Publicação
A Deep Learning integrated mortality model for Longevity Swap pricing
| Resumo: | This research empirically investigates the usage of Recurrent Neural Networks (RNN) to improve the accuracy of mortality rates forecasting within the context of Longevity linked securities pricing. The benchmark model in the mortality field is the classical Lee-Carter; the forecasting procedure of these model is often conducted with ARIMA models. I consider a fixed forecasting time horizon in order to compare the performance of Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) with different hyperparameter and data input choices against that produced by the best fitted ARIMA models. The results are then applied to Longevity Swap pricing in order to better estimates the premium of the derivatives contracts. The investigation is conducted for six countries, using mortality data from 1950 onwards, differentiating by gender. The research shows how RNN outperform the classical ARIMA models in the forecasting procedure. Although the advantages of RNN’s techniques are strictly bounded to the set of hyperparameter used for the comparison; the outcomes of such approaches can vary greatly using different input choices. In the end the results shows that an RNN approach can bring significant changes to the price of Longevity Linked securities. The research is in the first place one of the few to test the forecasting accuracy of Deep Learning methods accounting for alternative methodological, hyperparameter and data input choices. Afterwards the investigation demonstrate the necessity of revisit the classical mortality models in order to better estimates prices of derivatives contracts that are very useful in the context of Longevity risk. |
|---|---|
| Autores principais: | Salvo, Alberto Di |
| Assunto: | Mortality Deep Learning Long short-term memory Gated Recurrent Unit Lee-Carter model Longevity risk Longevity swap Longevity swap pricing |
| Ano: | 2022 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade Nova de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório Institucional da UNL |
| Resumo: | This research empirically investigates the usage of Recurrent Neural Networks (RNN) to improve the accuracy of mortality rates forecasting within the context of Longevity linked securities pricing. The benchmark model in the mortality field is the classical Lee-Carter; the forecasting procedure of these model is often conducted with ARIMA models. I consider a fixed forecasting time horizon in order to compare the performance of Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) with different hyperparameter and data input choices against that produced by the best fitted ARIMA models. The results are then applied to Longevity Swap pricing in order to better estimates the premium of the derivatives contracts. The investigation is conducted for six countries, using mortality data from 1950 onwards, differentiating by gender. The research shows how RNN outperform the classical ARIMA models in the forecasting procedure. Although the advantages of RNN’s techniques are strictly bounded to the set of hyperparameter used for the comparison; the outcomes of such approaches can vary greatly using different input choices. In the end the results shows that an RNN approach can bring significant changes to the price of Longevity Linked securities. The research is in the first place one of the few to test the forecasting accuracy of Deep Learning methods accounting for alternative methodological, hyperparameter and data input choices. Afterwards the investigation demonstrate the necessity of revisit the classical mortality models in order to better estimates prices of derivatives contracts that are very useful in the context of Longevity risk. |
|---|