Brazilian stock data prediction over univariate and multivariate LSTM models
Accurate prediction of non-linear data patterns has proven to be difficult, especially those which address stock data, erratic in nature. Every stock broker has employed its own strategy intending to resolve such conflict, many of which apply the use of Artificial Intelligence, referred by the literature as intelligent models. Stock market value represents a vast indicator of economic health, whereas precise prediction may anticipate index fluctuation and perhaps critical events. Hence, this work centres itself in two main predictive approaches over the Brazilian stock represented by its most distinguished index (ibovespa), applying a specific set of Artificial Neural Networks, known as LSTM (long short term memory); (i) Adaptability of a given univariate LSTM model to market fluctuation, evaluated over the COVID-19 quarantine period and (ii) Relevance of macro-economical features as support information to long-range prediction, over a multivariate LSTM model.