The Statistics Seminar speaker for Wednesday, April 13, 2016 is Maria Caterina Bramati, a visiting senior lecture of Statistics at Cornell University. She started her research in robust inference applied to panel data and in multivariate time series analysis. More recently, she has carried out some research on Sampling Theory, Bayesian Networks, Extreme Value Analysis. She also enjoys some applied work to Macroeconomics, Development Economics, Firms behaviour and market competition, Environmental, economic and social impact of Climate Change.

* Title*: Optimal Rank-based Tests for Block Exogeneity in VAR Models

* Abstract*: The knowledge of the dependence structure of multivariate time series has been the object of many studies. For example, this is particular relevant in the impulse response analysis which requires orthogonality restrictions on the noise. Moreover, when noise is block spherical, testing independence becomes particularly useful to learn the causal structure of the process.

This talk will present a class of locally asymptotically most stringent tests (in the Le Cam sense) for independence between two sets of variables in the VAR models. These tests are based on multivariate ranks of distances and multivariate signs of the observations that are shown to be asymptotically distribution-free under very mild assumptions on the noise. The class of tests derived is invariant with respect to the group of block affine transformations and asymptotically invariant with respect to the group of continuous monotone marginal radial transformations.

The test procedure presented here seems to be promising in the structural learning of Bayesian networks when the nodes of the Direct Acyclic Graph are the components of a stationary VAR(p) model.

*Refreshments will be served after the seminar in 1181 Comstock Hall.*