Analyst rating drift

An analyst upgrade is public the moment it is issued, but it is not absorbed all at once. Institutions with direct access react first, the buy-side follows, retail reads about it days later. If that sequence is slow enough, the price should keep moving in the direction of the change, and whoever buys after the first wave is the counterparty. Stickel (1995) and Jegadeesh and Kim (2010) document the pattern.

It was also the first strategy in this lab to arrive with a full pre-registration, a fixed universe and a multiple-testing threshold.

Pre-registered on 2026-05-23. Upgrades bought and downgrades shorted at the next open, with same-day conflicting actions skipped. Twelve variants: direction, any versus strong change, and a 5, 20 or 60-day hold, each with a 3×ATR stop and a 5×ATR target. The universe was 261 US large and mid caps with at least 50 rating events, 16,600 clean events in all. Design period 2014-2020, out-of-sample 2021-2026, and a Bonferroni threshold for 12 variants: t above 2.93.

By the rules written that morning, it passed. Six of the 12 variants beat 2.93 out-of-sample, the best at t 7.13 and +2.28% per trade, and 4 of 12 cleared every formal gate. All six winners were longs. All six shorts had the wrong sign: downgrades did not keep falling, they rose with everything else.

That asymmetry prompted one more check, not in the plan: subtract SPY over the same holding window. No variant survived it. The best alpha t was 2.17, with 0.61% per trade, and fewer than half the trades beat the index (46-49% depending on the variant). The check assumed a beta of 1, the instrument of the time, and was never re-derived with Jensen alpha. The verdict did not need it. The same day, alpha against the benchmark became a gate for every strategy that followed.