Three researchers from University College of London, who have spent eight years developing algorithmic trading (AT) systems and an algorithmic trading/risk platform, have written a fascinating review article on AT. Algorithmic trading refers to any form of trading using sophisticated algorithms and systems to automate all or most parts of the trading cycle. Despite the lucrative market, the implementation details of AT are hard to find, as the AT community is secretive and highly competitive in nature.

This article gives an excellent exposition on algorithmic trading and explains the different components. The trading cycle consists of pre-trade analysis, signal generation (selecting the right features), trade execution, post-trade analysis, risk management, and asset allocation (portfolio management). Since AT is data driven, many ideas include machine learning, statistical reasoning, and decision making. Developers use simulations, historical data, and optimizations to evaluate and improve the utility of their algorithms.

Another key feature of algorithmic trading is latency, which involves network location (relative to the stock exchanges) and the possibility of snooping on the data. This aspect is not dealt with in the article.