Publication: Improving Volatility Forecasting Of Garch-Type Models Using Indicator Saturation And Winsorization Approaches
No Thumbnail Available
Date
2025-04
Authors
Mohamed, Suleiman Dahir
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The traditional indicator saturation (is) approach in financial time series data analysis is good in detecting breaks, trend breaks, and outliers simultaneously. To improve the detectability of the is approach, the winsorization (win) method is proposed, which mitigates extreme tail observations. This thesis initially proposes a hybrid approach called the win-is strategy, which addresses the influence of extreme outliers in the tail and identifies breaks, trend breaks, and outliers in cryptocurrencies. The effect of winsorization depends on the chosen percentile and dataset attributes. On the other hand, garch models are frequently employed for volatility prediction. However, standard garch models assume normality and do not account for the underlying data features such as structural breaks, outliers, and trend breaks. Therefore, this study also aims to improve the predictive accuracy of garch-type models for volatility by integrating them with indicator saturation (is) and winsorization (win) approaches. The research uses bitcoin, ethereum, litecoin, tether, and ripple as financial data from november 2014 to june 2023, which exhibit significant and frequent price fluctuations, to evaluate the predictive ability of the developed garch-type models.
Description
Keywords
Financial engineering