Notes from the work while the answer is still uncertain.
The research log records hypotheses, mistakes, reactions to new evidence, and decisions that change the next experiment. It is less formal than a paper, but still part of the research record.
The weak-stock hypothesis looked more promising than the holdout eventually allowed.
The original intuition came from momentum. Classic evidence from Jegadeesh and Titman in 1993 showed that, over intermediate horizons, past winners tended to outperform past losers. That did not imply that every weak stock should keep falling, but it made persistence in relative weakness a reasonable place to start.
The first rebound-short experiment looked particularly promising. A stock could remain structurally weak, rebound from a recent trough, and then resume underperforming. At first, the historical result looked strong enough to imagine that some version of it might become a useful shorting strategy or even a source of alpha.
That version of the result did not survive the methodological audit. The rebound peak had been chosen using future information. Once that hindsight problem became clear, the result was rejected and the experiment was rebuilt around fixed Day 5, Day 10, and Day 20 landmarks.
The redesigned Day-20 candidate later passed the locked 400-stock holdout, but the relationship shrank substantially. Spearman correlation fell from about −0.188 in discovery to about −0.061 in validation.
Surprised, and a little disappointed. Passing the holdout was encouraging, but the shrinkage made it clear that the discovery result had made the effect look much stronger than it probably is.
The more useful interpretation is narrower. The evidence supports a modest sector-relative relationship. It does not establish a directional short strategy, absolute-price decline, or trading profitability after costs.
Why this kind of research is worth doing
Empirical research can matter even when the published specification never becomes a production trading strategy. Buy-side research often starts from an empirical observation, then changes the horizon, implementation, risk controls, data, or market. The value of a paper can therefore be the question it makes worth investigating, not only the exact rule printed in the final table.
That is one reason CMB Research will keep both successful and failed ideas visible. A rejected test can still point toward a better experiment, and a modest validated relationship can still be more useful than an impressive result built on a flawed design.
Narasimhan Jegadeesh and Sheridan Titman, “Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency,” The Journal of Finance, 48(1), 1993, pp. 65–91.
What remains unresolved
The main open question is economic mechanism. Why does the surviving relationship appear more clearly relative to sector peers than in the stock’s absolute return? The historical result is now frozen, so future work should either explain that mechanism or test it prospectively rather than continue tuning the same historical sample.