A0269
Title: Structural modelling of dynamic networks and identifying maximum likelihood
Authors: Christian Gourieroux - University of Toronto and CREST (Canada) [presenting]
Joann Jasiak - York University (Canada)
Abstract: Nonlinear dynamic models where the main parameter of interest is a nonnegative matrixcharacterizing the network (contagion) effects are examined. This network matrix is usually assumed to either have a limited number of nonzero elements (sparsity) or admit a reduced-rank Nonnegative Matrix Factorization (NMF). We follow the latter approach and develop new probabilistic NMF inference methods. We introduce a new Identifying Maximum Likelihood (IML) method for consistently estimating the identified set of admissible NMFs and derive the asymptotic distribution of this random set. We also propose a maximum likelihood estimator of the parameter matrix for a given non-negative rank and derive its asymptotic distribution and the associated efficiency bound.
