Discussion of high-frequency trading on MT5 - page 86
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
Apart from the obvious problems, there are pitfalls.
Let's say the DC has slow (in terms of pings, execution, etc.) LPs. So, you cross-connect with the LP and send it your order. It processes it and sends it to, say, a retarded LP. All sorts of stories about filters for retarded LPs will be omitted.
What happens in the end? As a result, you may be ahead of other clients of LP with your cross connection. But this will not solve the problem.
It's better to think then to do. Not vice versa, as in this case.
Have already contacted the DC. They have a server at Equinix NY4. The latency on all their LPs is less than 1 ms as the LPs themselves are sitting there at Equinix NY4.
Equinix is a well known company and their data centreEquinix NY4 is a very well known place, almost the whole financial system has servers there and trading is done within this site.
He does the resting with his hands. His video is on YouTube
At the moment, the concept oflocal instantaneous prediction on the current state of the market, where pending orders will not help, andslippage + execution + latencyleads to a loss.
We have: a neural network (a committee of networks) based on multiple anomalies found over a year and a half of research, makes forecasts + a trading robot (primitive) trades.
We have to move on to the next stage:
Predictive neural network (multistage) + Neurodynamic network (automated trading machine that takes into account slippages and sets different dynamic lot sizes depending on the market situation)
In general to work with limit orders you should make a multistep interval forecast, we should obtain a highly precise forecast for 1 minute in 10 second steps.
I.e. forecast of Level 2 state by 10 seconds, from +10 to 20 seconds, from 20 to 30 seconds, etc. with accuracy of 1 point.
In this case we may operate both market and limit orders, in which case the slippage factor will be built into the model.
The server will have to be installed in any case.
Maybe I have missed something. I understood from your posts and some other that MT5 is useless for HFT?
How do you test the bot?
How do I test?
At the moment the concept oflocal instant prediction on the current state of the market is implemented, here pending orders will not help, andslippage+execution+latencywill reduce to a loss.
You set a Limit order at calculated price at the moment of forecast.
When it reaches the set price you close it in the same way. Yes, there may be problems at exit (limit may not work), then you use 0 as an exit point.
It did not come - remove it (there may be some variations when), wait for the next prediction.
At least the entry problem will be solved.
Wait to talk about server rent again. Just try to implement logic on limiters...
I tried it. I have. It's not executing at all.
Can you use the same networks to predict slippage? All the data for that is supposed to be there.
Question #2 -- can you set limits outside of the spread? e.g. buy limit above the ask?
The picture is depressing...
So the conclusion is that you have to build up either the execution or the thick-skinned strategy, or implement the STP scheme yourself. Like a bogatyr near a rock where it says scribe everywhere :)
Do you have access to the tape?
There is datafeed and tradefeed for real.
Do you use it for analysis?
About the glass. I recall that hrenfx used it, among other things, to estimate execution slippage.
Very productive practice, as a successful arbitrator in a slightly different industry.