Comparative Study of Artificial Neural Network and ARIMA Models for Economic Forecasting

190 Pages·2011·1.47 MB·English

id5314390 pdfMachine by Broadgun Software - a great PDF writer! - a great PDF creator! - http://www.pdfmachine.com http://www.broadgun.com

Al- Azhar University (cid:150) Gaza
Faculty of Economics and Administrative Sciences
Department of Applied Statistics


Comparative Study of Artificial Neural
Network and ARIMA Models
for Economic Forecasting

By
Assem A. Yassen

A thesis submitted in partial fulfillment of requirements
for the degree of Master of Statistic

Under the Supervision of

Dr. Mahmoud Okasha
Associate Professor of Statistics

Sep 2011


A
















Comparative Study of Artificial Neural Network and
ARIMA Models for Economic Forecasting















B
Al-Azhar University (cid:150)Gaza
Deanship of Graduate Studies

Comparative Study of Artificial Neural Network and ARIMA Models for Economic
Forecasting

A thesis submitted in partial fulfillment of requirements for the
degree of master of statistics
By

Assem Ali Hasan Yassen

Committee of Evaluation

Title Signature

1. Dr. Mahmoud K. Okasha
Head of committee (cid:133)(cid:133)(cid:133)(cid:133)(cid:133)(cid:133)(cid:133).
Associate Professor of Statistics

2. Dr. Abdulla M. El-Habil
Internal examiner (cid:133)(cid:133)(cid:133)(cid:133)(cid:133)(cid:133).
Associate Professor of Statistics

3. Dr.Samir K. Safi
External examiner (cid:133)(cid:133)(cid:133)(cid:133)(cid:133)(cid:133).
Associate Professor of Statistics

Faculty of Economic and Administrative Science
Department of Statistics
Gaza (cid:150) Palestine
2011

C

Ú웦ׅÞʦׅį…Ú¬‰

ÖÓÍæËå4†°ßÞۍ†—¥¡Â˦ß
ÚìØÃÚØÃé£
ÚìÀÄׅį…Í¢³


Ê«æë‹¥æ«



i
ŇŒťƋō

ƓœŕƁŧŰŌƏƓœŕŗţŌ¿ƄƑƅŏ


ʼnŕƊŸƅŔƏŖŸśƅŔƓƊƏƄũŕŮƉƔŨƅŔʼnŔŪŷƗŔƓœŕƊŗŌƏƓśŠƏŪƑƅŏ 
ʼnŕŷŧƅŕŗƓƆŷ¿ŦŗśƇƅƓśƅŔŘŪƔŪŸƅŔƓśŧƅŔƏƑƅŏ 
ĺŔƌƈţũŪƔŪŸƅŔƒŧƅŔƏšƏũƑƅŏ 
ƓƆŸƅŔĺŔŧŸŗƉŕƄƒŨƅŔƉŬţƅŔƓŦŌũƄŨƅŕŗůŦŔƏĻŕŸƔƈŠƓśŔƏŦŌƏƓƊŔƏŦŏƑƅŏ 
řŬŔũŧƆƅƓśƆŰŔƏƈŖŗŬũƔŧƂƅŔ

ʼnŕųŸƅŕŗƓƆŷŔƏƆŦŗƔƇƅƉƔŨƅŔƉƔũƁƏƈƅŔƓśŨśŕŬŌƑƅŏ 
ʼnŔŪŷƗŔƓœŕƁŧŰŌ¿ƄƑƅŏ 
¿ƈŸƅŔŔŨƍŪŕŠƊŔƓžƓƊŧŷŕŬůŦŮ¿ƄƑƅŏ 


žžžžžžŶŲŔƏśƈƅŔŧƎŠƅŔŔŨƍƒŧƍŔʼnƛŎƍ¿ƄƑƅŏ




ii
DECLARATION

I certify that this thesis submitted for the Master degree is the result of my own research,
except where otherwise acknowledged, and that this thesis (or any part of the same) has
not been submitted for a higher degree to any university or institution.

Signed (cid:133)(cid:133)(cid:133)(cid:133)(cid:133)(cid:133)(cid:133)(cid:133)(cid:133)(cid:133)..
Assem Ali Hasan Yassen

Date: -----------------

iii
Acknowledgement

I would like to extend my sincere thanks and gratitude to all who helped me complete this
work and link to such a degree of knowledge. I would like to express my sincere
gratitude to my supervisor, Dr. Mahmoud . K. Okasha, for his constant support, and
encouragement throughout my studying. I want to thank my graduate committee
members: Dr. Mahmoud. K. Okasha, Dr. Abdalla. M. El-Habeel, and Dr. Samir. K. Safi
for help, support, and constructive suggestions, throughout my master program study.
I would also give my thanks to all the faculty members in the Departments of Statistics.
The friendship and support from all the graduate students and the department staff will be
kept in my heart for the rest of my life.
I also thank all members of my family, my relatives, colleagues, and all those who helped
me and supported me.

