When and how much do average forecasts improve predictive accuracy

Abstract

A composite forecast combines two or more individual forecasts into a single estimate by way of a number of different averaging schemes. The easiest way to combine forecasts is through using a simple average. In this paper, the authors show that in many instances the simple average of individual forecasts approximates the optimal combining scheme. Results are expressed in terms of the probability that a composite forecast will improve upon an individual forecast.

Keywords:

• Forecasting
• reducing error variance
• correlated forecast errors
• time series
• composite forecasting models