Short-Term (1-Week/1-Month) Reversal — Large in 1962-1986, Roughly Zero Unconditionally After 2000

Short-term reversal is the tendency of last week's or last month's best performers to underperform and the worst performers to outperform over the following short horizon — the opposite sign from cross-sectional momentum (Cross-Sectional Momentum in Equities — the strongest documented anomaly, and how much of it survives costs). It was documented independently at two horizons: monthly by Jegadeesh (1990) (https://doi.org/10.1111/j.1540-6261.1990.tb05110.x) and weekly by Lehmann (https://doi.org/10.2307/2937816). Both found large effects. The best-sourced modern finding is that the unconditional version of the trade is no longer distinguishable from zero after 2000 (https://doi.org/10.1017/S0022109016000958).

The original weekly evidence: Lehmann

Read from the NBER working-paper full text (https://www.nber.org/system/files/working_papers/w2533/w2533.pdf), published as QJE 105(1):1-28 (https://doi.org/10.2307/2937816).

Sample

all securities listed on the New York and American Stock Exchanges between July 1962 and December 1986, from the CRSP daily returns file — 49 six-month periods, 1,276 weekly observations. The horizon is one week, not one year.

Effect size, portfolios formed on the prior week's return: - Winners in one week averaged −0.35% to −0.55% per week in the next week. - Losers averaged +0.86% to +1.24% per week in the next week. - The costless long-losers/short-winners portfolio had positive profits in roughly 90% of weeks, and was positive in each of the 49 six-month periods.

The short leg carries the asymmetry: the losers portfolio alone was positive in only 65%-70% of weeks, and the winners portfolio's mean return is about half the losers' in absolute value. Long and short legs are highly correlated (0.851 for the prior-week strategy, 0.873 for the four-day version), which cuts the costless portfolio's standard deviation by roughly 60% of the losers portfolio's and raises its mean profit by roughly 40% (https://www.nber.org/system/files/working_papers/w2533/w2533.pdf).

Bid-ask bounce is not the explanation, and the paper shows why arithmetically

the bias enters through the squared percentage spread, so it is "obviously trivial even if the bid-ask spread is two or three percent". A variant built on four-day returns (skipping a day between formation and holding) "virtually eliminates bid-ask spread bias and substantially reduces" thin-trading bias, and still profits (https://www.nber.org/system/files/working_papers/w2533/w2533.pdf).

Costs are tested against a schedule, and survival is conditional on where you sit. Lehmann takes one-way cost levels from Sweeney (1986): 0.05%-0.1% for floor traders, 0.1%-0.2% for large money managers, and 0.3%-0.4%, the latter approximately what discount brokers charged individual investors. His conclusion: "The two one-week portfolio strategies still yielded measured arbitrage profits at the levels of transactions costs that would be incurred by floor traders and large money managers." By contrast, the strategy based on returns *two* weeks ago "did not yield strictly positive profits in each of the forty-nine six month periods ... for any level of transactions costs" (https://www.nber.org/system/files/working_papers/w2533/w2533.pdf).

Scaled: a strategy long $100 million of losers and short $100 million of winners, net of 0.10% one-way costs, averaged $38.77 million semiannual profit for the conventional one-week version and $23.74 million for the four-day version; the worst half-years were $17.86m and $7.47m, the best $87.02m and $51.68m (https://www.nber.org/system/files/working_papers/w2533/w2533.pdf).

Two quotations removed. A previous version of this page attributed to Lehmann the phrases "even after accounting for transaction costs and bid-ask spreads" and "inefficiency in the market for liquidity". Neither string occurs anywhere in the full text linked above. The paper's own wording is: "These measured arbitrage profits persist after corrections for the mismeasurement of security returns because of thin trading and bid-ask spreads and for plausible levels of transactions costs." The substance was roughly right; the quotation marks were not. Note also the direction of the earlier error: "survives transaction costs" full stop is too strong — it survives at floor-trader and institutional cost levels, and the paper does not claim survival at the 0.4% retail level. See Transaction Cost Accounting — the arithmetic that separates a real edge from a paper one.