Thank you all

iv
Abstract
The accuracy of forecasts using time series models on economic data, such as stock price
indexes, has recently received a great attention. The Box-Jenkins (Autoregressive
Integrated Moving Average) (cid:147)ARIMA(cid:148) models have been the most widely used models
for forecasting. These models give good forecasts for future observations but they are
unfortunately, not so accurate for many ones. This is because the forecasts converge to
the mean of the series after three or four forecast values. Recent studies suggest that
artificial neural networks (ANN) can be a promising alternative to the traditional method,
ARIMA, in forecasting especially in the case of nonlinear data and for forecasting many
future forecast values.
In this study the forecasting capabilities of Artificial Neural Networks and traditional
ARIMA models are compared. The comparison has been conducted using economic and
financial data through studying the efficiency of ARIMA and ANN models for modeling
and forecasting the daily data of Al-Quds Index for 3 years. The results was that the ANN
provides forecasts so close to the actual ones when the logarithmic transformation of the
original time series is used. However, the first difference for the log transformation of the
original data was required in ARIMA, but the results were not satisfactory.







v
κΨϠϣ
ϡΎѧϤΘϫϻ΍ ΰϛήΗ ΚϴΣ ˬ΍ΪΟήϴΒϛϞϜθΑ ΔϳΩΎμΘϗϻ΍Ε΍ήϴϐΘϤϟΎΑΆΒϨΘϟ΍ ΔϴϤϫ΃ ΓήϴΧϷ΍Δϧϭϵ΍ϲϓΕΩ΍Ωί΍
έΎόγϷΔϴγΎϴϘϟ΍ϡΎϗέϷ΍ ΔϠδϠγϞΜϣˬΔϴϨϣΰϟ΍Ϟγϼδϟ΍ ϰϠϋ΍ΩΎϤΘϋ΍ΔϳΩΎμΘϗϻ΍Ε΍ΆΒϨΘϟ΍ ϲϓ ΔϗΪϟ΍ ϰϠϋ
ϞѧϣΎϜΘϤϟ΍ϲΗ΍άѧϟ΍έ΍ΪѧΤϧϻ΍ ΝΫΎѧϤϧˬΰѧϨϴϜϨϴΟβϛϮѧΑΝΫΎѧϤϧΖѧϴψΣΪѧϘϟϭΔѧϴϨϣίΕ΍ήΘϔϟϢϬγϷ΍
ϞγϼδϟΎΑ ΆΒϨΘϟ΍ ϲϓήΒϛϷ΍ϡΎϤΘϫϻΎΑ ARIMAΝΫΎϤϧέΎμΘΧΎΑΔϓϭήόϤϟ΍ϭ ΔϛήΤΘϤϟ΍ΕΎτγϮΘϤϠϟ
ϩάѧϫϥ΃΢πѧΘϳϪѧϧ΃ϻ·ˬΔϴϠΒϘΘδѧϣΕ΍ήѧΘϓϲѧϓΔϠδѧϠδϟ΍ϢϴѧϘϟΕ΍ΆѧΒϨΗΝΫΎѧϤϨϟ΍ϩάѧϫϲѧτόΗΚѧϴΣˬΔϴϨϣΰϟ΍
