Unit:
Κατεύθυνση Στατιστική και Επιχειρησιακή ΈρευναLibrary of the School of Science
Author:
Karabateas Alexandros
Supervisors info:
Μελιγκοτσίδου Λουκία, Επίκουρος Καθηγήτρια, Τμήμα Μαθηματικών, ΕΚΠΑ
Original Title:
Predictive Regressions: Variable Selection and the Complete Subset Approach
Translated title:
Predictive Regressions: Variable Selection and the Complete Subset Approach
Summary:
This dissertation is concerned with the problem of controlling the estimation error in forecasting, having many potential predictor variables. While having to deal with a limited number of independent variables permits any strategy that includes and analyzes all of them, when this number gets higher (or equivalently the data sample is relatively short), it is important to limit the number of parameters or in other ways reduce the effect of the parameter estimation error. Otherwise, analysis can become from time intensive to impossible.
Complete Subset Regression is a simple and powerful method/technique for combining forecasts, first introduced by Elliott et al. (2013). In particular, for a given set of potential predictor variables, forecasts from all possible linear regression models that keep the number of predictors fixed are combined. This method is akin to a complex version of shrinkage which, in general, does not reduce to shrinking the Ordinary Least Squares estimates coefficient by coefficient.
Apart from the apparent savings in terms of computational effort, combinations of subset regressions can produce accurate forecasts compared to other conventional, still very well established, approaches.
Main subject category:
Science
Keywords:
Bayesian Model Averaging, Complete Subset Regression, Variable Selection, Model Selection, Regression Analysis, Bayesian Linear Regression, Predictive Distribution
