What Is Backtest Overfitting? Definition and Example
Backtest overfitting is when a strategy is tuned so closely to past data that it captures noise instead of a real pattern. Here is how to spot it, with a real example.
Alpha decay is the fading of an investing edge over time as more people find and trade it. Here is what the research shows and what it means for strategies you read about.
What is your own portfolio exposed to? Check it with Exent →
Alpha is return beyond what market risk explains. Alpha decay is that excess shrinking. It has two main causes. Some of the original edge was never real, a result of backtest overfitting. The rest gets competed away once enough money chases it.
See how this applies to your own holdings with an Exent portfolio check.
McLean and Pontiff, “Does Academic Research Destroy Stock Return Predictability?” (Journal of Finance, 2016) tracked 97 predictors from academic papers. Portfolio returns were 26% lower in the period after the study’s sample ended but before publication, and 58% lower after publication. The authors read the first gap as an upper bound on data mining and the rest as investors learning from the research.
Illustrative: a strategy that earned 6 points a year above its benchmark in a published backtest.
| Stage | Excess return a year |
|---|---|
| In the published backtest | 6.0 pts |
| If it falls 26% (out of sample) | about 4.4 pts |
| If it falls 58% (after publication) | about 2.5 pts |
| After fees and trading costs of 1–2 pts | about 0.5–1.5 pts |
By the time a strategy is popular enough to reach you, much of its edge may be gone. Treat published results as a ceiling. This applies to factor investing, screener rules and newsletter systems alike; see do newsletters beat the market?
Because part of a backtested edge is often luck, and the real part attracts capital once it is known. Trading by many investors pushes prices until the edge shrinks.
It varies. McLean and Pontiff found returns of published predictors averaged 58% lower after publication, with larger declines for predictors that had higher in-sample returns.
The same logic applies. Factor premiums have varied a lot over time, and there is no guarantee any will persist at its historical size.
Upload a screenshot of your holdings and get concentration, overlap, crash exposure and a risk-matched benchmark in about a minute.
Educational content, not investment advice. Examples are illustrative; past performance does not guarantee future results.