- Title
- Score driven volatility models based on skewed-t distributions
- Creator
- Janse van Rensburg, Stéfan
- Subject
- Gqeberha (South Africa)
- Subject
- Eastern Cape (South Africa)
- Subject
- GARCH Model
- Date Issued
- 2021-04
- Date
- 2021-04
- Type
- Doctoral theses
- Type
- text
- Identifier
- http://hdl.handle.net/10948/52010
- Identifier
- vital:43425
- Description
- Score driven (SD) conditional volatility models allow for rich volatility dynamics and realistic distributional assumptions. These models link the evolution of time-varying volatility to the shape of the predictive density, which often results in models that are robust against outliers. These models also allow for conditional skewness. The combination of skewness and robustness may improve in-sample fit, conditional volatility forecasts and tail-risk forecasts. Therefore, this study proposes novel SD conditional volatility models with Skewed-t distributed innovations. For some of these models, the formof skewness partially negates robustness against outliers. This result demonstrates that the formof skewness requires careful consideration in the specification of SD conditional volatility models. The study also expands upon earlier observations that the assumption of negative conditional skewness may induce a form of volatility asymmetry in SD conditional volatility models that is incompatible with the leverage effect typically observed in equity markets. A simulation experiment shows that neglected leverage, therefore, biases maximum likelihood estimates. Additionally, the study considers SD conditional volatility models that account for leverage effects and have Skewed-t distributed innovations. An empirical application to the daily Johannesburg Stock Exchange (JSE) / Financial Times Stock Exchange (FTSE) All Share Index returns demonstrates the utility of SD conditional volatility models with Skewed-t distributed innovations. Extensions of these models that account for leverage prove competitive with more traditional conditional volatility models in terms of in-sample fit and tail-risk forecasts. These results suggest that the models considered in this study are useful within the context of financial risk management.
- Description
- Thesis (PhD) -- Faculty of Science, School of Computer Science, Mathematics, Physics and Statistics, 2021
- Format
- computer
- Format
- online resource
- Format
- application/pdf
- Format
- 1 online resource (xxvi, 323 pages)
- Format
- Publisher
- Nelson Mandela University
- Publisher
- Faculty of Science
- Language
- English
- Rights
- Nelson Mandela University
- Rights
- All Rights Reserved
- Rights
- Open Access
- Hits: 966
- Visitors: 936
- Downloads: 0
| Thumbnail | File | Description | Size | Format |
|---|