Stock Market Anomalies: Why Most Disappear
After You Try to Trade Them.

The Anomaly Graveyard

In 2015, R. David McLean and Jeffrey Pontiff published what may be the most important paper in modern empirical finance.^[1] They collected 97 cross-sectional return predictors documented in top academic journals and tested what happened to them after publication.

The results were devastating. Out-of-sample — meaning in data that the original researchers never saw — anomaly returns declined by 26%. That alone would concern any practitioner. But the post-publication decay was far worse: returns fell by 58% on average after the papers were published and became widely known.

McLean and Pontiff decomposed this decay and found that roughly 32% of the total decline was attributable to publication-informed trading — traders reading the papers and exploiting the anomalies until they disappeared. The remaining decay came from a combination of statistical overfitting in the original research and structural changes in the market.

This means that if you read a paper documenting a new anomaly in the Journal of Finance and attempt to trade it, you should expect roughly half the reported return at best. In many cases, the expected return is zero or negative after costs. The academic anomaly literature is, in a very real sense, a graveyard — full of strategies that were alive when discovered but dead by the time you try to implement them.

The implications are profound. The entire enterprise of mining academic papers for trading signals is far less productive than most investors believe. The bar for an anomaly to survive publication, arbitrage, and transaction costs is extraordinarily high. Most do not clear it.

Why Anomalies Decay: Three Mechanisms

Understanding why anomalies disappear is more valuable than knowing which ones existed. There are three distinct mechanisms, and each has different implications for portfolio construction.

The data mining problem is the most insidious. Harvey, Liu, and Zhu (2016) documented that over 400 factors have been published in academic finance — a phenomenon they termed the "factor zoo." With so many variables tested, some will appear significant purely by chance. The conventional 5% significance threshold is laughably inadequate when hundreds of predictors are tested simultaneously. A proper Bonferroni-corrected threshold would require t-statistics above 3.0, not 2.0. Most published anomalies fail this higher bar.

The arbitrage mechanism is more straightforward but equally deadly. When a published anomaly has a genuine economic basis, capital flows into the strategy. This capital compresses the spread between the long and short sides of the trade, reducing the anomaly return. The speed of this process has accelerated dramatically: anomalies published in the 1990s decayed over 5-10 years; anomalies published in the 2010s can decay within months as quantitative funds systematically scan new research.

Structural change is the most difficult to diagnose because it requires understanding why an anomaly existed in the first place. The small-cap premium, for example, was enormous in the mid-20th century when small stocks were genuinely difficult to trade and information about them was scarce. As electronic trading, SEC disclosure requirements, and internet-based research eliminated these frictions, the economic rationale for the small-cap premium largely evaporated. The anomaly did not get arbitraged away — the reason for its existence disappeared.

The Transaction Cost Problem

Even anomalies that survive publication face a second gauntlet: transaction costs. Detzel, Novy-Marx, and Velikov published a landmark study in 2023 demonstrating that the standard approach to testing asset pricing models is fundamentally biased.^[2]

The core insight is that academic factor model tests ignore trading costs entirely. When researchers compare competing models — say, a 3-factor model versus a 5-factor model versus a 6-factor model — they evaluate which model best explains the cross-section of returns in a frictionless world. But in a frictionless world, factors that require enormous turnover look just as good as factors that require modest turnover.

This creates a systematic bias. The model selection process gravitates toward high-turnover factors because they have more opportunity to explain monthly return variation — even though implementing them would be prohibitively expensive. When Detzel, Novy-Marx, and Velikov re-ran the standard model horse races with realistic transaction costs, the results changed dramatically. The Fama-French five-factor model — which uses relatively low-turnover factors — dominated alternatives that appeared superior in the frictionless setting.

The practical implication is stark: many strategies that look profitable in academic backtests are actually negative after trading costs. A factor that generates 3% annual alpha but requires 200% annual turnover with 50 basis points of round-trip costs is actually destroying 7% of capital per year after friction. The net return is negative 4%. This is not a theoretical concern — it describes the majority of published anomaly-based strategies when implemented with realistic assumptions.

