Quantitative Trading

Quantitative trading is an approach to trading in which positions are chosen by rules derived from mathematical and statistical analysis of data. A model takes inputs such as prices, volumes, company accounts or news, and produces a forecast or a ranking that is turned into trades. Humans design, test and supervise the models, but they do not usually decide individual trades.

The approach is also called systematic trading. It is practised by hedge funds, proprietary trading firms, bank trading desks and a growing number of individuals. It overlaps with algorithmic trading, but the two are not the same: algorithmic trading refers to the automated placing of orders, which a discretionary investor may also use, while quantitative trading refers to how the investment decision itself is reached.

Quantitative strategies are built through a sequence of stages that most firms follow in some form.

• Data. Historical and live data are collected, cleaned and stored. Errors, missing values and adjustments for dividends and share splits must be handled before any analysis.
• Signal research. Researchers form a hypothesis about what predicts returns and test it statistically. A predictive variable is called a signal or alpha.
• Backtesting. The rule is simulated on past data to estimate its return, risk and costs. Backtesting is the main evidence for a strategy and also the main source of self-deception.
• Portfolio construction. Signals are combined into target positions, usually by an optimizer that balances expected return against risk, costs and limits on exposure.
• Execution. Orders are worked in the market by an execution strategy designed to limit market impact.
• Monitoring. Live results are compared with expectations, and risk limits are enforced automatically.

Most quantitative strategies fall into a few broad groups. Trend following buys assets that have been rising and sells those that have been falling. Mean reversion does the opposite over short horizons, betting that prices that have moved far from a reference level will return. Statistical arbitrage trades large baskets of related securities against each other while keeping little net exposure to the market. Factor strategies hold shares with characteristics such as low valuation or high past return. Volatility strategies trade options against the expected movement of the underlying asset, and market making earns the spread between buying and selling prices.

Strategies also differ by speed. High-frequency trading holds positions for seconds or less and depends on technology and low costs. Slower strategies hold for days to months and depend more on the breadth of the data and the quality of the research.

Results are judged on return relative to risk, not on return alone. The most common summary is the Sharpe ratio, the average excess return divided by its standard deviation. Researchers also examine the maximum drawdown, turnover, capacity and the correlation of the strategy with others already held. Capacity is the amount of capital a strategy can manage before its own trading erodes the profit. A fast strategy may earn a high return on a small amount of capital, and a slow one a modest return on a large amount.

The central danger is overfitting. With enough trials, a researcher will always find a rule that performed well in the past by chance. Defences include holding back data for out-of-sample tests, walk-forward optimization, limiting the number of parameters and requiring an economic reason for the effect. Related errors are look-ahead bias, in which the test uses information that was not available at the time, and survivorship bias, in which failed companies are missing from the data.

Costs are the second danger. Commissions, spreads, impact and financing can turn a profitable backtest into a loss, particularly for strategies that trade often. A third is decay: once a pattern becomes known, competition tends to reduce it.

Quantitative trading grew from the 1980s as computing became cheap and markets electronic. Firms such as Renaissance Technologies, D. E. Shaw, Two Sigma and AQR became closely associated with it.

Because many firms study the same data with similar methods, their positions can overlap. In August 2007 a number of equity funds running similar market-neutral strategies suffered large losses within days as they reduced positions at the same time, an episode often called the quant quake. Models also assume that the future resembles the past, and they can fail when a market regime changes. Operational risk is significant as well, since a software fault can send erroneous orders at machine speed. Regulators require automated traders to maintain pre-trade risk checks and the ability to halt their systems.