We consider a set of potentially misspecified structural models, geometrically combine their likelihood functions, and estimate the parameters using composite methods. In a Monte Carlo study, composite estimators dominate likelihood‐based estimators in mean squared error and composite models...