Systems challenges in high-frequency trading

Niall Dalton, Cantor Fitzgerald

High-frequency trading (HFT) involves fully automated trading systems that attempt to profit from short-term pricing inefficiencies or market liquidity imbalances. Such profit opportunities can last from microseconds to minutes (or even hours). Typically, HFT groups apply strategies in the automated market making, short-term statistical or volatility arbitrage, or liquidity detection areas.

This type of trading activity is inherently dependent on advanced computer systems to process large datasets, cope with message rates above 1 million events a second, execute algorithms at the fastest possible speed, while of course being correct and reliable in the face of inevitable failures.

Moreover, rapid improvements in technology, from smarter algorithms to low-level code optimizations and beyond, lead directly to increased profits. On the other hand, unreliable, poorly performing, unscalable or unpredictable jittery systems are one of the easier ways that a HFT trading group can put itself out of business.

After a brief introduction to HFT, this talk will explore a prototypical HFT architecture to illustrate the many systems engineering and research questions we tackle as we meet these challenges. In particular, I will focus on the programmability of these and future systems while meeting stringent performance, reliability, and efficiency goals.