A Giant Leap Of Speed

In the current arm race towards ever higher speed and ever lower latencies, new ultra-fast machines are just around the corner.


The name is FPGA, Field Programmable Gate Arrays. This techno-sounding name is a technology, used by the military for quite a few years already, which allows burning programs directly on a chip.


This can make any process hundreds of times faster in some cases. E.g., a FIX message could have all its components parsed at once, instead of through an iterative process.Deutsche Bank has just announced the use of such a technology for its equity trading machines. And as we all know, what happens in equities happens in FX, just a while later. I know of at least 3 banks which are currently looking at these technologies, although sometimes they just buy finished products from specialized (and expensive) suppliers.


FPGA techniques can have applications in both matching and routing engines, so we can expect blitz-aggregators any time soon as well.But what kind of speeds are we really talking about? Here are a few numbers:A new company I am in touch with, e.g., is setting up a matching engine which could potentially deal with 4 million trades per second, or a routing system able to deal with 50 million messages per second. A simple message parsing could take 100 nanoseconds. These numbers are a thousand times better than what most of the market deals with today. Yes Ma'am.

Incidentally, the above numbers are also better than most other FPGA speeds seen so far. You read it here first. Call me for more details.


Ok this is all good, but do we really need all these goodies? I believe we do, as the quest for good prices will go on no matter what.


Already, the probability that a good price issued by a bank in New York is traded on by a fund in London is similar to the probability of a cod in the Atlantic to die of old age.Algo-FX markets are similar to a restaurant, with diners sitting close to or far from the kitchen. When a waiter gets out of the kitchen, good dishes get grabbed first. If you are sitting at the back, you get to eat the leftovers.


What I mean by this is that your TCA will increase, and the cost of your TCA must be compared with the cost of these new technologies, which is a quantitative way of deciding whether you want them.

Some banks have already decided. Some funds too, from what I hear.

Write a comment

Comments: 7
  • #1

    Xavier (Wednesday, 06 July 2011 18:00)


  • #2

    starbrightpartners (Wednesday, 06 July 2011 22:13)


  • #3

    lefildargent (Wednesday, 20 July 2011 08:09)

    Have a look here http://www.jimdo.com/app/forum/viewtopic.php?f=3&t=11130&p=45806#p45806
    If it can help.

  • #4

    CaptainSlow (Sunday, 18 September 2011 13:02)

    And still, some of the best performing funds (and by this meaning people able to add absolute added value = returns * AUM) are staying away from anything that has to do with high frequency.

    The restaurant is a good analogy, if you add one detail: they are all waiting for the freshest burger in a McDonalds. In the Fat Duck, no one bothers to be close to the kitchen. ON THE CONTRARY. In the investment world, you could argue the same. Smart money shuns the short term noise!

  • #5

    CaptainSlow (Wednesday, 05 October 2011 22:32)

    No reaction to my smart comment???
    Probably means that the McDonalds eaters are sick in a ditch...

  • #6

    Xavier Alexandre (Thursday, 03 November 2011 11:25)

    Sure let me reply to this and sorry for delay. I actually do not agree completely with you. I do know some high frequency funds that do find great alpha in that sector. The trick is to eliminate the noise but there are techniques to do that. But there is a fine point. Reducing the noise uncovers arbitrage opportunities where some funds tend to concentrate, giving a bad name to the industry.

  • #7

    Beats By Dre Cheap (Monday, 12 November 2012 07:28)

    I saw this actually excellent post these days!