Even though data revisions are a normal feature of any statistical compilation process, such revisions are seldom taken into account or understood by users of statistics. This results in an over reliance on initially published preliminary estimates that are subject to change. Gross domestic product (GDP) estimates, for instance, are needed for the Monetary Policy Committee to set interest rates and the National Treasury to set budget limits, but because of the need for timely data, policy decisions are often based on preliminary estimates which are later revised as more comprehensive data become available. At the same time, changes to the initially published estimates may lead to adjustment measures being made to the assessment of the performance of the economy.
This brief focuses on revisions to South Africa’s quarterly GDP estimates for the period 1999 to 2013. Based on the study of revisions to the South African quarterly GDP growth rates the following conclusions are made. First, the initially announced estimates are most likely to be revised upward. This is because initial announcements of the quarterly GDP growth rates are on average underestimated. Second, a bias exists in the estimation of the initially announced quarterly GDP growth rates. This suggests that the estimates contain measurement errors that can be eliminated to become better estimates of the final or true value. The brief looks at why data revisions happen and highlights how the study of revisions can be used to evaluate the reliability of initially published estimates. It then provides an analysis of revisions to the South African quarterly GDP growth rates followed by recommendations.