Quantitative Trading
Quantitative Trading relates to the development strategies based on econometrics and mathematical models employng historical. While it was a prerogative of financial institutions and hedge funds, today it can be implemented, altough on a smaller scale, by small trading firms and individuals. In general, the strategy is completely automated (order generation, submission, and execution) but it is usually supervised.
In general, quantitative strategies may complement a core-satellite investment strategy by providing additional returns profiting from a specific trend or style and reducing the impact of fees in its active components (because of the augmented returns). Moreover, QT can be profitably used to implement research and obtain additional insights to discover interesting financial products.
Quantitative trading can be structured into four major components:
- Strategy Identification – Profitable strategies are organized into databanks (quantpedia, etc.) easily accessible; however, the optimized parameters and tuning methods are usually to be estimated.
- Strategy Backtesting – the viability of a strategy should be assessed by testing it on historical data; it is important to obtain clean and bias-free data. Specifically, corporate actions (splits, dividends adjustments, etc.) should be correctly merged with data, otherwise erroneous interpretations may arise.
- Execution System – Linking to a brokerage, automating the trading and minimizing transaction costs
- Risk Management – Optimal capital allocation, “bet size”/Kelly criterion and trading psychology
Because strategies extensively rely on data, major issues arise from biases occurring in data:
- Optimization Bias: adjusting or introducing additional trading parameters to force the strategy to perform in a specific way; such bias is also named curve fitting or data-snooping. Parameters can be entry/exit criteria, look-back periods, moving average smoothing parameter, or volatility measurement frequency. OB can be reduced by increasing the number of data points, minimizing the number of parameters, and performing sensitivity analysis on the parameters. The same performance results should be invariant to changing parameters.
- Look-Ahead Bias: Perhaps the most common bias, it results from incorporating future data into the simulation supplying the algorithm with information that will not be available when trading live. For instance, giving a 10 years dataset, training should occur in the first 5 years without including information from year 6 to 10 for the reason when trading live in year K there is no information available for year K+1.
- Survivorship bias occurs when dead firms are not included in the dataset used for training; in other terms, the strategy is trained on past winners only. However, when trading live, it is not possible to know who are the winners and the losers; moreover, the strategy is not tested on losers. Thus, datasets including dead firms or datasets including only recent data must be employed; additionally, some commodities and currencies (and relative future derivatives) are inherently bias-free.
- Psychological Tolerance Bias refers to the difficulty in accepting historical drawdown and duration in live trading even after accepting it when backtesting.
Strategies can be generally classified into four categories:
- Momentum/Trend Following strategies attempt to profit from an existing trend capitalizing on market volatility; in general, short-term buy/sell positions are open and closed to systematically exploit the trend and volatility.
- Sentiment Based strategies attempt to profit from investors perceptions, feelings, and emotions reflected in the traded volume and volatility. Examples include News Based Trading relying on linguistic analytical tools matching news with numbers to suggest a trading decision.
- Arbitrage strategies attempt to profit from market inefficiencies occurring before or after a corporate event (bankruptcy, acquisition, merger, spin-offs, dividends, earnings, etc.).
- Statistical Arbitrage strategies, or pairs strategies, are based on the mean reversion hypothesis. In general, such strategies attempt to profit from statistical mispricing but are also based on a sound economic justification; specifically, econometric techniques provide the signals to entry/exit positions.
Common measures of quantitative strategies are the maximum drawdown (the largest peak-to-trough drop in the account equity curve over a particular time period) and the Sharpe Ratio (the average of the excess returns over the free-rate divided by the standard deviation of those excess returns).
References
Chan, Ernest P. (2009). Quantitative trading: how to build your own algorithmic trading business
Chan, Ernest P. (2013). Algorithmic trading : winning strategies and their rationale