Modeling business turnover time series : a generalized least squares approach

Abstract

This dissertation studies the adequacy of generalized least squares estimation in a time series context. Performing regression analysis using time series variables is most likely to violate classical linear regression assumptions. The behaviour of the ordinary least squares estimator when these assumptions are not satisfied is reviewed, while generalized least squares estimation is employed as an alternative estimation method that can be used when the variables being studied are time series processes. The generalized least squares estimator is derived under different error covariance structures, and its properties are examined. We apply generalized least squares estimation to estimate the parameters of a linear model explaining the relationship of the turnover of a local betting company with a number of key performance indicators.