Computational Finance
Algorithmic trading applies econometric models, statistical analysis, and computational techniques to the study of financial markets and the development of systematic investment strategies. Once largely confined to financial institutions and hedge funds, these methods are now accessible to smaller firms and individual researchers, although their implementation requirements vary substantially.
Research typically combines economic hypotheses, historical market information, backtesting, execution rules, and risk management. Automated order generation and execution may be employed, but remain subject to operational controls and supervision.
The materials below are organized into three complementary areas: macroeconomic indicators for understanding economic conditions, options trading for derivatives and volatility strategies, and market data for empirical analysis. The methodological overview follows below.
Algorithmic Trading Methodology
Algorithmic trading can be organized into four principal components. The value of a strategy depends on the joint quality of its assumptions, data, implementation, and risk controls.
1. Strategy identification
Develop economically or statistically motivated hypotheses. Published strategy databases, such as Quantpedia, may inform research, but parameter choices and implementation assumptions require independent assessment.
2. Strategy backtesting
Evaluate historical behavior on appropriately adjusted datasets, including corporate actions, delisted securities where relevant, timing of information availability, and realistic transaction costs. Reserve independent data for out-of-sample evaluation.
3. Execution systems
Translate strategy rules into orders and connect to broker or market infrastructure, accounting for latency, liquidity, slippage, commissions, implementation failures, and the distinction between simulated and executable prices.
4. Risk management
Define portfolio exposure, position sizing, capital allocation, drawdown limits, and operational safeguards. Kelly-type sizing and other optimization methods depend on estimation uncertainty and should be assessed for robustness.
Biases in Strategy Development
Historical results can be misleading even when the underlying calculations are correct. Four recurring problems are particularly relevant:
- Optimization bias (curve fitting or data snooping). Repeatedly modifying entry and exit rules, lookback periods, smoothing parameters, or other choices may fit historical noise. Mitigations include parsimonious models, sensitivity analysis, nested validation, and genuinely unseen test data.
- Look-ahead bias. A historical simulation may inadvertently use data unavailable at the decision time. Every input should respect its actual publication or availability timestamp; chronological separation of training and evaluation periods is necessary but not always sufficient.
- Survivorship bias. Using only surviving firms or instruments overstates historical results when failed or delisted constituents would have been part of the investable universe. Point-in-time constituent selection and delisting information are important.
- Psychological tolerance and implementation risk. Drawdowns that appear tolerable in a backtest can be difficult to sustain in real time. Capital requirements, funding constraints, and decisions to suspend or modify a strategy should be considered before deployment.
Common Algorithmic Strategy Categories
- Momentum and trend following. Seek to exploit price persistence over specified horizons. Results are sensitive to regime changes, trading frequency, and transaction costs.
- Sentiment-based strategies. Translate investor sentiment, news, language signals, or related information into systematic indicators, with attention to timestamping and availability of data.
- Event-driven and arbitrage strategies. Analyze pricing discrepancies and event-related effects around mergers, earnings, dividends, restructurings, or other market events. Apparent discrepancies may reflect material execution risk.
- Statistical arbitrage and pairs trading. Examine relative-pricing relationships and possible mean reversion. Backtests must account for unstable relationships, short-sale constraints, and model breakdown.
Performance should be assessed using several complementary measures, including the Sharpe ratio, maximum drawdown, turnover, exposure, and robustness across market conditions. No single statistic adequately describes investment risk.
Explore Algorithmic Trading
Economic conditions and market regimes
How do economic growth, inflation, interest rates, monetary policy, and business cycles influence financial markets and the assumptions underlying trading strategies?
Explore macro indicatorsVolatility, pricing and dynamic hedging
How can options pricing, volatility dynamics, market making, expiration effects, and hedging methods be analyzed and evaluated under changing market conditions?
Explore options tradingData quality and quantitative research
What data sources, historical price adjustments, validation procedures, and analytical tools are required for reproducible research, reliable backtesting, and the evaluation of trading systems?
Explore market dataReferences and Further Reading
- Chan, Ernest P. (2009). Quantitative Trading: How to Build Your Own Algorithmic Trading Business. Wiley.
- Chan, Ernest P. (2013). Algorithmic Trading: Winning Strategies and Their Rationale. Wiley.
- Chan, Ernest P. (2018). “Optimizing Trading Strategies without Overfitting.” QuantCon presentation.
