again on the tester - page 7
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
Regarding real ticks: there is publicly available recorded historical tick data with volumes: tick arrival time (milliseconds), trading symbol, best Bid price and its volume, best Offer price and its volume. I.e., data that allow analyzing even multi-currency ticks simultaneously.
If you want to have historical data of the whole tick, you can record them by yourself. Such data will occupy approximately 10 times more space.
Why don't you give us ten links to these histories?
Yes even from a number of brokers that traders work with. There is not and never has been publicly available recorded tick data for serious periods of time. There is none as a class.
There are a lot of theorists around who have little understanding of the subject.
Has anyone published (as I asked) a table of real recorded ticks and simulated ones for at least one hour of terminal operation?
I had to do everything myself, in the archive:
Here is the graph of ticks superposition (red line of simulated ticks is superimposed over the blue line of real ticks):
Comparison of real and generated GBPUSD ticks in 1 hour (click to enlarge image)
As a result, there are minor differences of 1-2 points and omissions of price saws (sequences of the type 5-6-5-6-5-6-5-5 tester simulates once as 5-6-5).
Who is modelling better or for whom such modelling quality is lacking?
Why didn't you just provide a dozen links to these stories at once?
Yes more from a number of brokers that traders work with. There is not and never has been publicly available recorded tick data for serious periods of time. There isn't it as a class.
One is enough. Recorded tick history, as I wrote above, for two years (since January 2007). Very convenient way (as one wants) to get them via API. Publicly available.
About brokers with traders. This broker will also be available on MetaTrader4. And it will not be a DC, but rather a full-fledged broker on your own platform with an available initial and minimum possible deposit.... wait for the new year.
There are a lot of theorists around who have little understanding of the subject.
Has anyone published (as I requested) the table of real ticks recorded and simulated during at least one hour of terminal operation?
I had to do everything myself, in the archive:
Here is the graph of ticks superposition (red line of simulated ticks is superimposed over the blue line of real ticks):
Comparison of real and generated GBPUSD ticks in 1 hour (click to enlarge picture)
As a result, there are minor differences of 1-2 points and omissions of price saws (sequences of the type 5-6-5-6-5-6-5-5 tester simulates once as 5-6-5).
Who will model better or for whom such modeling quality is insufficient?
And you take parts of periods, where there were high tick volumes (more than hundred per minute). Especially often it can be observed at some crosses. Nobody is measuring anything here. I mentioned above one of nuances of possible inconsistency of results of strategy's performance in the tester with simulated ticks and with real ticks.
If I have time, I will generate OHLC+V on M1 from real tick data on Bid and provide comparisons of modelled and real ticks.
You are wrong to think that the testing experience of others is inferior to yours. I can assure you that if the quality of modelling had a bearing on a person's life, you would have a different attitude. And you would have more questions.
You do not want to answer my question about the adequacy of tick arrangement inside a bar. Pity. The drawings were given. The strategies that depend on them were given. But that is not the point. The issue is the generation quality. All to nothing ((
It is incorrect to accuse us that we have not collected ticks on Saturday, when the quotes do not go, at least it is not correct.
You have, I see, history is stored. You stored it in 2007.11.28 and now it's 2008.12.20. It means it is important for you (history) and you keep it a year has passed. But we do not need it. Why - it's noise. But excuse me, if the tick data doesn't make sense, then it doesn't exist in minutes and so on... Because all bars are built from ticks (or will you deny that too?).
I don't see the point of modelling at all. A model is always errors and inaccuracies, which is what they are trying to show you. Keep the tick history. Submit it to a tester. Slice it however you want, even by minutes, even by 1.5 minutes, you can slice it by equal number of ticks (equi volumes), you can slice it by price increment (kags, crosses zeros), etc.
I don't understand your insistence on admitting the obvious.
Regarding the comparisons of real and tester quotes you posted.
And made a comparison.
1. I checked if the tester works in the same way as you do. Yes, it is exactly the same. Generated 267 ticks. And it's a perfect match. Here's the figure.
Here is the formula.
It checks if price and time match. If yes, they are matched, then 1. And they are added together. There are 267 coincidences, i.e. all ticks coincided and this can be seen in the figure too.
0 coincidences, i.e. the tester created by you has never reproduced exactly one tick !!! And it didn't even match the number of ticks in real they are 333, generated 267 !!!! Where is the 20% of ticks ?
What kind of modelling adequacy are you talking about ? And what do you mean by this word ?
Sincerely Privalov.
