Using deep learning in stock price forecasting

• Autor(a) principal: Dietsche, Lucian Andreas Felix
• Data de Publicação: 2019
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Institucional do FGV (FGV Repositório Digital)
• Texto Completo: https://hdl.handle.net/10438/28601
• topic: Finance and accounting; Valuation; Equity valuation; Firm valuation; Financial ratios; Artificial intelligence; Neural network; Recurrent Neural Network; Finanças e contabilidade; Avaliação; Avaliação de ações; Avaliação de empresa; Índices financeiros; Inteligência artificial; Rede neural; Rede neural recorrente; Ciência política; Ações (Finanças) - Preços; Empresas - Avaliação; Inteligência artificial; Aprendizado do computador; Redes neurais (Computação)

The following paper investigates the possibility of using artificial intelligence, in particular a long short-term memory Network (LSTM), to forecast stock prices. As input data 59 different variables are chosen based on desk research and include: fundamental, technical, and macroeconomic data. The objective of the study is to use the selected independent variables to predict the stock return of the subsequent quarter of five retail companies listed on the Brazilian stock exchange (IBVOESPA). The research showed, that LSTM can be used to forecast stock price changes and an investment strategy based on the forecasts outperforms a buy and hold strategy of the same stock. Nevertheless, it should be said, that such an investment strategy is unlikely to have the same return in a real environment like it had in the backtesting. The reason for that is, that the number of data entries for each individual variable was not sufficiently large and the LSTM was not able to generalize the relationships. In other words, the superior performance of the algorithm may be due to overfitting of the model.