Spreading-invariant sequences and processes on bounded index sets (Q1591370)

A finite or infinite sequence \(\xi_1,\xi_2,\dots\) is spreadable (or spreading-invariant) if \((\xi(k_1),\dots, \xi(k_m))=^{\text{d}}(\xi_1,\dots, \xi_m)\) for every finite sequence \(k_1< k_2<\cdots< k_m\). Here \(\xi(i)= \xi\), and \(=^{\text{d}}\) denotes equality in distribution. A process \(X_t= X(t)\), \(0\leq t\leq 1\) or \(t\geq 0\), is called spreadable when it has spreadable increments, i.e. when all sequences \(X(ja)- X(ja-a)\), \(j= 1,\dots, n\), are spreadable. Spreadability generalizes exchangeability, where the above distinct \(k_1,\dots, k_m\) do not need to increase. For infinite sequences and processes on \([0,\infty)\) the two concepts coincide by Ryll-Nardzewski's and Bühlmann's theorems. For finite sequences this is not true (example). The paper extends theorems on exchangeability from the author's earlier papers: representation of the distribution as a mixture of extreme distributions (not always unique); conditions for equivalence of exchangeability and spreadability; (semi)martingale properties; convergence in distribution; tightness; preservation of spreadability; norm inequalities. The predictable sampling theorem: Let \(\xi_1,\dots,\xi_n\) be spreadable and \(\tau_1< \tau_2<\cdots< \tau_k\) predictable, i.e. \(\{\tau_j= k\}\in \sigma(\xi_1,\dots, \xi_{k-1})\). Then \((\xi(\tau_1),\dots, \xi(\tau_k))=^{\text{d}} (\xi_1,\dots, \xi_k)\). For spreadable \(X_t\) the theorem is stated in terms of stochastic integrals. Wald-type identities. They state that moments of \(\sum\xi_k \eta_k\), with \(\xi_1,\dots,\xi_k\) spreadable and \(\eta_1,\dots, \eta_k\) predictable, under conditions may be computed as if \((\xi_1,\dots, \xi_k)\) and \((\eta_1,\dots, \eta_k)\) are independent. Generalized to \(d\) dimensions and to continuous time in terms of stochastic integrals w.r.t. \(X_t\).