Abstract: A new method of combining economic forecasts, the odds-matrix
     approach, is tested using Clemen and Winkler's (1986) data on gross
     national product forecasts.  The results show that the combined forecasts
     derived from the odds-matrix method are more accurate than all of the
     methods examined in the Clemen and Winkler study for each of their 8 sets
     of forecasts.  Clemen and Winkler examined the following models: 1. 
     simple average, 2.  normal, 3.  independence, 4.  ordinary least squares,
     5.  Bayesian, and 6.  Bayesian independence.  Using examples of cases with
     negative weights and extreme forecasts, the results also show that the
     method provides combination weights and combined forecasts that are
     intuitively meaningful.  The findings support the extensive simulation
     results reported by Gupta and Wilton (1987), which showed that the
     odds-matrix method is especially useful when the amount of available data
     is sparse.