Forecasting multidimensional financial time series with multi-output least squares support vector regression

Forecasting financial time series has always been an area of great interest to both practitioners and researchers. Recently, machine learning techniques such as Neural Network(NN) and Support Vector Machine(SVM) has been studied intensively for stock prediction. However, most of the research done ar...

Full description

Saved in:
Bibliographic Details
Main Author: Leong, Wai Leong
Other Authors: PUN Chi Seng
Format: Final Year Project
Language:English
Published: Nanyang Technological University 2021
Subjects:
Online Access:https://hdl.handle.net/10356/146091
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
Description
Summary:Forecasting financial time series has always been an area of great interest to both practitioners and researchers. Recently, machine learning techniques such as Neural Network(NN) and Support Vector Machine(SVM) has been studied intensively for stock prediction. However, most of the research done are based on single-output stock prediction. To predict multiple stocks, the usual procedure is to train multiple single-output models, thus disregards the underlying cross relatedness among the stocks. In this project, a multidimensional stock prediction model is developed and evaluated using Multi-output Least Square Support Vector Regression(MLSSVR) which was proposed by Xu, An, Qiao, Zhu, & Li. Technical indicators are extracted from the historical stock prices and Independent Component Analysis(ICA) is used to extract the underlying/hidden information of the original indicators. The separated sources from ICA are then served as the inputs of MLSSVR to build the multidimensional stock prediction model. The proposed method is compared with training multiple single-output models namely the Linear Regression, LASSO, Support Vector Regression(SVR) and Least-squares Support Vector Regression(LS-SVR). The experimental results show that the proposed method perform on par to other models trained with the feature set incorporating information from other stocks.