Old news‎ > ‎

StaTalk on Computational Tools for Statistical Modelling

Department MEMOTEF, Sapienza Università di Roma
via del Castro Laurenziano 9, 00161, Roma (Italy)

StaTalk on Computational Tools for Statistical Modelling

Registration is now open (it is mandatory, even if free of charge): please, fill in this form

(for information, please contact the chair of the organizing committee at clara.grazian at uniroma1 dot it)


09.30-10.00 Opening

10.00-10.40 Christian Macaro (SAS System) "A mixture of heterogeneous models with time dependent weights"

Nowadays, the huge availability of data poses new and interesting questions to analysts and researchers in the scientific community. The use of established methods targeting scenarios with limited amount of information do no take advantage of the opportunities provided by the heterogeneous sources of information. We propose to analyse time series data as if generated by a mixture of heterogeneous models where the mixing weights can change over time. This approach is extremely flexible and appears to explain anomalies which are typically modelled with fat tails, outliers and structural breaks. An application shows how to model volatility in financial markets taking into consideration financial, macroeconomic and microeconomic shocks.

10.40-11.10 Coffee break

11.10-11.50 Clara Grazian (MEMOTEF Department) "Modelling financial dependence through approximate Bayesian inference"

Modelling the dependence among variables is a crucial but difficult task, especially in a multidimensional context. Often standard models used in the analysis greatly simplify the reality, by introducing assumptions that do not occur or difficult to be tested. For example, many analyses on returns of financial institutions use Gaussian assumptions to make the models tractable; however, it is now well-known that the returns tend to have a stronger correlation in periods of crisis than in good times. We will present models and computational statistical methods to treat such situations. These methods are based on a Bayesian analysis and on Monte Carlo approximations and will be applied on the returns of some Italian financial institutions.

11.50-12.30 Stefano Vaccari (MEMOTEF Department) "Pricing in the Presence of Social Learning: Modelling Information Aggregation in a Competitive Market"

We study a model where two or more competing firms launch different products at the same time on a market of consumers with heterogeneous quality preferences. The true quality of these products is initially unknown to the market. Consumers' purchases decisions depend on quality estimates they evaluate on the base of online reviews reported by prior purchasers. As information on the market varies with time, so do their quality estimates, leading to the so-called "Social Learning" process. The aim of this work is firstly to provide a theoretical framework in order to establish the general conditions that allow asymptotic learning to take place and secondly, from the sellers' point of view, to study whether pricing strategies that take account of the information spread are more effective relative to policies that do not.

12.30-14.30 Lunch break

14.30-15.10 Luca Tardella (Department of Statistical Science) "Alternative approaches for approximating marginal likelihood."

The computation of the marginal likelihoods is a crucial point for Bayesian model selection where the ratio of marginal
likelihoods of two competing models defines the well-known Bayes Factor. The problem has received a lot of attention in the literature and many alternative approaches have been proposed. Despite that it continues to be a formidable challenge in many complex models. We will review some of the simulation-based techniques which take advantage of Monte Carlo (MC) and Markov Chain Monte Carlo (MCMC) methods. We will focus particularly on one alternative marginal likelihood estimator, named Inflated Density Ratio estimator (IDR), originally proposed in Petris and Tardella (2007). The IDR estimator is an alternative implementation of the Generalized Harmonic Mean (GHM) (Newton and Raftery, 1994; Raftery et al., 2007). In particular, the IDR method, similarly to the original Harmonic Mean Estimator, recycles posterior simulation but relies on a different choice of the importance function, defined through a parametrically inflated version of the target density. We use some benchmark simulation plans and some complex models for real data in the field of system biology to explore its comparative merits and limits.

15.10-15.50 Andrea Tancredi (MEMOTEF Department) "A Bayesian approach for de-duplication, record linkage and inference with linked data."

We introduce a new Bayesian methodology for performing record linkage and regression analysis using the resulting matched units in a k lists framework with possible duplications within the same file. The linkage and duplication process is viewed as a way of clustering the records into distinct entities and random partition theory is used for calibrating the prior distribution on the cluster structure. We frame the record linkage process into a formal statistical model which comprises both the matching variables and other variables included at the inferential stage. This way, the researcher is able to account for the matching process uncertainty in inferential procedures based on probabilistically linked data, and at the same time, he/she is also able to generate a feedback propagation of uncertainty between the working statistical model and the record linkage stage. We argue that the feedback effect is essential to eliminate potential biases that could characterize the resulting post-linkage inference. We also show that this feedback effect is able to improve performance of the record linkage procedure. Practical implementation of the proposed method is based on standard Bayesian computational techniques. Although the methodology is quite general, we have restricted our analysis to the popular and important case of linear regression set-up for expository convenience.

15.50-16.30 Davide Di Cecco (ISTAT) "Towards the use of administrative sources in official statistics: Some examples of methodologies for administrative multisource integration utilized at ISTAT."

