Please use this identifier to cite or link to this item:
Author: Λουκέρης, Νικόλαος
Title: Επιλογή χαρτοφυλακίου μέτοχων και βελτιστοποίηση διαχείρισης του με μεθόδους τεχνητής νοημοσύνης και υπολογιστικής ευφυΐας
Alternative Titles: Portfolio selection and optimal portfolio management with methods of artificial intelligence and computational intelligence
Date Issued: 2016
Department: Πανεπιστήμιο Μακεδονίας. Τμήμα Λογιστικής & Χρηματοοικονομικής (ΛΧ)
Supervisor: Ελευθεριάδης, Ιορδάνης
Abstract: The scope of this Doctorate of Philosophy is triple: A) at first to investigate the Utility Function of investors in terms of overall characteristics and investor attributes to incorporate more detailed information that will offer an advanced Portfolio Theory based on Markowitz’s (1952) ideas, and B) secondly to examine thoroughly models of Artificial Intelligence in the domain of Neural Networks and Hybrid models, of Computational Intelligence and further Heuristics that will be implemented on more efficient Portfolio Selection by professionals or the academia C) Examining further alternative models in optimal Asset Allocation, and Risk measurement The fulfillment of the previous scope is achieved in case A) elaborating the Isoelastic Utility Fuction, and the Power Utility Functions (continuing past research on this one), and examining further higher moments: the fourth, the fifth and the sixth moments to incorporate more detailed information on the investor preferences towards asset allocation, B) investigating the Support Vector Machines, Time Lag Recurrent Networks, Recurrent Networks, Jordan Elmans, MultiLayer Perceptrons, Voted Perceptron, Radial Basis Functions, models of Neural Networks, their Neuro – Genetic Hybrids and the Regressions: Multinomial Logistic Regression-Logistc, Linear Logistic Regression-Simple Logistic, Logistic Model Trees, Additive Logistic Regression-Logitboost, Bayesian Logistic Regression, AdaBoost M1, whilst in Computational Intelligence Genetic Algorithms are examined, and Heuristics: Differential Evolution, Particle Swarm Otimisation, to define the best models in terms of performance and efficiency into Portfolio Selection and Financial Management. C) determine the ability of Three Factor Model to track the anomalies of average returns, introducing a more detailed portfolio creation, and examining alternatives on VaR and CVaR models(continuing past research). The innovations are A) the implementation of special forms of the Power Utility and Isoelastic Utility Functions, the incorporation of the fourth, the fifth and the sixth moments of the investors wealth on the utility function, B) the thorough evaluation of the SVM, TLRN, RNN, MLP, Voted Perceptron, RBFN Neural Nets and their Neuro-Genetic Hybrids architectures in various topologies from 0 to 10 hidden layers, the evaluation of MLR, LLR, LMT, ALR, BLR, AdaBoost M1 Regressions, of GA, DE, PSO Computational Intelligencee and Heuristic models to define the best model for Portfolio Selection and Corporate Evaluation. C) the extended portfolio creation following the Morningstar 3x3 bonds matrix and introducing a new 5x5 matrix further to the well known 3x2 FF’s matrix, and two different sorting methods, Proportional, and Value categorisation in three different cases of 6, 9 and 25 portfolios, whilst on Risk further backtesting and VaR and CVaR alternatives are examined. In Chanpter 1 the theoretical substratu on the Portflio Selection probem beyond higher moments. Chapter 2 describes th metyhodology I followed, the models I created and the data of the neural computation. Chapter 3 has the combine discussion of the Artificial Intelligence in Portfolio Selection. Chapter 4 discusses my past research during my MSc at Essex University, on the Heuristics and Utlity Functions. Similarly Chapter 5 describes my past research rom Essex Uni. on the Three Factor Model that has a close contribution.
Keywords: Επιλογή χαρτοφυλακίου
Ανώτερες ροπές
Νευρωνικά δίκτυα
Υβριδικά δίκτυα
Γενετικοί αλγόριθμοι
Θεωρία χαρτοφυλακίου
Time lag recurrent networks
Recurrent networks
Jordan Elman networks
Support vector machines
Generalized feed forward networks
Radial basis function networks
Multilayer perception
Logistic regressions
Adaboost m1
Information: Η βιβλιοθήκη διαθέτει αντίτυπο της διατριβής σε έντυπη μορφή.
Διατριβή (Διδακτορική)--Πανεπιστήμιο Μακεδονίας, Θεσσαλονίκη, 2016.
Περιλαμβάνει βιβλιογραφικές αναφορές (σ. 182-209).
Rights: Το ηλεκτρονικό αντίτυπο της διατριβής θα αποδεσμευτεί μετά τις 22/12/2019.
Αναφορά Δημιουργού-Μη Εμπορική Χρήση 4.0 Διεθνές
Appears in Collections:Τμήμα Λογιστικής & Χρηματοοικονομικής (Δ)

Files in This Item:
File Description SizeFormat 
LoukerisNikolaosPhd2016.pdf5.94 MBAdobe PDFView/Open

This item is licensed under a Creative Commons License Creative Commons