Algos Underperform in the Dark?


You may not be pleased to know that it is over two and a half years since the SEC issued proposals to “shed greater light on dark pools,” without action. SEC Chairman Schapiro introduced the intention to focus on dark pools in late October 2009 as follows:

 “We should never underestimate or take for granted the wide spectrum of benefits that come from transparency, which plays a vital role in promoting public confidence in the honesty and integrity of financial markets. Today’s focus on dark pools is just one part of our broader ongoing review of how the equity markets are structured.”

Thirty months have passed and dark pool usage is at an all-time high. As you all know, there are between 40 and 50 dark pools in the US equity markets. As you all know, the main reason for most of these pools is the serving of the owner’s economic interests (internalization and Alpha feeding), and not institutional customer interests. While some tend to trade blocks greater than 30,000 shares, most trade in the odd-lot to 200 share-size range.

While we have spent certainly a good amount of time talking about conflicted stock exchanges, a dark pool study we read over the weekend gave us a break from thinking about the stock exchanges. Informational Linkages Between Dark and Lit Trading Venues (Nimalendran and Ray, University of Florida) is a March 2012 study that uses confidential data provided to the authors by a large independent dark pool. We are guessing based on “clues” in the authors’ paper (a desk that works orders, large trading blocks, independent) that the supplier of the data is either Liquidnet or ITG, but this is just a guess. We digress anyways…

The data the authors examine includes three types of trades in their 100-stock sample:

– (1) Trades facilitated/worked by the crossing network desk (13%vol 36% of trades)
– (2) Trades negotiated by their customers (59% vol 1% trades)
– (3) Trades between a customer on one side, and “liquidity providers” on the other (28% vol 63% trades)


The study found that trades worked manually by the crossing network desk, where the CN employees had considerable discretion where and how to execute, transmitted very little data to the public markets (exchanges), and had very little impact. The study also found that institutions crossing large blocks on the crossing network, negotiating among themselves, also transmitted very little price-impacting data to the market place. However, the study did find that trades between customers and “liquidity providers”, constituting 28% of the volume and 63% of the trades, were the highest impacting and most damaging:

“These effects are greatest for trades involving crossing network members using algorithmic strategies to trade less liquid stocks, as opposed to trades in more liquid stocks, trades manually negotiated by members, and trades executed through the crossing network’s brokerage desk. This suggests that information in the crossing networks is concentrated in this sub segment (crossing network members using algorithmic strategies) of the market. The fact that these effects are seen on the quoting exchanges despite trades on the CN not leaving a traditional footprint suggest that either the informed trader on the CN or other informed traders with correlated information are concurrently trading on their information on the quoting exchanges.”

Let’s recap this study: Trades done in the dark which transmit the most information are algorithmic orders that are routed through dark pools indiscriminately. These trades are ones that shake the sweaty hands of “liquidity providers”, and account for nearly 2/3 of dark pool volume. Trades that transmit the least information are trades that are either controlled by the customers (negotiated blocks), or controlled by the crossing networks trading desk, where these employees know which pools are toxic and to be avoided, and who are directly accountable to their institutional client, and those trades combined are a little over 1/3 of dark pool volume.

This may make one re-think the use of algorithms that visit dark pools – especially as it has been nearly 30 months since the SEC’s proposed dark pool regulation without any action on their part.