Filippo Ascolani, a third-year doctoral student in statistics, is among the winners of the 2022 Hannan Graduate Student Travel Award, yearly assigned to outstanding Ph.D. students by the Institute of Mathematical Statistics. Very few winners of this prestigious award are affiliated to non-U.S. universities, among whom Filippo.
The award will support Filippo’s participation in a conference to be held in Puerto Varas, Chile, where he will present the paper “Nonparametric Priors with Full Range Borrowing of Information”, written together with Beatrice Franzolini (a Bocconi PhD alumna and currently post-doc in Singapore), Antonio Lijoi and Igor Prünster (Bocconi Department of Decision Sciences)
A crucial problem in statistics is how to combine data arising from different but related groups. In this case the model should be chosen so that it implies homogeneity within each group, but describes the possible heterogeneity between them. The mathematical concept that translates these requirements into the Bayesian context is called partial exchangeability, which means in simple terms that observations cannot be exchanged across different groups without altering the inferential results.
A great number of partial exchangeability models has been proposed in recent years, but the theoretical implications of this assumption have not been thoroughly studied: moreover, despite the great progress in the last two decades, most of the available proposals allow only non-negative correlations between groups, which may be overly restrictive in many contexts.
Filippo’s work introduces a new concept, that of “hypertie”, which represents a simple and direct measure of statistical dependence between groups. He also proposes a new class of models in which correlation can be negative and tuned on the basis of available information.
by Andrea Costa
Source: Bocconi Knowledge