A carregar...
Publicação
American put option pricing : a comparison between neural networks and least-square Monte Carlo method
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 1.62 MB | Adobe PDF |
Orientador(es)
Resumo(s)
Esta tese compara dois métodos de pricing de opções de venda Americanas. Os métodos estudados são redes neurais (NN), um método de Machine Learning, e Least-Square Monte Carlo Method (LSM). Em termos de redes neurais foram desenvolvidos dois modelos diferentes, um modelo mais simples, Model 1, e um modelo mais complexo, Model 2.
O estudo depende dos preços das opões de 4 gigantes empresas norte-americanas, de Dezembro de 2018 a Março de 2019.
Todos os métodos mostram uma precisão elevada, no entanto, uma vez calibradas, as redes neuronais mostram um tempo de execução muito inferior ao LSM. Ambos os modelos de redes neurais têm uma raiz quadrada do erro quadrático médio (RMSE) menor que o LSM para opções de diferentes maturidades e preço de exercício.
O Modelo 2 supera substancialmente os outros modelos, tendo um RMSE ca. 40% inferior ao do LSM. O menor RMSE é consistente em todas as empresas, níveis de preço de exercício e maturidade.
This thesis compares two methods to evaluate the price of American put options. The methods are the Least-Square Monte Carlo Method (LSM) and Neural Networks, a machine learning method. Two different models for Neural Networks were developed, a simple one, Model 1, and a more complex model, Model 2. It relies on market option prices on 4 large US companies, from December 2018 to March 2019. All methods show a good accuracy, however, once calibrated, Neural Networks show a much better execution time, than the LSM. Both Neural Network end up with a lower Root Mean Square Error (RMSE) than the LSM for options of different levels of maturity and strike. Model 2 substantially outperforms the other models, having a RMSE ca. 40% lower than that of LSM. The lower RMSE is consistent across all companies, strike levels and maturities.
This thesis compares two methods to evaluate the price of American put options. The methods are the Least-Square Monte Carlo Method (LSM) and Neural Networks, a machine learning method. Two different models for Neural Networks were developed, a simple one, Model 1, and a more complex model, Model 2. It relies on market option prices on 4 large US companies, from December 2018 to March 2019. All methods show a good accuracy, however, once calibrated, Neural Networks show a much better execution time, than the LSM. Both Neural Network end up with a lower Root Mean Square Error (RMSE) than the LSM for options of different levels of maturity and strike. Model 2 substantially outperforms the other models, having a RMSE ca. 40% lower than that of LSM. The lower RMSE is consistent across all companies, strike levels and maturities.
Descrição
Mestrado em Mathematical Finance
Palavras-chave
Valorização de Opções Redes Neurais Artificiais Métodos de Valorização Opções Americanas Aprendizagem de Máquina Aprendizagem Profunda Option Pricing Neural Networks Pricing Methods American Options Machine Learning Deep Learning
Contexto Educativo
Citação
Sequeira, Bernardo Pinto Machado Portugal (2019). "American put option pricing : a comparison between neural networks and least-square Monte Carlo method". Dissertação de Mestrado, Universidade de Lisboa. Instituto Superior de Economia e Gestão.
Editora
Instituto Superior de Economia e Gestão