Forums  > Trading  > algo trading frontier  
     
Page 1 of 1
Display using:  

nikol


Total Posts: 971
Joined: Jun 2005
 
Posted: 2020-03-19 11:43
Where is it?

Is it ultimately about speed (kernel level programming, FPGA's or maybe even AI/TPU's)?

doomanx


Total Posts: 57
Joined: Jul 2018
 
Posted: 2020-03-19 12:05
Assuming by the topics you mention we're talking about classic HFT here (latency arb and such). I think the frontier we're seeing is more an economics barrier; increasing competition demands more and more expensive infra while reducing profitability of the arbs that do exist. That's why you're seeing this consolidation at the top (Virtu/KCG etc.) and I think we'll continue to see more of this in the coming years.

There's a few firms without top 3 infra that compete, and what they are doing is essentially using your favourite term for forecasting (ML/stats...) to trade some uncertainty for latency. So you trigger faster but you might be wrong sometimes, with some clever modelling you (hope to) win via the law of large numbers. Still very high frequency, but have say 2-10 infra guys instead of 100.

nikol


Total Posts: 971
Joined: Jun 2005
 
Posted: 2020-03-19 12:19
Yes, I meant sub-millisecond domain.

Essentially, you say that "the frontier" means expensive infrastructure. I would expect so. But the question was more about specifics in tech and algo (type) space.

Of course, generic "specifics" )) . To not harm anybody.

> faster but you might be wrong sometimes, with some clever modelling you (hope to) win via the law of large numbers

That's the point. Time of reaction is a (measurable?) balance between diffusion uncertainty (~sqrt(Time of delay)) and approximation error (~1/Solution Time?). The snake has less IQ than the human, but has better chances to kill human without any tools.

doomanx


Total Posts: 57
Joined: Jul 2018
 
Posted: 2020-03-19 13:19
Sub-milli is not HFT these days but I get your point. I've been out that game for a few years now, but couple of things is people putting more complicated logic onto FPGAs and writing their own firmware/drivers for switches and such.

On your second point, true but the world 'reaction' implies you receive a stimulus to react to - the idea of forecasting the next tick is you don't need an update before you react, hence removing the dependence on a gazillion dollar network setup to get that quote first. You could decompose time to update quote = time to receive tick + time to react, forecasting allows you to reduce the first term without spending on infra.

ronin


Total Posts: 546
Joined: May 2006
 
Posted: 2020-03-19 13:33
HFT is about being first in the queue. If you are the first, you make money. If you are second, it's 50-50. Third or worse, you lose money.

And there are only so many queues.

So it's a bit like that African proverb - it's not about running faster than the lion, it's about running faster than the other guy.

Algo trading in general is a different story. All types of trading are becoming more and more algo.

So the joke parallel is like doing a PhD - every year you know more and more about less and less, and in the end you know everything about nothing.

"There is a SIX am?" -- Arthur

Its Grisha


Total Posts: 23
Joined: Nov 2019
 
Posted: 2020-03-19 14:16
> (kernel level programming, FPGA's or maybe even AI/TPU's)?

This is all at the point of colocation, but for latency arbitrage between exchanges, so much of it is about the network right? In the past year or two, there has been some very interesting information coming out about shortwave (AM) radio being beamed between US and Europe.

Sacrificing a ton of reliability and bandwidth for a tiny edge over fiber. Microwave is old news, and not going to make it across the ocean, so need to be using shortwave. Very cool engineering problem to be dealing with packet loss and such over 100 year old radio technology where the serialization latency can easily out scale the speed advantage if you are not careful.

Relevant links:
https://sniperinmahwah.wordpress.com/2018/05/07/shortwave-trading-part-i-the-west-chicago-tower-mystery/

https://stacresearch.com/STAC-Summit-13-Jun-2018-shortwave-trading

nikol


Total Posts: 971
Joined: Jun 2005
 
Posted: 2020-03-19 14:34
@Its_Grisha

Can quantum entanglement improve reliability of shortwave channels?
e.g.
https://arxiv.org/abs/quant-ph/9703013
and
https://www.sciencedaily.com/releases/2017/03/170329102514.htm



nikol


Total Posts: 971
Joined: Jun 2005
 
Posted: 2020-03-19 14:39
@ronin

>> HFT is about being first in the queue. If you are the first, you make money. If you are second, it's 50-50. Third or worse, you lose money.

Earlier (few months ago, somewhere in Sep19) in our discussion about number of allowed quote update frequency (reality in crypto exchanges) you said that I do not need very many... Or your notion relates to quote updates in Market Making ?

Ok. Lets broaden it a bit to 1 second scale. But what i really mean is frontier in inter-play between technological and algorithmic complexity.

Its Grisha


Total Posts: 23
Joined: Nov 2019
 
Posted: 2020-03-19 14:55
>Can quantum entanglement improve reliability of shortwave channels?

Had a look at those articles, not really qualified to answer what the barriers to practical use are. Concept makes sense though. Headlines would be funny if HFT firms were the first ones to use this mechanism in practice.

ronin


Total Posts: 546
Joined: May 2006
 
Posted: 2020-03-19 21:10
> Earlier (few months ago, somewhere in Sep19) in our discussion about number of allowed quote update frequency (reality in crypto exchanges) you said that I do not need very many...


Man - I have no idea what I said yesterday, let alone a few months ago...

But it does sound like something I would say. And I don't think it contradicts what I wrote here. If you are the first in the queue, you don't want to mess with the quote. If you are not first, there is no amount of messing with the quote that is going to help you.



"There is a SIX am?" -- Arthur
Previous Thread :: Next Thread 
Page 1 of 1