I repeat: "No sequence of ticks that happened is repeatable in the future". First, prove that any random sample (or selection) of ticks is worse than some "real" set of ticks. Exactly for the purpose of testing a strategy, not to accurately replicate the behaviour of that strategy in the past.
I've seen a picture similar to the one posted by Renat about real and simulated ticks before here, and it suited me fine. Exact repetition of a strategy in the past at the most "microscopic" level (supposedly "real ticks from a real DC") is quite capable of leading to dangerous illusions - especially if these are systems with small stoplevels.
Heck, accurate history does not exist in principle at all. Reasons:
1. There is no single source of forex quotes - no matter what the Masterforex followers say about the Consortium conspiracy.
2. Each DC selects suppliers it likes by some criteria, forms a common stream and then filters it somehow. This is "story 1" that goes to the trader. And this filtered flow, alas, is not the history (even if we focus on a single broker).
3 The so-called "History 1" is further normalized, i.e. further filtered, and then placed in the brokerage company archives. This "history 2" is presented as the history to be tested. This is the "real ticks", colleagues!
4. Where will the trader get the "true history" of ticks if there was a connection failure? And where will the brokerage company get it, if the server that receives its own quotes fell down?
I think that such a rigid emphasis on the indispensable accuracy of history at ticks level is just ridiculous - simply because such an "objective history" at such a small level does not exist. It turns out to be something analogous to the Heisenberg principle. Accurate history is a myth, if only because it is inherently statistical. In reality, there are too many factors that could change it quite significantly.
Another important point: the "true history" does not contain complete information that the creator of the system would like to have. Where in this history is the data about environment parameters, which (especially now) are changing very dynamically? And where is it reflecting the behavior of brokerage companies that return "trade flow is busy" or failing to open an order within several minutes on a calm market due to requotes?
My point is the following. The "fuzzy" reality (for example "EURUSD exchange rate at a given moment in time" - is in fact some random process with many actual realizations existing simultaneously) causes quite serious doubts in the expediency of storing "the exact tick history". And if there is a possibility in qualitative modelling of this history, it is better to use it - instead of storing a huge amount of information about a single realization of this process, which supposedly was in the past.
P.S. In the same thread I saw Renat's suggestion:
By the way, there is an idea to add requotes to the tester and serious slippage on movements.
Oh, how interesting. And the models have already been developed, Renat?
you're right Alexey... everything is as you say... and the quotes we see are indicative, and trades often do not follow them...
I am ready to agree with the developers that for testing purposes the suggested algorithms are quite reliable in modeling price behaviour...
But what should Prival do? I don't need it yet, but who knows, he may need it someday...
Renat писал(а) >>
Being in a society of technicians, play by the rules of technicians.
...I am sure that if you try to look for evidence, you will quickly retract your words.granit77 wrote >>
Since you are so offended by this, I will clarify my point to make it clearer that this is not a complaint about MT, but a subjective opinion.
Any simulation is a priori different from the real process and these differences may affect the result.
Therefore, it is logical to strive for bar-based strategies rather than getting into the search for flaws in tick modelling.
In this case, the testing is performed using real historical data and there is no room for doubt.
Victor, a couple of words in defense of the simulation algorithm:
Not so long ago, succumbing to the new trend of pipsing, I wrote a couple of TSs.
In the Strategy Tester and on the demo it is even OK with targets exceeding the bounds of pipsing.
On the real, I have problems in the constant struggle with the broker. But that is not the point.
EAs trade on M1 TF. So running on demo (where the broker is honest),
coincides 1:1 with running in the tester in "all ticks" model, where ticks are modelled.
I don't need it yet, but who knows, maybe I will need it someday...
Prival needs to get away from theoretical maths and get closer to practice.
Theoretical delusions are only good when they translate into results in a given topic. So far all I hear over the years is the standard attempts to dig something out of the tick noise. And it's all the pipers who get carried out of the market when trying to apply their "strategies" in the real world.
We simulate the tick development of bars qualitatively and accurately enough. We eliminate unnecessary ticks (often sawtooth movements of +-peeps) that do not affect the result. We do the practical work very well.
Every generation/wave coming into the market makes the same mistakes. Burrowing into ticks and trying to find the truth there are standard rookie mistakes. People realise that analysing the market and doing research is too difficult. People want to obtain everything quickly and immediately, without straining, and it is a direct way to mindless pipsing on the noise.
One is enough. Recorded tick history, as I wrote above, for two years (since January 2007). A very convenient way (as one would like) to retrieve them via API. Publicly available.
There is absolutely no tick history information on the above link. The link to the api is not "publicly available tick history".
I have no doubt you haven't even used that api in practice.