11.5.2016

Pozývame vás na seminár Oddelenia teoretických metód Ústavu merania SAV, ktorý sa bude konať vo štvrtok, 26. 5. 2016 o 10.00 hod. v zasadacej miestnosti ÚM SAV v Bratislave.

Výsledky svojho výskumu predstaví RNDr. Matúš Maciak, Ph.D., vedecký pracovník Katedry  pravděpodobnosti a matematické statistiky, Matematicko-fyzikální fakulta, Univerzita Karlova, Praha.

Abstract

Estimating of various types of structural changes in some unknown dependence structure is an important task in regression modelling approaches especially when dealing with more complex data where sudden changes can be naturally expected. Our estimation method is motivated by some machine learning ideas and using recent developments in statistic, post-selection inference especially, we propose a fully data driven estimation approach where the unknown model and possible change-points are estimated all at once. The estimation approach apriori considers all possible model alternatives which makes the method suitable in situations where no knowledge on the number or position of change-points is given in advance. The estimation approach allows for different structures of change-points to be considered and, as an easy and straightforward extension, I can also incorporate additional shape restrictions which might be useful in some practical scenarios. We also discuss some testing approaches to test such shape restrictions in the regularized model (e.g. testing monotonicity, convexity, etc.).