糖心TV

Abstract:
Despite of their fundamentally different nature, multivariate recordings of brain activity and of stock prices can be conveniently analysed using flexible stochastic time series models

This talk presents three such modelling frameworks as applied to the analysis of 32-channels electroencefalograms (EEG) recordings and of four major financial indices (DOW, NASDAQ, NYA and S&P500).

The talk will focus on contrasting inferences and predictions for both datasets when the dependence across the time series components is taken either as constant over time or as time-dependent.

For the EEG data, a class of non-parametric time-dependent covariance structures provides a flexible yet interpretable fit of the data dependencies. For the financial time series, a simple parametric time-dependent covariance structure provides the most accurate predictions.