On this page
An equity anomaly is a return pattern that appears unusual relative to a chosen pricing model, but a historical pattern is not yet a usable trading signal. A CAIA® candidate should test what the benchmark omitted, whether the signal was knowable before the trade, and whether a portfolio could implement it after costs and constraints. This three part test is more useful than memorising a list of anomalies as guaranteed sources of return.
Why an abnormal return is ambiguous
The first question is what “abnormal” means. An analyst compares a realised or simulated return with a model’s expected return. If the difference is positive, either the market price was temporarily wrong or the model failed to account for a risk exposure. A market efficiency test therefore tests the market and the benchmark together. You cannot establish mispricing merely by showing a regression intercept against a thin benchmark.
Suppose a strategy seems to outperform a broad equity index. Its holdings might tilt towards smaller companies, a sector or distressed businesses. If those exposures carry risks missing from the benchmark, the excess return may be compensation rather than an anomaly. A richer factor model may reduce the unexplained return. Conversely, a model with many fitted factors can explain away an interesting pattern without showing that it was tradable. The candidate’s job is to name the uncertainty, not to assert one interpretation from the intercept alone.
CAIA Association’s Level I curriculum companion explicitly places joint hypothesis testing and persistence beside accounting accruals, price momentum, earnings momentum, net share issuance and disclosed insider activity. It also asks candidates to understand factor models, pairs trading, short selling and limits to arbitrage. Those are connected stages: a proposed return pattern must survive measurement and implementation before it can be evaluated as a strategy.
Make the information date explicit
A signal must be available at the time the imagined portfolio would have acted. Published accounts arrive after the reporting period. An earnings announcement reaches the market at a specific time. A share repurchase plan is different from completed repurchases, and a disclosed insider trade becomes public after the underlying transaction. A backtest that stamps all these observations at the quarter end gives the strategy information it could not have possessed. This look ahead error can make a weak idea look remarkably precise.
For accounting accruals, compare reported earnings with operating cash generation and ask whether a large noncash component may reverse. That is an economic hypothesis, not a mechanical verdict on every firm with receivables or inventory. Growing businesses may legitimately carry more working capital. A candidate should separate the measurement from the interpretation: identify the accounting item, its disclosure date, the plausible reason it might predict later earnings, and the alternative explanations.
Price momentum asks whether recent relative winners continue to outperform over an intermediate horizon. Earnings momentum looks at how new earnings information and market prices adjust after an announcement. They sound similar, but the clock starts from different events. A price trend may incorporate many causes, while an earnings surprise is tied to a particular disclosure. Mixing the two in a factor score without checking overlap may count the same information twice. A useful explanation states what is ranked and when the ranking is refreshed.
Net share issuance asks whether a firm has expanded or reduced its share count over a defined period. A repurchase may convey management confidence, but it may also offset employee share awards or change capital structure without revealing future operating performance. Issuance may finance a promising investment or relieve distress. Public insider transactions likewise need context: a purchase and a sale can have different motives, and the law distinguishes disclosed transactions from trading on material nonpublic information. The observable event is a candidate signal, not a complete story.
An invented scorecard to see the steps
Consider an illustrative universe of ten unbranded companies. At the start of a month, after allowing for publication lags, rank each one by a price trend, its latest available accrual measure and net share issuance. Give the strongest available signal ten points and the weakest one point on each measure, reversing ranks where a lower raw value is treated as favourable. Add the three ranks to get a simple composite out of thirty. The arithmetic is invented to show an ordering process; it is not a finding about actual returns.
One company might score eight on trend, seven on accrual quality and six on issuance, for twenty one points. Another might score nine, two and three, for fourteen. The first ranks higher on this chosen composite even though the second has stronger recent price movement. That distinction matters: a multi signal model can prevent one spectacular signal from dominating, but only because the designer has chosen comparable ranks and weights. Change the definitions or weights and the ordering may change. The score is a decision rule, not proof of alpha.
Before interpreting the ordering, ask whether the ranks share hidden exposures. The highest scorers may all be small, illiquid firms. A strategy that buys them and shorts the lowest scorers could be an unrecognised size or liquidity bet. Check sector, market and other factor exposures, then examine the result outside the sample used to choose the score. The point is not to optimise until the backtest looks smooth. Repeatedly selecting the best historical specification increases the chance of fitting noise.
Can the portfolio actually be held?
A manager needs shares at executable prices. Turnover causes commissions, spread costs and market impact. A signal concentrated in small names may offer little capacity because orders themselves move the price. Shorting the opposite side also requires stock borrow and carries financing and recall risks. The SEC’s explanation of short sales describes borrowing, losses when a shorted price rises, and the pressure to cover during a squeeze. These are implementation conditions, not optional footnotes to a gross backtest.
A pairs trade tries to focus on relative performance by pairing securities with similar broad market exposure. If the apparent gap closes, the long leg can outperform the short leg even when both prices fall. But similar historical returns do not guarantee that the businesses remain economically comparable. A corporate event can permanently change the relationship, and borrow or funding costs can consume the spread. Market neutrality to one factor does not mean the pair has no risk.
A simple invented payoff illustrates the sign. Imagine a long position worth 100 units and an equally sized short at entry. If the long rises by 6 units while the shorted security rises by 2, the relative gross gain is 4 units before costs. If the long falls by 3 while the shorted security falls by 8, the relative gross gain is 5. If the relationship moves the other way, the trade loses even if the original spread looked unusual. Both examples are illustrative arithmetic, not observed performance.
A candidate sometimes assumes that short selling automatically removes market risk. It can reduce a chosen exposure when the long and short sensitivities are properly matched, but the match can break and idiosyncratic losses remain. A short position has a different loss profile from merely omitting a stock from a long portfolio. Borrow availability and close out pressure can force action at the least convenient time. Thus the implementation stage changes the possible return distribution, not just the headline expected return.
What persistence would require
A pattern can disappear because investors learn about it, because the underlying behaviour changes, or because the original sample was lucky. It can also persist when the uncomfortable side of the trade is costly or constrained. To argue persistence, identify an economic reason for the pattern, the investor who would take the other side, and why competition has not erased it. Do not infer that an anomaly will continue simply because a chart spans many past periods.
The question is especially sharp for signals based on public disclosures. If everyone can read the same filing, a durable return requires some friction, disagreement, risk or delayed adjustment that stops immediate arbitrage. The claim should be modest: a plausible mechanism makes further testing worthwhile; it does not establish future returns. The Omni CAIA® program page places these concepts in the wider syllabus, and the methodology page sets out the source standards behind the study material.
An exam ready way to answer
Start by naming the purported anomaly and its direction, then state the information that creates the signal. Explain why the return is abnormal only relative to a specified model. Next, describe how a manager could rank or pair securities using information that was available at the decision date. Finally, name at least one cost or constraint that could erase the apparent edge. If the question asks whether a historical result proves inefficiency, the answer is no: benchmark misspecification and data selection remain live explanations.
Keep the language conditional. Accruals, momentum, issuance and disclosed trades are hypotheses for evaluation, not instructions to buy or short particular securities. The curriculum’s value is in making you move from pattern to test to feasible portfolio. When those steps are kept distinct, you can explain why an attractive paper result may fail in practice and why a modest, well specified observation may still deserve careful investigation.
Sources
- Level I Curriculum Companion 2026 · caia.org · Retrieved
- Key Points About Regulation SHO · sec.gov · Retrieved



