A Flash Judgment

April 15, 2014

Flash Boys, the latest book from Michael Lewis (author of Liar’s Poker, Moneyball, The Blind Side, The Big Short, etc.), is a nonfictional account of the development of high-frequency trading (HFT) in U.S. equity markets, and of Brad Katsuyama’s quest to reform the system by creating a new trading platform, IEX, designed to resist the most damaging HFT strategies.

This commentary is not a book review, but a response – because I am mentioned in the book. My 2012 SEC comment and working paper, “Races, Rushes, and Runs: Taming the Turbulence in Financial Markets,” was considered by Katsuyama and the IEX team as they designed their new exchange. The paper proposes the use of a randomizing temporal buffer – a brief random delay on all incoming orders – to allow a buffered exchange to compete successfully with exchanges that are more exposed to HFT strategies. Here is Lewis’s account of the team’s reaction:

[O]ne professor suggested a “randomized delay.” . . . The Puzzle Masters [IEX's technical staff] instantly spotted the problem: Any decent HFT firm would simply buy huge numbers of lottery tickets – to increase its chances of being the 100-share sell order that collided with the massive buy order. “Someone will just flood the market with orders,” said Francis. “You end up massively increasing the quote traffic for every move.”

— Michael Lewis, Flash Boys p. 174 (62%)

Actually, I am not a professor; I’m a Visiting Scholar at GW’s Regulatory Studies Center. But I’m grateful for the promotion, and also appreciative of the fact that few other academic papers are even mentioned in the book. However, I thought then – and think now – that the “Puzzle Masters” were wrong about how a temporally buffered market would work. First, “massive” orders would not be monolithic; they would be broken into smaller pieces, each with its own random delay. Second, the system would not allow orders to be cancelled without also imposing a random delay on the cancellation, so that anyone “flooding the market” with exploratory sell orders would find those orders being crossed – i.e., being matched with the component parts of any buy orders with which they “collided.”

Crossing orders is exactly what a financial exchange is supposed to do. Could an HFT firm nonetheless use this flood-the-market strategy to uncover information about the existence of a large unfilled supply or demand? Sure, if the HFT firm was willing to accept the resulting trades. But the information it thereby gained about the state of the market would be partial, and would emerge at a pace that provided little advantage to speed-based trading strategies. The randomizing buffer system is not intended to hide information indefinitely, nor to prevent any market movement in response to large orders; it is simply intended to dampen the latency arbitrage that depends for its success on high-cost high-speed strategies.

I did meet with Brad Katsuyama as he developed his trading platform. In our conversations he was unsure whether the Puzzle Masters had uncovered a real vulnerability, but he gave three other reasons why IEX decided not to use what I had proposed: first, a random delay would sound risky to institutional customers that crave certainty; second, there would be challenging regulatory obstacles; third, the system would need to achieve substantial trading volume before it could demonstrate efficiency advantages. I agreed with all of those.

Meanwhile, in London, a group of banks was searching for ways to improve foreign exchange trading, which is far less regulated than U.S. equity trading and therefore more open to innovation. They built a randomizing delay mechanism into a new currency trading platform called ParFX. Since it first began trading in April, 2013, the company reports that the randomizing buffers are working as intended. Other exchanges are beginning to adopt the same technology. One thing that is helping randomizing buffers gain market share: Australian banks are demanding it. Without it, due to their geographic location, they are always on the wrong end of any latency race.

My paper is critical of HFT, but does not recommend regulating it. Instead, I conclude that randomizing buffers, and other technical improvements that mitigate HFT, will emerge and compete successfully: “Competition will be the best test of theory.” Cliff Asness and his colleagues at AQR Capital Management wrote a prominent defense of HFT in the Wall Street Journal (“High-Frequency Hyperbole”) last week. They conclude: “But let’s allow the issue to be decided by open competition, not by politics.” In that we have common ground.