From theory to practice - page 20

 
Alexander_K:

Vladimir - look at the pictures I have attached above on the subject. This is the actual flow of quotes from the ECN account.

Here are the statistics:


How can it be that Nikolay writes about average frequency of ticks coming once in 3 sec, and I got the average = 3 sec. We are absolutely strangers to him personally, and the data are the same?

Besides, there is some dip around 3 sec. as if the DC is deliberately cutting out the data.


This gap is artificial. I wrote about filters. Imagine that brokerage company receives information from all over the world, thousands of deals in a second, brokerage company aggregates them and generates one tick. But in addition to the fact that the brokerage company server should process the external liquidity, filter it and send quotes, the quote should hang for some time to avoid requotes.

That is, on the one hand, quotes should be marked as frequently as possible due to competition, and on the other hand, as seldom as possible in order to give an order without requotes.

Apparently, brokerage companies set filters for such frequency, and everything would be fine if it is stable, but it jumps depending on the market activity.

 
Nikolay Demko:

Just awesome!

Respect!!!

 
Vitaly Muzichenko:

Just awesome!

Respect!!!


Thanks to the unknown moderator for keeping an eye on the thread too.

I would have torn it down myself after a while so as not to clutter it up, but it's fine as it is.

 
Nikolay Demko:

Congratulations you have discovered the forex sessions )))

The first little rise at the beginning of the chart is Sydney (Australian session)

The second also not very big is Tokyo (Asian session)

At the end of the Asian session Europe opens and for about an hour they trade together.

And finally the last peak is the end of Europe and the beginning of the Americas.

Thanks for the verbal explanation of the identified quantitative properties. USDCHF is not the only instrument which has the dynamics of A(i) "repeating" in the course of a day. Different symbols reveal not only session boundaries but many other characteristic features of the rate movement during a day, which I do not know how to name, although I do not need to. We expect in 30 minutes the activity to increase 6 times, and we expect it. Here, for example, is a picture of the same minute activity over the same period for the other three instruments. You can see that the MXN has gone from being the most inactive in the morning to becoming the most active of the three. And the JPY is the opposite.


I am posting this figure here in hope that the formula given by Alexander for the moving average weight w=s^2/[2*sqrt((s^2+x^2)^3)] can be modified to take into account not only s value, but also the dependence of this value on time of day. This dependence, shown in the figure, is also one of the statistics. However, it is not scalar, but e.g. vector with dimension 1440.

P.S. This method of identifying statistical properties can also be applied to weekly activity. There are patterns there as well. But what appeals to me most is that the square root law is satisfied. If we measure activity on one timeframe, we don't need to measure it on all other timeframes, the conversion error is about 20%.

 
Nikolay Demko:

So, on the one hand, quotes should be marked as frequently as possible due to competition, and on the other hand, they should be marked as rarely as possible in order to give an order without stumbling on requotes.

For brokerage companies, slippage in order execution is an alternative to requotes.
 
Vladimir:

Thanks for the verbal explanation of the identified quantitative properties. USDCHF is not the only instrument that has the dynamics of activity A(i) "repeated" during the day. Different symbols reveal not only session boundaries but many other characteristic features of the rate movement during a day, which I do not know how to name, although I do not need to. We expect in 30 minutes the activity to increase 6 times, and we expect it. Here, for example, is a picture of the same minute activity over the same period for the other three instruments. You can see that the MXN has gone from being the most inactive in the morning to becoming the most active of the three. And the JPY is the opposite.


I am posting this figure here in hope that the formula given by Alexander for the moving average weight w=s^2/[2*sqrt((s^2+x^2)^3)] can be modified to take into account not only s value, but also the dependence of this value on time of day. This dependence, shown in the figure, is also one of the statistics. However, it is not scalar, but, for example, vector with dimension 1440.

P.S. This method of identifying statistical properties can also be applied to weekly activity. And there are patterns there too. But what appeals to me most of all is that the square root law is satisfied. If you measure activity on one timeframe, you don't need to measure it on all other timeframes, the conversion error is about 20%.


Again, Forex is looking at futures, therefore Yen is relatively more active in the Tokyo session, which is the most interesting for it, and Mexican is active in the American session, due to the local exchanges where futures are traded.

If someone asks me why forex looks at futures and not vice versa, it's an empirical fact.

And futures in turn look at options, so the main battles take place near option levels.

And by the way yes, I tried to put weights on bars depending on time of day, the chart becomes more normal, but there are still fat tails (I don't know how to say it, the distribution changes towards normal, but of course it does not reach it).

And then, it is not clear how to trade it, that is why I gave it up.

ZS I don't know how to express the idea correctly, if we assign weights to increments over time, then levels float, and levels are important in forex.

 
Alexander Sevastyanov:
The brokerage companies have an alternative to requotes - slippage during trade order execution.

The brokerage companies use everything: quote filtering, requotes, slippage, a little bit of everything.

It all depends on the server settings of a particular brokerage company.

 
Alexander_KMy draft of the dynamic model in the Vissim system for the EURJPY pair.

Alexander, don't you think your approach is overcomplicated? The same thing can easily be achieved in much simpler ways.

Look, here is a primitive system known to traders as "Bollinger channel" - a regular SMA and regular deviations from it, about 4 RMS. It gives roughly the same results as yours - trades are very rare, almost all to the upside. Same EURJPY, same historical range, trades in the same places.

And this is without any ticks, on 5-minute bars. The test is performed by opening prices, i.e. one tick in 5 minutes is taken. Without any speculation with sampling frequency.

Without any Fokker-Planck, Student's, and Vysokovsky-Petunin. Without any physic-mathematical pathos at all.

Such things have long been known among algotraders, no America has been discovered here. Everything here is extremely simple, i.e. the potential for improvement is huge. But of course the history test says nothing about future success.



 
Yuriy Asaulenko:

To paraphrase Winnie the Pooh - the Grail is such an object, that if there is one, it is immediately gone.

If several such Grails are placed on MOEX they will simply stop working and the Grail will disappear - there is not enough liquidity in the cup for several Grails. My guess is that it will last a bit longer on DC Forex.

Well, greed should be kept in check, then everything will be all right)) But DC Forex has another problem - they don't pay out the money.)

 
Dmitriy Skub:

Well, greed has to be contained, then everything will be fine)) At "DC Forex" another trouble - the money will not be paid)


Sounds like "I'm not going to do it because according to the second law of dermodynamics we all will die anyway" ))))

The first thing to do is to write a grail and then think with what means to squeeze our own money out of brokerage companies. For example, we may set a grail in a few brokerage companies where they may pay out a bit, take a bit from each, etc. There are plenty of strategies.