Skip to content
StockMarketAgent
Methodology · Market regime

Market regime model

A composite scorecard that classifies the equity market into a regime band by rolling up seven equally weighted sub-components into a single 0-100 score with explicit confidence.

Specification

Market regime model — operational spec

Top-of-funnel reading on the monthly outlook surface. Sets the prior for individual position sizing and is published with confidence so the implied fair-value band widens or tightens transparently.

Inputs
  • 10y minus 2y Treasury spread (FRED, daily)
  • NTM EPS revision diffusion across the index (sell-side consensus, weekly)
  • BBB option-adjusted credit spread (FRED, daily)
  • Breadth health score (this site, daily)
  • AAII / II sentiment composite, NDR-style smoothing (weekly)
  • Valuation dispersion percentile (this site, monthly)
  • Citi macro surprise index (daily)
Computation
  • Each sub-component is winsorized at the 1st / 99th percentile across the full multi-decade history.
  • Winsorized values are converted to a 0-100 percentile rank against their own history.
  • Equal weights are applied. Weighting by perceived informativeness back-fits to the most recent regime and is rejected.
  • Confidence is computed as 100 minus the dispersion of sub-component scores. High disagreement -> low confidence.
  • The composite is mapped to a four-band classification: early-cycle, mid-cycle, late-cycle, contraction.
Outputs
  • Composite score 0-100 with a regime band label.
  • Confidence number 0-100 used to size the implied fair-value band.
  • Seven sub-component scores with month-over-month deltas.
  • Composite implied fair value (low / mid / high) for the index.
Limitations
  • Equal-weighted by design. A regime where one sub-component is structurally dominant will be under-fit.
  • Confidence is an internal disagreement metric, not a true forecast variance.
  • Sub-component data is point-in-time but vendor revisions can leak look-ahead into the historical replay; treat the longest-tenure backtest with skepticism.
  • Not a market-timing signal. The model is silent on next-month returns.

Frequently asked

Is the regime model a market-timing signal?
No. The model has no return-prediction component. It is a regime classifier whose output sets the prior for position sizing — wider margin of safety in late-cycle / contraction regimes, tighter in early- / mid-cycle regimes. Treating it as a buy / sell signal is a category error.
Why are the sub-components equal-weighted?
Weighting by perceived informativeness back-fits to the most recent regime and produces overconfident reads at exactly the moment a new regime is starting. Equal weights are a deliberate constraint that trades a small amount of in-sample fit for considerably better out-of-sample stability.
How does confidence affect the implied fair-value band?
Low confidence widens the band. The convention is to translate the inverse confidence into a percentage band around the composite implied fair value, so a 40-confidence reading produces a materially wider published low-to-high range than an 80-confidence reading. Tightening the band when confidence is low is the opposite of what the model recommends.
What is the lookback for the percentile ranks?
The full available history of each sub-component, capped at the longest series we trust. Most sub-components run from the late 1980s. Newer indices use the longest contiguous series, with the start date documented in the monthly outlook footnotes.
See the full methodology hub for the rest of the model registry, or open the glossary entry for the headline definition.