Humboldt-Universität zu Berlin - Wirtschaftswissen\=schaftliche Fakultät

• Nonparametric Estimation of Nonstationary Time Series
  
• Dimension Reduction Techniques
• Semiparametric Methods
• Generated Regressors

Financial Econometrics

• Multivariate Risk, Systemic Risk
• Quantification of Systemic Risk
  

In this project we develop new econometric procedures for determining general cointegration type relationships of flexible functional form between stochastically nonstationary variables. Such methods allow in particular to shed new light on systematic open questions in two fundamentally different research fields of key Economic interest: Euler Equations and Foreign Exchange Markets. For our methods the form of the structural relation does not need to be pre-specified as linear or as of a certain parametric type but is non- or semiparametrically estimated from the data. Furthermore, the required framework of recurrent processes allows for very general data generating processes containing stochastically nonstationary processes such as unit and long memory processes, and all types of stationary processes. In particular the developed methods do not differ in form for stationary and nonstationary quantities, thus pretesting is not needed, avoiding the risk of misspecification. As nonparametric estimation in such a general setting requires large sample sizes especially when including several regressors, semiparametric methods are still flexible but improve on feasibility particularly for nonstationary data. Moreover, we aim at developing statistical testing procedures to assess the validity of existing parametric models.

The new methods allow for a new and general approach to nonparametric estimation of the Euler equations associated with dynamic utility maximization. We will obtain new insights on individual risk behaviour from a new look on consumption data using these techniques. In particular, the form of the marginal utility function can be estimated from the data, which was econometrically not possible before, but is of central interest for economic theory. We expect that the general model improves the practical performance of intertemporal optimization models providing a new understanding of some of the present puzzles.

With the availability of large high quality data sets and high frequency data on Foreign Exchange Markets, the new nonparametric techniques are ideal to provide new useful evidence on the true underlying structural model between different exchange rates and exchange rates and (macro)economic fundamentals. Besides that high frequency data might contain information not extracted so far, it is their sample size which permits to use fully nonparametric methods and therefore makes them especially appealing. Obtained results will improve our understanding of currency risk.