💰 quantitative finance

What Useful Alphas?

This paper argues that published academic equity anomalies have become practically useless for non-micro-cap portfolio managers in the 21st century, as their returns have diminished to negligible levels after 2005 when accounting for transaction costs and luck, indicating that public stock markets are highly efficient.

Original authors: Andrew Y. Chen, Ivo Welch

Published 2026-07-08

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Andrew Y. Chen, Ivo Welch

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are a treasure hunter looking for gold in a vast, ancient forest. For decades, academic explorers have published maps claiming to show exactly where the gold is buried. These maps promise that if you follow a specific path (a "strategy"), you will find a steady stream of gold coins every month.

Andrew Chen and Ivo Welch, in their paper "What Useful Alphas?", decided to test these maps. They didn't just look at the old maps; they went into the forest themselves to see if the gold was still there for a modern explorer with a specific type of vehicle.

Here is what they found, explained simply:

1. The "Old Map" vs. The "Real Forest"

The academic maps (published papers) were drawn using data from the "good old days" (before 2006) and included every single tree in the forest, even the tiny, hidden saplings in the deep woods (micro-cap stocks).

The Result on the Map: If you followed these old maps in the old forest, you found a lot of gold. The average "treasure" was about 48 cents per month for every $100 invested. This looked like a gold mine.

2. The Modern Reality Check

Chen and Welch asked: "What if we are a modern investor who can't drive a tiny jeep into the deep woods? What if we only drive on the main highways (the top 90% of the largest, most liquid stocks) and we only look for gold starting from 2006 onward?"

When they applied these two filters, the gold mine vanished.

The Highway Filter (No tiny stocks): By ignoring the tiny, hard-to-reach stocks, the gold disappeared by half.
The Time Filter (Post-2005): By looking only at recent years, the gold disappeared by another 60%.

The Final Result: When you combine both filters (big stocks only, recent years only), the "gold" left was only 7 cents per month.

3. The "Luck" Factor

The authors realized that even this tiny 7 cents might not be real gold; it might just be a lucky coin toss.

Imagine 200 monkeys throwing darts at a board. Even if they are blindfolded, some will hit the bullseye just by chance.
When the authors adjusted for this "monkey luck," the 7 cents of gold turned into almost zero.
Furthermore, the cost of driving the car (transaction fees) would have eaten up that last 7 cents instantly. So, for a real-world investor, the profit was negative.

4. Why Did the Maps Look So Good?

The authors broke down the mystery into a simple "Two-by-Two" puzzle:

The "Good Old Days" Effect: In the past, markets were slower, and information traveled slower. It was easier to find gold. Today, high-speed computers and algorithms find the gold instantly, leaving none for the rest of us.
The "Tiny Stock" Effect: The original maps included thousands of tiny, obscure companies. These are like hidden caves that big trucks (institutional investors) can't enter. The "gold" was mostly in these caves. Once you exclude them, the main highways are very efficient, and there's no easy money left.

5. The Few Survivors

Were there any strategies that still worked?

A few "profitability" strategies (looking at companies that actually make money) and some "momentum" strategies (betting on stocks that are already rising) showed a tiny bit of life.
However, even these were mostly just lucky hits. Once you account for the fact that we picked the "best" ones out of 200, their value dropped to near zero.

The Bottom Line

The paper concludes that the "magic formulas" found in academic textbooks are useless for modern, large-scale investors.

The Anomalies were real: The gold was there in the past.
The Gold is gone: It has been picked clean by technology and competition.
The Location was wrong: The remaining gold is in the "tiny stock" caves that big investors can't reach.

The Analogy: It's like trying to find a rare, cheap sandwich in a city. In 1980, you could walk down any street and find one for $1. Today, if you only walk the main avenues (where the big crowds are) and you look for sandwiches after 2006, you won't find any for $1. If you try to find them in the back alleys (tiny stocks), you might find one, but you can't fit your car there to buy it.

The Takeaway: If you are a professional investor managing a large portfolio today, the hundreds of "proven" strategies published in academic journals over the last 50 years will not make you money. The era of easy, free lunch in the stock market is over.

Technical Summary: "What Useful Alphas?"

Problem Statement
From 1973 to 2016, academic finance literature published over 200 firm-level characteristics (anomalies) predicting the cross-section of stock returns, often reporting impressive long-short returns of 50 to 100 basis points (bps) per month. However, a critical disconnect exists between these published findings and the reality faced by equity portfolio managers. Most institutional managers operate within a "non-microcap" universe (typically the largest 3,000 stocks or top 90% of market capitalization) and trade in the modern era (post-2005). The paper addresses the question: Which of these published anomalies remain profitable for managers who cannot trade small, illiquid micro-cap stocks in the 21st century?

