Strategies · Mean reversion
Short-term reversal: buying last week’s losers, tested
The weekly reversal strategy on the 30 Dow stocks, 2021 to 2026: the rules, a backtest before and after costs, and why 42 portfolio turnovers a year erase the edge.
On this page
Short-term reversal is one of the oldest findings in stock research: shares that fell the most over the past week or month have tended to do a little better than average over the next one, and the biggest winners a little worse. Two papers published in 1990, by Narasimhan Jegadeesh and by Bruce Lehmann, documented it in decades of US data. The usual explanation is that short bursts of selling push a price below where it would otherwise be, and whoever buys from the hurried seller is paid a small premium when the price drifts back.
A strategy follows directly: every week, buy the stocks that just fell the most. This lesson states that rule exactly and tests it on the 30 Dow stocks over five years. The test is useful mainly for one thing it shows very clearly. Before trading costs, last week's losers beat the group. After a modest cost, they trailed it badly, because the rule replaces most of its portfolio every week.
The idea
Suppose a large fund has to sell a stock quickly. Buyers will take the shares only at a discount, so the price drops more than the news, if there is any, justifies. Over the following days the pressure ends and the price recovers part of the drop. Anyone who bought during the selling earned the recovery. In effect they were paid for supplying liquidity, the same service a market maker sells through the bid-ask spread.
Ranking stocks by last week's return is a crude way to find where that pressure has been. Some of the biggest losers fell on real bad news and will not recover. Some fell because of temporary selling and will. The strategy buys the whole bottom of the list and relies on the second group being large enough.
Two features follow from the mechanism. The effect is small per trade, because it is a discount for liquidity, not a forecast about the business. And it is short-lived, so the portfolio has to be rebuilt constantly. Small gains and constant trading is the combination most exposed to costs.
The rules
Checked once a week, at the last close of the week:
- For each of the 30 stocks, compute the price change over the week just ended.
- Rank the 30 from lowest to highest.
- Hold the six lowest (the bottom fifth) in equal dollar amounts for the coming week. Sell whatever is no longer on the list and buy what is new.
- Repeat every week.
The academic versions also sell the winners short, which makes the portfolio market-neutral. This test is long-only, for two reasons: shorting adds borrowing costs the data cannot measure, and most individual investors cannot run 30 short positions. The winners are tracked as a separate long portfolio so the two ends can be compared.
A worked example of the turnover
Take a week when five of the six losers are new to the list, which is typical. Five positions are sold and five bought: five-sixths of the portfolio is sold and the same amount bought. At 0.10% per trade, that week costs 0.17% of the portfolio (0.10% on the five-sixths sold and 0.10% on the five-sixths bought). It looks like nothing. Repeated 52 times, it is almost 9% a year. In the test the average week replaced 81% of the portfolio.
What the backtest shows
The test, stated in full so it can be checked:
- Data. Weekly closes of the 30 current members of the Dow Jones Industrial Average from November 2021 to the week of November 9, 2026. The first four weeks are kept for the longer look-backs shown below, so every version runs from December 13, 2021: 256 weeks.
- Trading. Ranking and trading both use the week's last close.
- Costs. 0.10% of every dollar bought and every dollar sold.
- Comparison. All 30 stocks in equal amounts, rebalanced weekly at the same cost. That is the fair yardstick: it holds the same stocks without the ranking.
- Left out. Dividends and taxes. Every gain is short-term, so in a US taxable account taxes would cut the result further.
- Survivorship bias. The 30 stocks are today's members. Companies dropped from the index after doing badly are missing, and a rule that buys losers is exactly the rule that would have owned them. The results are too favorable for that reason.
| Measure | Last week's six losers | All 30, equal weight |
|---|---|---|
| Total return | 28.3% | 71.0% |
| Return per year (CAGR) | 5.2% | 11.5% |
| Deepest fall from a peak | −29.8% | −22.5% |
| Longest time below a peak | 136 weeks | 79 weeks |
| Time invested | 100% | 100% |
| Portfolio turnover | 42 times a year | 0.7 times a year |
After costs the rule earned less than half of what the plain basket earned, with a deeper fall. It was ahead of the basket in 123 of 256 weeks, which is no better than a coin.
