Choosing the most efficient model for the forecasting of financial resources

Within financial institutions, it is common practice to analyze the value of resources to obtain future results through prediction calculations. The calculation methodologies are the most diverse, from descriptive analysis and moving averages implemented in Excel spreadsheets to the most varied and powerful models in finance. These models are fundamental when it comes to creating diverse and complex systems that help simulate a range of investment scenarios and thus choose the best one among them. With the advent of Big Data and the significant increase of investments in new technologies, it has become more feasible to apply models derived from techniques called Machine Learning. These techniques have a high computational power to obtain the most accurate analysis and forecasts of financial market behavior. This paper aims to compare models based on the traditional methodological structure, using Time Series models, such as Autoregressive and Moving Average (ARIMA) and GARCH models. Furthermore, models with the Machine Learning approach, namely: Support Vector Regression (SVR) and Artificial Neural Network (ANN). These models were applied to some series of financial assets and some tokens. The results showed that for both types of assets, models with the machine learning approach performed better, but with different highlights for SVR and ANN respectively.