Introduction

Quantitative trading engines are sophisticated systems that scan markets, evaluate opportunities, and execute decisions across varying time scales. This guide explores their core architecture and design principles.

Prop-trading engines are less like a monolith and more like an ensemble of decision units that coordinate and operate at different time-horizons. There’s a trade-off between speed and sophistication, as the shorter the time-horizon for a decision is, the simpler the algorithm has to be.

At an ultra low-latency environment, rules have to be distilled and compressed into heuristics that can flip a signal in nanoseconds.

At high latency, you have a holistic view, but state transitions are fast compared to your ability to act on them. At low latency, you have tunnel vision but an incredible ability to act fast on new data.

Both signal opportunity and risk assessment have to be layered in that context.

Compute performance is largely a battle against the speed of light, and building a high-frequency trading infrastructure is largely about shortening the distance that electrical signals have to travel for a given action to occur. On a hardware level this means using FPGAs or ASICs. At the software level this means using kernel bypass, optimal data locality, non-blocking data structures, and so on.

Topologies vary depending on the strategies employed. Multi-venue strategies require cross-colo synchronization, or at the minimum distributed hot-path coordination. This arguably increases the complexity of the architecture.

In the rest of this guide, we will focus exclusively on the software aspects and will not cover topics that rely on specialised hardware infrastructure.