On non-negative modeling with CARMA processes (Q2633848)

In this paper, the authors analyze two types of Continuous-ARMA (CARMA) models, both of which have long been widely used in various applications, particularly in financial mathematics. These two models are Cox-Ingersoll-Ross (CIR) and Ornstein-Uhlenbeck (OU) processes. Let's remember how they are defined. Be \((\Omega,F,P)\) a complete probability space and be \((F_{t})_{t\geq 0}\) a complete and right-continuous filtration on \((\Omega,F,P)\), that is: \(F_{t}\) is a complete \(\sigma\)-algebra, \(F_0\subset F_{s}\subset F_{t}\subset F\) for \(0\leq s\leq t\), and \(F_{s}=\cap_{s\leq t}F_{t}\) for \(0\leq s\). A Lévy process, \(L=(L_{t})_{t\geq 0}\), is a stochastic process that satisfies the following properties: \(L_{t}\) is \(F_{t}\)-measurable for all \(t\geq 0\) (\(L\) is \((F_{t})_{t\geq 0}\)-adapted); \(L_0=0\) a.e.; \(L\) has independent increments from the past, i.e. \(L_{s}-L_{t}\) is independent from \(F_{u}\) if \(0\leq u\leq t<s\); \(L\) has stationary increments, i.e. \(L_{s}-L_{t}\) and \(L_{s+u}-L_{t+u}\) if \(0\leq t<s\) and \(u>0\) have the same distribution; \(L\) is continuous in probability, i.e. \(P(|L_{s}-L_{t}|>\varepsilon)\to_{s\to t}0\) if \(\varepsilon>0\) and \(0\leq s,t\). An Ornstein-Uhlenbeck (OU) one-dimensional process, \(X=(X_{t})_{t\geq 0}\), is defined by the following stochastic differential equation: \[ dX_{t}=-\theta X_{t}\cdot dt+\sigma \cdot dW_{t}, \] where \(\theta>0\) and \(\sigma>0\) are parameters and \(W=(W_{t})_{t\geq 0}\) denotes the Wiener process. The authors of this paper define the OU process a little more generally, changing the Wiener process to a Lévy process and working with \(p\)-dimensional processes. \(Y=(Y_{t})_{t\geq 0}\), is a CARMA one-dimensional process if \[ Y_{t}=b\cdot X_{t} \] with \(X=(X_{t})_{t\geq 0}\) being an OU process as defined before. In this case, \(X\) is said to be the state-space process associated to \(Y\). A Cox-Ingersoll-Ross (CIR) one-dimensional non-negative process, \(R=(R_{t})_{t\geq 0}\), is defined by the following stochastic differential equation: \[ dR_{t}=a(b-R_{t})\cdot dt+\sigma\sqrt{(R_{t})}\cdot dB_{t}, \] where \(a,b,\sigma>0\) are constants and \(B=(B_{t})_{t\geq 0}\) is a standard Brownian motion. In this article the authors first study the distribution function, the ACF and other properties of processes defined as sums of a finite number of independent CARMA processes. Later they analyze the connection between CARMA processes and Wishart process. Finally they consider a variant of the CARMA process by changing the Lévy process to a compound Poisson process. A compound Poisson process, parameterised by a rate \(\lambda>0\) and jump size distribution \(G\), is a process \(P=(P_{t})_{t\geq 0}\) given by \[ P_{t}=\sum_{i=1}^{N_{t}}D_{i}, \] where, \(N=(N_{t})_{t\geq 0}\) is a Poisson process with rate \(\lambda\), and \(D_i/i\geq 1\) are independent and identically distributed random variables, with distribution function \(G\), which are also independent of \(N\). In the paper \(G\) is the exponential distribution. The studied models are generally applied in financial mathematics although the authors present an example of the application of a CIR process to model wind speed in a paper by other authors on weather forecasting. This article is hard to read for anyone who is not a mathematician with extensive knowledge of stochastic processes. All the results are rigorously enunciated and proven. Some examples show the application of the processes studied.