The use of administrative data (AD) in National Statistical Institutes has been dramatically increasing in recent years, and has passed from an indirect use of AD as auxiliary information to a direct use in the production of official statistics. This change of perspective required to develop a new methodological framework for every aspect of data treatment. One of the main problem is that AD scoping and definitions differ from those of the statistician, so all observed variables are potentially affected by errors, but at the same time, several sources register the same phenomenon. Consequently, as we do not use survey data, unsupervised models based on Latent Variables have gained a certain attention, where the latent variable is to be interpreted as the “real”; value of the quantity of interest, and the observed variables are possibly erroneous measurements of that quantity. In this talk we present a pair of applications of this kind of models developed in ISTAT: one can be viewed as a Record Linkage problem with constraints. The other one make use of Latent Class models in a capture-recapture framework.

16.30-17.00 Closing


Cognome Nome Affiliazione
Amabili Lorenzo SIS
Andreini Paolo Università di Roma “Tor Vergata”
Barbieri Marilena Dipartimento di Economia, Università di Roma Tre
Benassi Federico Istituto Nazionale Italiano di Statistica
Bignozzi Valeria MEMOTEF, Sapienza Università di Roma
Bondarenko Maksim Juriovic Sapienza Università di Roma
Calì Camilla Università degli Studi di Napoli "Federico II"
Camerlenghi Federico Università Bocconi & Collegio Carlo Alberto
Cappuccio Fabio Università degli studi di Cassino e del Lazio Meridionale
Catania Leopoldo Università di Roma “Tor Vergata”
Ceccantoni Giulia MEMOTEF, Sapienza Università di Roma
Cerquetti Annalisa MEMOTEF, Sapienza Università di Roma
Cesarone Francesco Università di Roma Tre
Colusso Teresa Sapienza Università di Roma
D'Amore Gabriele Sapienza Università di Roma
D'Angelo Silvia Sapienza Università di Roma
D'Assante Luigi Università degli Studi di Napoli "Federico II"
De Angelis Simone MEMOTEF, Sapienza Università di Roma
Di Cecco Davide ISTAT
Di Gennaro Danilee MEMOTEF, Sapienza Università di Roma
Domenicano Ilaria DSS, Sapienza Università di Roma
Dotto Francesco DSS, Sapienza Università di Roma
Fop Michael University College Dublin
Foschi Flavio ISTAT
Fusco Davide Università degli Studi di Napoli "Federico II"
Galante Paolo Lelio Sapienza Università di Roma
Ghaly Simone Sapienza Università di Roma
Gentile Maria MIUR
Grazian Clara MEMOTEF, Sapienza Università di Roma
Iannoni Fabrizio Sapienza Università di Roma
Iuzi Orietta ISTAT
Jona Lasinio Giovanna DSS, Sapienza Università di Roma
Lipizzi Fabio ISTAT
Liseo Brunero MEMOTEF, Sapienza Università di Roma
Lytvyn Rostyslav Sapienza Università di Roma
Macaro Christian The SAS System
Maksimova Kseniia Sapienza Università di Roma
Mandolini Giorgio Maria Sapienza Università di Roma
Martini Alessandro ISTAT
Mastrantonio Gianluca Politecnico di Torino
Moretti Jacopo MEMOTEF, Sapienza Università di Roma
Naccarato Alessia Dipartimento di Economia, Università di Roma Tre
Paci Lucia Università degli Studi di Bologna
Padellini Tullia DSS, Sapienza Università di Roma
Parisi Antonio Università di Roma “Tor Vergata”
Pasquini Alessandra Sapienza Università di Roma
Patella Valeria Sapienza Università di Roma
Petrella Lea MEMOTEF, Sapienza Università di Roma
Polettini Silvia MEMOTEF, Sapienza Università di Roma
Rocchetti Irene ISTAT
Rocco Giorgia DSS, Sapienza Università di Roma
Sacco Armando MEMOTEF, Sapienza Università di Roma
Santi Flavia Sapienza Università di Roma
Scarsini Marco LUISS Guido Carli
Serpieri Carolina Sapienza Università di Roma
Serri Matteo MEMOTEF, Sapienza Università di Roma
Sorrentino Giulia Sapienza Università di Roma
Stefanucci Marco Sapienza Università di Roma
Tancredi Andrea MEMOTEF, Sapienza Università di Roma
Tangari Gioacchino University College London
Tardella Luca DSS, Sapienza Università di Roma
Totaro Simone Studente
Tuoto Tiziana ISTAT
Vaccari Stefano MEMOTEF, Sapienza Università di Roma
Varriale Roberta ISTAT
Volpini Andrea Sapienza Università di Roma

Young SIS,
Sep 13, 2016, 4:08 AM
Young SIS,
Sep 13, 2016, 4:09 AM
Young SIS,
Sep 13, 2016, 4:06 AM
Young SIS,
Sep 13, 2016, 4:04 AM
Young SIS,
Sep 13, 2016, 4:07 AM
Young SIS,
Sep 13, 2016, 4:09 AM