Services on Demand
Journal
Article
Indicators
- Cited by SciELO
- Access statistics
Related links
- Similars in SciELO
Share
Revista de Matemática Teoría y Aplicaciones
Print version ISSN 1409-2433
Abstract
SANDI-CORRALES, Ana Rosa. Predictive Modelling of losses in non-life insurance. Rev. Mat [online]. 2021, vol.28, n.1, pp.105-124. ISSN 1409-2433. http://dx.doi.org/10.15517/rmta.v28i1.12345.
Accident and health insurance with differentiated premiums for homogeneous risk groups was analyzed. The estimation of these premiums on previous opportunities was in univariate form, which has the limitation that when there are risk groups with few observations, the results are very volatile and omit the information that could provide predictive variables. Therefore, it was decided to estimate the expected claims (which are an input in the premium calculation) with three multivariate models: ordinary linear, additive and mixed linear. Several were used in order to compare their forecasting capability. Performance was acceptable within both the fit and test samples in the case of ordinary linear and additive models with a difference of about 1% from the real data. Linear mixed could not make predictions for combinations of predictors not observed in the fit data.
Keywords : insurance; pricing; predictive modelling; linear models; additive models; mixed models; software R..