Measure-Valued CARMA Processes

Fred Espen Benth, Sven Karbach, Asma Khedher · 2025

Read Measure-Valued CARMA Processes by Fred Espen Benth, Sven Karbach, Asma Khedher (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-08852.txt.

Introduction

Fred Espen Benth, Sven Karbach, and Asma Khedher build measure-valued CARMA flows in 2025. Such flows stack past pull and drift in one time rule on wide math spaces. Their space holds signed counts, and jumps drive the flow in place of smooth noise.

Proofs give life, cone hold, and mean and spread forms for the flow. Set terms mark when the flow stays still in law over time. Use spans field marks in space and time, with links to past CARMA kin. The frame bends to wide needs since its parts can be set case by case.

Catalog record

CREATORFred Espen Benth, Sven Karbach, Asma Khedher
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE114 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-08852

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Measure-Valued CARMA Processes Thanks: Fred Espen Benth and Asma Khedher are grateful for the financial support by the Research Foundation Flanders (FWO) under the grant FWO WOG W001021N Fred Espen Benth, Sven Karbach, Asma Khedher Abstract . In this paper, we examine continuous-time autoregressive moving-average (CARMA) processes on Banach spaces driven by Lévy subordinators. We show their existence and cone-invariance, investigate their first and second order moment structure, and derive explicit conditions for their stationarity. Specifically, we define a measure-valued CARMA process as the analytically weak solution of a linear state-space model in the Banach space of finite signed measures. By selecting suitable input, transition, and output operators in the linear state-space model, we show that the resulting solution possesses CARMA dynamics and remains in the cone of positive measures defined on some spatial domain. We also illustrate how positive measure-valued CARMA processes can be used to model the dynamics of functionals of spatio-temporal random fields and connect our framework to existing CARMA-type models from the literature, highlighting its flexibility and broader applicability. Keywords: CARMA, Linear state-space models, Measure-valued processes, Lévy subordinator, Banach space-valued processes, Stationarity. 1. Introduction In this paper, we introduce and analyze a class of measure-valued, Lévy-driven continuous-time autoregressive moving-average (CARMA) processes, which we refer to as measure-valued CARMA . These processes will extend finite-dimensional approaches to define ARMA-like time-series processes in a continuous-time context to an infinite-dimensional functional setting. More specifically, we generalize the concept of non-negative Lévy-driven CARMA processes to general separable Banach spaces by defining them as cone-invariant solutions to a particular class of continuous-time linear state-space models driven by Lévy subordinators. This framework encompasses existing CARMA models, such as real-valued CARMA processes [ 15 , 44 ] , their multivariate extensions [ 34 , 5 ] , and the Hilbert space formulation [ 10 ] . A particularly interesting case arises when the underlying Banach space is the space of finite signed measures, and the cone is the set of all positive measures. In this scenario, measure-valued CARMA processes can be used to capture the dynamics of (functionals of) spatio-temporal random fields. We provide a comprehensive analysis of this measure-valued setting, demonstrating how the short-memory and continuous-time attributes of CARMA processes adapt naturally to infinite-dimensional, cone-valued systems. 1.1. CARMA Processes and Linear-State Space Models On general state spaces, one can view CARMA processes as solutions to higher-order stochastic differential equations of the form \displaystyle\mathrm{D}^{p}X_{t}+\tilde{A}_{1}\mathrm{D}^{p-1}X_{t}+\ldots+\tilde{A}_{p}X_{t}=\tilde{C}_{0}\mathrm{D}^{q+1}L_{t}+\tilde{C}_{1}\mathrm{D}^{q}L_{t}+\ldots+\tilde{C}_{q}\mathrm{D}L_{t}, (1.1) where \mathrm{D}=\frac{\mathrm{d}}{\mathrm{d}t} , \{\tilde{A}_{i}\}_{i=1}^{p} and \{\tilde{C}_{j}\}_{j=0}^{q} are families of linear operators for p,q\in\mathbb{N} , and (L_{t})_{t\in\mathbb{R}} denotes a two-sided Lévy process. To interpret a higher-order stochastic differential equation of the form in ( 1.1 ), one draws inspiration from the analogous case in ordinary differential equations. There, a higher-order linear equation is reformulated as a higher dimensional first order system by introducing auxiliary state variables. A similar approach applies to ( 1.1 ), resulting in a linear state-space model with Lévy input process, whose transition operator is given by the companion block operator matrix of the characteristic polynomial of the differential equation. In this way, one can also define CARMA processes in general Banach spaces driven by Lévy processes, provided the resulting linear state-space model is well-posed in a stochastic strong sense, that is, one can define an Ornstein-Uhlenbeck (OU) process on a Cartesian product of the underlying Banach space. For reference, linear state-space models associated with CARMA processes in the multivariate setting are discussed in [ 16 ] (with cone-valued extensions in [ 5 ] ), and their adaptation to Hilbert spaces appears in [ 10 ] . In general separable Banach spaces, the feasibility of this approach critically depends on both the properties of the Banach space and the characteristics of the driving Lévy noise, since stochastic integration techniques are not universally available in all infinite-dimensional settings. In this paper, since our primary interest lies in positive measure-valued processes, we focus first on CARMA processes taking values in convex cones that are driven by Lévy subordinators. We show that, under suitable conditions on the model parameters, solutions to the CARMA linear state-space equations exist and remain within the cone. From a modeling perspective, this enables the construction of CARMA processes that evolve within cones of general separable Banach spaces. But more importantly, it provides a coherent solution concept for CARMA models in this general infinite-dimensional setting. In particular, we exploit the Pettis integral to define stochastic integrals with respect to the Poisson random measures associated with the driving Lévy process. However, in the case of non-separable Banach spaces (such as the space of finite Borel measures equipped with the total variation norm) one must either restrict to finite-dimensional Lévy subordinators or adopt a stochastically weak formulation of the CARMA stochastic differential equation. The latter approach is a common and natural alternative in the literature, which we briefly review and compare to our approach in the following section. 1.2. Measure-valued Processes We consider a different approach than considered in large parts of the measure-valued process literature [ 24 , 33 , 28 ] , where usually Markovian techniques are used to establish so-called superprocesses . Indeed, in [ 33 , Chapter 9] , the author showed the existence (in a stochastic weak sense) of a general class of processes taking values in M_{+}(E) , the cone of finite positive measures on some topological space E , called immigration superprocesses . We also refer to [ 21 ] , where measure valued affine and polynomial processes where investigated. In the analysis in [ 33 ] and [ 21 ] , the authors used the fact that M_{+}(E) endowed with the weak topology is separable, and locally compact when E is a locally compact Polish space [ 45 ] . This allows the use of the positive maximum principle to show the existence of the associated martingale problem. In our case, we consider the cone M_{+}(E)^{p} as a state space. If E is compact, then the cone M_{+}(E)^{p} equipped with the topology induced by the direct sum inner product \langle\cdot,\cdot\rangle_{p} is a locally compact Polish space as the finite product of locally compact Polish spaces is itself locally compact and Polish. Hence following similar derivations as for example in [ 21 ] , one could approach to prove existence of OU-processes by using the positive maximum principle. However, since our model is an OU process driven by a measure-valued Lévy subordinator, we instead use the theory on integration with respect to Banach-valued Lévy processes to prove the existence of an analytically weak and stochastically strong solution of the linear state-space equations directly. 1.3. Applications to Modeling Dynamics of Renewable Energy Markets CARMA processes are widely recognized for their tractability, interpretability and flexible autocorrelation structure, inherited by their discrete-time ARMA versions. These features have led to their application across diverse fields, including meteorology, engineering, and finance. In particular in the renewable energy domain, CARMA processes have been used to model (deseasonalized) weather and time-dependent climate variables, including wind speed [ 13 , 12 ] and temperature [ 12 , 11 ] and solar irradiance [ 22 , 31 ] . Moreover, in financial applications, CARMA models served as mean-reverting processes for volatility [ 43 , 14 , 5 ] and

Read the complete text