Statement

The latent regimes recovered by Markov regime-switching CIR-class term-structure models, when estimated on US Treasury yield data, are economically interpretable as business-cycle phases. Specifically: a high-volatility, low-mean-reversion regime concentrates in NBER recession dates, while a low-volatility regime dominates NBER expansions. This alignment provides post-hoc economic credibility for the latent regime mechanism.

Evidence summary

Bansal and Zhou (2002) compare the filtered probabilities of regime 1 vs regime 2 from their estimated regime-switching CIR model against the NBER business-cycle reference dates and observe substantial overlap between the high-volatility regime and recession periods. The comparison is qualitative — based on visual inspection and probability series, not a formal classification score against alternative regime labelings.

Conditions and scope

Holds only when all of the following are true:

  • US Treasury yields are the data source.
  • A regime-switching CIR-class model (2 states) is the estimator.
  • The reference is the NBER business-cycle dating committee.
  • The comparison is interpreted post-hoc and visually.

The claim is not asserted for: non-US markets, non-Treasury debt, structurally identified macro models, or formal hypothesis tests against alternative regime labelings.

Counter-evidence

None reported in the paper itself. Independent regime-switching macro-finance work (Hamilton 1989 on GDP, Ang-Bekaert-Wei on yields) reports similar alignments but with different intensities depending on data and model class. Whether the alignment is unique to CIR-class models or generic across regime-switching estimators is not established.

Linked ideas

(none yet)

Open questions

  • Is the regime/business-cycle alignment a robust feature of any reasonable latent-regime model on US yields, or is it specific to the CIR functional form?
  • Does the alignment survive structural identification of regimes against macro variables (inflation, monetary policy, output gap)?
  • For applications outside Treasuries (corporate bonds, commercial real estate, equity), does the regime classification still align with NBER cycles, or do the regimes pick up sector-specific cycles instead?