Indicadores líderes, redes neuronales y predicción de corto plazo

Abstract

This paper shows a procedure to constmct a short run predictor for the GDP. We use theBaxter & King filter to decompose the monthly GDP on its three components: seasonal, business cycle and iong-run trend. Furthermore we estimate and forecast the businesscycle using a set of leading economic variables. We propose that the complicated relationshipsamong this variables and the business cycle are well captured by a non linearartificial neural network model. The other components are estimated using standardeconometric techniques. Finally, the three components are added to obtain an indicatorfor the future behavior of the GDP. The prediction shows an aceptable leve1 of reliability,so the index can be used to take decisions in the private or public sector. The mainadvantage of the index is its faster availability relative to the oficial statistics.