skip to main contentskip to main menuskip to footer Universität Bielefeld Play Search

Forschung

© Universität Bielefeld

Research

© Universität Bielefeld

Research in the research group econometrics is partly funded by the DFG in the corresponding research projects or organized in dissertation projects.

Currently our research agenda includes topics in three areas:

1. Time series analysis   

Time series denotes observations of economical process over time. Many times these observations are taken in equidistant time steps like days, weeks, quarters, months or years.

Such data sets often are modelled using linear dynamic models such as autoregressions, which explain future observations as weighted averages of current and past observations.

The next level of generality are ARMA models (autoregressive moving average) or the in a certain sense equivalent linear state space models. These are used in the classical Box-Jenkins framework.

In order to estimate this class of the models classical methods use the pseudo maximum likelihood optimization. Recently heuristic approaches have been proposed in the form of so called subspace procedures. The properties of the corresponding estimators are well researched in the stationary case, but in particular for unit root (random walk like) processes a number of questions exist.

2. Discrete Choice Models

In many areas ranging from economics to transportation science deciders must choose between a finite number of alternatives. Models representing such choice situations are called discrete choice models.

The predominant approach introduces a latent utility function as the main rationale for taking the choice. Hereby deciders alocate a utility depending on the characteristics of the alternatives, the preferences of the deciders and additional factors that the modeller does not know or does not want to model. These additional factors are included as random utility contributions, providing the name "random utility models (RUMs)". Depending on the specification of the random terms this leads to multinomial logit (MNL) or multinomial probit (MNP) models.

Recently the focus of much research is on modelling unobserved heterogeneity in preferences including by a random mixture of the parameters. Our group examines approaches for mixed MNP models for which the group of Chandra Bhat suggested estimation using the maximum approximate composite maximum likelihood (MacML) paradigm. The properties of this approach are largely unknown.

3. Applications of Discrete Choice Models

In addition to developing estimation and inference methods we also investigate the application of these models in the area of transportation science. In particular the modelling of human mobility is at the forefront of our interests.

Zum Seitenanfang