Monday’s Analyst Calls: Buy-Side Noise or Actionable Signal? My Backtest Says Proceed Carefully
Monday’s analyst research calls — Block, Boise Cascade, Equifax, FirstEnergy, Lockheed Martin, Lumentum, NetApp, Roper, Spotify, and a handful of tech names — arrived on schedule, as they do every trading day, and the market absorbed them with the usual enthusiasm of a factor model ingesting stale data.
Here’s the uncomfortable statistical reality: sell-side price targets have roughly the predictive power of a 50-day moving average crossed with a coin flip. Academic work going back decades ( Barber, Lehavy, McNichols, Trueman’s studies on analyst recommendations being the canonical reference) shows buy ratings exhibit weak, decaying post-publication drift that evaporates once you control for size, momentum, and liquidity. If you’re routing analyst upgrades into your signal stack as a standalone alpha source, your backtest is telling you something your ego isn’t.
That said, these calls aren’t useless — they’re just mispriced as *information* rather than *direction*. A batch of upgrades or downgrades across a sector is a sentiment feature. In Python, this is trivial: pull the recommendation events, convert to a signed score, lag appropriately to avoid look-ahead bias, and test incremental Information Coefficient against your existing model with `alphalens` or a custom `pandas` pipeline. Most of the time the marginal contribution is negligible. Occasionally — particularly in low-coverage names where a single downgrade moves consensus estimates — it isn’t.
The names themselves offer a lesson in factor dispersion. Lumentum and the optical/photonics complex trade on AI datacenter capex narratives, which means they’re high-beta to a theme with wide analyst target dispersion — precisely the conditions where recommendation noise is loudest and signal-to-noise worst. Lockheed Martin, by contrast, sits in a low-volatility, dividend-heavy pocket where analyst calls barely move the tape; defense names respond to contract flow and budget cycles, not brokerage sentiment. Block and Spotify are momentum-heavy retail favorites where downgrades function more as short-term liquidity events than information shocks — exploit the overreaction, not the rating.
FirstEnergy and Equifax are the interesting cases: regulated utility and data-centric financials respectively, where estimate revisions actually matter because earnings are relatively modelable. If you’re going to use analyst work anywhere, it’s there — in the *estimate revision* channel, not the price target channel. Consensus EPS revision momentum remains one of the more durable post-earnings-announcement-drift-adjacent signals in the literature.
Practical takeaway: scrape the calls, tag the tickers, lag the events, and run a proper event study before letting any of it touch your book. The analysts did their job — publishing. Your job is verifying whether it carried information. Most Mondays, the Sharpe contribution rounds to zero, and the honest quants say so out loud.