A0503
Title: Extreme events detection for volatility prediction
Authors: Andrea Montanino - University of Naples Parthenope (Italy) [presenting]
Giovanni De Luca - University of Naples Parthenope (Italy)
Abstract: The predictive power of extreme events such as financial bubbles and flash crashes on volatility is evaluated in cryptocurrencies and the Big Tech market. To precisely identify periodically collapsing bubbles and flash crashes, the backward supremum augmented Dickey-Fuller (BSADF) test is applied for date-stamping expansion and collapse phases. Volatility is then modeled using the most appropriate GARCH specification, with dummy variables included in the mean equation to capture each event type. The results show that these dummies significantly enhance predictive power not only for the reference asset but also cross-asset; for example, bubbles and crashes in Bitcoin anticipate volatility in other cryptocurrencies. Finally, a Diebold-Mariano test confirms that adding these dummies as external regressors yields a statistically significant forecasting improvement compared to the benchmark model without dummies. Understanding the link between extreme events and volatility is essential for both financial institutions and investors, as it informs risk management and portfolio allocation strategies.
