magistrsko delo
Nina Kos (Author), Marjetka Krajnc (Mentor)

Abstract

Cilj podjetij in investitorjev na finančnih trgih je bolj ali manj enak, doseganje čim večjih dobičkov. To je glavni razlog za razvoj metod, ki bi čim bolj natančno napovedale donose investiranih sredstev. Sprva je gibanje cen vrednostnih papirjev veljalo za povsem nepredvidljivo, kar je vodilo do razvoja teorije učinkovitega trga kapitala. Tekom let pa so empirične raziskave na finančnem področju pokazale, da na trgu obstajajo določene anomalije. S pomočjo analize gibanja cen v preteklosti si investitor lahko zastavi trgovalno strategijo, ki mu bo na dolgi rok prinašala nadpovprečne donose v primerjavi s trgom. Obstaja veliko metod za analizo in napovedovanje, a z razvojem računalništva v ospredje stopa strojno učenje. Ena izmed najbolj razširjenih metod strojnega učenja so umetne nevronske mreže. S pomočjo strojnega učenja je mogoče natančneje modelirati anomalije na trgu kapitala in iskati morebitne povezanosti med cenami vrednostnih papirjev, kot pa z uporabo tradicionalnih metod. V magistrskem delu je predstavljeno teoretično ozadje predvidljivosti v gibanju cen delnic in metode napovedovanja gibanja cen delnic. Poudarek je na integriranem avtoregresijskem modelu s premikajočim povprečjem (ARIMA model) in metodi nevronskih mrež. ARIMA model je kombinacija linearnih modelov časovnih vrst, in sicer avtoregresijskega modela (AR model) in modela premikajočega povprečja (MA model). Nevronske mreže pa so inteligentni sistemi, ki posnemajo delovanje živčnih celic v možganih. Zgrajene so iz umetnih nevronov in njihova glavna lastnost je sposobnost učenja povezave med vhodnimi in izhodnimi podatki. Omenjeni metodi sta uporabljeni na dejanskih podatkih ameriškega trga, in sicer na podatkih indeksa Dow Jones Industrial Average (DJIA) iz obdobja 2009 - 2014. Indeks DJIA kotira na newyorški borzi in na borzi NASDAQ ter zajema trideset najpomembnejših delnic newyorške borze. S pomočjo metod smo napovedali gibanje vrednosti indeksa ob koncu dneva v letu 2014. Ob primerjavi dobljenih rezultatov metod smo ugotovili, da metoda nevronskih mrež da veliko boljše rezultate napovedi kot ARIMA model. Na podlagi analize dobljenih rezultatov bi investitorju predlagali, da si pri napovedovanju gibanja cen delnic in s tem pri izbiranju najboljše strategije trgovanja z delnicami, pomaga z metodo nevronskih mrež.

Keywords

delnice;predvidljivost;napovedovanje;časovne vrste;ARIMA model;umetne nevronske mreže;algoritem vzvratnega razširjanja napake;

Data

Language: Slovenian
Year of publishing:
Typology: 2.09 - Master's Thesis
Organization: UL FMF - Faculty of Mathematics and Physics
Publisher: [N. Kos]
UDC: 519.2
COBISS: 18458713 Link will open in a new window
Views: 1112
Downloads: 308
Average score: 0 (0 votes)
Metadata: JSON JSON-RDF JSON-LD TURTLE N-TRIPLES XML RDFA MICRODATA DC-XML DC-RDF RDF

Other data

Secondary language: English
Secondary title: Forecasting DJIA index using traditional methods and artificial neural networks
Secondary abstract: The goal of companies and investors in financial markets is more or less the same, achieving the highest possible profits. This is the main reason for the development of methods, which would predict returns of invested assets as accurately as possible. Initially, the movement of securities prices was considered completely unpredictable, which led to the development of efficient capital market theory. During the years, empirical researches in the financial field has shown, that the capital market does not meet all assumptions of efficient capital market and that there exist certain anomalies. With the help of the analysis of past price movement, investor can set a trading strategy that will bring him above-average returns compared to the market in the long run. Many methods have been developed for analyzing the movement of securities prices and forecasting prices in the future. With the development of computer science, machine learning is becoming more and more popular. One of the most commonly used methods of machine learning are artificial neural networks. Machine learning enables more precise modeling of anomalies in the capital market and easier definition of possible connections between securities prices, compared to the traditional methods. In the master thesis the theoretical background of predictability in the movement of share prices and methods for forecasting their movement in the future are presented. The focus is on Autoregressive integrated moving average model (ARIMA model) and artificial neural networks. ARIMA model is a combination of time series linear models, namely autoregressive model (AR model) and moving average model (MA model). Artificial neural networks are intelligent systems that imitate the functions of nerve cells in the human brains. They are built from artificial neurons and their main characteristic is the ability to learn the connection between input and output data. Both mentioned methods are applied on the actual USA market data, namely on Dow Jones Industrial Average (DJIA) data from the period 2009-2014. DJIA index trade on the New York Stock Exchange (NYSE) and on NASDAQ. It consists of thirty major shares traded on the NYSE. We made the forecast of the DJIA index value movement in the year 2014. The comparison of the obtained results has shown, that artificial neural networks provide better results than ARIMA model. Based on our findings, we would propose an investor to choose artificial neural networks for share prices movement forecasting. Consequently, the investor could choose the best trading strategy.
Secondary keywords: shares;predictability;forecasting;time series;ARIMA model;artificial neural networks;backpropagation algorithm;
Type (COBISS): Master's thesis/paper
Study programme: 0
Thesis comment: Univ. v Ljubljani, Fak. za matematiko in fiziko, Oddelek za matematiko, Finančna matematika - 2. stopnja
Pages: VIII, 69 f.
ID: 10962496