Conferencistas Internacionais

Regression models under time-varying coefficients and locally stationary variables. Applications to Finance

Eva Ferreira Garcia (Universidad de Bilbao, España)

Motivation:
There are many situations where the parameters in a regression model cannot be assumed to be constant and they should be considered as time-varying. In such situations, we can assume a parametric structure for the coefficients depending on time or use a nonparametric approach. In this course, we analyze the problem in the latter case, where no parametric distribution is assumed for the coefficients.

Estimation:
To estimate the coefficients, kernel estimators will be use in order to solve the normal equations to minimize the mean average squared error in the regression model. Moreover, the covariates need not be stationary, which is a usual, but not realistic, assumption in many practical situations.

Different regression models can be analyzed as: shape restrictions, seasonal constraints and unknown covariates, that need to be previously estimated.

Real application:
As a real example, an application to Finance will be presented where a nonparametric procedure is developed to estimate and test conditional beta pricing models which allows for flexibility in the dynamics of covariances and market prices of risk. This case is very interesting since the theory behind imposes no restriction for the shape of the risk premia and, therefore, the nonparametric estimation procedure fits here very naturally.