An inverse problem for infinitely divisible moving average random fields (Q2316341)

One denotes by \(\Lambda\) a stationary infinitely divisible independently scattered random measure with Lévy characteristics \((a_0,b_0,v_0)\), where \(a_0\geq 0\), \(b_0\in R\) and \(v_0\) is the Lévy density. Denote with \(X=\{X(t);t\in R^n\}\) the moving average of an infinitely divisible random field on \(R^n\), \(X(t)=\int _{R^d}f(x-t)\Lambda(dx)\), \(t\in R^n\), with Lévy characteristics \((a_1,b_1,v_1)\) and \(f\) a simple function expressed as \(f=\sum_{k=1}^nf_k\mathbf{1}_{\Delta_k}\) for certain fixed bounded pairwise disjoint Borel sets \(\Delta_1,\dots,\Delta_n\) and with \(n\) real non zero coefficients \(f_1,\dots,f_n\). The aim of the paper is the study of nonparametric estimation of the Lévy characteristics \((a_0,b_0,v_0)\). Section 2 of the article is devoted to preliminaries and in Section 3 the inverse problem to be treated in the paper is described and this is: given \(N\) observations \(X(t_1),\dots,X(t_N)\), \(t_i\in R^d\), \(i=1,\dots,N\) of the random field \(X\), estimate the Lévy triplet \((a_0,b_0,v_0)\) of the \(\mathbf{ID}\) random measure \(A\). The formulas for \(a_1\), \(b_1\) and \(v_1(x)\) provide estimations for \(a_0\) and \(b_0\) and \(v_0\). Since estimations for \(a_0\) and \(b_0\) are easily to be obtained, the study concentrates on the equation in \(v_1(x)\) and \(v_0(.)\). Existence and uniqueness results are discussed in the fourth section, and estimates for the characteristic Lévy triplet of \(X\) for pure jump infinitely divisible fields are given in Section 5. Section 6 is devoted to the development of estimation methods for the Lévy density \(v_0\). Three different estimation approaches are presented. The first one is the plug-in approach, the second one is the Fourier approach and the third one is the orthonormal basis approach. In the last section of the article, the numerical performance of the estimators is discussed.