The lesson for individual investors is to be deeply skeptical of any strategy that requires frequent trading, trades illiquid securities, or targets small-cap stocks where bid-ask spreads are wide. The most implementable anomalies are precisely the ones that require the least trading: broad factors like Quality, Value, and Momentum applied to liquid, large-cap stocks with quarterly or monthly rebalancing.

Anomaly Timing: Speed Kills

Even when an anomaly is genuine and survives transaction costs, there is a timing problem. Bowles, Reed, Ringgenberg, and Thornock published a crucial finding in 2024: anomaly returns are heavily concentrated in the first month after information release.^[3]

Their study examined a broad set of well-known anomalies and measured when, precisely, the excess returns accrued. The concentration was striking — the vast majority of the anomaly premium was earned in the narrow window immediately following the arrival of new information (earnings announcements, analyst revisions, macroeconomic data releases). By the time the information was 30 days old, most of the return had already been captured.

This finding has uncomfortable implications for portfolio managers who rebalance monthly or quarterly. If the anomaly return is earned in the first few days after a signal fires, a portfolio that waits until the next scheduled rebalance date misses the majority of the premium. The anomaly "works" in theory but cannot be captured at the frequency most investors operate.

Speed, then, is not merely an advantage — it is a prerequisite. Quantitative hedge funds with daily or intraday rebalancing capability can capture anomaly returns that are structurally inaccessible to monthly rebalancers. This creates a natural hierarchy: the fastest traders extract the return first, leaving diminished scraps for slower participants. For individual investors without institutional infrastructure, this means the effective anomaly premium is substantially smaller than academic estimates suggest.

The exception is anomalies driven by slow-moving behavioral biases rather than information events. Value and Quality, for example, do not depend on rapid response to news. They capture long-duration mispricing driven by investor overreaction, anchoring, and preference for lottery-like payoffs. These slow anomalies are the ones accessible to investors with modest rebalancing frequency — and they form the backbone of any sensible factor portfolio.

The Short-Sale Constraint

Most academic anomaly studies construct long-short portfolios: buy the stocks in the top decile of the signal and sell short the stocks in the bottom decile. The reported "anomaly return" is the spread between these two legs. But in practice, the short side of many anomalies accounts for the majority of the return — and the short side is precisely where implementation is most difficult.

Muravyev, Pearson, and Pollet (2025) demonstrated that when short-sale costs are properly accounted for, the abnormal returns on anomaly portfolios are eliminated entirely.^[4] This is not a marginal effect. The short-side alpha — which often represents 60-80% of the total long-short spread — evaporates once you include the cost of borrowing shares, the risk of recall, and the fee drag of maintaining short positions.

The natural experiment provided by Regulation SHO confirms this mechanism. Chu, Hirshleifer, and Ma (2020) studied what happened when the SEC temporarily relaxed short-sale constraints on a random subset of stocks.^[5] When shorting became easier, anomaly returns on those stocks weakened by 72 basis points per month — a massive decline that confirms the short-sale constraint was propping up the apparent anomaly premium.

The implication is that many anomalies are not really about the long side outperforming. They are about the short side underperforming so severely that it creates a statistical spread. If you only implement the long side — as most individual investors must — you capture a fraction of the reported premium, often an economically insignificant fraction.

This is why BCR operates as a long-only model. We are not ignoring the short side out of ignorance — we are acknowledging that the short-side premium is illusory for anyone without a prime brokerage account, low borrowing costs, and the operational infrastructure to manage short positions. The honest question is not "what is the long-short spread?" but "what does the long-only leg return after costs?" The factors in our model — Quality, Momentum, Value, Investment, Stability, Short Interest — all have documented long-side premia that survive independently of the short leg.

Which Anomalies Survive?

After filtering for data mining, transaction costs, timing decay, and short-sale constraints, which anomalies remain standing? The survivors share two non-negotiable characteristics: a clear economic rationale and implementability on the long side.

