Abstract
We develop a general construction for nonlinear Lévy processes with given characteristics. More precisely, given a set $\Theta$ of Lévy triplets, we construct a sublinear expectation on Skorohod space under which the canonical process has stationary independent increments and a nonlinear generator corresponding to the supremum of all generators of classical Lévy processes with triplets in $\Theta$. The nonlinear Lévy process yields a tractable model for Knightian uncertainty about the distribution of jumps for which expectations of Markovian functionals can be calculated by means of a partial integro-differential equation.
Introduction
In classical probability theory and stochastic analysis, we typically operate under the assumption of a fixed, known probability measure. This framework allows us to model randomness—such as the price of an asset or the arrival of insurance claims—provided we know the underlying "rules" (e.g., the drift, volatility, and jump frequency). However, in many complex systems, we encounter Knightian uncertainty: a situation where the probability model itself is unknown or misspecified.
In "Nonlinear Lévy Processes and their Characteristics," Ariel Neufeld and Marcel Nutz address this challenge by providing a rigorous mathematical foundation for stochastic processes that remain robust under model uncertainty. By extending the concept of Lévy processes to a nonlinear setting, the authors create a framework capable of handling uncertainty in drift, diffusion, and, crucially, complex jump structures.
The Challenge of Model Risk
A standard Lévy process—such as Brownian motion or a Poisson process—is characterized by a single triplet $(b, c, F)$, representing its drift, diffusion (volatility), and jump measure, respectively. While previous research into $G$-expectations and $G$-Brownian motion successfully addressed uncertainty regarding volatility, these models often struggled with the inclusion of jumps, particularly those of "infinite variation" where the process jumps infinitely often within any finite time interval.
Existing frameworks were often restricted to specific types of jumps or required stringent assumptions that limited their applicability in finance and insurance. The primary goal of Neufeld and Nutz is to provide a unified construction that accommodates a broad range of jump behaviors while maintaining the computational tractability required for practical applications.
Methodology: Sublinear Expectations and Optimal Control
To account for model uncertainty, the authors move away from a single probability measure. Instead, they consider a set of possible measures $\mathcal{P}$. Within this set, they define a sublinear expectation $\mathcal{E}(\cdot)$, which essentially calculates the "worst-case" expectation across all candidate models:
$$\mathcal{E}(X) = \sup_{P \in \mathcal{P}} E^P[X]$$
Construction on Skorohod Space
The construction is performed on the Skorohod space of càdlàg paths (right-continuous with left limits), which is the natural environment for processes with jumps. The authors frame the problem through the lens of optimal control. Here, the "controls" are the laws of semimartingales whose local characteristics are constrained to remain within a prescribed set of triplets $\Theta$.
This approach requires significant measure-theoretic "heavy lifting." The authors prove that the resulting sets of measures are sufficiently well-behaved (analytic and measurable) to support the use of dynamic programming principles, ensuring the process is mathematically consistent across time.
Key Insights and Theoretical Results
1. The Supremum Generator
The defining feature of a nonlinear Lévy process is its generator. In a classical setting, the generator is a linear operator. In this nonlinear framework, the generator $G$ is defined as the supremum of all linear generators associated with the triplets in the set $\Theta$. This operator acts as the "engine" of the process, capturing the maximum possible rate of change across all potential models.
2. Stationary Independent Increments
One of the paper's most significant findings is that the defining characteristics of Lévy processes—stationary and independent increments—can be preserved even under uncertainty. The authors demonstrate that the "uncertainty" regarding the process's behavior in a future interval is independent of its past trajectory, and the range of possible outcomes depends only on the duration of that interval.
3. Connection to Nonlinear PIDEs
The authors establish a vital link between abstract probability and computable analysis. They prove that the value function (the worst-case expectation of a future payoff) is the unique viscosity solution to a nonlinear Partial Integro-Differential Equation (PIDE). For a bounded set of triplets $\Theta$, the expectation $v(t, x) = \mathcal{E}(\psi(x + X_t))$ solves:
$$\partial_t v - \sup_{(b,c,F) \in \Theta} \left\{ \text{Generator}(b,c,F) v \right\} = 0$$
This result transforms a complex problem of supremum expectations into a deterministic calculus problem, allowing for the application of numerical methods to price options or assess risk under uncertainty.
Conclusion and Future Directions
Neufeld and Nutz successfully provide an existence theorem (Theorem 2.1) for nonlinear Lévy processes for any measurable set of triplets $\Theta$. By including examples such as Poisson processes with uncertain intensity and nonlinear $\alpha$-stable processes, they demonstrate the versatility of their framework.
While the paper marks a significant advancement, it also identifies areas for further research. The current construction assumes the set $\Theta$ is constant over time; extending this to "nonlinear semimartingales," where the set of possible models evolves based on the state of the process, remains a promising frontier. Additionally, while the authors construct the process, a full representation theorem for the infinite variation jump case remains an open challenge in the field of stochastic analysis.
Summary updated