Dear SEC: Beware the Law of Averages
At last week’s STA Washington DC Market Structure Conference, SEC Chairman White gave a speech titled Focusing on Fundamentals: The Path to Address Market Structure (we sent out the text to you last week). The speech was very well-received by the industry in attendance. It began by highlighting that the success or failure of modern markets depends on investors and public companies, and she went on to note that the number of US listed companies has declined from 8,000 in 1997 to less than 4,900 today, despite rising price levels. Today’s fragmented and complex market structure undoubtedly has played a major role in that decline, and it has been a source of worry at the SEC well before critics made it an issue.
Consider that even before incidents such as The Flash Crash:
– In January 2010 the SEC published its Concept Release on US Equity Market Structure, and asked for comments. – Barely just two years since REG NMS had been implemented.
– In October 2013, nearly 4 years after those issues were highlighted by the SEC, markets are even more complex and fragmented, with little substantive action by the SEC, save for reactionary band-aids.
– With all today’s talk about “complete and holistic reviews”, we will wait even longer than the 4 years that we have waited thus far.
The HFT and exchange lobbyists have certainly gummed up the process, and have no doubt earned their pay.
Chairman White went on to say the she will focus on three fundamentals:
– Operational integrity.
– Identify and test long-standing market structure assumptions.
– Decisions must be based on evidence, and not self-interest and anecdotes.
That last fundamental lays the groundwork for some MIDAS-related data that the SEC plans to release to the public this week:
“Part of this initiative will be to disseminate data and related observations drawn from MIDAS that address the nature and quality of displayed liquidity across the full range of U.S.-listed equities –from the life-time of quotes, to the speed of the market, to the nature of order cancellations.”
Chairman White makes it a point to illustrate that MIDAS has helped them see that nearly 67% of stock orders rest on the book for greater than half of one second (or that a third of all stock orders are cancelled within a half second) on average, and that fleeting liquidity is therefore not so fleeting:
“These findings not only provide an empirical basis for measuring and tracking the speed of today’s markets, but also suggest that even short-lived quotes are generally accessible by at least some traders.”
However, there is something funny about using plain averages and making assumptions based on them. We would like to offer you an analogy.
Imagine that someone asked you to ascertain if there was a traffic problem in Time Square NYC, and you collected data on cars flowing through the area. Imagine there being very little car traffic between 12:00 midnight – 5:30 am. Would an average car flow rate that included that 12:00-5:30 am “dead period” be distortive? Would it be relevant in helping you arrive to a conclusion about traffic conditions at Time Square during rush hours?
At the same STA Washington Conference, Manoj Narang, developer of the SEC’s MIDAS system, made the point that regulators need substantial expertise and skill interpreting the data – and not just possessing the systems. Maybe he can advise the SEC against drawing incorrect conclusions from correct data, and the appropriate use of averages. While we eagerly await the SEC release of MIDAS data concerning order cancellations and “fleeting liquidity” later this week, we hope to see the following:
1) That the SEC did not just take average cancellation rates on the limit order book across the entire trading day.
2) That the SEC looked at cancellation rates at each and every 1 minute time interval throughout the day. The data might show that rates at 9:31 am might differ from rates at 1:00 pm and the close.
3) That the SEC looked across market cap and price levels, with appropriate data for each. An order cancellation rate on BAC is different from an order cancellation rate in ZQK. An order cancellation rate in PCLN is different than one in T.
That the SEC has taken the first step and licensed Tradeworx’s MIDAS software is a good first step, even though it does not take into account where orders travelled on their way to a final execution, and even though it does not take into account the relatedness of dark pool activity to the lit markets. Let’s just hope that they do interpret data appropriately, and not draw conclusions based on simple averages. Such conclusions might turn out to be quite average in quality.