An Analysis of Media 24's “Economist of the Year” Forecasting Competition

Abstract

We rank the forecast accuracy of 21 individuals and 8 institutions which have participated in Media 24's “Economist of the Year” forecasting competition for at least 10 of the past 14 years. Participants predict 12 macroeconomic and financial variables at the start of the year, and update their predictions during the course of the year. Over the 14-year period the “consensus forecast” (i.e. the average of the various forecasts) has outperformed all other participants. There is substantial variation in predictive accuracy among participants. Having won the competition at some point is not an indicator of producing consistently accurate forecasts.

Notes

1 The benchmark bond for the long-run yield was as follows: R153 from 1998 to 2008, R157 from 2010 to 2012, and R186 since 2013. The long-run yield was not forecast in 2009. The inflation rate was defined as the year-on-year change in the following definitions of the CPI: CPIX from 2004 to 2007, and headline inflation in metropolitan and urban areas since 2008.

2 We used a Student's t test of equivalence of two means. The average normalised ranking of each participant in Table 3 is calculated from more than 100 individual year- and variable-specific normalised ranking observations. For participants who have taken part in all 14 years, the number of observations on which the average normalised ranking is based is 156. This is calculated as 14 x 12 minus 12 variable-year combinations for which a variable was not part of the competition. The test statistic for the hypothesis that the average normalised ranking (ANR) of participant i is equal to the ANR of Dennis Dykes (ANR_DD) is .