University of Minnesota
Bootstrap-Based Prediction Inference for Linear Autoregressive Models with GARCH-Type Heavy-Tailed Errors
Abstract
dc:description.abstractIn many real-world applications, time series data rarely follow simple linear patterns or standard distributional assumptions. Instead, they often exhibit features such as persistence, volatility clustering, nonlinearity, and heavy-tailed behavior. In financial markets, for example, stock returns and exchange rates frequently experience extreme movements, sudden crashes, and sharp rallies that occur far more often than would be predicted under a normal distribution. Such heavy-tailed innovations can severely affect predictive performance, as conventional forecasting methods tend to underestimate the probability and impact of rare but significant events. We propose a generalized AR forward bootstrap method for (h)-step predictions and prediction intervals that can handle stationary and nonstationary AR processes, as well as AR processes with GARCH type-heavy-tailed errors having finite or infinite variance.
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Incoom-Koomson, Kojo
Subjects
dc:subject × 6Rights
- Language dc:language.iso
- en
Identifiers
dc:identifier.*- Dc Identifier Citation
- Incoom-Koomson, Kojo. (2026). Bootstrap-Based Prediction Inference for Linear Autoregressive Models with GARCH-Type Heavy-Tailed Errors. Retrieved from the University Digital Conservancy, https://hdl.handle.net/11299/283010.