Why alpha decay matters
It determines whether a historical strategy remains worth trading after publication, competition, costs, capacity constraints, and changes in the underlying economic relationship.
How it is applied
Researchers compare rolling and out-of-sample performance, signal strength by holding horizon, turnover, implementation cost, crowding, live results, and behavior across regimes and universes. Researchers measure how predictive performance changes between signal observation and trade execution, and as holding periods lengthen. Decay curves inform rebalance frequency, capacity, and acceptable implementation delay. Analysis should be performed out of sample and net of realistic costs.
Portfolio example
A factor generated 4% annual excess return before publication but only 1% afterward, while trading cost remained 1.5%. Its implementable alpha has effectively disappeared. A signal forecasts 1.0% gross excess return when traded next day, 0.6% after five days, and 0.2% after twenty. If round-trip cost is 0.3%, the twenty-day implementation has negative expected value despite retaining some gross signal.
How to interpret it
Decay can occur across time, after a signal event, or as assets grow. Lower recent return does not prove permanent death, but requires a fresh economic and statistical case. Fast decay favors rapid execution but can increase cost and crowding. Slow decay may support larger capacity and lower turnover. Decay can reflect information becoming public, competitors trading, or a signal correcting prices as intended.
Limitations and common misconceptions
Short histories, data revisions, changing volatility, and poor execution can imitate decay. Repeatedly modifying a model after weak results can create overfitting rather than restore genuine alpha. Backtests can underestimate delay, impact, borrow, and regime change. Measured decay may arise from overfitting or overlapping observations. A historically persistent signal can disappear once capital, market structure, or data definitions change. Decay should be estimated separately by market, liquidity bucket, signal strength, and regime. A blended average can hide that the strongest opportunities disappear fastest. Researchers should simulate actual production timestamps, including data release, cleaning, portfolio construction, approval, and execution. Capacity is reached when marginal impact and delay consume expected alpha, not when a backtest merely reaches a chosen asset level.
Sources and further reading
- Quantitative MethodsCFA Institute