Relative efficiency of regression using original data or first differences - case of autocorrelated disturbances

Abstract

In two recent articles, Geary and Tillman have compared the efficiency of using data in absolute form or in the form of first differences in regression analysis of economic time series. Both authors take a highly unfavourable view of the widely-used method of first differences: Geary concludes that this method will usually be highly inefficient and suggests instead the inclusion of a time trend along with the independent variable X in its original form. Tillman extends the discussion to the case of multiple regression, and argues that taking first differences is not an appropriate way of overcoming the difficulties associated with multicollinearity among the independent variables.