Good HFT, Bad HFT: Dividing Line between Predatory and Passive Strategies
This is the first in a series of blog posts on high-frequency trading (HFT) and will examine an aspect of HFT that should be of concern to investors. The next post will discuss one trading venue’s attempt at addressing these concerns.
The much-maligned practice of HFT is an example of how an umbrella industry term — negative in connotation, deserved or not — can serve to cloud the debate. There are many types of HFT, from passive forms such as arbitrage or market-making, which are typically found to have benign or positive impacts on market quality, to more complex strategies. In fact, markets are completely dependent on automated trading now, much of which qualifies as HFT in the sense that it is fast, computer-driven, and fleeting, even though it is largely passive and benign.
The more contentious debate revolves around so-called predatory HFT strategies, even though the distinction between passive and predatory HFT is rarely made explicit. By ignoring this distinction, critics of HFT often weaken their arguments by allowing the other side of the debate to cite market quality improvements that are actually associated with a completely different set of (passive) HFT strategies.
The most philosophically troubling HFT strategy is very similar to the longstanding problem of front-running, and is referred to as latency arbitrage. While the traditional practice of front-running sees the broker trade ahead of his or her client and is illegal, latency arbitrage sees principal traders take advantage of faster connections to exchanges, relative to other market participants, and is not illegal. However, the practice is generally considered to be against the spirit of fair markets and typically manifests itself as ghost liquidity and price slippage, which are major concerns to institutional investors.
To illustrate latency arbitrage, consider an investor that splits a block trade and sends two buy orders to exchanges A and B where B is geographically further away from the trader. The trade reaches exchange A in, say, 1 millisecond and executes, while it takes 2 milliseconds to reach exchange B, located in the next town. However, the co-located HFT servers at exchange A receive notification of the first trade 0.5 milliseconds after the trade happens, leaving them with half a millisecond to act on this information before the second trade reaches exchange B. How could they take advantage of this?
One possibility is that high-frequency traders design algorithms to estimate the probability that the original trade is part of a larger block trade. If these algorithms conclude that a large block trade is being worked in the market, HFTs know to expect buy orders on exchange B. The HFT would then use its network speed advantage over the trader to cancel its sell orders on exchange B and post new sell orders at a higher price. If the HFT can send these amendments in 0.3 milliseconds, by the time the second trade finally arrives at exchange B, 0.2 milliseconds later, it will find that the price has gone up! The investor executing the block order will therefore experience either ghost liquidity, wherein posted sell orders on exchange B disappear before they can be executed, and/or price slippage (the trades are executed on exchange B at prices higher than the ones observed when the trade was initiated).
These issues raise the ire of institutional investors in particular who are more likely to lose large sums of money than retail investors due to the volumes traded. In response, several former executives of Royal Bank of Canada departed to set up the IEX dark pool that is designed to neutralise latency arbitrage. The way in which IEX evens the playing field is by placing a 38-mile fibre optic coil between its co-located HFT servers in Secaucus, NJ, and its exchange-matching engines in Weehawken, NJ. While IEX trades bypass the coil (contained in a space the size of a shoe box) and connect with other exchanges using low-latency links (the slowest and farthest being NYSE, 320 microseconds away), co-located signals suffer a 350-microsecond delay going in and out of IEX. In this way, a trade posted and partially filled on IEX cannot be front-run on its way to another exchange because the HFT signal will always arrive later than IEX’s signal.
In my next blog post, I’ll look at IEX more closely and describe a different feature of the trading venue that should concern investors.
If you liked this post, consider subscribing to Market Integrity Insights.
Photo credit: iStockphoto.com/xijian