Before and after costs
| Portfolio | No cost | 0.10% per trade | 0.25% per trade |
|---|---|---|---|
| Last week's six losers | 14.3% a year | 5.2% a year | −7.1% a year |
| Last week's six winners | 11.1% a year | 2.3% a year | not run |
| All 30, equal weight | — | 11.5% a year | not run |
This table is the lesson. Before costs, the losers beat the basket by almost three points a year and beat the winners by more than three. The reversal effect was there. A cost of one tenth of a percent per trade, a level a retail investor in large stocks can roughly achieve, removed nine points a year and turned a lead into a deficit. At a quarter of a percent the strategy lost money while the stocks it traded rose.
Over the whole test the rule paid costs equal to 41% of its starting capital.
The two halves of the period
| Period | Six losers, after costs | All 30, equal weight |
|---|---|---|
| December 2021 – May 2024 | −2.1% | 18.6% |
| May 2024 – November 2026 | 31.0% | 44.2% |
It trailed in both halves. Year by year, the gross effect was not steady either:
| Year | Losers, no cost | Losers, after costs | Winners, after costs | All 30 |
|---|---|---|---|---|
| 2022 | −6.0% | −13.3% | −26.6% | −10.7% |
| 2023 | 14.6% | 5.4% | 27.7% | 20.3% |
| 2024 | 16.8% | 7.5% | 4.3% | 20.5% |
| 2025 | 19.0% | 9.3% | 8.9% | 14.2% |
| 2026 (to November 9) | 22.8% | 14.1% | 5.3% | 12.7% |
In 2023 the winners beat the losers by a wide margin: that year rewarded strength, not weakness. Even before costs, the losers trailed the basket in 2023 and 2024.
Neighbouring settings
| Version (all at 0.10% per trade) | Return per year | Deepest fall | Turnover a year |
|---|---|---|---|
| Three losers, one-week look-back | 6.1% | −37.5% | 47 times |
| Six losers, one-week look-back | 5.2% | −29.8% | 42 times |
| Ten losers, one-week look-back | 6.0% | −26.3% | 36 times |
| Six losers, two-week look-back | 2.1% | −31.0% | 29 times |
| Six losers, four-week look-back | 8.4% | −26.3% | 21 times |
None reached the basket's 11.5%. The four-week version came closest, and the reason is in the last column: it trades half as often. Before costs it earned 13.1% a year, slightly less than the one-week version's 14.3%, so the improvement after costs came from lower turnover, not from a stronger effect.
Where it fails: the losing periods
- Costs, every week. This is the main failure and it is not a period, it is permanent. The effect is a fraction of a percent per week and the cost of capturing it is of the same size.
- Falling markets. From January 3 to September 26, 2022 the rule fell 29.8%, against 22.5% for the basket. In a broad decline the week's worst stocks are often the ones the decline is about, and they keep falling.
- Momentum years. In 2023 last week's winners returned 27.7% after costs and the losers 5.4%. When a few strong stocks lead for months, a rule that sells strength and buys weakness is on the wrong side every week.
- Bad news is not a liquidity discount. A stock that drops 15% on a failed product or a cut forecast is on the losers list for a reason. The rule cannot tell that drop from a temporary one, and stocks often keep drifting lower after bad earnings news.
- Where the effect lives. Avramov, Chordia and Goyal found that reversal profits were largest in stocks with high turnover and low liquidity, the ones that cost the most to trade, and that the profits were smaller than the likely trading costs. Thirty of the largest US companies are the cheapest stocks to trade and also where the effect is weakest.
Compared with its neighbours
| Approach | Holds | Turnover | What the test showed |
|---|---|---|---|
| Short-term reversal | Last week's six worst of 30 | 42 times a year | 14.3% gross, 5.2% after costs |
| Two-day RSI rule on one fund | One fund, 16% of the time | 29 round trips in two years | Costs took a fifth of the gain |
| All 30 held equally | Everything | 0.7 times a year | 11.5% a year |
The two-day RSI rule bets on the same kind of rebound in a single price. It trades less, so costs hurt it less. Neither removes the basic arithmetic: a small edge per trade divided by a fixed cost per trade.
How Finabase does it
No Finabase tool runs a ranking strategy or tells you which stocks to trade. The product shows the ranking itself and nothing more.
- Performers. The Dow performers page, linked above, lists the index's stocks with the biggest gains and losses over a chosen range. Set to five days, its "worst" list is close to the list this rule would hold. It is a table of what happened, not a signal.
