Averages of Forecasts: Some Empirical Results

By Spyros Makridakis, Robert L. Winkler. 1983

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Predicting Wimbledon 2005 tennis results by mere player name recognition (2007, unknown): “Makridakis and Winkler (1983), as well as Winkler and Makridakis (1983), increased the accuracy of time series predictions by averaging across different forecasting models.”
Combining Rainfall-Runoff Model Outputs for Improving Ensemble Streamflow Prediction (2006, background): “This weighted average may be particularly attractive in this study where only two forecasts are considered to be combined because, as concluded by Makridakis and Winkler 1983, the simple average is generally successful if several methods, probably more than two, are combined.”
Combining forecasts: A review and annotated bibliography (1989, unknown): “In follow-up studies, Makridakis and Winkler (1983) discussed the impact of the number of forecasts included in a simple average, and Winkler and Makridakis (1983) addressed the problem of using 'optimal' weighted averages, using techniques from Newbold and Granger (1974) for estimating the weights.”
Improving forecasting by estimating time series structural components across multiple frequencies (2014, background): “Combining has been considered widely to be beneficial for forecasting (Bates and Granger, 1969; Makridakis and Winkler, 1983; Clemen, 1989).”
Tourism forecasting: To combine or not to combine? (2007, methodology): “Makridakis and Winkler (1983) applied the simple average combination to a number of models and tested the effectiveness of this simple forecasting combination technique.”
An evaluation of neural network ensembles and model selection for time series prediction (2010, background): “Makridakis and Winkler [ 4 ] suggest that using averages of forecasts provides impro ved forecasting accuracy and that the variability of accur acy among different combinations decreases, as the number of methods in the average increases.”
Forecasting with temporal hierarchies (2017, background): “Forecast combinations are widely regarded as beneficial, leading to a reduction of forecast error variance (Bates and Granger, 1969; Makridakis and Winkler, 1983; Clemen, 1989; Timmermann, 2006).”
Performance of Combined Double Seasonal Univariate Time Series Models for Forecasting Water Demand (2009, unknown): “Firstly, we compute the optimal combination of the forecasts using weights by the inverse of the MSE of each of the individual methods (see Makridakis and Winkler, 1983 ), as fol- lows: is the sum of the post-sample forecast mean squared errors of the three methods.”