Rank-based statistics can be improved by considering both ranks and signs, leading to more efficient results in statistical analysis. By incorporating signs along with ranks, a new class of statistics is developed that enhances the efficiency of semiparametric inference in regression and time series models. The signs contribute separately to the variance, potentially increasing efficiency. This approach allows for better estimation of parameters like the median without assuming symmetry, leading to more accurate statistical analysis.