A1101
Title: Bayesian nonlinear expectation for time series modelling and its application to Bitcoin
Authors: Tak Kuen Siu - Macquarie University (Australia) [presenting]
Abstract: A two-stage approach to parametric nonlinear time series modelling in discrete time is proposed with the objective of incorporating uncertainty in the conditional mean and volatility. A reference time series model is specified and estimated in the first stage. In the second stage, Bayesian nonlinear expectations are introduced to incorporate model uncertainty in prediction by specifying a family of alternative models. The construction of Bayesian nonlinear expectations for prediction is based on closed-form Bayesian credible intervals evaluated using conjugate priors and residuals of the estimated reference model. Using real Bitcoin data, including some periods of Covid 19, the proposed method is applied to forecast Bitcoin returns and evaluate Bitcoin risks under three major parametric nonlinear time series models, namely the self-exciting threshold autoregressive model, the generalized autoregressive conditional heteroscedasticity model, the stochastic volatility model.