Methodology for Forecasting Election Results Using Combination of Analytical and Trend Modeling Methods

Abstract

The study is devoted to forecasting political party ratings using short time series based on the VCIOM electoral polls data. The authors examine three groups of methods: trend modeling, expert assessments, and analytical modeling.

A review of the theoretical background and the results of empirical experiments shows low accuracy of forecasts produced solely by the trend modeling methods. To improve their accuracy, the authors propose a method of adjusting forecasts by logarithmic approximation. This method bases on hierarchical clustering that uses a vector of the coefficients of the approximated equations and describes the most similar electoral situation in the past. Then the reminder calculated as the difference between the sum of the predicted values and 100% is proportionally redistributed among the participants of the election campaign. To forecast political party ratings where the time series values do not exceed 5%, it is preferable to use the average growth rate method.