- Backtest. The Backtest tool works on one symbol at a time, buying a fixed $10,000 per indicator signal with no trading cost. It cannot rank a group of stocks, and because it charges nothing to trade, it would hide exactly the problem this page is about.
- Strategy. The Strategy tool replays if/then allocation rules over two years. Its conditions compare indicators and prices; there is no "the six lowest of these 30" step.
Frequently asked questions
Do last week's losers really outperform?
Before costs, slightly, in this test: 14.3% a year against 11.5% for all 30 stocks and 11.1% for the winners, from 2021 to 2026. The academic studies from 1990 found the same pattern in much longer and broader data. After a 0.10% cost per trade, the losers returned 5.2% a year.
Why do costs matter so much more here than for other strategies?
Because of turnover. A strategy that trades once a year pays its cost once. This one replaced 81% of its holdings every week, so it paid the cost about 84 times a year on the buys and sells combined. A small cost multiplied by that count is large.
Is the reversal the opposite of momentum?
They work on different horizons. Over a week to a month, losers have tended to bounce (reversal). Over three to twelve months, winners have tended to keep winning (momentum). Researchers measuring momentum usually skip the most recent month for that reason.
Could a professional trade this profitably?
Firms that provide liquidity for a living, with costs far below a retail investor's, earn a version of this premium. That is a business built on speed, scale and very low costs, not a rule for an individual account.
Summary
Buying last week's worst six Dow stocks earned 14.3% a year before costs from December 2021 to November 2026, more than the 11.5% from holding all 30. Charged 0.10% per trade, it earned 5.2%; charged 0.25%, it lost 7.1% a year. The rule replaces four-fifths of its portfolio every week, so a cost too small to notice on one trade consumed most of the return. The sample is short, contains only large companies and leaves out the ones that failed. What it shows reliably is the arithmetic: a strategy's edge has to be measured against what it costs to trade, and for short-term reversal the two are about the same size.
Check your understanding
4 questions on this lesson. Each answer comes with the reason, and nothing is scored.
What does the weekly reversal rule hold?
- The six stocks that rose the most last week
- The six stocks with the lowest return last week, in equal amounts
- All 30 stocks, weighted by last week’s return
Show the answer
B. The six stocks with the lowest return last week, in equal amounts The rule ranks the 30 stocks by the past week’s price change and holds the bottom fifth for one week, then repeats. The winners are the other end of the list.
Before costs the losers earned 14.3% a year; after a 0.10% cost per trade they earned 5.2%. What caused the difference?
- The rule replaces most of its portfolio every week, so a small cost is paid many times
- The cost was charged twice by mistake
- Dividends were removed from the second figure
Show the answer
A. The rule replaces most of its portfolio every week, so a small cost is paid many times The portfolio turned over about 42 times a year. A tenth of a percent on each buy and each sell, on roughly 81% of the portfolio every week, adds up to about nine points a year.
A strategy has a small edge per trade and must trade very often. What is the most important number to know before judging it?
- Its win rate
- Its return in its best year
- Its realistic cost per trade compared with its gain per trade
Show the answer
C. Its realistic cost per trade compared with its gain per trade When the gain per trade is a fraction of a percent, the cost per trade decides whether anything is left. This rule made money at no cost, little at 0.10% and lost at 0.25%.
The test uses the 30 stocks that are in the Dow today. Why does that make a buy-the-losers rule look better than it was?
- Today’s members are larger than past members
- Companies removed after doing badly are missing, and those are the ones this rule would have bought
- The index changes its weights every week
Show the answer
B. Companies removed after doing badly are missing, and those are the ones this rule would have bought This is survivorship bias. A rule that buys the weakest stocks would have held the ones that later failed and were dropped, but the data contains only the survivors.
Sources and further reading
- Narasimhan Jegadeesh, “Evidence of Predictable Behavior of Security Returns,” Journal of Finance 45(3), 1990 — reversal in monthly returns
- Bruce N. Lehmann, “Fads, Martingales, and Market Efficiency,” Quarterly Journal of Economics 105(1), 1990 — reversal in weekly returns
- Doron Avramov, Tarun Chordia and Amit Goyal, “Liquidity and Autocorrelations in Individual Stock Returns,” Journal of Finance 61(5), 2006 — reversal profits against trading costs
- Finabase chart data — the 30 stocks of the Dow Jones Industrial Average, weekly bars November 2021 to November 2026; rule and results computed by Finabase
