Adapted Law Invariance and Time-Consistent Dynamic Risk Measures

Abstract: In static risk measurement, law invariance expresses the principle that the risk of a position should depend only on its distribution, and not on the particular probability space on which it is represented. In a dynamic setting, the same principle leads naturally to adapted law invariance: the risk assessment should depend only on the probabilistic structure of the financial position together with the way information about it is revealed over time. We show that, for time-consistent risk measures, adapted law invariance is equivalent to a recursive one-step conditional-law representation. More precisely, assuming Fatou regularity, the one-step risk evaluations are exactly conditional lifts of static law-invariant risk measures, and the full dynamic risk measure is obtained by backward composition of these one-step maps. Convexity and coherence of the dynamic risk measure are characterized by the corresponding properties of the static one-step risk measures. This identifies adapted law invariance as the dynamic counterpart of ordinary law invariance. It also clarifies the strength of terminal-law invariance, as it appears in the rigidity theorem of Kupper and Schachermayer: it does not distinguish risks with the same distribution but different times of resolution. We further obtain an adapted Kusuoka representation in the coherent case and establish an extension of the Kupper--Schachermayer theorem.

• The paper establishes a sharp characterization where each one-step conditional risk evaluation depends solely on its conditional law via a static law-invariant risk measure.
• It extends static law invariance to capture temporal information flow, leading to a nested adapted Kusuoka representation that generalizes traditional risk measures.
• The findings broaden the class of admissible dynamic risk measures, offering practical insights for risk management in finance and insurance.

Adapted Law Invariance and Time-Consistent Dynamic Risk Measures: An Expert Analysis

Law Invariance in Risk Measurement: From Static to Dynamic Settings

The principle of law invariance is a foundational axiom in the theory of monetary risk measures, positing that the risk of a given stochastic position depends solely on its distribution and not on the specifics of the underlying probability space. In a static context, this is formalized by requiring the risk measure $\bar\rho$ to satisfy $\bar\rho(X) = \bar\rho(\widetilde X)$ whenever $X$ and $\widetilde X$ are random variables with the same law. Statistically, this principle is justified by the inaccessibility of the sample space; only the distribution of observed outcomes is informative.

Extending law invariance to a dynamic (multiperiod) setting is subtle. In filtered probability spaces $$(\Omega, \mathcal{F}, \mathbb{P}, (\mathcal{F}_t)_{t=0}^N)$$, not only the terminal law of a claim matters, but also the information flow—how uncertainty about payoffs resolves over time. Two positions may share the same terminal distribution yet present fundamentally different risk profiles due to disparate information revelation timelines. This motivates the refined notion of adapted law invariance, which augments the law with the filtration, encoding the sequential structure of information.

Time-Consistency and Adapted Law Invariance

Dynamic risk measures are mappings $(R_t)_{t=0}^N$ with $$R_t: L^\infty(\mathcal{F}_N) \to L^\infty(\mathcal{F}_t)$$, satisfying monotonicity, cash additivity, normalization, and a terminal condition. The property of time consistency requires $R_t(X) = R_t(R_{t+1}(X))$ for all $t$, ensuring that risk assessments evolve coherently in time.

The paper establishes a sharp characterization of time-consistent, relevant dynamic risk measures that also satisfy adapted law invariance and the Fatou property. The central result demonstrates that adapted law invariance for a time-consistent risk measure is equivalent to having each one-step conditional risk evaluation depend only on the conditional law of the next-period position, via a static law-invariant risk measure. Explicitly,

$$S_t(Y) = \sigma_t(L(Y|\mathcal{F}_t)), \quad Y \in L^\infty(\mathcal{F}_{t+1}),$$

where each $\bar\rho(X) = \bar\rho(\widetilde X)$0 is a static law-invariant risk measure on distributions, and the entire dynamic risk measure is constructed by the backward iteration of these one-step conditional maps.

Furthermore, the convexity or coherence of the dynamic measure is inherited from these static risk maps; thus, the global property is reduced to a stepwise property. The results are formalized in Theorem 1 (Main characterization of relevant time-consistent risk measures) of the paper.

Comparison with Terminal-Law Invariance and Rigidity Theorems

The Kupper–Schachermayer rigidity theorem, which operates under terminal-law invariance (i.e., invariance under the ordinary marginal law without regard for the filtration), established that the only relevant, normalized, time-consistent dynamic risk measures under this axiom are conditional entropic risk measures (or their extremal cases: conditional expectation or conditional essential supremum). The present paper clarifies that terminal-law invariance is far more restrictive than adapted law invariance, as it forces identification of positions that differ in the timing of information but have the same terminal law.

The authors further contribute by providing a finite-horizon (two period) proof of the Kupper–Schachermayer theorem, directly exhibiting the entropic form as a consequence of the assumptions. Under adapted law invariance, however, the class of admissible dynamic risk measures is substantially richer: the risk depends on both the law and the filtration's information structure, providing a more faithful representation of risk in multiperiod settings.

The Adapted Kusuoka Representation

In the coherent case (convex and positive homogeneity), the static Kusuoka representation (Kusuoka, 2001) states that law-invariant coherent risk measures are characterized on the law space by mixtures of Average Value-at-Risk (AVaR) functionals. The present work extends this to the dynamic setting under adapted law invariance and time consistency, yielding a nested adapted Kusuoka representation. At each time $\bar\rho(X) = \bar\rho(\widetilde X)$1,

$\bar\rho(X) = \bar\rho(\widetilde X)$2

with $\bar\rho(X) = \bar\rho(\widetilde X)$3 a convex, weakly closed set of measures on $\bar\rho(X) = \bar\rho(\widetilde X)$4. The entire dynamic risk measure is formed by iterated composition of these conditional AVaR operators. This form is notable for its capacity to represent a large spectrum of dynamic, time-consistent, coherent, adapted-law-invariant risk measures, significantly generalizing the entropic case forced by terminal-law invariance.

Implications and Future Developments

The implications of this work are both structural and practical:

• Structural: The complete characterization situates adapted law invariance as the natural symmetry principle for dynamic risk assessment in filtrated probability spaces, adequately refining the static concept to capture the temporal aspect of risk revelation.
• Practical: The adapted Kusuoka representation yields tractable, implementable forms for dynamic risk measures that can be harnessed in finance, insurance, and risk management, where sequential information flow is essential.
• Contrast with terminal-law invariance: The results decisively show that terminal-law invariance is unnatural for filtered models where temporal resolution of information is fundamental.

This framework is also broadly compatible with dynamic extensions to robust and distributionally ambiguous settings, as well as with applications to dynamic risk sharing, portfolio selection under risk constraints, and stochastic optimal control.

Directions for future research may include:

• Relaxing the finite-time, discrete framework to continuous-time filtrations, potentially requiring advanced measure-theoretic tools (e.g., martingale problems, BSDEs).
• Exploring computational schemes for the nested Kusuoka form, particularly for high-dimensional or path-dependent contingent claims.
• Extending the concept to multi-agent or market-consistent risk measurement in incomplete markets, where different agents may have different informational filtrations.

Conclusion

This work provides a mathematically rigorous and comprehensive foundation for the theory of adapted law invariance in dynamic risk measurement, resolving the interplay between information flow and risk symmetry. The main characterization theorem and the adapted Kusuoka representation significantly expand the class of admissible time-consistent dynamic risk measures beyond those allowed by terminal-law invariance, offering a refined modeling tool for time-evolving risk assessment in financial and insurance applications (2607.04392).