University of the Aegean
A study of time series with time-varying variance: a practical approach with applications in actuarial science and finance
Abstract
dc:descriptionA stochastic time series is a type of time series where the future values can only be determined in terms of a probability distribution. If this probability distribution is constant over time, then the time series is said to be stationary. A less strict condition for stationarity requires that at least the level and variance of the time series be constant over time. Researchers employ various tests to check for non-stationarity in the level of a time series, but more often than not they neglect to investigate non-stationarity in its variance when conducting applied research. In fact, regarding time series variance, the primary research emphasis is on modeling autoregressive conditional heteroscedasticity, typically using a variety of ARCH-GARCH models. It is essential to differentiate between two key concepts: variance non-stationarity, often referred to as heteroscedasticity, and conditional heteroscedasticity. Heteroscedasticity implies a functional relation between the variance of a series which is non-stationary in its level and its mean level. This entails non-stationarity in the variance, and the variance is neither conditionally nor unconditionally constant. Consequently, the process is non-homogeneously non-stationary in the sense of Box and Jenkins and cannot be made stationary by simply differencing it. To address variance non-stationarity, one approach is to apply power transformations, such as the well-known Box and Cox transformations. On the other hand, conditional heteroscedasticity, often described using ARCH or GARCH models, signifies that while the conditional variance varies over time, the unconditional variance remains constant. As a result, the series is stationary in the second moment. In the present Ph.D. thesis, the focus is on the series with non-constant variance both conditionally and unconditionally, covering to a certain extent a gap in that area, as the existing research work is relatively scanty. Indeed, even though it is crucial to deal with non-constant variance in time series modeling, there is a shortage of comprehensive theoretical research on its detection and correction. Moreover, in practical applications, the treatment of non-stationary variance is not only insufficient, as the choice of a specific transformation is often arbitrary, but also, as is documented in Chapter 2, occasionally biased towards over-rejection of the null hypothesis of unconditionally constant variance. The aim of this Ph.D. thesis is to present a formal econometric approach that not only identifies non-stationary variance and suggests appropriate transformations for correction, but also is robust to the specific partitioning of a time series, which is a necessary step for conducting the test, and the possible presence of outliers. The importance of employing this approach in the fields of macroeconomics, actuarial science and finance is extensively examined and supported in Chapters 3, 4 and 5 respectively.
Degree
thesis:*- Grantor dc:publisher
- University of the Aegean
- Year dc:date
- 2023
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Galanopoulos, Nikolaos
- Γαλανόπουλος, Νικόλαος
Subjects
dc:subject × 25- Εφαρμοσμένη ανάλυση χρονοσειρών
- Μετασχηματισμοί χρονοσειρών
- Ακραίες τιμές
- "Γραμμικοποίηση" χρονοσειρών
- Αποτελεσματικότητα της αγοράς υπό την μορφή ασθενούς ισχύος
- Προβλέψεις μακροοικονομικών χρονοσειρών
- Προβλέψεις αναλογιστικών χρονοσειρών
- Ποσοστά θνησιμότητας
- Applied time series analysis
- Time series transformations
- Outliers
- Time series "linearization"
- Weak-form market efficiency hypothesis
- Forecasting of macroeconomic time series
- Actuarial time series forecasts
- Mortality rates
- COVID-19
- Μαθηματικά
- Φυσικές Επιστήμες
- Στατιστική και Πιθανότητες
- Μοντελοποίηση και Προσομοίωση
- Mathematics
- Natural Sciences
- Statistics and Probability
- Modeling and Simulation
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/54988