Publication: Predicción de valores en mercados financieros
Loading...
Official URL
Full text at PDC
Publication Date
2019
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citations
Exportar
Abstract
En este trabajo se van a comparar diversos métodos para predecir el futuro de ciertos índices bursátiles a partir de los datos históricos de otros o de él mismo. En concreto, se seleccionan uno índices bursátiles a los que se les aplican distintos experimentos. En primer lugar, utilizamos el método ARIMA, en el cual se utiliza el pasado de la misma serie temporal de la cual se quiere predecir el futuro y se compara con el sencillo método de Naïve, consistente en proyectar el último valor conocido. Posteriormente, se realizan experimentos con distintos métodos de regresión (Regresión Lineal, Gradient Boosting Regressor, Random Forest Regressor y Voting Regressor) utilizando incrementos en lugar de los valores reales, para lo que es necesario hacer un preajuste de los datos de entrada. Además, incorporamos varias variables independientes para mejorar la predicción. Como último método de predicción se realiza un experimento utilizando redes neuronales, en concreto se trata de percepción multicapa con backpropagation.
Finalmente, se comparan y analizan los resultados de todos estos métodos a través de las medidas de error típicas y se exponen las conclusiones finales.
In this work we will compare different methods to predict the future of certain stock indices from the historical data of others and itself. Specifically, one stock index is selected to which different experiments are applied. First, we use the ARIMA method, which uses the pasts of the same time series from which we want to predict the future and compares it with the simple Naïve method (consisting of projecting the last known value). Subsequently, experiments are performed with different regression methods (Linear Regression, Gradient Boosting Regressor, Random Forest Regressor and Voting Regressor) using increments instead of real values, for which it is necessary to make a preset of the input data. In addition, we incorporate several independent variables to improve predictions. The last method of prediction is an experiment using neural networks, specifically multi-layer perceptron with backpropagation. Finally, the results of all these methods are compared and analyzed through the typical error measures and the final conclusions are reveal.
In this work we will compare different methods to predict the future of certain stock indices from the historical data of others and itself. Specifically, one stock index is selected to which different experiments are applied. First, we use the ARIMA method, which uses the pasts of the same time series from which we want to predict the future and compares it with the simple Naïve method (consisting of projecting the last known value). Subsequently, experiments are performed with different regression methods (Linear Regression, Gradient Boosting Regressor, Random Forest Regressor and Voting Regressor) using increments instead of real values, for which it is necessary to make a preset of the input data. In addition, we incorporate several independent variables to improve predictions. The last method of prediction is an experiment using neural networks, specifically multi-layer perceptron with backpropagation. Finally, the results of all these methods are compared and analyzed through the typical error measures and the final conclusions are reveal.
Description
Trabajo de Fin de Grado, Universidad Complutense, Facultad de Informática, Departamento de Sistemas Informáticos y Computación, Curso 2018/2019