Quantitative Strategies
From hypothesis to signal
A quantitative strategy specifies a universe, a signal, a portfolio construction rule, and an execution assumption. Signals are typically ranked cross-sectionally (momentum, value, quality, low volatility) or used in time series (trend following, mean reversion). The two families are regime dependent: momentum tends to work over intermediate horizons and mean reversion over short horizons in liquid instruments. Portfolio construction translates scores into weights, subject to risk limits and turnover constraints.
Backtest hygiene
Most apparent alpha dies in the details. Survivorship bias omits delisted names; look-ahead bias uses data unavailable at decision time, including restated fundamentals; data snooping arises when many variants are tested and the best is reported. Transaction costs, slippage, borrow costs, and market impact must be charged at realistic levels, especially for high-turnover signals. Proper practice uses out-of-sample and walk-forward testing, holds parameters fixed across folds, and reports the number of trials so multiple-testing corrections can be applied.
Metrics and traps
Sharpe ratio, information ratio, Sortino ratio, turnover, and maximum drawdown summarize performance. Sharpes are not comparable across sampling frequencies if autocorrelation or non-normality is present, and a high in-sample Sharpe from a large parameter search is usually an artifact. Traps also include annualizing with the wrong period count and ignoring capacity.
Worked example: a strategy earns 0.8% per month with 3% monthly standard deviation. Monthly Sharpe = 0.8/3 = 0.267. Annualized = 0.267 x sqrt(12) = 0.92. If the same edge came from 200 tested variants, treat 0.92 as an upper bound and demand out-of-sample confirmation before allocating capital.