the storm of emotions set in, which can be seen in this thread :))))))))

 
Alexander_K2:

Greetings, Alexander!

1. I am also interested in the fact that nobody has described this distribution in market models so far. How so? Some Gauss, Laplace, Cauchy... And further on ALL the models that are out there at the moment, you can not look at the word "at all". I can repeat the following - there is a relationship between two successive quotes, the so-called "memory". We have 2 degrees of freedom. The increments form a non-Markov chain the probability distribution of which must be described by the t2-distribution. It has to be there. Nothing else is there and can't be! Only this is a discrete analogue of the continuous t2-distribution which you can read about, e.g. herehttp://www.mathnet.ru/php/archive.phtml?wshow=paper&jrnid=znsl&paperid=1692&option_lang=rus.

One gets the impression that you use the word "repeat" in the sense of "prove", "prove". It is as if you believe in such justifications. That's the way it's done in the media these days, but why here?

At your link I managed to "read" only the phrase "There is no information". Maybe others were more fortunate?

 
Vladimir:

You seem to use the word "repeat" in the sense of "prove", "justify". And it is as if you yourself believe in such justifications. This is the way it is done in mass media nowadays, but why here?

At your link I managed to "read" only the phrase "There is no information". Maybe others were more fortunate?

http://www.mathnet.ru/links/4d855cb05856b9210d332bf0ee475aef/znsl1692.pdf
 
Vladimir:

You seem to use the word "repeat" in the sense of "prove", "justify". And it is as if you yourself believe in such justifications. That's the way it's done in the media nowadays, but why here?

I only managed to "read" the phrase "no information" on your link. Maybe others have had better luck?


Vladimir, I think you are well aware that this is the case. The evidence was hard for me personally. Do you think that I do everything in one day? If you don't believe me, go and see for yourself.

 
Alexander_K2:

Greetings, Alexander!

1. I can repeat the following - there is a link between two quotations coming in consecutively, a so-called "memory". We have 2 degrees of freedom. The increments form a non-Markov chain the probability distribution of which must be described by the t2-distribution. It has to be there. Nothing else is there and can't be!

2. ... ...at the receipt ofeach tick it should be pipsed. Well, the spread and commissions will simply reduce your profit.


Alexander, thanks for the detailed answers.

  1. If I understand correctly, the conclusion about tick distribution is based on the fact that the chain is non-markovian, i.e. there is a "memory" and connection of the present/future with the past. Do you accept this as an axiom, without proof? I also tend to think that there is some "memory" because, for example, levels, trend lines and other technical tools work more often than not. But what is the quantitative measure of this "memory" and whether we can blindly take it as the basis?
  2. "Pipsing at every tick"? Are you sure about that? There are broker/dealers that will help you to do it on the demo and you will make a profit even with random entries: they will shift quotes in your favour, open/close with a positive slippage for you. But on the real market it will be exactly the opposite. And my question about the LFL that wasn't asked wasn't about curiosity. Any TS with the LFL within the spread is doomed to fail on the real market.
I sincerely wish you success, but I'm afraid you're making the mistake of investigating tick increments specifically. In my humble opinion, instead of (Bid+Ask)/2 and samples in equal or exponentially distributed time intervals, it's better to switch to filtering/sampling the price increments by a fixed value of about 1-2 average spread, in fact to the simplest threshold filter. What is the disadvantage of counting at fixed intervals? Because price can go back and forth by an appreciable amount during that interval. And it may not be an accidental spike, but you will miss it. The threshold filter will greatly reduce the amount of data for analysis by filtering out tick noise generated by the market and broker, and will not miss significant price movements. And another advantage is that increments of 1-2 spreads can no longer be traded theoretically, but practically. And with your "quantile function and confidence levels" apparatus, even more so.
 
Alexander Sevastyanov:

Alexander, thank you for your answers.

  1. If I understand correctly, the conclusion about the distribution of tick returns is based on the fact that the chain is non-markovian, i.e. there is a "memory" and a present/future link to the past. Do you accept this as an axiom, without proof? I also tend to think that there is some "memory" because, for example, levels, trend lines and other technical tools work more often than not. But what is the quantitative measure of this "memory" and whether we can blindly take it as the basis?
  2. "Pipsing at every tick"? Are you sure about that? There are broker/dealers that will help you to do it on the demo and you will make a profit even with random entries: they will shift quotes in your favour, open/close with a positive slippage for you. But on the real market it will be exactly the opposite. And my question about the LFL that wasn't asked wasn't about curiosity. Any TS with the LFL within the spread is doomed to fail on the real market.
I sincerely wish you success, but I'm afraid you're making the mistake of investigating tick increments specifically. In my humble opinion, instead of (Bid+Ask)/2 and samples in equal or exponentially distributed time intervals, it's better to switch to filtering/sampling the price increments by a fixed value of about 1-2 average spread, in fact to the simplest threshold filter. What is the disadvantage of counting at fixed intervals? Because price can go back and forth by an appreciable amount during that interval. And it may not be an accidental spike, but you will miss it. The threshold filter will greatly reduce the amount of data for analysis by filtering out tick noise generated by the market and broker, and will not miss significant price movements. And another advantage is that increments of 1-2 spreads can no longer be traded theoretically, but practically. And using your "quantile function and confidence levels" apparatus, even more so.

As for the point 2 - I completely agree. That's why I did not do it.

Thanks for the post! I will have to think about it.

 
Thanks for the link, very interesting.
 
Alexander_K2:

On point 2, I agree completely. That's why I didn't do it.

For the post - Thanks! I'll have to think about it.

But here you are wrong.

Your way is right, there are just so many mistakes in the practical application of the theory and in the formulas.

And you can only correct yourself if you combine four things: theory, trading, programming in specialized trading languages, and seeking the advice of experienced traders.

And do not forget that an SELL order is opened on bid, and BUY order is opened on ask, while ask-bid is a spread. What are you going to trade using tick data as initial data?