Optimal weights depend on the conditional variance of the vector martingale increment and its partial derivative with respect to parameters φ and can be estimated using two step approach. It can be shown that the MEF method produces strictly more efficient estimators than GMM (except in the case when they coincide). MEF approach can be further extended to account for mixed frequency data and latent variables. Data at lower frequency (e.g. output observed quarterly instead of monthly) can be treated as if some observations are missing. The authors suggest to replace the missing values by conditional predictions under the model at hand given the actual observations. Latent variables (e.g. daily rental rate of capital) can be simulated using the process implied by the model. The simulated paths can later be used to compute the conditional expectation needed for the MEF approach.