Regime-Shift Risk Premium

Definition

The regime-shift risk premium is the component of an asset’s expected excess return that compensates investors for bearing the risk that the underlying Markov regime will change. Formally, in a regime-switching no-arbitrage model, it is the part of the equilibrium expected return that arises from the wedge between the physical transition matrix ${\Pi }^P$ and the risk-neutral transition matrix ${\Pi }^Q$ — i.e., from the market price of regime-shift risk, separate from the within-regime market price of factor risk.

Intuition

In a regime-switching pricing model with state $s_t$, an asset’s value depends on the future regime path. There are two distinct sources of risk:

1. Diffusive factor risk within a regime — the same volatility risk you have in any single-regime affine model, captured by an essentially-affine market price of risk $\lambda (X_t,s_t)$.
2. Regime-jump risk — the possibility that next period the model jumps from regime $i$ to regime $j\ne i$, which can change the level, volatility, and persistence of factor dynamics in a discontinuous way.

If investors are risk-averse to (2), they demand compensation for it. The mathematical embodiment of that compensation is a Radon–Nikodym density ${\xi }_{ij}$ that twists ${\Pi }_{ij}^P$ into ${\Pi }_{ij}^Q$. The expected excess return on an asset can then be decomposed into a factor-risk component (driven by $\lambda (X_t,s_t)$) and a regime-shift component (driven by the gap between ${\Pi }^P$ and ${\Pi }^Q$).

A common but incorrect simplification in early regime-switching pricing models is to set ${\Pi }^Q={\Pi }^P$ (i.e., zero regime-shift risk premium) for tractability or because it “looks like” a complete-markets assumption. DSY 2007 makes the case that this assumption substantively understates long-end term premia and assigns regime-shift compensation to the wrong place in the model.

Formal notation

Let $M_{t+1}$ be the pricing kernel and split it as

$M_{t+1}=exp(-r_t)\cdot D_{t+1}^{factor}\cdot D_{t+1}^{regime}$

where:

• $D_{t+1}^{factor}=exp(-(1/2){\lambda }^{\prime }\lambda -{\lambda }^{\prime }{\epsilon }_{t+1}^P)$ is the within-regime essentially-affine density ($\lambda =\lambda (X_t,s_t)={\lambda }_0(s_t)+{\lambda }_1(s_t)X_t$).
• $D_{t+1}^{regime}={\xi }_{s_t{s}_{t+1}}(X_t)$ is the regime-shift density.

The risk-neutral transition matrix is then

${\Pi }_{ij}^Q(X_t)={\Pi }_{ij}^P(X_t)\cdot {\xi }_{ij}(X_t)/{\Sigma }_k{\Pi }_{ik}^P(X_t)\cdot {\xi }_{ik}(X_t)$

with normalization across $j$ so each row of ${\Pi }^Q$ sums to one. The market price of regime-shift risk is parametrized by the ${\xi }_{ij}$ function — typically exponential-affine in $X_t$:

$log{\xi }_{ij}(X_t)={\eta }_{ij,0}+{\eta }_{ij,1}^{\prime }{X}_t$

with ${\eta }_{ii,0}={\eta }_{ii,1}=0$ (no premium for staying put) as a normalization.

The decomposition of the one-period expected excess holding-period return on an $n$-period bond becomes:

$E_t[hp{r}^{(n)}]-r_t=(factor-riskpremium)+(regime-shiftriskpremium)$

where the regime-shift component is large when (a) $\Vert {\Pi }^Q-{\Pi }^P\Vert$ is large, (b) regime change has a big effect on bond prices through the regime-specific loadings $(A_{n-1}(j),B_{n-1}(j))$, or (c) both.

Variants

• Zero regime-shift premium (${\xi }_{ij}\equiv 1$, so ${\Pi }^Q={\Pi }^P$): historical default in early regime-switching pricing models. Tractable but materially understates long-end compensation in DSY 2007’s empirical work.
• Constant regime-shift premium (${\eta }_{ij,1}=0$, so ${\xi }_{ij}$ constant in $X_t$): twists ${\Pi }^P$ into a constant ${\Pi }^Q\ne {\Pi }^P$, allowing a nonzero average regime-shift premium but no time variation.
• Affine regime-shift premium (DSY 2007): $log{\xi }_{ij}(X_t)$ affine in $X_t$, giving a time-varying regime-shift component that responds to the macro state.
• Compound-regime extensions (current CRE asset pricing project): with two independent binary sub-chains, the regime-shift risk premium decomposes into two sub-premia (e.g., monetary policy regime risk vs wage rigidity regime risk), each priced separately.

Comparison

Specification	Time-varying premium	Closed form	Identification
${\Pi }^Q={\Pi }^P$ (zero premium)	n/a	Yes	n/a
Constant ${\xi }_{ij}$	No	Yes	Easy (just shifts mean term premium)
Affine $log{\xi }_{ij}(X_t)$ (DSY 07)	Yes	Yes (with log-linear ${\Pi }^Q$ approx)	Moderate — needs regime transitions and yield-curve variation

When to use

• The asset class has a long enough horizon that regime-change probabilities materially affect cash flows (long-maturity bonds, real estate with long hold periods, infrastructure).
• You suspect that earlier regime-switching pricing models systematically understate the term premium, especially at long maturities or during transition windows.
• You want to interpret time-varying term premia in terms of what kind of risk is being priced (factor-vol risk vs regime-jump risk), not just an unstructured single number.

Known limitations

• Identification is thin when in-sample regime transitions are few — even DSY 2007’s two-regime estimate of ${\lambda }^{regime}$ rests on a small number of Volcker-era and post-Volcker switches.
• The standard parametrization is exponential-affine in $X_t$; richer parametrizations exist but trade off identification.
• The regime-shift premium can absorb unrelated mispricing if the within-regime factor model is misspecified — the two channels are not always cleanly separable in finite samples.

Open problems

• A clean prior / regularization for ${\eta }_{ij,\cdot }$ that prevents the regime-shift premium from soaking up factor-model misspecification.
• Identification of separate regime-shift premia for compound (independent sub-chain) Markov models — does the product structure give one premium per sub-chain or one premium per compound state?
• The regime-shift risk premium under exponential-quadratic pricing factors — the affine recursion in DSY 2007 has a closed-form analog under quadratic factors (Riccati), but the regime-shift density derivation has not been written down for the quadratic case in the public literature, to my knowledge.

Key papers

• dai-singleton-yang-2007-regime-shifts-term-structure — canonical derivation, decomposition, and empirical evidence that the regime-shift risk premium is economically large for long-maturity US Treasury bonds and that setting it to zero meaningfully distorts the term premium.

My understanding

The regime-shift risk premium is the most economically important and least mechanically obvious piece of the regime-switching DTSM apparatus. Two things matter for the current CRE asset pricing project:

1. The premium is non-trivial — DSY 2007 makes a strong empirical case that it is large at long maturities. CRE has very long effective hold periods, so the premium should be at least as important here as in long Treasuries.

2. Compound regimes need separate prices. The current project decomposes $s_t$ into two independent binary chains. The regime-shift kernel should in principle carry two prices (one per sub-chain) plus possibly an interaction term. How best to identify these from the data is open and not directly covered by DSY 2007 (which has a single 2-state chain).

The literature gap on regime-shift pricing under quadratic pricing factors is the most directly project-relevant open problem here.