A preliminary white-paper written by Arne Breuer, Hans-Peter Burghof, and Julian Stitz was sent to me by Nick via Twitter. The paper was written in January of last year, addressing data from 02/22/2010 to 2/26/2010 and the topic involved is more than relevant 16 months later.  The main study is the lifetime of a limit-order that is sitting on the order book at the exchanges. Recognizing the need to analyze HFT behavior on the nano-second level shows that the researchers understand how fast and complex these systems are required to be, thereby enabling them to offer insight into the effects on the order book caused by high-frequency trading.

The authors have found evidence (which many of us already knew) regarding the increase of HFT activity in ETF’s over stocks since ETF NAV is easily calculated where equity fair value is more difficult to determine.  At any given millisecond an unmentionable number of algorithmic strategies are able to calculate the price of the ETF given the current price of its constituents.  By gauging the order books, these algo’s are able to determine the level of participation in a given ETF.  Since more limit orders means quicker changes in market conditions, these algo’s must be able to perceive changes, calculate actions and adjust/send/cancel their orders rapidly.

For proof of this heightened activity, the authors display this chart whereby the dotted line represents the average limit order revision times for ETFs and the solid line is stocks.  You can clearly see that more noise and more algo’s are in ETF’s than stocks as displayed through the increasing revision of orders.  This jives with the notion HFT are more active in structured products

Consider this nugget from the paper:

“[HFT are able to] calculate the optimal limit order for the then-changed order book. For the deleting market participant, the order-revision time can be infinitesimally small or even negative, if the new order is routed more quickly than the deletion message.  Of course, only high-frequency engines can perform actions in such short intervals.  Even though human traders might know exactly the properties of their next limit order they plan to place after they delete their old one, they will not be able to place it in a matter of milliseconds, probably not even in a tenth of a second”

Many times the authors reference the flash crash, comparing the periods of increasing limit-order revisions and highlight that though in the morning most behavior was “normal”, meaning fucked up as usual, until 2:45 rolled around when the limit-order revision increased rapidly, and destroyed asset prices as the market moved so fast, most machines just shut off.

The authors finish with a focus on May 6, 2010.  Readers are advised to scour Nanex’s data trove (here, here, here, here, and here) before going any further into the flash crash if they have not done so yet.  Each line on each graph represents a 15-minute time interval which is explained in the caption:

Changing and adjusting limits orders can be deemed as quote-stuffing since so many cancellations occur on the nano-second level driving exchange processing down and slowing down other algo’s that now must read and process these new cancelled orders, placing them at a “small” time disadvantage.

The dynamics of the market on a daily basis for active managers has changed dramatically and for human traders to compete they must be well aware of these dynamics.  The money that is now in actively trading structured products has no concern of the fair value of the underlying assets.  This is why monitoring the activity of the derivates allows some predictability of the underlying, much in the same way the e-mini SP 500 sell-off on May 6, 2010 lead the SPY derivate then the individual companies. Goodbye value based market, hello speed based market.  #GoodLuckHuman

Final Note:  The authors used data provided by NASDAQ with time stamp precision of one billionth of a second.  The SEC determined in their flash crash report (which has been recorded for history on the VPRO Money and Speed App for iPad) that looking inside of one second is useless and will create too much noise thereby distracting from their investigation.  That’s why I call them Phantom Regulators.

 

Tags: , , , , ,