A0791
Title: Sentiment regimes for predicting tail-risk: Tree-structured conditional autoregressive value-at-risk
Authors: Christoph Hirt - University of St Gallen (Switzerland) [presenting]
Abstract: The empirical importance of sentiment and attention measures is evaluated for conditional quantile forecasting of U.S. equity return processes. To this end, Tree-CAViaR is proposed, a data-driven approach to select among threshold nonlinear CAViaR specifications. Similar to tree-structured models within the GARCH and HAR families, Tree-CAViaR offers a fully data-driven approach to identify structural changes in the quantile process by selecting threshold-regime CAViaR models through binary splitting. Results show that Tree-CAViaR effectively detects distinct tail-risk regimes, defined by exogenous predictors such as the CBOE volatility index (VIX), leading to improved out-of-sample forecasting performance.
