Analyst revisions

Definition

Analyst revisions — changes in earnings estimates, price targets, and ratings — are the professional expectations layer in motion. The documented facts: revisions DRIFT (upgrades follow upgrades — analysts anchor and adjust incrementally, bias-anchoring institutionalized), revision momentum carried return information in classic samples (earnings-estimate revisions were one of Chan-Jegadeesh- Lakonishok’s momentum ingredients), and recommendation changes moved prices with post-event drift (Womack 1996). The LEVEL of ratings is nearly information-free; the CHANGE is the signal.

How it works / structure

• The revision cascade: company guides or prints → estimates move → the stock’s consensus trajectory shifts — revision BREADTH (how many analysts moved) and MAGNITUDE (how far) are the standard signal dimensions (fa-guidance-estimates is the layer being revised).
• The documented anchoring: analysts revise in strings of small steps toward the eventually-correct number (under-reaction, documented across decades) — the mechanical basis of revision momentum; PEAD (event-earnings) is its price-side twin.
• Ratings-level worthlessness: the distribution sits overwhelmingly at buy/hold (sell ratings are rare — conflict-of-interest structure is documented); the level carries almost nothing, changes and their direction carry the information (Womack: upgrades drifted +2-3%, downgrades −4-5% over following months in his sample).
• Engine-executable form: revision-breadth score (net up-revisions / total, rolling window), estimate-momentum percentile, days-since-revision freshness, and the conflict flags (banking relationships are disclosed — parsed where available).

When it applies

Momentum-family strategies (revision momentum is the fundamental-information variant — strategy-momentum); earnings positioning (revision trajectory INTO the print conditions the surprise reaction — fa-guidance-estimates); forward-multiple hygiene (fa-multiples-comparables — the denominator’s momentum matters as much as its level).

Risk profile & failure modes

• Post-publication decay: revision-momentum returns attenuated in recent large-cap samples like every published anomaly (lens-quantitative); current-sample replay required.
• Crowded-name saturation: mega-caps with 40 analysts revise continuously — signal density concentrates in mid-caps with moderate coverage.
• Level-reading regression: “consensus is buy” as thesis support — the level’s documented emptiness makes this pure decoration.
• Reflexivity at turns: revision strings follow the business cycle with a lag; the drift’s END (peak revisions) clusters near cycle turns — momentum’s usual crash shape in fundamentals clothing.

Evidence & limits

Womack (1996) and Chan-Jegadeesh-Lakonishok (1996) are the peer-reviewed anchors; under-reaction/anchoring in estimates is documented across the accounting literature; ratings- distribution conflicts are documented and disclosure- regulated. Effect sizes are sample-bound and decayed — quoted as such.

Falsifiable-thesis examples

Illustrations only, not signals:

• “The top revision-breadth decile of sector S will outperform the bottom decile over the next quarter” — falsified by the cohort pair.
• “X, with three upgrades this month, will not be downgraded by any covering analyst within the quarter (string- persistence thesis)” — falsified by a downgrade.

Cross-references

• The layer revised: fa-guidance-estimates; the price twin: event-earnings (PEAD)
• The strategy family: strategy-momentum
• The bias mechanized: bias-anchoring
• The lens: lens-sentiment (professional expectations in motion)