Forecasting stock markets using machine learning : forecasting the PSI-20 index using a machine learning approach

Castelli, MauroDinis Oliveira, André2017-06-062017-06-062017-05-30http://hdl.handle.net/10362/21452Project Work presented as the partial requirement for obtaining a Master's degree in Statistics and Information Management, specialization in Risk Analysis and ManagementPredicting nancial markets is a task of extreme di culty. The factors that in uence stock prices are extremely complex to model. Machine Learning algorithms have been widely used to predict nancial markets with some degree of success. This Master's project aims to study the application of these algorithms to the Portuguese stock market, the PSI-20, with special emphasis on genetic programming and the introduction of the concept of semantics in the process of evolution. Three systems based on genetic programming were studied: STGP, GSGP and GSGP-LS. The construction of the predictive models is based on historical information of the index extracted through a blooberg portal. In order to analyze the quality of the models based on genetic programming, the nal results were compared with other Machine Learning algorithms through the application of signi cance statistical tests. An analysis of the quality of the results of the di erent algorithms is presented and discussed.engGenetic programmingStock marketsMachine learningGeometric semantic operatorsForecastingForecasting stock markets using machine learning : forecasting the PSI-20 index using a machine learning approachmaster thesis201702312