Monday’s Analyst Calls: Signal or Noise? My Screen Weighs the Evidence
Monday’s analyst research calls arrived as usual: a list of tickers — Block, Boise Cascade, Equifax, FirstEnergy, Lockheed Martin, Lumentum, NetApp, Roper, Spotify — attached to ratings that, in aggregate, carry roughly the predictive power of a coin flip with a slight upward drift caused by the fact that almost nobody issues Sells.
The base rates are well documented. Buy ratings outnumber Sell ratings on Wall Street by something like 5-to-1, and the average price target is set so a stock “has upside” roughly 90% of the time. If your signal generator produced outputs this skewed, you’d delete the feature and move on. Yet these calls move intraday prices, which means they’re tradeable — not as alpha, but as a short-horizon volatility event. The empirical pattern: price target changes and downgrades produce abnormal returns concentrated in the open-to-10:30 window, decaying within a session or two. That’s an execution problem, not a conviction problem. If you’re trading it, you’re front-running reaction flows, not agreeing with the analyst.
A few of the names deserve specific skepticism. Lockheed Martin — the analyst upgrade typically cites backlog and defense budgets, both of which are already priced in by anyone running even a naive factor screen on government revenue exposure. Spotify gets calls driven by narrative: podcast monetization, subscriber growth, “platform story.” Narrative-driven calls correlate poorly with next-quarter earnings surprises, and if you backtest “analyst price target increase” as a factor against forward 12-month returns, the edge mostly evaporates after controlling for momentum — meaning the analysts are often just re-rating stocks that already ran, which is momentum laundering, not research.
The industrials and data-center adjacent names — NetApp, Lumentum, Boise Cascade — are where the calls at least reference observable fundamentals: order books, optical transceiver demand, housing starts. That’s a better starting point, but the analyst’s timing is usually lagged. The earnings-announcement drift literature (post-earnings announcement drift, the one anomaly that’s survived decades of out-of-sample testing) suggests you’d do better systematically trading the *reaction to results* than the reaction to someone’s opinion about future results. `yfinance` gets you the prices, `pandas` gets you the event windows, and an hour of work gets you a backtest that tells you whether these calls ever mattered for a given name. Spoiler: usually they don’t, except in the first forty minutes.
The deeper issue is selection bias in the coverage itself. The tickers that make Monday’s roundup are the ones with catalysts — earnings, guidance, M&A chatter — which means the sample is conditioned on volatility, not on information content. Any aggregate statistic about “analyst accuracy” computed on this subset is survivorship-adjacent and mostly measures the news cycle, not analyst skill.
The actionable takeaway: treat analyst calls as an event calendar, not an edge. Log the tickers, timestamp the calls, measure the drift, and only allocate where the event-study evidence — not the analyst’s conviction — supports it. The market pays for processing, and the analyst already published their processing for free. Free is rarely where the alpha lives.