Aplicación de support vector machine y long short term memory para el pronóstico del precio de las acciones del banco macro de argentina y bancolombia.

Abstract

The main purpose of the current development project has been to acquire the ultimate accuracy of state-of-the-art Artificial Intelligence (AI) algorithms like Machine Learning (ML) and Deep Learning Models (DL), more specifically Long Short-Term Memory (LSTM) and Support Vector Machine (SVM), in the forecasting of time series. During the last decade, immersion in AI models has opened new paths in different fields of study. Within this area, financial time series forecasting has gained great interest due in large part to new approaches that allow ML and DL models to achieve. These new approaches range from improving the predictive capacity, in comparison with statistical models such as ARIMA, SARIMA, GARCH, etc., allowing the transfer of knowledge to new data, to the application of new techniques for forecasting stock prices such as investor’s sentiment analysis. One of the main reasons for the great popularity of ML and DL models in the last decade is because they hold an outstanding forecasting capability, superior to that achieved by the statistical models mentioned in the previous paragraph. Another great advantage is that they have a large number of hyperparameters, which can be tailored to meet the specific characteristics of the problem that needs to be investigated and tackled. The novelty of the current investigation lies at the fact that LSTM and SVM will be applied to forecast the stock price of two Latin America Financial Institutions listed in the New York Stock Exchange, hopping this kind of investigation serves as an inspiring work to implement such algorithms in different issues of the financial landscape and others as well. As a way to evaluate the performance of the two AI algorithms, 3 error metrics will be implemented such as Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE) and Mean Absolute Error (MAE).