Measuring the impact of predatory high-frequency trading strategies on best execution starts by defining HFT. But the impact of competition for passive liquidity can be measured, along with the price impact. Analysis of Canadian equity data shows varying levels of predatory behavior among different groups of securities, and the results can be further aggregated by venue and broker.
Our understanding of the various activities of high-frequency trading has evolved to the point where we now speak of “good” HFT and “bad” HFT. The “bad” HFT is, in turn, often referred to as “predatory.” At the same time, a generally accepted definition of HFT remains elusive, whether “good,” “bad” or “predatory.” In this article we attempt to provide some definitions and, more important, some measurements using Canadian public equity data to measure the level of predatory HFT activity.
First, let’s provide a working definition of HFT: HFT comprises the set of strategies that profit by exploiting the inherent alpha in market structure.
While not everyone may agree with this definition, it does provide a starting point to embark on a quantitative study. Examples of various specific HFT strategies are provided in Table 1, below.
Next, let’s consider the “bad/predatory” HFT that has gotten so many people worked up, to the point of some calling the equity markets “rigged.” It is certainly not possible to know exactly what everyone means when they use the term, so here I will make some assumptions about what I believe they mean and then give the definition, and then we will attempt to measure the extent of predatory HFT.
Predatory HFTis the set of strategies that compete with (some would say “front run”) other active orders for consumption of the available liquidity with the goal of unloading the positions acquired at a profit after the market has moved favorably (due to the amplified impact of the other active orders).
For example, if a large consumer of liquidity attempts to access multiple venues that are displaying orders, then a predatory HFT strategy would attempt to access the remaining available liquidity and then offer it back to the original order at inferior prices (on Table 1, below, the predatory strategies would be latency arbitrage and order flow detection). The predatory HFT strategy would have the effect of amplifying the market impact of the active order.
Table 1. A taxonomy of HFT strategies is provided along with stakeholder perspective regarding whether the particular strategy is good, bad or neutral. It is clear that not only are there a wide range of HFT strategies but also varying perspectives from stakeholders regarding the strategies. (Source: The Hidden Alpha in Equity Trading, Oliver Wyman)
There may be other types of predatory HFT strategies in addition to what is described above, but we will focus here on the “front running” predatory HFT strategy defined above.
The data used consist of Canadian equity trades and quotes obtained from the information processor (IP) feed and archived by Market Data Authority. The trades contain identifiers for the buying and selling broker, as well as the venue of execution. There also are condition codes associated with each trade to allow for identification of odd lots and specialty crosses (among other things). The timestamps on the trades are centiseconds, while on the quotes, millisecond resolution is provided on the IP feed. Market Data Authority archives the data and all ticks for listed securities that trade on Canadian equities markets are available.
Board-lot trades that are executed during continuous trading hours (9:30 ET – 16:00 ET) are considered. Crosses, odd-lot trades and trades that occur when the market is locked or crossed are excluded.
Each trade is compared with the prevailing bid/ask prices to determine where it occurred relative to the prevailing bid or ask. If the trade occurred at the bid price or lower, then the buyer is considered passive and the seller is considered active. If the trade occurred at the ask price or higher, then the seller is considered passive and the buyer is considered active. Codes that identity executing brokers are provided on each trade, along with the venue of execution. In some cases, the broker codes indicate “anonymous” brokers, which is an option that any broker can utilize.
Events in which at least two active brokers executed trades having identical time stamps are identified by querying the data. Further filters are applied identifying events in which at least one active broker received executions on multiple venues (for example, by using a spray router).