讲座编号：jz-yjsb-2016-y002
讲座问题：Maximum likelihood estimation of the State Space Cointegration Model: the optimization problem and the role of parametrization
主 讲 人：Bernard Hanzon 教授 爱尔兰考克大学
讲座时间：2016年03月03日（周四）下昼14:30
讲座所在：阜成路东校区一号楼241室
加入工具：理学院统计专业的研究生及青年西席
主理单位：理学院
Bernard Hanzon ，爱尔兰考克大学（University College Cork) ,数学学院教授 ，数学主席 ，数学系主任。
曾划分在阿姆斯特丹自由大学（VrijeUniversiteit Amsterdam ，荷兰）和代尔夫特手艺大学（Technical University Delft ，荷兰）任副教授和讲师。Hanzon教授于1986年获得鹿特丹伊拉斯姆斯大学（Erasmus University Rotterdam ，荷兰）博士学位。他的研究兴趣包括金融数学 ，金融计量经济学 ，数学系统和控制理论 ，代数优化等。
主讲内容：
In economic and financial time series analysis, the phenomenon of cointegration is well-known. The concept was introduced by Granger and Engle (for which they received the Nobel prize). The concept concerns the effect that in a non-stationary economic or financial time series, one may find linear combinations of outputs which are stationary. This can be interpreted as the effect of an equilibrium relation between the outputs involved. In practice the equilibrium relation does not hold exactly, but rather there is a drive towards the equilibrium relation over time, in case the observed variable values deviate from the equilibrium relation. A well-known form of the model that stresses this interpretation is the so-called error-correction formulation of the model. In the literature up till now the VAR (vector-autoregressive) model has been dominant in the cointegration literature. In two papers, available on SSRN, the Th.Ribarits and the speaker have generalized the prevailing VAR approach to state-space models with, by introducing the state-space error correction model for cointegration analysis, and by describing a methodology for finding the Maximum Likelihood Estimator for this model. In this talk we will explain how the state-space error correction model can be derived and how one can approach the problem of finding the corresponding maximum like lihoodestimator by using a special suitable parametrization of the model involved. The parametrizations emphasized in the talk are different from those used in the SSRN papers. The new approach uses a more flexible class of linear dynamical models to analyse cointegration.
We argue that especially for models of high-frequency data the new approach can be preferable over the classical VAR approach,as it allows for larger delays without having to use very large order models. If time permits some remarks will be made about the possible application to global models of financial liquidity.
The talk is based on joint work with Thomas Ribarits (EIB, Luxembourg) and joint work with Ralf Peeters (Univ Maastricht, Netherlands) and Martine Olivi (INRIA, Sophia Antipolis, France).