master's thesis
EKONOMENTRIJSKI MODELI ZA PREDVIĐANJE RAZINE CIJENA I VOLATILNOSTI NAFTE

Vicko Perasović (2016)
University of Split
Faculty of economics Split
Metadata
TitleEKONOMENTRIJSKI MODELI ZA PREDVIĐANJE RAZINE CIJENA I VOLATILNOSTI NAFTE : Diplomski rad
AuthorVicko Perasović
Mentor(s)Josip Arnerić (thesis advisor)
Abstract
Ovaj rad bavi se problematikom modeliranja cijene i volatilnosti sirove nafte na dva seta podataka: desetogodišnji vremenski niz s tjednim opservacija te petogodišnji vremenski niz s dnevnim opservacijama. Započinje se analizom fundamentalnih karakteristika proizvoda i tehnikama deskriptivne statistike. Nakon pobližeg proučavanja dinamike vremenskih nizova, testiranjem stacionarnosti, sezonalnosti i strukturnih lomova, potrebnim transformacijama podaci se pripremaju za modeliranje. Na temelju odabranih informacijskih kriterija biraju se optimalni ARIMA modeli čije su karakteristike detaljnije proučene. Biranjem nezavisnih varijabli i testiranjem korelacije, gradi se optimalni model linearne regresije koji se potom dodatno razrađuje. Kao nastavak na ARIMA proces, modelira se osnovni GARCH model te se vrši naknadna procjena modela. Modeliraju se i različite varijante GARCH-a: apARCH, eGARCH i gjrGARCH. Konačno, različiti ARIMA modeli se međusobno uspoređuju na temelju MSE i MEA kriterija. Rezultati ne pokazuju prednost u odabiru konkretnog modela, čime se naglasak prebacuje na odabir jednostavnijih modela, kao što je AR(1). GARCH modeli se također uspoređuju na temelju istih kriterija. Rezultati pokazuju znatnu prednost u korištenju apARCH modela s naglaskom na desetogodišnji vremenski niz. Regularni sGARCH predstavlja drugu najbolju alternativu. GARCH modeli pokazuju bolje rezultate u modeliranju vremenskih nizova sirove nafte u odnosu na ARIMA modele zbog volatilne prirode proizvoda.
GranterUniversity of Split
Faculty of economics Split
PlaceSplit
StateCroatia
Scientific field, discipline, subdisciplineSOCIAL SCIENCES
Economics
Quantitative Economics
Study programme typeuniversity
Study levelgraduate
Study programmeBusiness Studies
Academic title abbreviationmag. oec.
Genremaster's thesis
Language Croatian
Defense date2016
Parallel abstract (English)
This paper examines the problematic of modeling crude oil price and volatility on two sets of data: a ten year time series with weekly observations and a five year series with daily observations. It begins by analyzing the fundamental characteristics of the product and by employing methods of descriptive statistics. After closely studying the dynamics of the series by testing for stationarity, seasonality and structural breaks, with the necessary transformations the data is prepared for modeling. Based on the selected information criterion, the optimal ARIMA model is chosen, from whoms characteristics are studied further. By choosing independent variables and testing for correlation, an optimal linear regression model is built which is then furtherly analyzed. As an addition to the ARIMA process, a basic GARCH model is built, followed by postestimation. Different variants of GARCH are also modeled: apARCH, eGARCH and gjrGARCH. Finally, different ARIMA models are compared based on the MSE and MEA criteria. The results don't show an advantage in choosing a concrete model, which therefore stresses the importance of choosing a simpler model, like AR(1). GARCH models are also compared based on the same criterion. The results show an advantage in using the apARCH model with emphasis on the ten year time series. The regular sGARCH is the second best alternative model choice. GARCH models show better results in modeling crude oil time series in comparison to ARIMA models because of the high volatility nature of the product.
Parallel keywords (Croatian)nafta ARIMA model GARCH model
Resource typetext
Access conditionOpen access
Terms of usehttp://rightsstatements.org/vocab/InC/1.0/
URN:NBNhttps://urn.nsk.hr/urn:nbn:hr:124:372911
CommitterIvana Gizdić