Analyst Calls Tracked: Block to Spotify—Screening Which Upgrades Actually Move the Needle
Sell-side research calls are a fascinating dataset — if you enjoy studying noise. Monday’s roundup of analyst actions (Block, Boise Cascade, Equifax, FirstEnergy, Lockheed Martin, Lumentum, NetApp, Roper, Spotify) is a reminder that most rating changes carry roughly the predictive power of a coin flip with a lag and a bias problem.
The academic literature is consistent: analyst upgrades and downgrades do produce short-horizon drift, but the edge decays fast, transaction costs eat most of it, and the drift is partly a underreaction artifact you can replicate mechanically without paying for the call. If you’re trading rating changes, the signal worth modeling isn’t the rating — it’s the size of the earnings revision behind it. Target price changes following EPS estimate revisions show materially stronger post-event drift than ratings changes alone. In Python, this is a straightforward event-study: pull revisions from an estimates feed, compute abnormal returns over a -20/+60 day window versus sector benchmark, and test drift persistence with Newey-West adjusted t-stats. Most people skip the adjustment and declare victory on overlapping-window autocorrelation. Don’t be most people.
A few names in Monday’s batch illustrate why factor context matters more than the call itself. Lumentum: optics names are high-beta plays on AI datacenter capex, and any analyst action there is mostly a proxy for the same crowded factor exposure — your covariance with NVDA-adjacent momentum does the work, not the analyst. FirstEnergy: regulated utilities trade on rate-case outcomes and duration sensitivity; a ratings call is noise around a rates-driven factor load. Lockheed: defense budgets are legislative processes, not analyst opinions. If your model doesn’t condition on those exposures, you’re trading the factor and attributing it to the call.
Block and Spotify are the interesting ones from a signal standpoint, because both carry wide analyst dispersion — and high dispersion is itself a documented predictor of post-earnings drift and negative anomalies (see Diether, Malloy, Scherbina, 2002). When the street can’t agree, information uncertainty is priced slowly. That’s a tradeable inefficiency; a price-target bump is not.
Practical takeaway: treat analyst calls as a metadata source, not an alpha source. Backtest them properly — point-in-time ratings data, survivorship-bias-free universe, cost assumptions that include slippage on the opening print after publication. `pandas` plus a clean event-study harness will tell you in an afternoon whether the calls add anything after controlling for momentum, size, and revision momentum. The answer is usually no, but at least you’ll have the regression to prove it.