Signal Decay: How Trading Edge Erodes Over Time

Key Takeaways

• Signal decay is the erosion of a trading signal's edge over time as capital arbitrages away the anomaly after it is discovered or published.
• McLean and Pontiff (2016) studied 97 stock-return predictors and found post-publication returns averaged 58 percent below in-sample returns, with the largest decay hitting the most loudly advertised anomalies first.
• Using the full historical sample average as a forward return expectation is one of the most common mistakes in systematic investing; post-2005 and trailing 5-year performance should be weighted more heavily.
• Not all decayed signals are permanently dead; value had a brutal 2010s and a strong 2022, suggesting the decay was a cycle rather than structural extinction.

What It Is

A signal is a rule that maps observable data to an expected return, such as "buy the cheapest decile on price-to-book" or "short stocks with the weakest six-month momentum." Signal decay is the observed erosion of that expected return through time, measured by comparing the signal's performance before and after a reference date.

The reference date is usually the publication of a research paper, the launch of a popular factor ETF, or a structural change in the market. The simplest test is to split the full sample into pre and post periods and compare the annualised return, Sharpe ratio, or information coefficient in each.

The Intuition

Any persistent signal is, by definition, money sitting on the sidewalk. In a market with thousands of professional arbitrageurs, money on the sidewalk gets picked up. Once a signal is published or commercialised, capital flows into the trade, pushes prices toward the signal's prediction, and shrinks the future edge.

Three mechanisms drive most decay. Arbitrage compresses obvious mispricings as quant funds add the factor to their books. Crowding means many investors now hold similar positions, so exits become costly and realised returns fall below theoretical returns. Structural change shifts the market itself, for example through decimalisation, the rise of ETFs, or regulatory changes to short selling.

The uncomfortable implication is that a backtest run on the full historical sample almost always overstates what you will earn going forward. Using unadjusted historical returns as a forecast is one of the most common mistakes in systematic investing.

How It Works

The clearest empirical study is McLean and Pontiff (2016), who examined 97 stock-return predictors documented in academic finance. They split each predictor's history into three windows: the in-sample period used by the original paper, the out-of-sample period between the end of that sample and publication, and the post-publication period.

Their headline finding:

Average predictor return, in-sample:     100%  (baseline)
Out-of-sample, pre-publication:           74%  (-26%)
Post-publication:                         42%  (-58%)

The 26 percent drop between in-sample and out-of-sample reflects overfitting and data-mining bias. The further drop after publication is attributed to investors learning the signal and arbitraging it away. Decay is larger for predictors whose in-sample returns were highest, consistent with arbitrage capital flowing toward the loudest anomalies first.

Detection in practice uses three tools. Rolling alpha, estimated on a trailing 3 or 5 year window, shows whether the signal's edge is still present recently. Sub-period tests split the sample into decades and compare. Out-of-sample validation reserves the most recent segment of data, never touched during research, as the honest forward estimate.

Worked Example

Consider the classic value factor: long cheap stocks on price-to-book, short expensive ones. A researcher backtesting from 1963 to 2024 sees an impressive annualised premium. Split by decade, the picture changes.

From 1963 to 1999, the long-short value portfolio earned roughly 4 to 5 percent per year on average. From 2010 to 2020, the same rule was essentially flat, with a prolonged drawdown that lasted more than a decade. A forecast built on the full-sample mean would have missed this regime change entirely.

The prudent response is not to abandon value outright. Some signals recover, and value has shown signs of life after 2020. But you would down-weight the signal in your forward expectations, combine it with other factors, and size it based on its recent information coefficient rather than its 60-year average.

Common Mistakes

1. Using the full historical sample as forward expectation. A signal that averaged 8 percent for 40 years may earn 2 percent next year. Report post-publication, post-2000, and trailing 5 year performance alongside the headline number and weight them more heavily in forecasts.

2. Ignoring the post-2000 factor-investment boom. Factor ETFs, smart beta products, and quant hedge funds absorbed far more capital after 2000. Signals that worked before this expansion faced much thinner arbitrage capital. A decent rule of thumb is to test whether the signal still pays after 2005.

3. Dismissing decay as normal noise. Every signal has noisy years. Decay is structural, not cyclical. If the trailing 10 year Sharpe is half the trailing 30 year Sharpe and crowding into the factor is visible in positioning data, that is evidence, not noise.

4. Assuming every decayed signal is permanently dead. Some factors mean-revert. Value had a brutal 2010s and a strong 2022. Permanent death requires a structural reason, such as the signal being fully priced in or the underlying cause disappearing. Otherwise, the decay may be a drawdown.

5. Stacking many decayed signals and calling it diversification. Combining ten signals that each lost their edge post-publication does not produce a working portfolio. Check each input for current edge, not just historical edge.

Frequently Asked Questions

Q: What is signal decay in simple terms? Signal decay is the gradual weakening of a trading rule's predictive power as more investors learn about it and trade it. When capital flows into the same trade, prices adjust faster and the historical edge shrinks or disappears entirely.

Q: How does signal decay affect investment decisions? It means you cannot use a factor's long-run historical return as an expectation for what you will earn going forward. The realistic forward estimate needs to weight recent performance more heavily, particularly the post-2005 period when factor investing became widely commercialized.

Q: What is a real-world example of signal decay? The classic value factor earned 4 to 5 percent per year on average from 1963 to 1999. From 2010 to 2020, the same rule was essentially flat for a full decade, the kind of prolonged underperformance that would never have appeared in a forecast built on the full 60-year average.

Q: How can investors adjust for signal decay when building a portfolio? Test whether the signal still pays after 2005, check trailing 5-year and trailing 10-year performance separately from the full-sample average, and combine the signal with others that are imperfectly correlated so that any single signal's decay is partially offset.

Q: How is signal decay different from a normal drawdown? A drawdown is temporary underperformance during a bad cycle; the edge eventually recovers. Decay is structural erosion caused by arbitrage capital permanently compressing the anomaly. The difference becomes clearer over time: a drawdown reverses, while a decayed signal stays flat or continues to underperform even during regimes that historically would have rewarded it.

Sources

1. McLean, R.D. and Pontiff, J. (2016). "Does Academic Research Destroy Stock Return Predictability?" The Journal of Finance, 71(1), 5-32. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2156623
2. McLean and Pontiff (2016). Journal of Finance, Wiley Online Library. https://onlinelibrary.wiley.com/doi/abs/10.1111/jofi.12365
3. "When do systematic strategies decay?" Quantitative Finance, Taylor & Francis. https://www.tandfonline.com/doi/full/10.1080/14697688.2022.2098810
4. Boston College Carroll School. "Award-winning Pontiff Paper Shows Investors Care about Academic Research." https://www.bc.edu/bc-web/schools/carroll-school/news/2017/pontiff-paper.html

Disclaimer

This article is educational content only and is not financial advice. Nothing here is a recommendation to buy, sell, or hold any security. Consult a licensed advisor before making investment decisions.