Around half of all the buying and selling of U.S. shares is by computer algorithms that measure time in nanoseconds, or billionths of a second. Those high-frequency trading algorithms help make trading cheaper, indeed effectively free for members of the general public. But there are expensive speed races at the heart of HFT, races that reduce the savings to the public stock markets.
Equally worryingly, because these races take place in the public markets, there’s an incentive to avoid their costs by doing private deals off-market. This makes perfect sense for those involved. The trouble is that these deals are parasitic on the public markets: they use the prices in those markets to set the private-deal price.
And there is a risk that the parasite will devour the host. In May, Gary Gensler, chair of the Securities and Exchange Commission, told the House Committee on Financial Services that the proportion of U.S. stock trading that takes place on public exchanges such as the New York Stock Exchange had fallen to just 53% in January, down from around 60% only a couple of months previously. Although the exchanges have subsequently recaptured some of their lost market share, a renewed decline would threaten the capacity of public markets to set informative prices.
Brilliant work by University of Chicago economist Eric Budish and colleagues has revealed just how common speed races in public markets are. Using London Stock Exchange messaging data that are not normally publicly available, they have shown that the average stock in the FTSE 100 index experiences on average one such race every minute of every trading day. Given that hundreds of companies’ stock is traded in London, the speed races are essentially nonstop.
What gives rise to speed races of this kind? A good example is what happens when there is a price change in the futures contract universally known to professional traders by its ticker symbol, ES. In theory, the ES (which is traded in the Chicago Mercantile Exchange’s computer data center in the city’s suburbs, and not to be confused with Eversource Energy, a small company that trades on the New York Stock Exchange with the same two letters) tracks the S&P 500 stock-market index In practice, though, the ES often changes price before either the underlying shares or the SPY, the exchange-traded fund that tracks the same index.
An HFT algorithm trading shares and/or ETFs therefore has a considerable advantage if it receives ES price data even a microsecond – a millionth of a second – faster than its rivals.
In 2011, a fast trading system could react in 5 microseconds. By 2019, a trader told me of a system that could react in 42 nanoseconds.
Most U.S. share and ETF trading takes place in data centers in northern New Jersey, making speed in the transmission of prices from Chicago crucial to success or failure in HFT.
All high-frequency traders are aware of interconnections of this kind among the prices of financial instruments, and all know that the shifting balance between bids and offers in electronic order books is a useful short-term predictor of prices. And they know that everyone else knows this too. Winning often comes down to being fastest.
One of the first high-frequency traders I spoke to, in 2011, told me that a fast system was one that could react in 5 microseconds. When I met him again five years later, he said “now you have to be under a microsecond.” By 2019, a trader told me of a system that could react in 42 nanoseconds.
Today’s speeds can’t be achieved with standard computer systems. You now need the specialized silicon chips known as FPGAs, field-programmable gate arrays. You have to pay whatever it costs to site your FPGAs as close as possible to exchanges’ computer systems and for those FPGAs to receive the fastest possible data feeds. Fiber-optic cables, the standard as recently as 2010, are now far too slow. You need to use microwave or millimeter wave wireless links, and lasers on towers or tall buildings are now employed to flash stock-market data between the New Jersey data centers.
The trouble with speed races is not just what they cost, but that the same HFT firm or small group of firms can keep winning them, and in the long run that hurts competition. Over the past decade, I’ve watched firm after firm disappear, most prominently Getco, once a dominant presence in HFT. Only a couple of the HFT firms that have replaced them (XTX Markets and Headlands Technologies) approach anything like the scale of those that have vanished.
Furthermore, for an HFT firm that can win the technologically hugely demanding races in public markets, the role of being a wholesaler, privately executing “retail” trades by members of the general public, counts almost as a relaxing hobby. The two largest wholesalers, Citadel Securities and Virtu (both with roots in HFT), are responsible, Gensler reports, for more trading than any exchange other than Nasdaq. Citadel handles a whacking 47% of all retail orders.
Wholesalers pay brokers to send them their customers’ orders for execution. This “payment for order flow” sounds like a Mafia racket, but it actually works reasonably well for customers. It means that your broker can let you trade for free, and SEC regulations dictate that you have to receive at least the best price available at that moment on the public markets. Indeed, the wholesaler will normally give you a better price.
That, however, epitomizes the problem: an arrangement that is beneficial for those directly involved in can slowly suck the lifeblood from the public markets. This issue is manifest not just in the trading of shares but also, for example, in Treasurys. There, some HFT firms have partnered up with the big primary-dealer banks to privately execute the banks’ orders.
It will be hard for Gensler and other regulators to make progress unless they address the underlying speed races. Budish has a radical solution: replace today’s continuous trading, in which microsecond or even nanosecond advantages can be decisive, with frequent single-point-in-time auctions that incentivize competing on price, not speed.
The way forward is a controlled experiment in which the SEC tries this idea out on a limited scale. If the experiment succeeds, then perhaps the markets can reap the benefits of the automation of trading without the damaging side effects of speed races.
In March, the SEC approved a proposal from the Chicago Board Options Exchange to allow it to hold ad hoc stock auctions in parallel with continuous trading. The experiment that I would like to see would involve a more thoroughgoing change, but I hope that the SEC’s decision indicates a willingness to move in that direction.
Donald MacKenzie is professor of sociology at the University of Edinburgh and the author of “Trading at the Speed of Light: How Ultrafast Algorithms Are Transforming Financial Markets.”