Chapter 13. Statistical Arbitrage in High-Frequency Settings

Statistical arbitrage (stat-arb) exploded on the trading scene in the late 1990s, with PhDs in physics and other "hard" sciences reaping double-digit returns using simple statistical phenomena. Since then, statistical arbitrage has been both hailed and derided. The advanced returns generated before 2007 by many stat-arb shops popularized the technique. Yet some blame stat-arb traders for destabilizing the markets in the 2007 and 2008 crises. Stat-arb can lead to a boon in competent hands and a bust in semi-proficient applications.

The technique is a modern cousin of a technical analysis strategy utilizing "Bollinger Bands" that was used to indicate maximum highs and lows at any given point in time by plotting a two-standard deviation envelope around the simple moving average of the price. Despite the recent explosive popularity of stat-arb strategies, many misconceptions about the technique are prevalent. This chapter examines the stat-arb technique in detail.

At its core, stat-arb rests squarely on data mining. To begin with, stat-arb analysts sift through volumes of historical data with the objective of identifying a pervasive statistical relationship. Such a relationship may be between the current price level of the security and the price level of the same security in the recent past. The relationship may also be between price levels of two different securities, or the price level of one security and the volatility of ...

Get High-Frequency Trading: A Practical Guide to Algorithmic Strategies and Trading Systems now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.