Econometric forecasting at mixed frequencies: Application to Mexico's economy. Part II: Combined use of high frequency and low frequency models

Abstract

In a two-part paper, we apply to the Mexican economy the methodology of forecasting at mixed frequencies developed by Klein and Kushnirsky [2]. We construct two macroeconometric models for Mexico – high frequency, with monthly observations, and low frequency, with annual observations. Our methodology demonstrates how the high frequency model can be used for periodic adjustments of selected key indicators in the low frequency model. The methodology includes a theoretical foundation, methods for the adjustment of model structures and properties, procedures for achieving mutually consistent solutions, and step-by-step applications. The computational procedures are based on: (a) employment of a loss function in order to force the solution of the low frequency model to come close to the solution of the high frequency model and(b) construction of "exact" joint solutions, with values of key variables in the low frequency model imported from the high frequency model. We illustrate the applications by generating: (a) independent solutions of the high frequency and low frequency models; (b) solutions of the low frequency model with the use of a loss function; and (c) solutions of the low frequency model with key variables such as GDP and the price index imported from the high frequency model.