Multi-Objective Frequentistic Model Averaging with an Application to Economic Growth

In the Frequentistic Model Averaging framework and within a linear model background, we consider averaging methodologies that extend the analysis of both the generalized Jacknife Model Averaging (JMA) and the Mallows Model Averaging (MMA) criteria in a multi-objective setting. We consider an esti­mator arising from a stochastic dominance perspective. We also consider aver­aging estimators that emerge from the minimization of several scalarizations of the vector criterion consisting of both the MMA and the JMA criteria as well as an estimator that can be represented as a Nash bargaining solution between the competing scalar criteria. We derive the limit theory of the esti­mators under both a correct specification and a global misspecification frame­work. Characterizations of the averaging estimators introduced in the context of conservative optimization are also provided. Monte Carlo experiments sug­gest that the averaging estimators proposed here occasionally provide with bias and/or MSE/MAE reductions. An empirical application using data from growth theory suggests that our model averaging methods assign relatively higher weights towards the traditional Solow type growth variables, yet they do not seem to exclude regressors that underpin the importance of factors like geography or institutions.