Please use this identifier to cite or link to this item: http://hdl.handle.net/10609/107367
Title: A comparison between LSTM and Facebook Prophet models: a financial forecasting case study
Author: González Mata, Alejandro
Tutor: Isern, David  
Others: Ventura, Carles  
Abstract: Finance forecasting is one of the main applications of Machine Learning in general and Artificial Neural Networks in particular. Long Short-Term Memory (LSTM) networks have been proven specially useful in this field and in other time series problems. However, less literature has been written about the suitability and performance of the Facebook Prophet model, which is based on an additive model that blends non-linear trends with configurable seasonalities. This project aimed to build and compare two predictive models, one based on a LSTM network and another one based on the Prophet algorithm, to analyse which of them performed better in forecasting tasks, particularly in predicting the price of the S&P500 index. In order to compare these two models, a trading simulator module was built as a backtesting platform where the predictions of the models could be applied to determine its economic performance. After building the two models, it was demonstrated that the LSTM model achieved better results, and was proven a decent predictor compared to other benchmark trading strategies. The Prophet model also showed positive returns of investment, but its accuracy as a predictor was not as high. The project results also suggest that a backtesting platform is a convenient feature when dealing with forecasting enterprises, since it allows to detect asymmetries between the fitness of a model during the testing phase of the algorithm and its later performance in a simulated trading environment.
Keywords: LSTM
Facebook Prophet
financial forecasting
Document type: info:eu-repo/semantics/bachelorThesis
Issue Date: Jan-2020
Publication license: http://creativecommons.org/licenses/by/3.0/es/  
Appears in Collections:Bachelor thesis, research projects, etc.

Files in This Item:
File Description SizeFormat 
agonzalez_mataTFG0120memory.pdfMemory of TFG1,44 MBAdobe PDFThumbnail
View/Open
agonzalez_mataTFG0120presentation.pdfPresentation of TFG309 kBAdobe PDFThumbnail
View/Open