ϭ΃ΙϼѧΛΪѧόΑΔѧϴϨϣΰϟ΍ΔϠδϠδϟ΍ςγϮΘϣϰϟ·ΏέΎϘΘΗϭΔϠϳϮσΔϴϨϣίΕ΍ήΘϔϟΔϴϓΎϜϟ΍ΔϗΪϟΎΑΖδϴϟΕ΍ΆΒϨΘϟ΍
ϢϴϘϠϟΕ΍ΆΒϨΗϊΑέ΃
ANNΝΫΎѧϤϨΑ˱΍έΎμѧΘΧ΍ΔѧϓήόϤϟ΍ ΔϴϋΎϨτѧλϻ΍ΔϴΒμѧόϟ΍ ΕΎϜΒθϟ΍ ϥ΃ ϰϟ·ΔΜϳΪΤϟ΍ΙΎΤΑϷ΍ήϴθΗϭ
ΔѧλΎΧϭ ˬΆѧΒϨΘϟ΍ ϲѧϓΰѧϨϜϨΟβϛϮΑ ΝΫΎϤϧϡ΍ΪΨΘγΎΑΔϳΪϴϠϘΘϟ΍ΔϘϳήτϟ΍ϦϣϞπϓ΃ ϼϳΪΑ ϥϮϜΗϥ΃ ϦϜϤϳ
.ΔѧѧѧϠϳϮσΔѧѧѧϴϨϣίΕ΍ήѧѧѧΘϔϟΔϠδѧѧѧϠδϟ΍ϢϴѧѧѧϘΑΆѧѧѧΒϨΘϟ΍ϑΪѧѧѧϬΑϭˬΔѧѧѧϴτΨϟ΍ήѧѧѧϴϏ ΕΎѧѧѧϧΎϴΒϟ΍ ΔѧѧѧϟΎΣ ϲѧѧѧϓ
ΔϴΒμѧόϟ΍ΕΎϜΒθѧϟ΍ϦѧϣϞѧϛϕήѧσϡ΍ΪΨΘѧγΎΑΆѧΒϨΘϟ΍ Ε΍έΪѧϗ ϦϴѧΑΔѧϧέΎϘϣ ΖѧϳήΟ΃Δѧγ΍έΪϟ΍ ϩάѧϫ ϲѧϓϭ
ΝΫΎѧϤϨϟ΍Γ˯Ύѧϔϛ Δѧγ΍έΩ ϝϼѧΧϦѧϣ ϚѧϟΫϢѧΗϭ ˬΔѧϳΪϴϠϘΘϟ΍ΎѧϤϳέ΃ΰѧϜϨϴΟβϛϮѧΑΝΫΎѧϤϧϭ ΔϴϋΎϨτѧλϻ΍
ΔϴϣϮϴϟ΍ ΕΎϧΎϴΒϟ΍ϡ΍ΪΨΘγΎΑ ϚϟΫϭˬΓΩΪόΘϣΔϴϨϣίΕ΍ήΘϔϟαΪϘϟ΍ήηΆϣ ΔϤϴϘΑ ΆΒϨΘϟ΍ϰϠϋΎϬΗέΪϗϭΔϘΑΎδϟ΍
(ANN ΝΫΎѧϤϧϡ΍ΪΨΘѧγΎΑΕ΍ΆѧΒϨΘϟ΍ ϥ΃ ΕΎѧϧέΎϘϤϟ΍ϩάѧϫΞ΋ΎΘϧΖϨϴΑΪϗϭˬΕ΍ϮϨγ 3 άϨϣ αΪϘϟ΍ ήηΆϤϟ
ΝΫΎѧϤϧΐѧϠτΘΗΎѧϤϨϴΑ ˬ ΔѧϴϨϣΰϟ΍ΔϠδѧϠδϟ΍ϢΘϳέΎѧϏϮϟ ϡ΍ΪΨΘѧγ΍ϢΘѧϳΎϣΪѧϨϋ ΔѧϴϠόϔϟ΍ ϢϴѧϘϟ΍ϰѧϟ·˱΍ΪѧΟΔѧΒϳήϗ
ΕΎѧѧϧΎϴΒϠϟΔѧѧϴϤΘϳέΎϏϮϠϟ΍ ΔѧѧϠϳϮΤΘϟ΍˯΍ήѧѧΟ·ΪѧѧόΑϚѧѧϟΫϭϰѧѧϟϭϷ΍ΔѧѧΟέΪϟ΍Ϧѧѧϣϕϭήѧѧϔϟ΍ΏΎδѧѧΣ(ARIMA
ΔΑϮϠτϤϟ΍Ε΍ΆΒϨΘϟ΍ ϰϠϋϝϮμΤϠϟΔϴοήϣ ϦϜΗϢϟΓήϴΧϷ΍ΝΫΎϤϨϟ΍ϡ΍ΪΨΘγΎΑΞ΋ΎΘϨϟ΍ϦϜϟ ˬΔϴϠλϷ΍



vi
Table of Contents
Subject Page NO
List of Tables x
List of Figures xi
Abbreviations xiii
Chapter one
1(cid:150) Introduction 1
1.1(cid:150) Rationale 1
1.2(cid:150) The data 2
1.2.1(cid:150) Palestine Market Stock Exchange 3
1.2.2(cid:150) Al-Quds Index 3
1.3(cid:150) Research problem 4
1.4(cid:150) Research objective 4
1.5 (cid:150) Research Methodology 5
1.6(cid:150) Literature review 5
1.7(cid:150) Organization of research 8
Chapter Two
2(cid:150)Artificial Neural Networks 9
2.1 (cid:150) Introduction 9
2.2 (cid:150) Historical Overview of NN 9
2.3 (cid:150) Neural Networks Models 11
2.3.1 (cid:150) Feed-Forward Neural Network 11
2.3.2 (cid:150) Multilayer Feedforward Neural Networks 12
2.3.3 (cid:150) Recurrent Neural Networks 15
vii