AUEB Library - Digital Repository

PYXIDA Institutional Repository
and Digital Library

Username
Password

Collections :	Ιδρυματικό Αποθετήριο ΟΠΑ / AUEB Institutional Repository Σχολή Οικονομικών Επιστημών / School of Economics Τμήμα Οικονομικής Επιστήμης / Department of Economics Μεταπτυχιακές Εργασίες / Postgraduate dissertations

Title :	Prediction of stock market index movement using machine learning techniques

Creator :	Impraimakis, Filippos

Contributor :	Kyriazidou, Ekaterini (Επιβλέπων καθηγητής) Dimelis, Sophia (Εξεταστής) Sakellaris, Plutarchos (Εξεταστής) Athens University of Economics and Business, Department of Economics (Degree granting institution)

Type :	Text

Extent :	57 p.

Language :	en

Abstract :	It goes without saying that the ability to predict the direction of stock/index is of paramount importance for the viability of the companies and individual investors. An accurate prediction of the sign of a stock index is an effective hedging strategy that can mitigate the risk level of companies. In essence, setting the risk is a mean that can yield a more efficient allocation of the companies' capital. In this study, eight different classification techniques were employed for the determination of the direction of the S&P 500. Two different approaches were used as inputs, first for the acquired principal components generated from PCA and second for our existing dataset. This comparison showed that Principal Component Analysis (PCA) negatively affect our results, except the KNN algorithm. Our experimental results verified the superiority of Support Vector Machines (SVM) in predicting financial time series.

Abstract :

It goes without saying that the ability to predict the direction of stock/index is of paramount importance for the viability of the companies and individual investors. An accurate prediction of the sign of a stock index is an effective hedging strategy that can mitigate the risk level of companies. In essence, setting the risk is a mean that can yield a more efficient allocation of the companies' capital. In this study, eight different classification techniques were employed for the determination of the direction of the S&P 500. Two different approaches were used as inputs, first for the acquired principal components generated from PCA and second for our existing dataset. This comparison showed that Principal Component Analysis (PCA) negatively affect our results, except the KNN algorithm. Our experimental results verified the superiority of Support Vector Machines (SVM) in predicting financial time series.

Subject :	Principal Component Analysis (PCA) Ridge & Lasso regression Random forests Nested cross-validation Trading Machine learning Prediction of stock market index movement Financial forecasting Neural networks Support Vector Machines (SVM)

Date Issued :	28-02-2019

Date Submitted :	08-06-2019

Access Rights :	Free access

Licence :

File: Impraimakis_2019.pdf

Type: application/pdf

Login