par De Mol, Christine 
Référence Recent Advances in Econometrics and Statistics: Festschrift in Honour of Marc Hallin, Springer Nature, page (493-510)
Publication Publié, 2024-01
Référence Recent Advances in Econometrics and Statistics: Festschrift in Honour of Marc Hallin, Springer Nature, page (493-510)
Publication Publié, 2024-01
Partie d'ouvrage collectif
| Résumé : | In a previous paper by Conflitti et al. (Int J Forecasting 31:1096–1103, 2015), we devised a simple multiplicative iterative algorithm to compute the weights of an optimal combination of density forecasts. However, the question of convergence of this algorithm was not addressed. In the present work, we provide a detailed convergence proof, both for the cost function and for the sequence of iterates. We then show that a similar algorithm can be used for a classic problem in statistics, namely, the estimation of a mixture of known probability densities. The algorithm can be generalized to treat the case where the data are affected by a known noise, i.e., a density deconvolution problem. Finally, we address the situation of an unknown noise, assuming that it is itself a mixture of known densities, and we propose an adapted alternating blind deconvolution algorithm, the convergence properties of which are also analyzed. |
