Πανεπιστήμιο Μακεδονίας
Προβλεψιμότητα στις σχέσεις μεταβλητότητας των τιμών πετρελαίου και των χρηματιστηριακών αποδόσεων με τη χρήση των μονομεταβλητών πολυμεταβλητών και υβριδικών υποδειγμάτων GARCH
Abstract
dc:descriptionThe majority of the literature examines the impact of oil markets on stock exchanges, while there is little research regarding the opposite direction. The conclusions of these studies vary depending on whether they use symmetric or asymmetric shocks in the oil prices or whether they focus on unexpected changes in oil prices. This doctoral dissertation attempts to determine the relationship between the volatility of the financial markets of the Group of Seven (G7) and the oil market, taking into account possible asymmetries and structural breaks, but also to predict the volatility of these markets. Through various financial-econometric and hybrid models, we try to determine the existence or non-existence of a causal relationship between the aforementioned markets and oil prices.In the context of the empirical analysis of the dissertation, the stock indices of the Group of Seven (G7), the United States of America (Dow Jones Industrial Average), France (CAC 40), Germany (DAX), Italy (FTSE MIB), Japan (Nikkei225), United Kingdom (FTSE 100) and Canada (S&P/TSX) were used. Furthermore, two global benchmark crude oils were utilized, namely, the Brent and West Texas Intermediate. The survey data has weekly frequency, covers the period 07/01/1998 to 27/12/2017 (a total of 1043 observations for each of the 9 time-series), and is collected from the Yahoo Finance and Energy Information Administration databases. Also, in order to take into account the impact of the Great Recession, the entire empirical period extending from January 7, 1998 to December 27, 2017 is divided into two periods of equal length: pre-financial crisis (07/01/1998 – 26/12/2007, 521 observations) and post-financial crisis (02/01/2008 – 27/12/2017, 522 observations). To achieve the objectives of the dissertation, the Box-Cox transformation was used to minimize the sum of the squares of the residuals in all time-series. For the examination of the volatility spillovers between the stock markets and the oil prices, four multivariate models were utilized: the BEKK-GARCH model, the Asymmetric BEKK-GARCH model, the VARMA-GARCH model, and the Asymmetric VARMA-GARCH model. Regarding the structural breaks in variance, bivariate asymmetric BEKK-GARCH models were applied for the variance equation, and twenty-one bivariate models were estimated, each including the oil and stock market returns under consideration. Then we investigate the structural breaks in the variance of the time-series through the modified ICSS algorithm and after their detection, we introduce them as dummy variables in the variance equation of the symmetric and asymmetric BEKK models using the whole sample. Also, for the prediction of the models, the Box-Jenkins methodology and the hybrid models that are a combination of ARIMA models with GARCH models were applied. Asymmetric EGARCH and GJR-GARCH models were used, to be then compared with the artificial neural networks (ANN) and specifically with the hybrid model Type I (ANN-GARCH) and the hybrid model Type II (GARCH-ANN). For the effectiveness of the predictions in the aforementioned models, both the static (in-sample) and the dynamic forecast (out-of-sample) with the respective evaluation indicators were used. The results of the dissertation showed that the Box-Cox transformation fails to convert the returns so that they are distributed normally. The correlations between the stock and oil markets are not high, which means that the presence of oil as an asset in a stock portfolio can lead to greater diversification benefits. Comparing the periods before and after the financial crisis, the results of the dissertation suggest that according to the symmetrical models and regardless of the choice of oil type, in all cases except the Japanese stock market, the financial crisis weakened the transmission of shocks and volatility between the markets. In essence, oil markets have become more resilient to volatility originating from stock markets and vice versa. In addition, according to the asymmetric models, a corresponding weakening of the transmission of volatility between the markets is uncovered. In this case, however, it is important to emphasize the fact that the financial crisis strengthened the oil market WTI in the US market but also increased the influence of the US and German markets in the Brent oil market. For the volatility forecasts, four models were used, EGARCH, neural network, and two hybrid models (ANN-EGARCH and EGARCH-ANN). The results of the in-sample forecasts (short-term predictions) showed that the neural network is the most suitable while increasing the forecasting period to 25, 50, and 100 weeks, the hybrid model ANN-EGARCH produces the most accurate predictions. However, little importance should be given to the in-sample forecasts, as those out-of-sample are the object of interest of the dissertation. In this case, both hybrid models showed their superiority over the asymmetric GARCH and the neural network, which suggests that investors, in order to predict more accurately the volatility of the indices they are interested in, should resort to combinations of GARCH and neural networks (hybrid models). The findings of this dissertation can be an important source of help not only in future studies dealing with similar issues but also in financial market participants, portfolio managers, and policy-makers, since understanding these causal links can lead to crucial changes in energy planning, portfolio diversification as well as energy risk management and prevention, given that oil plays a critical role in international asset hedging strategies for various economic factors. Finally, in the 8 appendices of the dissertation, there is, in addition to the basic concepts of the mathematics of the dissertation and the data, the programming for the processing of the data so that all the weekly observations have the closing price of Wednesday. The programming for the calculation of the parameter λ of the Box-Cox transformation was done, among others, through the programming language R. The construction of the logarithmic returns of the multivariate GARCH models and the returns (simple-log-Box-Cox) of the ARIMA models for all stock indices was done with the programming language R. The estimation of the ARIMA models was programmed in E-views software, while the multivariate models and multivariate models with structural breaks, as well as the modified ICSS algorithm were programmed in RATS software. Finally, the programming of the hybrid models was done by combining the programming language R and the E-views software.
Degree
thesis:*- Grantor dc:publisher
- Πανεπιστήμιο Μακεδονίας
- Year dc:date
- 2021
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Kartsonakis-Mademlis, Dimitrios
- Καρτσωνάκης-Μαδεμλής, Δημήτριος
Subjects
dc:subject × 33- Χρηματιστηριακοί δείκτες των G7
- Τιμές του αργού πετρελαίου Brent
- WTI
- Μεταβλητότητα
- Μετασχηματισμός Box-Cox,
- Μονομεταβλητά-πολυμεταβλητά-υβριδικά υποδείγματα (συμμετρικά-ασύμμετρα)
- Διαρθρωτικές μεταβολές
- Τροποποιημένος αλγόριθμος ICSS
- Νευρωνικά δίκτυα
- Προβλέψεις
- Αιτιότητα Granger
- Γλώσσες προγραμματισμού
- Λογισμικά πακέτα
- G7 stock indices
- Brent crude oil prices
- Volatility
- Box-Cox transformation
- Univariate-multivariate-hybrid models (symmetric-asymmetric)
- Structural breaks
- Modified ICSS algorithm
- Neural networks
- Forecasts
- Granger causality
- Programming language
- Software
- Οικονομικά και Επιχειρήσεις
- Κοινωνικές Επιστήμες
- Οικονομετρία
- Στατιστικές, Πιθανότητες και Αβεβαιότητα
- Economics and Business
- Social Sciences
- Econometrics
- Statistics, Probability and Uncertainty
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/48932