Dienstag, 27.04.2021, 12-13 Uhr - Vortrag über zoom
Rouven Michels, M.Sc.
Universität Bielefeld
The Reaction to News in Live Betting
Although live betting markets are comparable to financial markets, these markets have not been investigated empirically, likely due to appropriate data not being publicly accessible. Corresponding research on financial markets meanwhile is extensive and (e.g.) covers how news affect markets. Still, live betting markets in sports are even better suited for such analysis for various reasons. This talk demonstrates how we contribute to filling this gap by analysing high-resolution live betting data provided by a large European bookmaker. The data set comprises second-by-second stakes for all matches of one season of the Bundesliga. We develop a modelling framework within which betting behaviour is driven by an unobservable (latent) market activity level. We additionally incorporate covariates comprising team and match characteristics as well as information on in-game dynamics (goals, live odds) to examine whether there are specific reactions to news in live betting markets. Our results indicate that goals increase the market's activity level underlying the stakes placed on the team that scores and that unexpected match dynamics attract bettors.
The talk will be held in English.
Dienstag, 11.05.2021, 12-13 Uhr - Vortrag über zoom
Dr. Christoph Kiefer
Universität Bielefeld
Treatment Effects on Count Outcomes With Latent Covariates
The effects of a treatment on a count outcome can be assessed using a Poisson or negative binomial regression, where treatment effects are defined as the difference between the expected outcome under treatment and under control. These treatment effects can to date only be estimated if all covariates are manifest (observed) variables. However, some covariates are latent variables that are measured by multiple fallible indicators (e.g., depression). In such cases, it is important to control for measurement error of covariates in order to avoid attenuation bias and to get unbiased treatment effect estimates. In this talk, we present a new approach to compute average and conditional treatment effects in regression models with a logarithmic link function involving multiple latent and/or manifest covariates. Building on a multigroup SEM framework for count variables instead of the generalized linear model, count regression models with multiple latent and/or manifest covariates can be estimated. Then, average and conditional treatment effects are computed using analytical formulas based on moment-generating functions. We provide an illustrative example of our approach and evaluate a cognitive training in elderly people controlling for their (latent) pretest depression and locus of control. The model and effect estimation for the illustrative example are carried out with open-source software packages in R.
The talk will be held in English.
Dienstag, 25.05.2021, 12-13 Uhr - Vortrag über zoom
Prof. Dr. Göran Kauermann
Institut für Statistik der Ludwig-Maximilians-Universität München
Statistics in the COVID-19 pandemics – Challenges, Frustrations and Opportunities
The talk reports about the work of the COVID-19 Data Analysis Group (CODAG) at LMU Munich (https://www.covid19.statistik.uni-muenchen.de/index.html). The group started working on COVID-19 infection data as early as in March 2020, just when the pandemic stated and focussed on various aspects. This included the development of surveillance tools and prediction models. Nowcasting techniques were used to account for reporting delays and imputation techniques were developed to analyse time effects on the onset of the disease rather than on the testing date. We also looked at excess-mortality and the influence of mobility on disease spread. Besides methodological challenges our work was increasingly accompanied by frustration. Data quality, data accessibility and limited impact of our work on policy makers marked a turning point, when we started issuing a biweekly report. This found extended coverage in public media and even brought our colleagues to TV interviews. The talk illustrates some of our research and provides an aerial view on the pandemic from a statistician’s point of view.
The talk will be held in English.
Dienstag, 08.06.2021, 12-13 Uhr - Vortrag über zoom
Prof. Dr. Dietmar Bauer
Universität Bielefeld
Modeling Seasonality in Dynamic Linear Models: Testing For Seasonal Unit Roots
In many time series seasonality is not well represented by deterministic seasonal patterns, as the seasonal pattern slowly changes over time. In such cases an alternative are seasonal unit root models that are based on stochastic trends and cycles. Such processes can be modelled using linear dynamic state space models. These model class is hard to work with as estimation is hampered by many local maxima in the Gaussian likelihood. Alternatively so called subspace methods can be used of which canonical variate analysis (CVA) is a prime example. In this talk we propose statistical tests for the specification of stochastic cycles in a multivariate time series model in the context of CVA. The methods are demonstrated using a Kaggle data set on electricity consumption.
The talk will be held in English.
Dienstag, 22.06.2021, 12-13 Uhr - Vortrag über zoom
Prof. Dr. Kai Nagel
Institut für Land- und Seeverkehr (ILS) der TU Berlin
From mobility modelling to virus dynamics: Attempting to predict the consequences of policy interventions via detailed agent-based modelling
The talk will be held in English.
Dienstag, 06.07.2021, 12-13 Uhr - Vortrag über zoom
Aktuelle Forschungsthemen im ZeSt