The tutorials will take place on Friday the 5th of December 2014 at at the central builidng of the Scuola Superiore Sant'Anna, Piazza Martiri della Liberta' 33, Pisa (see location in a map). The registration for the tutorials will take place in the same building. Indications on how to reach the rooms will be placed. The number of participants to the tutorials is limited and restricted only to those who attend the conference. For further information send an email to info@CMStatistics.org.
Title: Band pass filtering and Wavelets analysis: Tools for the analysis of inhomogeneous time series.
Prof. D. Stephen G. Pollock, University of Leicester, UK. Email: Contact
Summary: A wavelets analysis provides a means of analysing non-stationary time series of which the underlying statistical structures are continually evolving. It is an analysis both in the time domain and in the frequency domain.
The tutorial will begin by describing the effects of digital filtering in the time domain and the frequency domain. It will proceed to provide the generalisation of the Shannon sampling theorem that is appropriate to bandpass filtering. This theorem establishes a relationship between continuous signals and their corresponding sampled sequences that is essential to a wavelets analysis. Once this background has been provided, the theories of Dyadic and non-Dyadic wavelets analysis can be described in detail.
Summary: I will review core perspectives of Bayesian modelling, analysis and forecasting of multivariate time series using state-space models. Following a basic review of dynamic linear models, sequential filtering, forecasting and batch analysis in a Bayesian framework, I will touch on several widely-used classes of structured, multivariate dynamic models: this will include time-varying vector autoregressions, matrix-variate models, dynamic factor models, and multivariate volatility models. I will then contact some areas of development over the last decade that respond to the need to define increasingly parsimonious models as time series dimension increases; this will include sparsity modelling using graphical models and other approaches. I will discuss examples from a range of application areas including econometrics, finance, and biomedical studies, and conclude by contacting more recent modelling developments and current research frontiers.
Summary: This tutorial covers the main statistical approaches to modeling networks of dependence (predictive) relationships between multiple blocks of manifest variables through linear composites. These approaches include Partial Least Squares Path Modeling but also its most recent variants and alternatives such as Generalized Structured Component Analysis and Regularized Generalized Canonical Correlation Analysis among others. The presented methods aim at optimizing different criteria related to correlations/covariances between composites, explained variances of endogenous blocks, or redundancies between blocks in coherence with the prediction flow specified by a path diagram.