RISK 2013 INICIO

Extendida Fecha Límite de Envío: 25 de Junio de 2013 !!

RIESGO 2013

5ª Reunión de Investigación en Seguros y Gestión del Riesgo

Gran Canaria, 17 y 18 de Octubre de 2013

El Departamento de Métodos Cuantitativos en Economía y Gestión de la Universidad de Las Palmas de Gran Canaria, en colaboración con el Grupo de Investigación del Riesgo en Finanzas y Seguros de la Universidad de Barcelona, tienen el gusto de invitarle a la 5ª Reunión de Investigación en Seguros y Gestión del Riesgo (RIESGO 2013). El evento se celebrará en Gran Canaria, los días 17 y 18 de Octubre de 2013. Como anteriores ediciones, celebradas en Barcelona (2005), Cantabria (2007), Madrid (2009) y Sevilla (2011), RIESGO 2013 constituye un encuentro de referencia para los investigadores y profesionales de las Ciencias Actuariales y Financieras. 


COMITÉ ORGANIZADOR (risk2013@ulpgc.es)

Emilio Gómez Déniz, U. de Las Palmas de Gran Canaria

Montserrat Guillén Estany, Universidad de Barcelona

Francisco José Vázquez Polo, U. de Las Palmas de G.C.


KEYNOTE SPEAKER

 Professor José Garrido

Dpt. of Mathematics and Statistics, Concordia University (Canada)

Title: Uses and abuses of generalized linear models for Insurance losses

Abstract: Predictive modeling has changed the way actuaries analyze insurance losses. The use of generalized linear models (GLM’s) for risk classification and ratemaking is now standard and has been fueled, to some extent, by the readily available statistical packages (SAS, Emblem) that can fit GLM’s to insurance losses. GLM’s are now being used also in retention models, to predict the probability that a policy holder will renew the insurance policy, as a function of the renewal premium increase (decrease).

Leaving the familiar independent-and-identically distributed (iid) world, in which actuaries have been trained, to move onto the uncharted territories of predictive models has been a bumpy ride, patched by abusive uses of statistical models. For instance, in the iid world it was common for actuaries to model loss frequency and severity separately, combining them at the end in an aggregate loss model. This is still common now to use separate GLM’s for frequency and severity, and simply multiply the estimated GLM expectations to estimate aggregate losses, as if the frequency and severity processes were independent, even if both are estimated on the same policyholder data, sometimes even using exactly the same rating variables.

In this talk we illustrate some of these abuses and discuss multivariate GLM’s that can used for insurance data that exhibits some forms of dependence.