The last time Hackerfall tried to access this page, it returned a not found error. A cached version of the page is below, or click here to continue anyway

Flash Boys in the US Treasury Market Medium

Flash Boys in the US Treasury Market

Every bank Ive spoken to recently has had the same complaint: The liquidity in the screen isnt real anymore. Traders used to be able to grab the entire bid or offer, but are no longer able to. The complaint (usually said while yelling at IT over the phone) goes something like this:

I saw $400M five years offered on the screen. I go to lift the entire offer and I only get $225M back. I just lost a $100k chasing the quote. Fix this f piece of s- technology!

The argument is a bit of a straw man in that it makes the assumption that it used to be just as easy to access the same quantity (reality: there used to be much less on the bid/offer) at the same bid/offer spread (reality: spreads used to be wider). However, it is true that the markets have fundamentally changed. The dominant liquidity providers have shifted from human to machine. A machine can change its quote thousands of times faster than a human can.

So whats actually happening when the trader sees $400M on the offer and goes to lift all of it? The first thing to understand is that the $400M is not a quote from a single venue, but an aggregated quote of multiple underlying venues. It is presented as a single quote to simplify the number of boxes the trader has to look at in order to understand the market depth.

The order book that the trader sees

When the trader wants to grab all $400M, the smart order router underneath sends two individual orders: One order for $175M to Venue B and another order for $225M to Venue E. The typical result: The order that goes to Venue E is filled in its entirety and the order that goes to Venue B misses as the liquidity at that price point evaporates. All of this happens in the blink of an eye and its incredibly frustrating to the trader who is just trying to get their risk off. If the number on the screen isnt the real number then how can you trust the screen at all?

Behind the Curtain

Orders take a finite amount of time to go from the banks data center (which is in Manhattan or NJ or Connecticut or Chicago or some other place not at the Venue B or Venue E data center because that would mean youd have to put money into the technology budget) to Venue B and Venue E (which are in NJ). For nearly all banks this time is measured in milliseconds. Optimizing the speed of a network is about the last thing on the list of an overworked technology staff. The un-optimized network means that orders dont land at Venue B and Venue E at the same time, but multiple milliseconds apart.

In the scenario of the trader lifting the $400M aggregated offer, the first order lands at Venue E and has all $225M filled. There is no longer any liquidity remaining at that price level available at Venue E.

The first order lands at Venue E and removes all $225 on the offer

The trade information for Venue E gets disseminated to all the participants within a hundred microseconds through a fast market data protocol. Speedy market participants then race from Venue E to Venue B to either pull their own quote (if they were resting orders) or remove other participants orders at that price level. As you might imagine this is an all-or-nothing scenario and the advantage is to the fastest participant. Electronic traders spend a lot more time optimizing their network than banks do.

Traders who see the order book change on Venue E race to Venue B to pull their own offers and remove others

Finally, the traders order for $175M arrives at Venue B. The bonds offered at that price point are gone. The trader receives no fill and is forced to rest their order.

The traders buy order now rests on the bid

The trader now has to send another order at the next price level in order to complete the $400M order. As you can imagine, this could even take multiple price levels to achieve, as everyone now knows how badly the trader wants the liquidity.

Flash Boys

This is exactly the same phenomenon that Brad Katsuyama discovered at RBC, as described in Michael Lewis Flash Boys. He could no longer grab all of the shares available on the screen in a single trade and instead had to engineer a smarter order router to place orders more intelligently.

Why would the electronic traders who see trades on one venue race to another venue to pull quotes or remove liquidity? In addition to optimizing the network, electronic traders understand how participants interact with the markets pretty well. Its their job after all. If the entire stack on Venue E is removed, then it is highly likely the entire stack of Venue B will be removed either by the same trader or other electronic traders. With just a few days of market data you could compute that probability pretty accurately.

Some actual numbers

As the crow flies, its about 30km between the Equinix NY2 data center (where BrokerTec is co-located) and the NASDAQ Carteret data center (where eSpeed is co-located). If the speed of light is still the 299,792,458 meters/second from my physics courses, then that means trade information takes about 100 microseconds to go from one data center to another on the new microwave networks.

The difference between 100 microseconds and multiple milliseconds is an eon in the electronic trading world. An electronic trader could go back and forth five times if the differential is only 1 millisecond. The differential is usually far worse than 1 millisecond.

One final thought. The scenario described is for two venues. One of the more recent trends has been a proliferation of new inter-dealer brokers. As you onboard new venues you increase the number of networks where the arrival timing has to be managed. Mathematically, this makes it exponentially more likely that the trader will miss at least part of their order. Its no wonder that traders dont get the liquidity they see on the screen anymore!

Continue reading on