Neural network pricing of american put options

• Autor(a) principal: Gaspar, Raquel
• Data de Publicação: 2020
• Outros Autores: Lopes, Sara D., Sequeira, Bernardo
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10400.5/20015
• topic: Machine learning; Neural networks; American put options; Least-square Monte Carlo

In this paper we use neural networks (NN), a machine learning method, to price American put options. We propose two distinct NN models – a simple one and a more complex one. The performance of two NN models is compared to the popular Least-Square Monte Carlo Method (LSM). This study relies on market American put option prices, with four large US companies as underlying – Bank of America Corp (BAC), General Motors (GM), Coca-Cola Company (KO) and Procter and Gamble Company (PG). Our dataset includes all options traded from December 2018 to March 2019. All methods show a good accuracy, however, once calibrated, NNs do better in terms of execution time and Root Mean Square Error (RMSE). Although on average both NN models perform better than LSM, the simpler model (NN model 1) performs quite close to LSM. On the other hand our NN model 2 substantially outperforms the other models, having a RMSE ca. 40% lower than that of the LSM. The lower RMSE is consistent across all companies, strike levels and maturities.