Statement
In a Markov-switching rational-expectations model jointly fit to bond and CRE prices, anticipated regime transitions in the monetary-policy block produce quantitatively first-order movements in expected commercial real-estate cap rates and in spreads on CRE-collateralized mortgages. Specifically, an anticipated transition from an Active-Flexible regime into a Passive-Rigid regime implies an expected 100–150 bp decline in cap rates (a 15–20% increase in property values), and the mirror move generates the opposite effect of roughly the same magnitude. CRE mortgage default spreads at high LTV differ by over 100 bp across extreme policy scenarios, and the value of existing CRE loan portfolios can move by 20–30% on a single regime change.
Evidence summary
The evidence is the model-implied impulse-response exercise reported in Figures 6–8 (cap rates) and Figures 10–12 (mortgages) of the source paper. For each of the four origin regimes the model is initialized at the conditional mean of the macro state, then a single regime change is forced one quarter in and the resulting expected cap-rate / mortgage-spread / loan-portfolio-value trajectory is traced out 40 quarters forward. The estimated risk premia ordering (Table 8) corroborates the price ordering: Passive-Rigid has the lowest real-estate risk premium and the highest model-implied prices, Active-Flexible the opposite. The mortgage exercise additionally documents that Apartment mortgages have the smallest spreads — consistent with both their lower NOI volatility and the GSE-subsidy hypothesis the authors flag.
Conditions and scope
- Risk-neutral / model-internal counterfactuals, not natural experiments. The “100–150 bp cap rate decline” is the change in the model’s expected forecast of cap rates conditional on a forced regime change, not an identified causal effect of an observed regime change.
- Sudden, deterministic regime changes in the impulse-response design; in practice the regimes are smoothly evolving with persistence > 0.95 per quarter, so the literal magnitudes are upper-bounds for any realistic policy shock.
- No zero lower bound, no unconventional policy. The QE era is forced into the four-regime structure as a Passive-Rigid classification; whether this is a faithful representation of QE is an open empirical question.
- The mortgage exercise abstracts from prepayment. Real CRE mortgages have prepayment options and yield maintenance / defeasance covenants that are not modeled.
Counter-evidence
- The 2005–2010 model–price gap (CRE prices persistently above their model-implied values across all asset categories) is direct evidence that the regime-switching model does not capture all of the variation in cap rates during the most policy-relevant period.
- The model has a single common CRE risk factor
Z_t. Cross-sectional dispersion in cap rates that is driven by sector-specific shocks (e.g. the office crisis post-2020) is forced into idiosyncratic measurement error and may bias the regime-effect estimates. - The Apartment mortgage spread is suspiciously low and the authors themselves attribute part of it to the GSE subsidy — i.e. some of the model’s apparent regime sensitivity for Apartment mortgages may be partially absorbing a non-regime institutional effect.
- The likelihood is profoundly sloppy (project measurement: condition number
~10²⁵–10²⁶on the Hamilton surrogate), so the magnitudes of the estimated regime sensitivities have wide credible intervals that the paper’s bootstrap procedure does not fully reflect.
Linked ideas
- out-sample-cre-cap-rate-forecasting — out-of-sample cap rate forecasting tests whether anticipated regime changes have predictive power beyond in-sample fit
Open questions
- How much of the impulse-response magnitude survives an extension to a prepayable-mortgage model?
- Does the cap-rate impulse response replicate on alternative CRE price series (Green Street, RCA)?
- What does the model predict for the 2022–2024 Fed tightening cycle, and does that prediction agree with realized cap-rate movements out of sample?
- Can the regime-switching CRE risk premium be identified from a CRE-only panel (without bonds), or is bond information necessary for identification?