The Truth Behind Broker Routing: Don’t Believe The Hype

When we have asked most traders what their ideal liquidity seeking algorithm does, we usually hear something like find the contra side of an order as quickly as possible, preferably without moving the price too much. If a trader was working an order manually-that is, without using an algorithm, liquidity seeking would also be described in terms of urgency, but, a fairly typical answer for a non-urgent order would be, check the blotter scrapers for naturals, then find active contras in dark pools, and finally clean up the order by crossing the spread and taking in lit markets.

If buy-side traders were constructing liquidity seeking algorithms, this is how they would behave-and the best algorithms feel like an extension of a good traders thought process. Unfortunately, as many of us know (and a recent SEC fine confirmed), just because an algorithm should behave this way doesnt mean that it will.

A Real-Life Example of Bad Routing

Here is client data that shows a broker’s conflicted routing strategy.

In this example, 3,100 shares were executed in Venue A as a result of sending 1,200 shares (fully filled) and then 1,900 shares (fully filled) 2 milliseconds later.

The concerning part is that19 routesto other venues took place before the algorithm went back to Venue A. Of these 19 routes,7 were to the algorithm providers own venue, which resulted in only 100 shares executed.

Later in the routing sequence, even after getting full fills back at Venue A – the router still decides to go back and check their own pool and a new venue (Venue H).

All things being equal, executing in fewer places is preferable to executing in more places. Sources of liquidity should be tapped until they are dry, especially if that liquidity can be accessed without adverse price movements.

So clearly this type of routing should be a concern to all buy-side traders and requires analysis to uncover.

A Deeper Dive

For a deeper look at this brokers routing behavior, we analyzed 3 months of this client data to compare the brokers routes (and fills) to its own venue versus to Venue A. The X axis is a measure of fills and the Y axis on this chart is a measure of routes. The data is scatterplot of all the routes – blue dots represent routes to the brokers own venue and red dots represent routes to Venue A, which provides a large amount of liquidity on a very small % of routes.

The concentration of blue dots between 30-60% of routes shows that the broker accesses their own pool often, even though there are rarely enough fills to support this activity. The red dots are almost always below the 5% value on the Y axis (Routes), even when Venue A is providing 20% or more of the fills. The chart clearly shows that there is a very weak relationship between fills and routes,the opposite of how we would expect a liquidity seeking algorithm to behave.

Also, we would postulate that if the routing sequence for this broker were to swap Venue A for the brokers own dark pool, that this client would find more liquidity with less information leakage – which would be in the clients best interest.

What Can the Buy Side Do?

Algorithms like the one described earlier have poor venue level performance, a negative impact on performance, and leave traders frustrated. Brokers explanation of complex heat maps and dynamic order routing are rarely supported by data, and data analysis is the only way the buy-side can truly understand, and correct, the nature of inefficient broker routing.

We often read self-serving articles about the demise of the buy side trader, with grand quotes such asA decade from now, maybe sooner, buy-side trading desks will be staffed almost exclusively by technologists, according to a top trading technology executive

We disagree. The current structure, where inefficient algorithms are prevalent, makes an educated buy-side trader more valuable than ever. It is up to the buy-side trader to:

  1. Help get the data from the broker (many times through their relationship with the broker)
  2. Help analyze the data to spot differences between their trading objectives (which are unique to them) and the broker’s routing decisions.
  3. Most important is to use the analysis to hold the broker accountable to make the proper choices, validated by data, to how they handle orders on behalf of the client.

This needs to be done by the trader — we dont expect the broker to respond to an algorithm making these requests!

Babelfish Analyticsis an US institutional equity trading venue and routing analytics service