In the article author considers the situation in which several forecasts of the same variable are available. The forecasts was marked on basis of the causal-descriptive models for economic variable having the form of time series with seasonal fluctuations. Author creates new forecast of the same variable – the combined forecast which should be burdened with the smallest error. The author analyses four methods of creating combined forecasts as a weighted average and examines the efficiency of combined forecasts in comparison with individual forecasts. In the majority of the examination cases combined forecasts marked two methods: artificial neural networks and variance-covariance have smaller prediction errors than their component forecasts. It appears that the results of empirical research confirmed the higher efficiency of combined forecast in comparison with individual forecasts.