Publication: Estudio de la longevidad aplicando Redes Neuronales Artificiales
Loading...
Identifiers
Publication date
2021-07-12
Defense date
2021-07-12
Authors
Advisors
Journal Title
Journal ISSN
Volume Title
Publisher
Impact
Export
Abstract
Uno de los riesgos asociados a la vida humana que más interés despierta en el campoactuarial es el estudio del riesgo de longevidad. Este riesgo se define como la probabilidadde que las personas puedan sobrevivir más allá de lo esperado, generando una crecientepreocupación en el mercado asegurador del negocio de seguros de vida debido a laposibilidad de la subestimación de las reservas, lo cual implica un riesgo de déficitde recursos financieros para cumplir las obligaciones de pago futuras. Una forma demitigación de este tipo de riesgos es la proyección de la mortalidad de la población anivel país, permitiendo al país o población asegurada estructurar sus planes de pensiones,o sirviendo de asistencia a las entidades aseguradoras en procesos de pricing o reserving.A lo largo del tiempo se han desarrollado diferentes técnicas y modelos orientados a lapredicción de la mortalidad. Entre ellos se encuentran modelos paramétricos, como losconocidos modelos clásicos o modelos no paramétricos como el modelo de P-splines.El desarrollo de técnicas más avanzadas, como las basadas en la Inteligencia Artificial,han permitido un estudio de la longevidad desde un nuevo paradigma, el cual podría darlugar al nacimiento de modelos que arrojen una predicción más precisa que la aportadapor los métodos hasta ahora utilizados. Una de estas técnicas es la desarrollada en elpresente trabajo, basada en el estudio de la longevidad utilizando “Redes NeuronalesArtificiales” , a partir de ahora RNA. Las RNA cuentan con una base matemáticacompleja, así como un conjunto de parametrizaciones como el número de neuronas encada capa, tipo de aprendizaje, funciones de activación etc, que hacen que la estimaciónde dichos paramétros se basen en procesos de prueba y error.La estructura del trabajo es la siguiente. En el punto 2 desarrollaremos teóricamente losmodelos clásicos hasta ahora más utilizados para el estudio de la longevidad. El punto3 está destinado a obtener un conocimiento más profundo sobre las Redes Neuronales,desde sus orígenes hasta su estructura. En el punto 4 nos centraremos en la metodologíautilizada para el desarrollo del trabajo. El punto 5 se centra en el análisis de la obtenciónde los datos necesarios para el estudio.En el punto 6 analizaremos los resultados obtenidosde nuestro estudio y finalmente en el punto 7 expondremos las conclusiones obtenidas.
One of the most interesting human life associated risks in the actuarial field is thelongevity risk study. This risk is defined as the probability of persons living longerthan expected, increasing the insurance market’s concern because of the possibility ofreserve underestimates, which could lead to a risk of financial resources deficit to fulfillfuture liabilities. A way of mitigating these kinds of risks is to forecast the mortalityrate at country level, allowing the country or population safeguarded structure in theirpension scheme, or supporting insurance pricing or reserving process. Over time differenttechniques and models have been developed to predict mortality rates. Between themwe can find parametric models as classic models or non-parametric models as P-splines.The advanced of development techniques, such as Artificial Intelligence, has allowedlongevity to be studied from a new point of view, which could lead to the birth of newmore precise predictions models than the more using models until now. One of thesetechniques is the focus of this study, based on the longevity study using “Artificial NeuralNetworks” (ARN). ARN has a complex mathematical base, as well as a number ofparameters as the neuron’s numbers in each layer, learning process, activation functionsetc. which means that the estimation of these parameters are based on a trial and errorprocess.The project’s structure is the following. In the second chapter we’ll develop the moreused classical models from a theorical point of view. In the third chapter we’ll earn adeeper knowledge and understanding of Neural Networks, from their origins to theirstructure. in the fourth chapter we’re going to develop the methodology of our project.In the fifth chapter we’re going to analyse the source of the needed data for our project.In the sixth chapter we’re going to analyse the results of our project and finally, in theseventh chapter, we’re going to present our conclusions.
One of the most interesting human life associated risks in the actuarial field is thelongevity risk study. This risk is defined as the probability of persons living longerthan expected, increasing the insurance market’s concern because of the possibility ofreserve underestimates, which could lead to a risk of financial resources deficit to fulfillfuture liabilities. A way of mitigating these kinds of risks is to forecast the mortalityrate at country level, allowing the country or population safeguarded structure in theirpension scheme, or supporting insurance pricing or reserving process. Over time differenttechniques and models have been developed to predict mortality rates. Between themwe can find parametric models as classic models or non-parametric models as P-splines.The advanced of development techniques, such as Artificial Intelligence, has allowedlongevity to be studied from a new point of view, which could lead to the birth of newmore precise predictions models than the more using models until now. One of thesetechniques is the focus of this study, based on the longevity study using “Artificial NeuralNetworks” (ARN). ARN has a complex mathematical base, as well as a number ofparameters as the neuron’s numbers in each layer, learning process, activation functionsetc. which means that the estimation of these parameters are based on a trial and errorprocess.The project’s structure is the following. In the second chapter we’ll develop the moreused classical models from a theorical point of view. In the third chapter we’ll earn adeeper knowledge and understanding of Neural Networks, from their origins to theirstructure. in the fourth chapter we’re going to develop the methodology of our project.In the fifth chapter we’re going to analyse the source of the needed data for our project.In the sixth chapter we’re going to analyse the results of our project and finally, in theseventh chapter, we’re going to present our conclusions.