Extreme values in FGM random sequences (Q1283921)

The Farlie-Gumbel-Morgenstern (FGM) class of distributions was introduced to model a type of dependence in multivariate distributions. Namely, a FGM random sequence \( \{ X_i\), \(i \geq 1 \} \) is defined by the univariate marginals \( F_i \sim X_i \) and a symmetric function \( a(\cdot) \) such that the joint distribution function of \( X_{i_1} ,\dots, X_{i_n} \) is \[ H_{i_1 ,\dots, i_n} (x_1,\dots,x_n) = \prod_{h=1}^n F_{i_h} (x_i) \Biggl\{ 1+ \sum_{1\leq j < k \leq n} a(i_j,i_k)(1-F_{i_j}(x_j))(1-F_{i_k}(x_k))\Biggr\}. \] The authors show that this dependence structure is rather restrictive, so that the partial maxima \( M_n = \max (X_1,\dots,X_n) \) of a FGM sequence, properly normalized, behave asymptotically like the maxima of independent rv's. This result is proved also for vectors in \( R^d \).