Alpha That Is Really Beta in Disguise
A backtest with a great Sharpe ratio can still be nothing more than a known risk factor wearing a new name. Before calling a result alpha, you have to show it survives after the obvious exposures are stripped out.
Prerequisites: Who Is On the Other Side of the Trade?
Every year, someone presents a strategy with a beautiful Sharpe ratio that turns out, on inspection, to just be a small-cap tilt, or a short-vol bet, or leveraged momentum, relabeled. This is not usually dishonest — it is what happens when a signal is validated only against a flat benchmark instead of against the risk factors it is quietly correlated with. The strategy is not fake. It is just not what the researcher thinks it is.
Why this happens so easily
A cross-sectional signal built from, say, low analyst coverage will tend to load on small size, because small stocks are undercovered. A signal built from recent price strength will load on momentum. Neither loading is a bug in the signal-construction code; it is a real, mechanical relationship between the signal and a well-known factor. The trouble is that "well-known factor" means the return is not new information about mispricing — it is compensation for a risk the market has already priced, that thousands of other funds are also collecting. Calling it alpha overstates how much the research contributed and understates how correlated the strategy is with everything else in the book.
The check that catches it
Regress the strategy's returns on a standard factor set — market, size, value, momentum, quality, low-vol, whatever the shop's model uses — and look at what's left over, the residual. If the loadings on known factors explain most of the return and the leftover (the true alpha) is statistically indistinguishable from zero, the "edge" was beta wearing a costume. If a meaningful, significant residual remains after the known factors are accounted for, there is a real claim to novelty — though even then, the honest caveat is that it may simply be a factor nobody has named yet, not necessarily true idiosyncratic skill.
| Signal you think you built | Factor it often turns out to load on |
|---|---|
| Low analyst coverage | Small size |
| Recent 6–12 month price strength | Momentum |
| High earnings yield | Value |
| Low realised volatility | Low-volatility / quality |
| Recent IPO underperformance | Small size, illiquidity |
A worked example
A researcher builds a "quality" signal from return-on-equity and finds an annualised Sharpe of 1.1 gross. Running the strategy's return series against a five-factor model shows a market beta of 0.15, a size loading of −0.4 (it's short small caps), and a quality-factor loading of 0.9. After removing the contribution of those three exposures, the residual return has a t-stat under 1. The signal was, almost entirely, a leveraged bet against small caps expressed through a quality screen — not new information, a repackaging of exposures the market already prices.
A high Sharpe ratio on a signal that has never been regressed against a factor model is not evidence of skill — it is an unopened question. Skipping this check is the single most common way "novel alpha" turns out to be an old factor with new packaging.
If a signal's Sharpe ratio survives factor-adjustment but drops a lot, that drop is not a failure — it is the honest size of the discovery. Report both numbers; a gross Sharpe without the adjusted one alongside it is close to meaningless to anyone allocating capital.
Related concepts
Practice in interviews
Further reading
- Chincarini & Kim, Quantitative Equity Portfolio Management
- Harvey, Liu & Zhu (2016), ...and the Cross-Section of Expected Returns