Clearly, if the weight function is constant, RNE is unity; values of RNE substantially less than unity reflect very unequal weights, and signal possible numerical inaccuracies in estimating the mean of h(y).
In principle, there is an RNE computable for every function h(y) of interest, whereas the Lorenz curve, Gini coefficient and [[omega].sub.m] depend only upon the single set of weights.
The greater the distortion, the more unequal the weights and the lower the RNE, so we use weight-inequality and RNE measures to assess the "lack of fit" of the moment restrictions.
The first panel of Figure 1 displays the time series of RNE for the 1-step ahead predictive density (normalized by subtracting the corresponding futures market forecast).
For the September 1993 forecasts the RNE values are uniformly high and the two histograms almost lie on top of each other, suggesting a close correspondence between the model's predictions and the futures market forecasts in each period.
The first panel of Figure 5 displays the time series of the RNE computed for the Taylor-rule restriction applied to one-quarter-ahead forecasts.