PhD Candidate: Bernhard van der Sluis
Start: Fall 2023
Modern datasets often contain data about many individuals over time. This creates a tradeoff for inference: do we use all the data to construct a one-size-fits-all model, or do we separate the data to obtain a more targeted policy? In other words, how do we deal with unknown degrees of heterogeneity while still exploiting the richness of modern datasets? In my research, I partly address this question by developing new inference methods and specification tests for both new and existing models.
In particular, I focus on modern econometric techniques including nonparametric statistics and Markov switching models to uncover how relationships between variables evolve over time and test whether the change over time is the same across individuals. To quantify the uncertainty in the estimates, I develop bootstrap and simulation algorithms to compute standard errors. Furthermore, clustering is a powerful tool to identify groups of individuals that exhibit similar patterns, enabling more targeted analysis than aggregate models allow. Going one step further, for datasets with a multi-dimensional structure where individuals are indexed along more than two dimensions, I develop tensor data models to capture complex dependencies across individuals.
The methods developed in this project are motivated by, and applied to, empirical questions where heterogeneity is central. In macroeconomics, I use them to study differences in business cycle dynamics across regions and the heterogeneous spillover effects of U.S. monetary policy. In microeconomics, I use these methods to uncover variation in housing price dynamics in the Netherlands. Beyond economics, I apply the methods to study climate trends, herding behavior in financial markets, and the structure of implied volatility surfaces in financial options markets. Several of the tools are available as open-source software and have been adopted by researchers in both academia and industry.
