Events

Please Note: This seminar will be given online.

Please Note: This seminar will be given online.

Personalized Statistical Learning and Network Analysis with Sparsity for Actuarial Use

Please Note: This seminar will be given online.

Novel contributions to statistical methods for meta-analyses of diagnostic or screening test accuracy studies

Please Note: This seminar will be held online.

Mapping My Dad's Peaks: An Introduction to Web Scraping and Regex

Please Note: This seminar will be given in person.

Informative censoring

Please Note: This seminar will be given online.

Whatever Happened to Design-Based Inference?

What exactly should we think about appropriate analyses for designed experiments and why? If conditional inference trumps marginal inference, why should we care about randomisation? Isn’t everything just modelling? The Rothamsted School held that design matters. Taking an example of applying John Nelder’s general balance approach to a notorious problem, Lord’s paradox, I shall show that there may be some lessons for two fashionable topics: causal analysis and big data. I shall conclude that if we want not only to make good estimates but estimate how good our estimates are, design does matter.

Please Note: This seminar will be given in-person.

Recalcitrant Betas: Intraday Cross-Sectional Distributions of Systematic Risk

High-frequency financial data allows for efficient estimation of assets’ exposures to systematic risk, provided these exposures do not vary significantly at high frequencies.  We develop a test for deciding whether this is the case. The test is constructed for a panel of high-frequency asset returns, with the size of the cross-section and the sampling frequency increasing simultaneously. It is based on a comparison of the empirical characteristic functions of estimates of the assets' factor loadings at different parts of the trading day, formed from local blocks of asset returns and the corresponding factor realizations. The limiting behavior of the test statistic is governed by unobservable latent factors in the asset prices. Empirical implementation of the test to stocks in the S&P 500 index and the five Fama-French factors, as well as the momentum factor, reveals different intraday behavior of the factor loadings: assets' exposure to size, market and value risks vary systematically over the trading day while the three remaining factors do not exhibit statistically significant intraday variation. Moreover, we find diverse, and for some factors large, reactions in the assets' factor loadings to major economic or firm specific news releases. Finally, we document that time-varying correlations between the observable risk factors drive a wedge between the time-of-day pattern of market betas, estimated with and without control for the other observable risk factors.

Please Note: This seminar will be given in-person.

Max-linear Graphical Models for Extreme Risk Modelling

Graphical models can represent multivariate distributions in an intuitive way and, hence, facilitate statistical analysis of high-dimensional data. Such models are usually modular so that high-dimensional distributions can be described and handled by careful combination of lower dimensional factors. Furthermore, graphs are natural data structures for algorithmic treatment. Moreover, graphical models can allow for causal interpretation, often provided through a recursive system on a directed acyclic graph (DAG) and the max-linear Bayesian network we introduced in [1] is a specific example. This talk contributes to the recently emerged topic of graphical models for extremes, in particular to max-linear Bayesian networks, which are max-linear graphical models on DAGs. 

In this context, the Latent River Problem has emerged as a flagship problem for causal discovery in extreme value statistics. In [2] we provide a simple and efficient algorithm QTree to solve the Latent River Problem. QTree returns a directed graph and achieves almost perfect recovery on the Upper Danube, the existing benchmark dataset, as well as on new data from the Lower Colorado River in Texas. It can handle missing data, and has an automated parameter tuning procedure. In our paper, we also show that, under a max-linear Bayesian network model for extreme values with propagating noise, the QTree algorithm returns asymptotically a.s. the correct tree. Here we use the fact that the non-noisy model has a left-sided atom for every bivariate marginal distribution, when there is a directed edge between the the nodes.

For linear graphical models, algorithms are often based on Markov properties and conditional independence properties. In [3] we characterise conditional independence properties of max-linear Bayesian networks and in my talk I will present some of these results and exemplify the difference to linear networks. 

Please Note: This seminar will be given in person.

Optimal Subgroup Identification

Please Note: This seminar will be given virtually.

Stable Variable Selection with Knockoffs