Method of building the forecasting model with dynamic parameters

Abstract

The forecasting problem is rightly considered to be one of the most important and interesting tasks. It is also one of the most difficult ones, since it is associated with such problems as the impact of volatility of input factors on the forecast process, a large number of suitable models and others. In this paper, a method that tries to deal with the first two questions is suggested, as it dynamically selects a set of models based on specified inputs and takes into account the changing nature of factors significance / impact. The approach designed by the authors allows moving from the global to globally-local methods, allowing to receive a more qualitative forecasting model (in terms of minimum error estimation of the forecasting model), applying it to the existing global practices. The described algorithm was tested on a set of samples that are publicly available on the Internet. The proposed method was compared with two other methods of forecasting models building for 11 different sets of data, and it had the lowest average mean square error.