Penalized quantile regression for risk assessment

Proponente Lea Petrella - Professore Ordinario
Sottosettore ERC del proponente del progetto
Componenti gruppo di ricerca
Componente Categoria
Paolo De Angelis Componenti il gruppo di ricerca
Maria Giuseppina Bruno Componenti il gruppo di ricerca
Componente Qualifica Struttura Categoria
VALENTINA RAPONI scholarship Imperial College Business School, South Kensington Campus, Kensington, London SW7 2AZ Altro personale Sapienza o esterni
LA PORTA ALESSANDRO laureando Sapienza University of Rome Altro personale Sapienza o esterni
GHISLAINE GAYRAUD Professor Université de Technologie Compiegne Altro personale Sapienza o esterni

In this project we consider the problem of building a quantile regression model to analyze risks in medical malpractice contest when the right predictors explaining the phenomenon have to be selected. As well known, quantile regression is an important method for implementing the impact of regressors on the whole conditional distribution of a response variable in particular on its tails. For this reason quantile regression is particular attractive when risk assessment is the main concern i.e. when the events are extreme. How to select the best predictors and estimate the right parameters in model building is critically important and here we will consider the penalized regression approach propose by Tibishirani in 1996 focusing the attention on the lasso and the elastic net framework. We will use those tools in the quantile regression contest building specific algorithms to implement the frequentist and the Bayesian statistical inference. Once the methodological part is implemented, we will use those tools in the medical malpractice risk assessment which is particular important for the risk managers of health care institutions involved in measuring frequencies, claims intensities of damages and amount of costs for them. Since several variables are involved in explain those phenomena it seems quite interesting to choose the right ones by using a penalizing quantile regression approach. This would be, to the best of our knowledge, the first quantile regression approach to face the medical malpractice issue.


© Università degli Studi di Roma "La Sapienza" - Piazzale Aldo Moro 5, 00185 Roma