CMStatistics 2018: Start Registration
View Submission - CFE
A0664
Title: Predicting bubble collapse using non-causal models Authors:  Elisa Voisin - Maastricht University (Netherlands) [presenting]
Abstract: Locally explosive episodes have long been observed in financial and economic time series and such dynamic features are well captured by mixed causal-non-causal autoregressive models. They incorporate both lags and leads of the variable of interest and are characterised by heavy-tailed distributions. The aim is to evaluate various methods' accuracy of predicting the probabilities of a crash during a bubble. Overall, three distributions are considered, a Cauchy and two Student-$t$, one with finite and the other one with infinite variance. Furthermore, numerical and, when possible, statistical approaches are employed. Conditional predictive densities do not always admit a closed-form expression; while for Cauchy, results can be obtained directly, for Student-$t$ ones, approximations need to be made. To do so, sample- and simulations-based methods are used. The empirical analysis only considers MAR(r,1) processes, and uses the MAR(0,1) as a benchmark.