A joint model for predicting total price variation
PhD Students Mia Holmfeldt and Marcus Larson
There are two important issues, which should be considered when analyzing jumps in returns, the jump occurrence and corresponding jump size. Empirical results document new dynamic features, such as clustering and time dependence, both in the jump occurrence as well as jump size compared to earlier studies using parametric models. This recognizable pattern in jump dynamic can be used to improve forecasts of the future price movements.
We rely on a two-stage estimation procedure where the initial step is to apply the non-parametric approach to detect the realized jumps along with its size and then simultaneously model the time between jumps, the jump size and the continuous price variation using a multivariate semiparametric GARCH (Generalized Autoregressive Conditional Heteroskedasticity) model. Given the model's ability to predict the future return variation its application within risk management should be promising. We therefore evaluate the model within a value-at-risk (VaR) framework.