After obtaining prediction errors
by comparing model estimates with actual returns, we computed and compared mean
error rates, and degree of error dispersion, i.e., standard deviations, for each of the four
ratio form models. If rank transformations of ratios that proxy for risk characteristics of
firms are in fact the best specified of the four ratio forms we examined, mean square error
and standard deviation should be lowest for the model estimated using rank transformations
of ratios.