Methodology
The authors utilize the open-source Chen and Zimmermann (2022) dataset, which reproduces nearly 200 published cross-sectional return predictors using standardized code. The study employs a rigorous filtering and decomposition approach:

Universe Filters: The authors apply two primary filters to the stock universe, separately and jointly:
- Rank Filter (N): Restricting to the top $N$ stocks by market capitalization (e.g., top 3,000).
- Percentage Filter (%): Restricting to stocks comprising the top $X\%$ of total market capitalization (e.g., top 90%).
- Baseline Universe: The "Standard" filter combines both, retaining the intersection of the top 3,000 stocks and the top 90% of market cap (approx. 3,000 large, liquid stocks).
Temporal Split: The sample is divided into two eras: "Through 2005" (December 2005 and earlier) and "Post-2005" (January 2006 onward). This split captures the impact of decimalization, algorithmic trading, and high-frequency trading.
Sparse Signal Handling: For signals defined on small subsets (e.g., short interest), the authors drop signals with insufficient data pre-2005. Post-2005, if a portfolio leg contains fewer than 20 stocks, the return is set to zero, simulating a real-world decision to abstain from trading due to lack of diversification.
Statistical Adjustment: To distinguish genuine signal from luck, the authors apply an empirical Bayes shrinkage adjustment (Efron, 2010). This calculates the variance of $t$ -statistics across the cross-section of anomalies; if the variance is close to 1 (the variance of pure noise), the observed returns are adjusted downward to reflect the proportion attributable to luck.

Key Results
The analysis yields a stark contrast between the "original paper" environment and the "practitioner's reality":

The "Original" Environment (Pre-2005, All Stocks): The median long-short return across ~200 anomalies was 48 bps per month.
The Practitioner's Reality (Post-2005, Top 90% Market Cap): When restricting to the Standard universe (Top 90% cap) and the Post-2005 period, the median return collapses to 7 bps per month.
- The median $t$ -statistic drops to 0.45.
- The median CAPM alpha is 9 bps with a $t$ -statistic of 0.64.
- Only 67% of anomalies retain positive returns.
Decomposition of the Decline: A two-by-two analysis reveals that the 85% reduction in median returns is driven by two multiplicative factors:
1. Time Effect (Post-2005): Restricting the period to post-2005 reduces returns by ~60%, attributed to post-publication decay and the rise of trading technology.
2. Size Effect (Non-Microcap): Restricting the universe to the top 90% of market cap reduces returns by ~50%, as the most significant mispricings were concentrated in illiquid micro-caps.
Luck vs. Signal: In the Post-2005 Standard universe, the cross-sectional variance of $t$ -statistics is 1.09, nearly identical to what pure luck would produce. After applying the shrinkage adjustment, the "signal share" is only ~8%. Even the best-performing anomaly (raw 66 bps) shrinks to a luck-adjusted return of roughly 6 bps.
Survivors: The few anomalies that retained modest returns were predominantly profitability signals (e.g., cash-based operating profitability, gross profitability) and a few momentum variants. However, even these were largely explained by selection bias and luck after adjustment.
Transaction Costs: The authors note that even minimal transaction costs would eliminate the remaining 7 bps median return, rendering the strategies economically unviable.

Key Contributions

Empirical Reality Check: The paper provides the first comprehensive quantification of how much "alpha" remains for large-cap managers after accounting for both the post-2005 market regime and the exclusion of micro-caps.
Two-by-Two Framework: It introduces a simple, memorable framework for evaluating any published anomaly result by asking two questions: "What stock universe was used?" and "What sample period?" This clarifies why impressive historical returns often fail to translate to current, investable strategies.
Synthesis of Decay and Liquidity: The paper synthesizes the concepts of post-publication decay and liquidity constraints, demonstrating that they act together to eliminate the profitability of standard anomalies in the investable universe.

Significance and Claims
The paper concludes that for non-micro-cap portfolio managers in the 21st century, the vast majority of published academic anomalies are useless.

Statistical vs. Economic Significance: The anomalies were "true discoveries" in a statistical sense (not data mining artifacts), but they have been "traded away" in the segment of the market that matters most for institutional investors.
Efficiency: The evidence suggests public stock markets (specifically the large-cap segment) are highly efficient.
Modest Scope: The authors explicitly state they do not claim that no quantitative strategy can be profitable. They acknowledge that dynamic factor timing, machine learning combinations, and proprietary data strategies may offer returns beyond the standard monthly-rebalanced portfolios studied here. However, regarding the specific question of whether the 200+ standard published anomalies work in large stocks today, the answer is "not much."

The paper serves as a sobering reminder that statistical significance in historical, all-stock samples does not equate to economic implementability for modern, large-cap portfolio managers.