Discussion of article "The algorithm of ticks’ generation within the strategy tester of the MetaTrader 5 terminal" - page 16

[Тимур]

Rosh:
Depending on the number of ticks in a given minute.

And if there are 70 ticks in a minute bar, which model will be chosen?

[Rashid Umarov]

tima_btl:
And if there are 70 ticks in a minute bar, which model will be chosen?

The article talks about it:

Tick generation is performed by reference points
.

Intermediate ticks between reference points are generated according to the following rules:
.

• If the number of ticks is greater than the number of points between anchor points, a "sawtooth" is generated.
• If there are enough points between the reference points, a linear sequence of ticks is generated.

[hrenfx]

I assume that the developers did some experimental and research work:

1. They recorded real ticks (of course, the reality of ticks on FOREX should be spoken about with great caution) and OHLCV+Spread M1.
2. We modelled ticks from OHLCV+Spread M1 and compared them with the recorded ones.
3. As a result of the comparison, some statistical characteristics were obtained (sampled expectation of the difference, its RMS).
4. Based on the comparison, appropriate conclusions were made about the confidence level of the selected tick model.
5. We left the golden mean between the best in item 4 and the least resource-intensive one.

It was logical to approach the choice of the artificial tick model in this way. And all this can be done independently:

1. Everyone can now get the statistical characteristics (p.3) of the modelling that works now.
2. Everyone can develop his own tick modelling algorithm and show its statistical characteristics.
3. If the model turns out to be better than the official one, the developers can adopt it.

That's why the following seems not silly:

1. Developers announce an open-ended contest for the best model of artificial ticks.
2. The prize is paid for each, for example, 5% improvement of statistical parameters of the new model in comparison with the previous one.
3. Upon receiving the prize, the contestant gives all the rights to the algorithm to the developers.

Several "birds" are killed at once:

1. No "fog" of rumours about the chosen model.
2. Open desire of the developers to improve.
3. Understanding that one of the best possible modelling models is being used.
4. Open clear statistical characteristics of the current model.

It should be understood that excellent tick modelling is not a "panacea". Market order slippages, order execution model (ECN, STP, ECN/STP) and other factors will offset the excellent properties of the chosen model. It is obvious that MT5 is not designed for high-frequency trading on various kinds of markets. Therefore, some "thick-skinnedness" of tick modelling justifies itself in 99% of cases of trading on the retail market.

P.S. In fact, the contest approach can be used not only in the case of tick modelling. But also in other areas. For example, improving the compression of quotes history.

P.P.S. All of the above concerns modelling of ticks for one financial instrument. Modelling ticks for several financial instruments at once is much more complicated because of synchronization, arbitrage and other issues.

[---]

a good idea, and a sensible one at that.

[Тимур]

Rosh:

The article talks about it:

Tick generation is done by reference points

Intermediate ticks between reference points are generated according to the following rules:

• If the number of ticks is greater than the number of points between anchor points, a "sawtooth" is generated.
• If there are enough points between the reference points, a linear sequence of ticks is generated.

Can you tell me more about how the "sawtooth" is generated?

For example, there are 10 points between the reference points. Total number of ticks is 50, how will ticks be generated in this case?

[William Kreider]

I'm pretty much confused on this Algorithm. I understand parts of it and then other parts i'm don't.

It looks like in regards to support points it basically takes the Volume of the historical bar and if it's higher than 11 (11 being the max number of support points) then use 11 however if this is not true then what is the formula for computing the number of support points. 

Better yet any more material on this would be great. I've beat google down to within a bit of it's binary life and have only been able to come up with two documents pertaining to this 'Miracle' Algorithm. I don't mind reading documents

thanks 

[Ivan Kornilov]

Renat:

We don't plan to providetick history - it's technical suicide.

Dukasokpy why not provide it through its jForex, despite the fact that they all work on slower jave'e. Conclusion technically it is quite realisable if there was a desire - and unfortunately there is none.

[Renat Fatkhullin]

You forgot to point out that Dukas doesn't have even close to the technological infrastructure like in MT5. They go the cheap and unworkable way - they replace the whole solution with an attempt to provide an api.

We have ticks in raltime, generation of sufficiently accurate tick history - too. At the current level of technology, trying to provide deep tick for the mass market is suicide.

[hrenfx]

Renat:
At the current level of technology, trying to provide deep ticking for the mass market is suicide.

Once again it sounds bizarre, given the existence of free super-successful peer-to-peer network protocols since the days when MT4 wasn't even in the plans, let alone MT5.

The excellent ready-made free BitTorrent infrastructure has long been able to distribute huge amounts of data just for the mass market.

No one prevents a broker from posting in torrent networks a single updated torrent file with an archive of any quotes (even Level2) for any period of time.

P.S. At the moment the technologies of Dukascopy and FXCM for distribution of their tick history are very shortsighted and weak indeed. However, in the same FXCM-tester there is a standard possibility (via SDK - a natural excellent limiter from the weakly understanding) to use custom tick history.

P.P.S. Collecting a full-fledged tick history is a complex technological process. And despite this is only the basis of analysis. Then there are various own custom filters-conditions of history, taking into account individual peculiarities of order execution, obtaining tick history in real-time, etc. I.e. we are always talking about creating at least a custom tick history.

[Mykola Demko]

hrenfx:

Once again it sounds strange, given the existence of free super-successful peer-to-peer network protocols since the days when MT4 was not even on the horizon, let alone MT5.

The excellent ready-made free BitTorrent infrastructure has long been able to distribute huge amounts of data just for the mass market.

No one prevents a broker from posting in torrent networks a single updated torrent file with an archive of any quotes (even Level2) for any period of time.

P.S. At the moment the technologies of Dukascopy and FXCM for distribution of their tick history are very shortsighted and weak indeed. However, in the same FXCM-tester there is a standard possibility (via SDK - a natural excellent limiter from the weakly understanding) to use custom tick history.

P.P.S. Collecting a full-fledged tick history is a complex technological process. And despite this is only the basis of analysis. Then there are various own custom filters-conditions of history, taking into account individual peculiarities of order execution, obtaining tick history in real-time, etc. I.e. we are always talking about creating at least a custom tick history.

+++++

Renat, give up, you will be chiselled again and again anyway, and if you refuse, you will get blackmail.

The proposal is quite sensible, especially since downloading tick history can be made optional for now, and then you can completely switch to cutting history from ticks in the terminal, as you are cutting it from M1 now.

ZЫ But you will be able to put one more star on the shield "we have a full-fledged deep tick history", maybe for someone it is really important!!!!