Skip to content
StockMarketAgent
Methodology · Earnings revisions

Revision diffusion index

A measure of whether sell-side analyst earnings estimates are being revised up or down across the index, expressed as the share of constituents with positive net revisions over the lookback.

Specification

Revision diffusion index — operational spec

Negative readings turn two to four months before a recession-driven earnings recession shows up in reported numbers, which is why this is the heaviest-weighted sub-component of the regime composite when interpreted alongside breadth.

Inputs
  • NTM EPS estimates for every covered constituent (sell-side consensus, daily)
  • Index membership and market-cap weights at the snapshot date
  • Trailing 4-week revision window (configurable)
Computation
  • For each constituent, the trailing 4-week change in NTM EPS estimate is classified as up, flat, or down.
  • Diffusion = (count up minus count down) divided by the total covered.
  • Computed twice: once equal-weighted, once market-cap-weighted.
  • A divergence flag fires when cap-weighted diffusion is positive but equal-weighted is negative — a classic narrow-leadership warning.
Outputs
  • Equal-weighted revision diffusion index (-100 to +100).
  • Market-cap-weighted revision diffusion index (-100 to +100).
  • Divergence flag with magnitude.
  • Sector-decomposed contribution to net diffusion.
Limitations
  • Coverage is biased toward larger names; smaller constituents can have stale estimates that mute the signal.
  • Estimate revisions cluster around earnings prints, producing a quarterly seasonal pattern that needs to be controlled for in the historical replay.
  • The signal is forward-looking but lags the underlying economic data that drives it by a few weeks.

Frequently asked

Why publish both equal-weighted and cap-weighted versions?
They tell different stories. Cap-weighted is what drives index returns; equal-weighted is what is happening to the median constituent. A meaningful gap between the two is itself a signal: cap-weighted positive while equal-weighted negative is a classic narrow-leadership warning that historically precedes regime change.
How long is the trailing window?
Four weeks by default. Shorter windows are too noisy; longer windows lag inflection points. Four weeks is the empirical sweet spot in our backtests, but the parameter is published so the reader can rebuild the index over a different window if desired.
Does this front-run reported earnings?
Yes, by construction. Sell-side analysts revise estimates as they receive new information from guidance, macro data, and competitor prints. The diffusion index aggregates those private updates into a public number that typically turns 2-4 months before reported earnings confirm the trend.
See the full methodology hub for the rest of the model registry, or open the glossary entry for the headline definition.