The Lighthouse · November 2026
The 53-Sharpe Mirage
When losers bounce—and why strength still matters. We rebuilt an extraordinary academic result, watched the daily version fail, and found a weekly long-only effect that deserves investigation without yet deserving capital.
The dazzling daily strategy was not tradable truth. Our daily long/short recreation lost money. Weekly, sector-relative, long-only reversal produced a credible research lead—but survivorship, execution, and point-in-time data limitations still stand between evidence and implementation.
Executive summary · 2-minute read
One irresistible claim. Three increasingly useful conclusions.
Daily long/short did not survive
The two daily long/short variants returned −9.86% and −14.55% annualized in the broad current universe under the common assumptions.
Weekly long-only became interesting
Sector-relative weekly reversal returned 22.83% annualized with a 0.882 Sharpe—but also a 47.27% maximum drawdown.
Strength remains primary
The production-oriented question is not “buy weak stocks.” It is whether a temporary pullback improves entry timing among stocks already demonstrating strength.
Central thesis
The real discovery was how easily a conclusion changed.
Timeframe, short exposure, sample construction, transaction costs, and portfolio breadth each changed the answer. A backtest is not a verdict; it is a controlled argument whose assumptions must remain visible.
The number that deserved suspicion
What did the MIT paper actually claim?
Amir Khandani and Andrew Lo’s 2007 paper, What Happened to the Quants in August 2007?, used a simple contrarian portfolio to diagnose a market dislocation. It was not presenting a finished retail system. The rule bought stocks that lagged the group and shorted stocks that led, scaling positions with the distance from the average.
For 1995, the paper reported a 53.87 annualized Sharpe ratio and 1.38% average daily return in the all-stock portfolio. Its average portfolio included 4,781 stocks and rebalanced daily. The authors explicitly described the return as unrealistic after transaction costs, market impact, short-sale constraints, and other institutional limitations. They also showed that returns declined substantially between 1995 and 2007.
A Sharpe above 53 is not an invitation to believe. It is an invitation to audit.
The number earns attention. The report earns trust by explaining why it was not a usable performance claim.
What we tested
Completed signal. Next-open trade. Six variants.
We used split-adjusted U.S. daily histories from WealthVelocity’s long-term databases and current classifications from Stocks.mdb. The global production databases are intentionally compact and were not used as historical research archives.
U.S. history only
Common stocks with acceptable current quality flags, current price of at least $5, and average volume of at least 100,000 shares.
No same-close fiction
The signal used a completed close, entered at the next adjusted open, and exited at the following period’s adjusted open.
Capped exposure
Individual positions were capped at 10% in the baseline. Long/short and long-only variants were tested separately.
Turnover charged
The baseline deducted five basis points for every 1.0 of portfolio turnover; wider cost scenarios were tested later.
The test inherits today’s eligible securities and classifications. It does not yet reconstruct inactive and delisted stocks or historical classifications. That creates current-universe survivorship and look-ahead risk.
Sample versus full universe
The daily idea failed. The weekly result depended on breadth.
The 100-stock sample intentionally selected a small number of highly liquid names within each sector. The broad run loaded 3,239 eligible securities; 3,193 had sufficient history for the baseline.
| Strategy | 100-stock sample | Full universe | Full Sharpe | Full max drawdown |
|---|---|---|---|---|
| Daily all-stock long/short | −12.09% | −9.86% | −0.461 | 96.14% |
| Daily sector-relative long/short | −5.04% | −14.55% | −1.118 | 96.73% |
| Daily sector-relative long-only | 2.47% | 13.34% | 0.546 | 53.01% |
| Weekly all-stock long/short | 1.01% | 13.83% | 0.793 | 37.91% |
| Weekly sector-relative long/short | −0.14% | 10.84% | 0.834 | 29.84% |
| Weekly sector-relative long-only | 4.46% | 22.83% | 0.882 | 47.27% |
The daily long/short drawdowns approached total loss. Weekly results were materially stronger, but the best return still carried a deep drawdown and aggressive turnover.
Robustness suite
The effect weakened with friction—but did not vanish immediately.
The baseline portfolio turned over approximately 1.405 times its capital each week. That is an aggressive execution requirement. The simplified cost sweep shows why headline return and implementation return are not the same thing.
Annualized result by turnover cost
Cost is charged per 1.0 of portfolio turnover. This is a sensitivity test, not a stock-level spread and impact model.
