Abstract. Trading the sign of analyst revisions to earnings estimates is decayed, sensitive to cost and close to price. Any residual edge sits in the second moment. Filed as a view rather than a result because the test needs a paid feed.

Keywords: earnings estimates, revision momentum, analyst dispersion, factor crowding, positioning, data economics.

1. The consensus signal

Stocks whose consensus estimates are revised upward tend to outperform over the following weeks, a drift as old and as documented as the drift after earnings surprises (Givoly and Lakonishok, 1979; Chan, Jegadeesh, and Lakonishok, 1996). The signal is clean, slow moving and vendor-packaged, and that legibility is the problem.

2. Why not the consensus way

Three reasons. Crowding: a signal legible enough to be catalogued in the factor zoo is in the price by the time it is clean. Decay under cost: a first-moment signal turns over often and bleeds its edge to transaction cost under a margin of safety. Proximity to price: revisions and prices move together, so trading the sign is trading momentum renamed.

The three claims, and the record each rests on.

ClaimWhat the record says
Revision momentum is crowdedOne of more than 400 return predictors catalogued in the published factor zoo (Harvey and Liu, 2019)
Price-adjacent families decay under costThe 158 technical trading signals and the within-sector portfolios, tested and retired elsewhere on this site
Positioning and flow are what surviveThe two signals that have cleared this pipeline’s bar both read positioning and flow, not past returns
The pipeline’s own revision attemptError status in the database of every idea tested here; recorded cause a vendor API timeout; not retried, for the reason in section 4

3. Where I would look instead

The second moment is less picked over. Revision dispersion, the disagreement across analysts around a revision, and the flow forced when a number gaps, are mechanisms rather than price patterns. I would test those, not the sign.

4. The data economics

A clean history of estimate revisions is a paid vendor feed. The free proxies rebuild revisions from price, re-importing the first-moment crowding the exercise exists to avoid. Naming the data that would settle the question beats faking it. Refusing the contaminated test is not an absence of work. It is the work.

5. The view

I would trade the disagreement around the revision and the flow it forces, and put no capital on either until it cleared the margin of safety that has retired every price signal tested here. A view is a hypothesis with a bill attached. This note states both.

This is a position, not a backtest, and it is labelled as one. An untested claim is filed here as a view, never dressed as a result. The discipline that refuses to promote a statistically significant finding before it is stress-tested is the one that refuses to manufacture a finding from convenient data.

References

  1. Chan, L. K. C., Jegadeesh, N., and Lakonishok, J. (1996). Momentum Strategies. Journal of Finance, 51(5), 1681–1713.
  2. Givoly, D., and Lakonishok, J. (1979). The Information Content of Financial Analysts’ Forecasts of Earnings. Journal of Accounting and Economics, 1(3), 165–185.
  3. Harvey, C. R., and Liu, Y. (2019). A Census of the Factor Zoo. SSRN Working Paper 3341728.