Humboldt-Universität zu Berlin >> Wirtschaftswissenschaftliche Fakultät


Subproject B1

Dynamic Semiparametric Modelling
Head of Project : Prof. Dr. Wolfgang Härdle

+49 (0)30 2093-5631


+49 (0)30 2093-5649



C.A.S.E. – Centre for Applied Statistics & Economics
Institute for Statistics und Econometrics
Wirtschaftswissenschaftliche Fakultät
Humboldt-Universität zu Berlin
Unter den Linden 6
10099 Berlin, Germany



NameTitle Email
Burdejova, PetraM.Sc.
Klinke, SigbertDr.
Lu, Meng-JouM.Sc.
Melzer, AwdeschM.Sc.
Petukhina, AllaM.Sc.
Quian, Ya
Thu, Hien Pham
Trimborn, Simon M.Sc.
Xu, XiuM.A.



In the modelling of many economic questions one meets three challenges: first of all an accurate functional relationship of the measurable parameters is often unknown, secondly also an empirical established relations are affected by temporal changes, thirdly the functional relations generate extremely high dimensional data structures. In the modelling the risk factors of financial markets all three challenges arise at the same time. It is like that for example in case of (infinitely dimensional) state price density (SPD), (and connected with it stochastic discount factor (SDF); pricing kernel), which can be implied by observed option prices. It is not sure if one can determine the functional form and its dynamic behavior for liquid markets is not known. The same applies for the dynamics of the implied volatilities (IV) surface, which can be clearly derived by the Black-Scholes formula from the option price structure.The knowledge of the dynamics of these measurements allows to form stochastic models for volatility and makes possible to study the risk aversion of an investor with connection to macro-economic data.

In this project we examine the approach of dynamic semiparametric factor models (DSFM). DSFMs are almost universally applicable tools of the modern statistics to understand high-dimensional, complex and dynamic phenomena, to analyze and to forecast. DSFMe can be numerically implemented in low dimensions, since their nonparametric factor structure are controlled by loading time series. The main aim of this research project is the development and implementation of the statistical theory of this class of models and their application to the dynamics of IV surfaces and SPDs as well as to panel data. High-iterative numerically intensive procedures e.g. maximum Scoring will be used and relations with alternative procedures e.g. Support Vector Machines examined. The empirical calibration are done from the data sources (e.g. high frequency option price time series, macro time series), which are provided in connection with the FEDC (TP D).

Home | Projekte | RDC

Contact CRC649 | Impressum