Alpha (α)
Excess return over a benchmark after accounting for beta; a measure of skill or edge in backtesting.
Definition
Alpha = R_strategy − (R_f + β × (R_benchmark − R_f)). It is the intercept in the CAPM regression of strategy excess returns on benchmark excess returns. Alpha represents return not explained by exposure to the benchmark.
Why it matters for backtesting
- Edge: Positive alpha suggests the strategy adds value beyond passive exposure.
- Attribution: Helps separate market (beta) from idiosyncratic (alpha) return.
- Overfitting: Backtest alpha that is very high or stable is often not robust; test out-of-sample and across regimes.
Limitations
- Depends on the benchmark; wrong benchmark distorts alpha.
- Single-factor model; multi-factor models (e.g. Fama–French) give different alpha.
- Alpha can be negative even when absolute returns are good if the benchmark did very well.
Linked concepts
Beta, Sharpe ratio, information ratio, factor models.