A 1% cap reduced annualized return to 11.72%, but also reduced maximum drawdown from 47.27% to 29.96%. Its 0.797 Sharpe remained close to the baseline’s 0.882.
| Bucket | Annual | Sharpe | Drawdown |
|---|---|---|---|
| Q1 lowest | 18.52% | 0.962 | 35.13% |
| Q2 | 16.62% | 0.798 | 40.71% |
| Q3 | 16.76% | 0.738 | 43.91% |
| Q4 highest | 17.20% | 0.730 | 45.79% |
| Bucket | Annual | Sharpe | Drawdown |
|---|---|---|---|
| Q1 lowest | 15.38% | 0.758 | 38.80% |
| Q2 | 19.04% | 0.866 | 41.17% |
| Q3 | 19.49% | 0.853 | 43.53% |
| Q4 highest | 16.40% | 0.768 | 43.74% |
| Period | Annualized | Sharpe | Maximum drawdown | Weekly turnover |
|---|---|---|---|---|
| 2008–2012 | 29.00% | 0.904 | 47.27% | 1.401 |
| 2013–2017 | 19.22% | 1.286 | 19.24% | 1.406 |
| 2018–2022 | 22.95% | 0.761 | 44.82% | 1.407 |
| 2023–Sep. 2026 | 19.26% | 0.890 | 23.76% | 1.399 |
All quartiles and subperiods were positive under the same assumptions. That weakens—but does not eliminate—the explanations that one liquidity pocket or one lucky episode created the full result.
Why this is not a WealthVelocity pick
Mean reversion and trend persistence make different promises.
WealthVelocity seeks a constructive hierarchy: exchange conditions, local and global sector behavior, stock trajectory, liquidity, timing, and data quality. Reversal begins with a stock that has fallen behind and expects a limited recovery.
WealthVelocity
Strength selection
- Buys demonstrated strength
- Expects trend persistence
- Holds while evidence remains constructive
- Requires an actionable timing state
- Primary selection philosophy
Research strategy
Weekly reversal
- Buys relative underperformance
- Expects partial mean reversion
- Uses a limited recovery window
- Requires a measurable peer dislocation
- Alternate strategy
Leaders, laggards, and a century-old warning
Livermore treated group divergence as evidence—not automatic value.
In Edwin Lefèvre’s 1923 Reminiscences of a Stock Operator, the Livermore-inspired narrator observes that people often buy a stock simply because it lagged while its group advanced. He warns against the shortcut: “beware of buying a stock that refuses to follow the group-leader.”
A brief pullback can create a reversal opportunity. Persistent failure to participate can reveal that something is wrong before the reason is public. Former leaders that fall and fail to recover can also warn that the market’s character is changing.
How should this stock act, given its group and the market—and what does it mean when actual behavior violates that expectation?
The compatible global experiment
Strength first. Pullback second. Forward only.
Because the global historical databases are intentionally compact production resources, we should not manufacture a retroactive global backtest. The responsible next step is a weekly forward shadow signal.
Establish the strong universe
Start with current global stocks that pass WealthVelocity’s exchange, sector, trajectory, liquidity, timing, and data-quality gates.
Measure the temporary lag
Calculate the weekly sector-relative pullback only after the strength universe is fixed.
Observe multiple recovery windows
Track one-, two-, four-, and eight-week outcomes without pretending a signal traded before it existed.
Reject persistent non-participation
A temporary lag may improve entry timing. Failure to rejoin the sector becomes a negative signal, not a reason to average down.
Point-in-time universe reconstruction, inactive and delisted stocks, realistic stock-level spreads and market impact, country and exchange breakdowns, removal of extreme winners, and a genuine out-of-sample shadow period.
The Lighthouse view
A statistical effect can deserve investigation without deserving capital.
The daily implementation failed. The weekly implementation became interesting. The sample and broad universe told different stories. Robustness tests made the easy dismissal harder, but did not erase the unresolved biases.
Weekly reversal now has a place in the research laboratory and in Quant Institute education. WealthVelocity’s primary selections remain stocks that have already demonstrated that their trend is real.
Less sensational than 53.87. More informative because its timing, costs, drawdown, and limitations are visible.
The goal is not a magical equation. It is a test that can survive the next question.
Sources and research record
Primary evidence.
Research cutoff: October 29, 2026. Historical evidence is U.S.-only. Eligibility and classifications are based on the current database state. Backtested results are hypothetical, depend on stated assumptions, and do not represent actual trading. Informational research only; not individualized investment advice or a recommendation to buy or sell any security.