Un Test Conjunto de Superioridad Predictiva para los Pronósticos de Inflación Chilena

Optimality under quadratic loss implies that forecasts built using a large information set should perform al least as well as forecasts built using a more restricted and nested information set. In this article we use a joint test of superior predictive ability to test this optimality condition for the term structure of several Chilean inflation forecasts coming from the following sources: Bloomberg, Consensus Economics, the Survey of Professional Forecasters and an average of selected seasonal univariate models. We do this by taking advantage of the fact that these sets of forecasts are built at different moments in time and, more importantly, using different and nested information sets. Our results indicate that the null hypothesis of optimality under quadratic loss cannot be rejected when Mean Squared Error is used to evaluate the term structure of the forecasts. Nevertheless, when the joint test is carried out to evaluate the term structure of the Mean Squared Forecasts, as suggested by Patton and Timmermann (2010), the joint test rejects the null hypothesis of optimality. Further analysis of our results reveals that this rejection is associated with a violation of an orthogonality condition that should be satisfied when forecasts are optimal. Moreover, this violation seems to stand both across different sources of forecasts and across different forecasting horizons. This suggests that there is room for improvement in the term structure of Chilean inflation forecasts.