Dust flux, Vostok ice core

Dust flux, Vostok ice core
Two dimensional phase space reconstruction of dust flux from the Vostok core over the period 186-4 ka using the time derivative method. Dust flux on the x-axis, rate of change is on the y-axis. From Gipp (2001).

Monday, July 11, 2011

More evolutionary adventures in HFT

The version of my recent post that appeared on Zero Hedge had this article from Nanex appended to it.

I don't disagree with the gist of the Nanex article and freely admit that they have far more experience in these matters than do I--furthermore, they have more timely access to much better data and have more computation power than I command. I point all this out simply to say that I do not claim to have written that portion of the post--the decision to append it to my article was made by the operators of Zero Hedge. Their website--their decision.

Now about that Nanex article . . .

CQS is on the losing side of history. The ability to increase the ability of hardware to deal with a volume of information pales in comparison to the ability to generate more noise by digital means. In fact, given the rates of increase proposed by CQS (33% last week, another 25% in October), I would guess that CQS is increasing their capacity several orders of magnitude more slowly than the rate at which algo traders can increase their data transmission.

From Nanex:

The first chart shows one micro-burst of activity as it occurred on July 5th, 2011, which is the first day after CQS capacity was increased by 33% to 1 million quotes/second. The second chart shows how this same data might look if we imposed the capacity limits that were in effect prior to this trading day. 
As you can see, the delay is significant. If this activity level occurred using the limits that existed on July 1st, the delay would have lasted approximately 200ms. We came up with three possible explanations:

1. July 5th just happened to be 33% more active than any trading day in history.2. The capacity limits from before July 5th were at least 33% too low.3. An algo is testing how much more quote noise it needs to generate to cause the same effect as before.

Obviously #3 is the case. The algo traders are merely toying with the authorities. No matter how much they increase the capacity of the lines, the algo traders have the ability to immediately saturate them and create latency on demand. 

The only possible fix is placing some sort of limit on the life of an offer.

No comments:

Post a Comment