The Austrian National bank funds Michael Pfarrhofer’s research project on “Inference with Bayesian nonparametric models in the presence of measurement errors and outliers” at the PLUS Department of Economics.

Data is often imperfect and incomplete when it is first released. A related problem are outlying observations such as those during the recent Covid-19 pandemic, which may severely affect parameter estimates in conventional linear models. On a related note, there may be changes in transmission channels of fiscal or monetary policies, which linear models are not equipped to detect. Particularly for structural and predictive inference, measurement errors and outliers may severely bias parameter estimates and thus yield misleading and suboptimal policy implications. This project aims to develop new Bayesian nonparametric methods that are robust to nonlinear relationships in the data, measurement errors and outliers. In addition, the proposed framework is capable of handling highly nonstandard data that is potentially irregular and sampled at different frequencies than key economic series of interest. The proposed research extends state-of-the-art econometric methods for high-dimensional nonlinear time series analysis (used for both scenario analysis and predictions). A key contribution is a flexible model for measurement errors that captures the full history of vintages for a set of crucial economic indicators.