The ability to determine the source of returns in a high-frequency strategy is an important part of portfolio and risk analysis. The authors present a high-frequency equity attribution framework that allows for gains or losses to net market exposure, multiple-dimension bucket exposures, long and short stock-specific effects, market-making effects, and exchange-traded fund trading activities.
Complications arise when a traditional attribution framework, which analyzes long-only positions and returns relative to a benchmark, is applied to high-frequency strategies. A portfolio that uses these strategies can have its value fluctuate substantially during the day, and the exact timing of a particular trade can affect the strategy’s outcome. Additionally, exchange-traded fund (ETF) arbitrage strategies are often a large component of high-frequency trading activity. The authors extend traditional attribution analysis to include these strategies. They use a basic absolute return model that does not use a benchmark per se and begins by defining a universe and incorporating the distinct effects for attribution: net exposure effect, bucket exposure effect, and stock-specific effect. After verification, they modify the model to include market-making and arbitrage effects.
How Is This Research Useful to Practitioners?
Attribution analysis allows clients, managers, and consultants to gain a better understanding of the effectiveness of their investment decisions on performance. But traditional attribution frameworks are not conducive to analyzing high-frequency strategies, the results of which generally depend on the speed of execution and ultra-low latency. These strategies usually exhibit very high trading turnover and take few, if any, overnight positions because they are focused on the accumulation of small short-term profits.
A high-frequency trading strategy’s portfolio positions may fluctuate by the second. In a high-frequency context, even the concept of stock return can be unclear. The portfolio can have a $3 million long position one moment, a $200 million short position the next moment, and a flat, or zero, position the next. This fluctuation makes traditional (low-frequency) attribution models highly inaccurate because the use of a benchmark portfolio is not applicable. The authors overcome this problem by defining an arbitrary anchor value on which to base the portfolio weights.
Arbitrage and market-making profits are incorporated into the attribution framework. The addition of these profits is helpful for risk managers who want to ensure that high-frequency trading is adding sufficient value to justify the additional risk and fees that high-frequency strategies entail.
How Did the Authors Conduct This Research?
To begin, the authors examine low-frequency attribution models by reviewing the relevant literature. They move away from a benchmark framework by determining a reference universe for high-frequency trades. They define five distinct effects for attribution: net exposure effect (the percent the portfolio is net long or short), bucket effect (long or short in defined sectors), stock-specific effect (broken down into long and short effects), market-making effect, and arbitrage effect. The authors present a simple example of the first three effects and show both notional return and dollar return attributions.
The authors introduce the market-making effect to handle fast arbitrage profits and to overcome the accompanying timing challenges. For example, in the case of a short time interval coupled with a long holding period, the effect will measure the effectiveness of placing the order at exactly the correct time. If the time interval is longer with a short holding period, the effect will fail to pick up some of the action because the action is faster than the “shutter speed” of the measurement.
Arbitrage is included as a component of the attribution model because ETFs are frequently arbitraged and can be broken down into individual stock components for attribution. When an ETF is restored to its fundamental value, an extra arbitrage return is received.
All five components are placed together in the final high-frequency equity performance attribution model. The authors demonstrate this model by presenting an expanded portfolio example that includes ETFs and market making effects.
Traditional (long only) attribution analysis will not work for portfolios that use high-frequency trading. The authors’ methodology can be adopted by the investment community because it extends the traditional attribution framework. The authors present their sample attribution results in easy-to-understand tables. This approach to attribution would benefit consultants, risk managers, and others who need to review the effectiveness of asset strategies. High-frequency trading also occurs in futures, options, bonds, and foreign-exchange trading. This framework could be extended to incorporate these types of assets as well. Additionally, the disadvantage of traditional performance attribution analysis is that it does not directly account for the type of risks in high-frequency trading. Extension of this research into risk attribution would be very useful.