Abstract :


The Bank of Israel Research Department's inflation forecast is an important input in the formulation of monetary policy. The one-month ahead forecast is currently based on a simple average of five models which project the change in the Consumer Price Index for the upcoming month. This paper proposes to weight the models' projections differentially, and to assign one of them a negative weight. The new method offers greater precision of the forecast—an improvement of 50 percent in the mean square error in out of sample tests. The results also indicate that including the model which was assigned a negative weight is preferable to removing it from the weighting. The paper presents a theoretical basis which illustrates the inherent benefit of using negative weighting, which derives from utilizing the positive correlation between model errors. Furthermore, professional literature and empirical results which support the theoretical basis are presented.
*  I would like to thank Dr. Edward Offenbacher, Sigal Ribon, Alon Binyamini, Itamar Caspi, Yossi Yakhin, Tanya Suhoy, and Nadav Steinberg for their helpful suggestions.