Monday’s Analyst Calls: Block, Equifax, Lockheed—Signal or Noise? Data Says Check.
Monday’s sell-side research calls arrived with the usual fanfare: Block, Boise Cascade, Equifax, FirstEnergy, Lockheed Martin, Lumentum, NetApp, Roper, Spotify, and assorted technology names. I skimmed them the way a quant reads any analyst note — looking for the hypothesis, the sample size, and the backtest. Predictably, two of the three were missing.
Analyst price-target changes are a fascinating dataset precisely because they’re mostly noise dressed as signal. Studies going back decades (Womack 1996, Barber et al. 2001, and the follow-up literature since) show short-term drift after upgrades and downgrades, but the effect decays fast — typically within 20 to 60 trading days — and after transaction costs, the edge shrinks to something your broker’s commission structure will eat alive. If you’re building a systematic strategy around Monday’s calls, you’re essentially trading a signal with a half-life shorter than your settlement cycle.
That doesn’t make the calls useless; it makes them features, not alpha. Analyst revision breadth — the net number of estimate changes across a universe — is a documented factor with modest but persistent explanatory power, and it’s trivially computable in Python. Pull consensus data, diff the estimate distributions week over week, normalize by coverage, and you’ve got a revision-momentum signal you can stack into a multi-factor model alongside value, quality, and momentum. `pandas.diff()` on a rolled median estimate is doing more honest work here than any single “Top Pick” headline.
Take the names in today’s batch. A Lockheed upgrade is a defense-cycle story — order backlog, defense budget appropriations, and margin guidance — all of which are quantifiable directly from 10-K filings and government procurement data before any analyst re-rates the stock. Spotify’s calls tend to orbit subscriber-growth revisions, a metric with known seasonality that naive models routinely overfit. Block remains the poster child for narrative-driven re-rating: high beta to crypto sentiment, near-zero correlation stability, and a price history that will happily destroy any model trained on a bull-market window. Train on 2021 data and watch it die in 2022. Walk-forward validation isn’t optional; it’s the only thing separating a strategy from a story.
NetApp and Equifax are the boring ones, and boring is where factor models actually earn their keep. Enterprise-hardware names trade heavily on IT-spend cycles and backlog conversions; Equifax on employment data and mortgage volumes — both of which are macro inputs you can model directly rather than waiting for an analyst to translate them into a target with false precision to the cent.
The actionable takeaway: don’t trade the call, trade the revision flow the call represents. Rank a universe by 4-week net estimate revision momentum, tilt long the top decile, short or underweight the bottom, control for sector exposure, and size positions off realized volatility, not conviction. Rebalance on a schedule determined by signal decay analysis, not analyst publication calendars. And for anyone about to backtest this on five years of data with 14 free parameters: your in-sample Sharpe is a lie, and the out-of-sample drawdown is the audit.
The market generated thousands of analyst ratings today. The ones worth anything were already priced in by lunchtime — the rest were content marketing with a price target attached.