Modeling Dependence Structure via Spatial Parameter and Variable Selection

• Activity Number: 551
• Type: Invited
• Date/Time: Wednesday, August 3, 2011 : 2:00 PM to 3:50 PM
• Sponsor: IMS
• Author(s): Chunrong Ai *+ and Meixia Meng
• Companies: University of Florida and Shanghai University of Finance and Economics
• Keywords: Dependence structure ; Variable Selection ; Spatial Parameter ; Asymptotic Properties

Abstract

Modeling financial data dependence properly is critical for risk management. However the existence of large number of financial instruments makes such endeavor difficult. There are three ways to overcome this difficulty. First is to specify a factor model. Second is to specify a Copulas model. Third is not to restrict dependence coefficients and let data decide via a variable selection technique. None of these approaches exploit researcher's prior knowledge about the dependence structure. For example, we all know that stock prices move more closely within sectors than between sectors. This kind of prior knowledge is not incorporated by the previous approaches. Spatial modeling on the other hand permits such knowledge and moreover incorporates other factors such as macroeconomic factors into dependence structure. In this paper, we shall model dependence structure with spatial parameter. But unlike the standard spatial models, we shall not restrict the number of spatial coefficients and instead allow data to determine via a variable selection technique. We shall derive the asymptotic properties of the proposed estimation and illustrate the procedure with an example.