Analyst-upgrade cluster

Thesis (rejected): When 3+ distinct analyst firms upgrade the same name within a 30-day window, the consensus has materially shifted and other institutional positioning follows for 1-3 months. Stickel 1995, Womack 1996 on analyst recommendation revisions predicting returns; clustering filters out single-analyst noise.

Code: research/analyst_upgrade_cluster_validate.py Report: research/analyst_upgrade_cluster_report.json (n=681)

Why it failed: Edge is negative in the short term and ~zero longer term. The signal is lagging - by the time 3 firms have upgraded, the move is already in.

Hit rates 54-60 % look reasonable but the edge vs random non-event days on the same names is basically zero. Analyst clustering is mostly driven by quarterly review cycles around earnings, which we already capture in PEAD; the residual signal after that overlap is too small.

Verdict: Not added. Suggests an even more leading signal might work (e.g. single analyst upgrade with the biggest price-target change from the highest-conviction firm), but that's a different shape of signal.