1. 01

   Quality / Profitability

   Novy-Marx (2013) showed that gross profitability is a powerful predictor of returns — a finding that has survived extensive out-of-sample testing and replication across international markets. The economic rationale is intuitive: companies that convert revenue into profit at high rates have durable competitive advantages. The market underprices the persistence of profitability, creating a systematic premium for patient investors. Critically, the Quality premium is earned entirely on the long side.^[6]

2. 02

   Momentum

   Jegadeesh and Titman (1993) documented the momentum effect — stocks that have performed well over the past 6-12 months continue to outperform. This anomaly has survived 30+ years of out-of-sample testing, replication across geographies and asset classes, and intense scrutiny from skeptics. The economic explanation combines behavioral underreaction to good news and institutional herding. While momentum is subject to violent reversals (momentum crashes), the long-term premium remains robust.^[7]

3. 03

   Value

   Fama and French (1993) formalized the value premium — cheap stocks outperform expensive ones. While the value premium has experienced its worst-ever drawdown in the 2018-2020 period, it has recovered substantially and remains one of the most documented phenomena in finance. The economic rationale (compensation for distress risk, or behavioral overreaction to bad news) is well-established. The long-side value premium, while smaller than the long-short spread, is positive and statistically significant over multi-decade horizons.^[8]

4. 04

   Investment (Conservative Minus Aggressive)

   Fama and French (2015) added the Investment factor to their model, showing that companies with conservative asset growth outperform aggressive expanders. The economic logic: empire-building managers destroy value through dilutive acquisitions and overinvestment. Companies that invest judiciously compound capital more efficiently. This factor has low turnover and high implementability.

5. 05

   Low Volatility / Stability

   The low-volatility anomaly — boring stocks beat exciting ones — contradicts basic CAPM theory but has been documented across virtually every equity market in the world. The explanation is behavioral: investors overpay for lottery-like, high-volatility stocks, depressing their forward returns. Defensive, stable companies are systematically underowned and underpriced.

6. 06

   Short Interest (Informational)

   Aggregated short-interest data provides a window into informed pessimism. Stocks with very low short interest benefit from the absence of informed negative views. While the short side of this signal (heavily shorted stocks underperform) is the stronger leg, the long side — stocks with minimal shorting activity — carries a modest but persistent positive premium, particularly when combined with Quality and Momentum filters.

Notice what these surviving anomalies have in common. They are all relatively low-turnover. They all have clear behavioral or risk-based explanations. They all generate meaningful long-only premia. And they have all survived decades of out-of-sample testing and scrutiny. This is not a coincidence — it is the filter that separates real phenomena from statistical artifacts.

What This Means for Your Portfolio

The anomaly graveyard teaches a clear lesson: most of what appears to work in backtests does not work in practice. The gap between theoretical alpha and implementable alpha is enormous, and it is widened by transaction costs, timing decay, short-sale constraints, and data mining. The investors who generate persistent excess returns are not the ones chasing the latest published anomaly — they are the ones disciplined enough to stick with the few factors that have survived every stress test the market has thrown at them.

For individual investors, the practical framework is straightforward. First, be deeply skeptical of any strategy that requires shorting. The short-sale premium is real but inaccessible to most investors, and claiming it as expected return is dishonest. Second, favor factors with low turnover. High-turnover strategies look good in frictionless backtests but bleed capital through transaction costs in the real world. Third, demand an economic explanation. If someone cannot explain why an anomaly exists in plain language, it is probably data-mined.

Fourth, diversify across surviving factors rather than concentrating in one. Quality and Momentum have a correlation near zero, which means combining them produces a smoother return stream than either alone. Value and Stability provide further diversification. The whole point of a multi-factor model is that individual factors have periods of underperformance, but a diversified composite is far more robust.

Fifth, keep costs low. Rebalance at moderate frequency — monthly or quarterly — rather than chasing every signal update. Use liquid, large-cap stocks where bid-ask spreads are narrow. Avoid exotic instruments, leveraged positions, and small-cap names unless you have a specific, well-documented edge in those segments.

This is exactly the philosophy underlying BCR's 6-factor composite model. Each factor — Quality, Momentum, Value, Investment, Stability, and Short Interest — meets every survival criterion identified by the academic literature: clear economic rationale, robust out-of-sample evidence, long-only implementability, and low transaction cost sensitivity. The model is deliberately simple because complexity is the enemy of robustness. The goal is not to find the next exotic anomaly before it disappears. The goal is to systematically harvest the durable premia that have survived decades of scrutiny, arbitrage, and market evolution.