Analyzing the Influence of Market Event Correction for Forecasting Stock Prices Using Recurrent Neural Networks.

Understanding financial behavior, particularly in the stock market, has become a crucial task in recent years due to its significant impact on the global economy. One of the areas that addresses the relationship between finance and computer science to generate prediction models in this field is known as stock market prediction. This area aims to forecast the behavior of different stocks in the financial market. One of the most well-known and used techniques is Deep Learning, which is composed of different structures of deep neural networks that enable the learning of non-linear models. In this study, we utilized open data from the largest companies in Brazil, namely Petrobras and Itaúsa, provided by BovDB, a stock quotes dataset of all companies in the Brazilian stock exchange between 1995 and 2020. The data from the considered stocks were processed by means of a recurrent neural network aiming to analyze the influence of a price correction method that takes into account the impact of previous market events, e.g. an Ex-bonus, on the training and validation results produced by the RNN model. The experimental results show a good affinity of the model with temporal data, as well as a positive influence on noise reduction and fewer prediction errors, achieving an error reduction up to 26% considering Petrobrás.