Modelo de inteligência artificial para previsão do índice S&P500

Abstract

The global financial market is responsible for the circulation of a huge amount of money, estimated at more than 30 trillion dollars in 2022. Currently, after the pandemic, wars and inflationary instability, there is great uncertainty surrounding the future and security that currencies and businesses that currently dominate the market can provide for investors and people in general. In response to these tensions, more and more people are looking to invest their money, and diversify such investments, in the stock market in order to protect their savings and income, or for the more adventurous, seek new sources of income in the ups and downs of the market. . In view of this entry of new entrants to the market with no experience and other older investors who may be having difficulty finding profitable investments in this new economy, this work aimed to develop a model capable of predicting the value of the S&P500 index , which evaluates and estimates the market value of the 500 largest publicly traded companies operating in the United States of America. In addition to acting as a case study, the choice of the index comes from its breadth and diversity of assets that compose it, which makes it widely used as a gauge of the general state of the market, which tends to follow with optimism or pessimism according to trends shown by the index. The proposed model differs from those previously developed by other authors due to its emphasis on developing a lightweight system that can be locally retrained by the user without the need for intensive resources, which was made possible by the use of the Facebook Prophet tool in conjunction with Decision Trees. The technologies used primarily consist of Facebook Prophet, which was the tool that proved to be the most suitable for the application of this solution and decision trees, which complement the multivariate needs of the model, and for various functionalities, Python libraries such as Sklearn. The results of the model developed showed a slight improvement compared to the best of the other models tested with different components, but given its performance in conjunction with the simulated results, it was concluded that the proposed objectives were achieved satisfactorily.