LeatherSagiKnowledgebase

Expected Returns and Expected Growth in Rents of Commercial Real Estate

5 min read

Problem

Commercial real estate prices fluctuate considerably across both locations and time. The fundamental present-value relation implies that observed cap rate variation should reflect variation in future expected returns, future expected rent growth, or both. Despite this, little was known about the dynamics of CRE returns and rent growth rates across property types and metropolitan areas.

Key idea

Using a “dynamic Gordon model” (Campbell-Shiller 1988b present-value decomposition applied to CRE), the authors establish that the cap rate — the ratio of rent to price, analogous to the dividend-price ratio for equities — captures time variation in expected returns for apartments, retail, and industrial properties, but NOT for office properties. The key structural insight is that the correlation between expected returns and expected rent growth differs across property types: for offices, the two are highly positively correlated, which causes the cap rate to lose its predictive power for returns (the two effects cancel in the cap rate). For the other three property types, expected returns dominate the cap rate variation, generating economically significant return predictability.

Method

  • Data: Transaction-based (not appraisal-based) commercial real estate data from NCREIF across 53 U.S. metropolitan areas at quarterly frequency (1994Q2-2003Q1), with a subset of 21 MSAs at semiannual frequency (1985Q2-2002Q4). Four property types: apartments, retail, industrial, office.
  • Model: Structural present-value model with AR(1) expected returns and AR(1) expected rent growth rates:
    • r_{t+1} = r_bar + x_t + zeta^r_{t+1} (returns)
    • Delta h_{t+1} = g + tau*x_t + y_t + zeta^h_{t+1} (rent growth)
    • x_{t+1} = phi*x_t + zeta^x_{t+1} (state variable)
    • y_{t+1} = psi*y_t + zeta^y_{t+1} (rent-specific state)
  • Estimation: Reduced-form predictive regressions of returns and rent growth on lagged cap rates, with structural parameters backed out from the reduced-form coefficients. Panel regressions with cross-sectional analysis conditioning on population density, land-use restrictions, and local economic variables.

Results

  • Apartments, retail, industrial: Cap rates positively and significantly predict next-period returns (beta_r > 0) but do NOT significantly predict rent growth. The return predictability generates economically significant price movements.
  • Offices: Cap rates do NOT predict returns but DO weakly track expected rent growth (lambda > 0). This is because the structural parameter tau (the exposure of rent growth to the expected-return state) is highest for offices, causing the discount-rate and cash-flow effects to offset in the cap rate.
  • Cross-sectional analysis: Return predictability is stronger in MSAs with higher population density and more restrictive land-use regulations, where supply is constrained and price variation is dominated by demand/discount-rate shocks rather than supply adjustments.
  • Portfolio allocation: A mean-variance investor who accounts for CRE return predictability optimally allocates 12-44% to CRE (depending on risk aversion and rebalancing frequency), compared to 0-5% ignoring predictability.

Limitations

  • Sample period (1994-2003 quarterly, 1985-2002 semiannual) ends before the mid-2000s CRE boom and 2008 bust, limiting the coverage of the most volatile period.
  • NCREIF transaction data may still contain appraisal-related biases since cap rates are partly derived from appraised values.
  • The structural model assumes AR(1) dynamics for expected returns and rent growth, which may be too restrictive for capturing regime-switching behavior.
  • The cross-sectional analysis (53 MSAs) has limited power for identifying causal effects of supply constraints vs. demand characteristics.
  • The portfolio allocation exercise assumes CRE positions are divisible and diversified across types and locations, which is unrealistic for most institutional investors.

Open questions

  • Does the office cap-rate predictability puzzle persist in out-of-sample data covering the 2003-2025 period, including the GFC and COVID?
  • Can the failure of cap rates to predict office returns be resolved by adding regime-switching dynamics to the expected-return process?
  • What is the structural economic explanation for why office rent growth covaries more strongly with discount rates than other property types?

My take

This paper is a key empirical reference for the CRE asset pricing project. Its central finding — that CRE cap rate variation is dominated by discount-rate news for three of four property types, with offices being the exception — is the reduced-form stylized fact that the Leather-Sagi MS-RE model is designed to explain structurally. The AR(1) present-value framework used here is the methodological precursor to the state-space approach in our project, though our model takes the next step of embedding the dynamics in a no-arbitrage NK macro framework with Markov switching. The cross-sectional result (predictability strongest where supply is constrained) suggests that the single-factor macro model may miss important supply-side heterogeneity. The Duca-Ling (2015) paper explicitly cites and builds on this framework, using survey data instead of transaction data to measure the same quantities.

Graph View

Backlinks