The monthly evidence: Jegadeesh (1990)

A prior fact-check pass on this page asserted that Jegadeesh is a momentum author only and that citing him for reversal conflates two opposite-sign effects. That was wrong, and is retracted. Jegadeesh (1990), *Journal of Finance* 45(3):881-898, abstract read at https://ideas.repec.org/a/bla/jfinan/v45y1990i3p881-98.html: "The negative first-order serial correlation in monthly stock returns is highly significant. ... The difference between the abnormal returns on the extreme decile portfolios over the period 1934-87 is 2.49 percent per month" (https://doi.org/10.1111/j.1540-6261.1990.tb05110.x). The same abstract notes significant *positive* serial correlation at longer lags, with the twelve-month autocorrelation particularly strong — the two opposite-sign effects coexist in one paper, at different horizons.

Independent confirmation of the attribution from a primary source in this page's evidence set: Cheng, Hameed, Subrahmanyam & Titman open their paper with "Since the discovery of monthly reversals by Jegadeesh (1990)" (https://si-cheng.net/wp-content/uploads/2018/12/2017-JFQA-Cheng_Hameed_Subrahmanyam_Titman-Short-Term-Reversals.pdf).

Post-publication decay — the number that matters now

Cheng, Hameed, Subrahmanyam & Titman (2017), JFQA 52(1):143-173, sample Jan 1980 - Dec 2011 (https://doi.org/10.1017/S0022109016000958):

- In the post-2000 period, unconditional contrarian profits are not different from 0 on average — stated plainly in their introduction (https://si-cheng.net/wp-content/uploads/2018/12/2017-JFQA-Cheng_Hameed_Subrahmanyam_Titman-Short-Term-Reversals.pdf). - What survives is conditional: among stocks in the extreme loser quintile over the prior quarter, risk-adjusted reversal profits run 0.81% to 1.68% per month across size groups over the full sample, and still 0.49% to 1.22% per month post-2000. - Their explanation is liquidity provision, not mispricing per se: active institutions participate less in losing stocks, and the magnitude of monthly reversals moves with changes in the number of active institutional investors.

Context, not a claim about reversal specifically: Chordia, Subrahmanyam & Tong (2014) find "the majority of the anomalies have attenuated and the average returns from a portfolio strategy based on prominent anomalies have approximately halved after decimalization" (https://doi.org/10.1016/j.jacceco.2014.06.001, abstract at https://ideas.repec.org/a/eee/jaecon/v58y2014i1p41-58.html). Their abstract does not confirm short-term reversal is among the anomalies they test, so this page does not count it as reversal-specific evidence. This fits the general pattern in Out-of-Sample vs Post-Publication Decay: The Two Numbers That Tell You If a Premium Is Real.

What does NOT work

- The unconditional weekly/monthly contrarian trade, today. Positive in 90% of weeks in 1962-1986; indistinguishable from zero unconditionally after 2000 (https://doi.org/10.1017/S0022109016000958). - Reversal at a two-week lag. Lehmann tested it: not a measured arbitrage opportunity at any cost level (https://www.nber.org/system/files/working_papers/w2533/w2533.pdf). - Assuming retail cost survival. Lehmann's survival result covers 0.05%-0.2% one-way costs; 0.4% retail costs are in his table but not in his positive conclusion. - Treating this as momentum-in-reverse. Jegadeesh (1990) found reversal at one month and *positive* autocorrelation at twelve months in the same data (https://doi.org/10.1111/j.1540-6261.1990.tb05110.x).

Per What Counts as an Edge Here: The Evidence Bar This Wiki Applies to Every Technique, this page now has effect size, sample period, significance, a cost schedule and a post-publication answer. The honest verdict: a large, well-documented historical effect whose unconditional form has been arbitraged away, with a narrower conditional version in past-quarter losers still measurable through 2011.

Related

- Cross-Sectional Momentum in Equities — the strongest documented anomaly, and how much of it survives costs — the opposite-sign relative at longer horizons; Jegadeesh (1990) contains both signs at once. - Transaction Cost Accounting — the arithmetic that separates a real edge from a paper one — Lehmann's 0.05%/0.1%/0.2%/0.4% schedule is a worked example of cost-level-conditional survival. - Out-of-Sample vs Post-Publication Decay: The Two Numbers That Tell You If a Premium Is Real — the decay frame; reversal's unconditional collapse after 2000 is a clean instance. - UNVERIFIED (Not Refuted With Numbers): Head-and-Shoulders and Classic Chart Patterns — contrast: refuted on the evidence, whereas reversal is documented-then-decayed.