Alpha
What is Alpha?
Alpha is the estimated return not explained by the benchmark or risk factors included in a specified performance model.
Alpha reporting should state model, benchmark, factors, frequency, period, annualization, risk-free proxy, intercept significance, and costs. A positive estimate can disappear with a better model or another window. Dashboards should show confidence and residual diagnostics rather than presenting alpha as a directly observed manager profit.
Out-of-sample persistence and investability matter more than the best-fitting historical regression alone.
Returns-based results should be reconciled with holdings and decision attribution. If alpha depends on one extreme month, one stale private mark, or a narrow start date, that sensitivity should be visible before users interpret it as durable skill.
Regression meaning
In a single-factor model, alpha is the intercept from regressing portfolio excess returns on benchmark excess returns. It estimates average return beyond what beta exposure would imply. Multi-factor alpha remains after controlling for additional factors such as size, value, momentum, quality, duration, credit, or currency under the chosen specification.
Model dependence
Alpha is not observed directly. It changes with benchmark, factors, period, frequency, risk-free rate, currency, and data treatment. An omitted risk factor can appear as alpha, while an inappropriate benchmark can penalize a sound mandate. Statistical noise can produce positive estimates even when no repeatable skill exists.
Example
If a portfolio's realized excess return is 8% and its modeled factor exposures explain 6%, the simplified unexplained amount is 2%. Regression alpha is estimated across many observations rather than one annual subtraction and may differ after compounding and annualization. Confidence intervals can include zero despite a positive point estimate.
Gross, net, and implementation
Gross alpha before fees, transaction costs, financing, and tax may not reach the investor. Capacity can erode an apparent edge as assets grow. Timing, stale marks, survivorship, and backfill can inflate historical estimates. Evaluate live net results and decision attribution alongside the regression rather than relying on a favorable backtest.
Interpretation
Positive alpha can indicate skill, luck, model error, hidden exposure, or data problems. Review economic rationale, persistence, breadth, significance, drawdowns, and out-of-sample evidence. Negative alpha does not automatically imply incompetence if the benchmark is unsuitable. Never compare alpha estimates calculated from materially different models as though they were one standardized return.
Sources and further reading
- Portfolio Performance Evaluation, CFA Institute