| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 4.9 MB | Adobe PDF |
Orientador(es)
Resumo(s)
As propensões atuais da mortalidade têm trazido novos desafios em diversas áreas. Para compensar
isso, têm surgido vários estudos relacionados com a previsão da mortalidade, a fim de projetar as
tendências futuras da mortalidade e assim reduzir o nível de incerteza associado. Porém, a esperança
média de vida tem sido constantemente subestimada. Propusemos a avaliar o desempenho de vários
modelos na modelação e projeção da mortalidade portuguesa, desagregada por género, entre 1980
a 2010. Concentramos na estrutura de modelos generalizados de idade - período - corte, com o
intuito de incorporar o efeito de coorte e averiguar a existência de gerações que compartilham as
mesmas características. E adicionalmente, estudamos o impacto do risco de longevidade numa
carteira de anuidades vida, através da mensuração do Value at risk e Expected Shortfall. Entre os
modelos discutidos o de Renshaw & Haberman apresentou o melhor ajuste e desempenho preditivo
para a construção de índices de mortalidade para os homens e para as mulheres, com base nos
Critérios de Informação e erros de previsão, respetivamente. Através da análise dos resíduos,
detetamos para os modelos sem o parâmetro de coorte, uma estrutura sistemática na população
masculina. Indicando assim, a inércia destes modelos em capturar o efeito de coorte, que por sua
vez, leva à possibilidade da existência de um efeito de coorte, conforme já revelado em estudo
anteriores. Não obstante, analisamos a incerteza nos parâmetros, mediante a construção de
intervalos de confiança, através do método de Bootstrap sobre os resíduos para os modelos de Lee &
Carte e Renshaw & Haberman, onde averiguarmos que o índice de coorte é o principal responsável
pela incerteza na projecção da mortalidade. No geral, esses resultados sugerem que apesar da
relevância do efeito de coorte na modelação da mortalidade, o impacto do mesmo se dissipe ao
longo do tempo.
The current mortality propensity has led to new challenges in many areas. To compensate this, several mortality prediction studies have been conducted to forecast future mortality trends and thereby reduce the associated uncertainty. However, the average life expectancy was being constantly underestimated. We proposed to examine the performance of different models in modelling and forecasting Portuguese mortality from 1980 to 2010, split by gender. We focused on the structure of generalized age-period-cohort models, such we could consider the cohort effect and check the existence of cohorts with the same characteristics. In addition, we examine the impact of longevity risk on a pension fund portfolio by measuring value at risk and expected shortfall. Among the models discussed, Renshaw & Haberman's model provided the best fitting and predictive performance for the men and women, based on Information Criteria and prediction errors. By analysing the residuals, we found a systematic structure in the male population for the models without cohort parameters. This indicates the inertia of these models in capturing the cohort effect, which in turn leads to the possibility of a cohort effect, as shown in previous studies. Nevertheless, we have analysed the uncertainty in the parameters by using the bootstrap method for residues for the Lee & Carte and Renshaw & Haberman models to establish confidence intervals where the cohort parameter is the main cause of uncertainty in the mortality projection. Overall, these results suggest that despite the relevance of the cohort effect in modelling mortality, the influence of the cohort diminishes over time.
The current mortality propensity has led to new challenges in many areas. To compensate this, several mortality prediction studies have been conducted to forecast future mortality trends and thereby reduce the associated uncertainty. However, the average life expectancy was being constantly underestimated. We proposed to examine the performance of different models in modelling and forecasting Portuguese mortality from 1980 to 2010, split by gender. We focused on the structure of generalized age-period-cohort models, such we could consider the cohort effect and check the existence of cohorts with the same characteristics. In addition, we examine the impact of longevity risk on a pension fund portfolio by measuring value at risk and expected shortfall. Among the models discussed, Renshaw & Haberman's model provided the best fitting and predictive performance for the men and women, based on Information Criteria and prediction errors. By analysing the residuals, we found a systematic structure in the male population for the models without cohort parameters. This indicates the inertia of these models in capturing the cohort effect, which in turn leads to the possibility of a cohort effect, as shown in previous studies. Nevertheless, we have analysed the uncertainty in the parameters by using the bootstrap method for residues for the Lee & Carte and Renshaw & Haberman models to establish confidence intervals where the cohort parameter is the main cause of uncertainty in the mortality projection. Overall, these results suggest that despite the relevance of the cohort effect in modelling mortality, the influence of the cohort diminishes over time.
Descrição
Dissertation presented as the partial requirement for obtaining a Master's degree in Statistics and Information Management, specialization in Information Analysis and Management
Palavras-chave
Mortalidade em idades avançadas Modelos GAPC Bootstrap Risco de Longevidade StMoMo Old Age Mortality GAPC Models Bootstrap Longevity Risk
