Abstract:
|
Multiple time series are commonly used in econometrics. These methods present serious difficulties in high dimensions as the number of parameters will tend to overwhelm any finite data set. We propose a way of using statistical learning methods to model high dimensional time series. The methods used are the LASSO, projection pursuit regression, and copulas. These provide a way to linking nonparametric univariate estimates into multivariate models using information from energy statistics.
|