High-frequency trading (HFT): how algorithms trade in sub-microseconds

Trading with algorithms, and the infrastructure that lets trades execute with very low latency

An HFT fund earns on speed and precision

Thousands of participants trade on an exchange at the same time, their buy and sell orders are collected in the order book, and the price is formed from it. Discrepancies keep appearing in this flow: an asset costs differently on two venues for a moment, or news is already reflected in the price of one instrument and has not yet reached a related one. Each such discrepancy can be made money on if the system is faster and the model more precise than everyone else's.

An example of two basic strategies

Market making. The algorithm keeps orders on both sides of the order book and earns on the difference between buying and selling. The risk: the price moves away while the position is open. So precision decides here: models predict the price movement fractions of a second ahead, and the orders are moved in advance.
Arbitrage. The algorithm buys where the asset is cheaper and sells where it is more expensive, until the prices on the two venues converge. The window lives for fractions of a second, and everyone sees it. So speed decides here: our own network stack, low-latency infrastructure, an order in microseconds.

People work on speed and precision

Engineers build the systems that execute the strategies and models of quant researchers and ML specialists

Systems. The infrastructure keeps the path from an event on the exchange to an order within micro- and sub-microseconds, even when there is ten times more data than usual. For this there is our own network stack, IPC and data formats. The same infrastructure stores exchange data, computes models and runs backtests, so both speed and precision depend on it.

People. Our quants are not from quantum physics =) Quantitative research is the study of exchange data: find a pattern, test it and turn it into a model that the algorithm executes. Behind this are applied mathematics, probability theory, stochastics and statistics, machine learning (ML/DL) from gradient boosting to neural networks, time series, market microstructure, optimisation theory, experiment design.