System approach to construction of regression models on time series
Abstract
A modified approach to mathematical models building of various processes using time series is proposed. The notion of model structure that allows to properly select a specific structure in the process of model construction is formulated. It is proposed to split the process of model construction in the following five steps: process analysis, detection of non-linearity, selection of model structure for possible candidates, selection of parameter estimation methods and estimation of the candidates, and selection of the best fit model using a set of statistics. The approach proposed has been tested successfully on a set of examples of econometric model building.Downloads
Issue
Section
Mathematical methods, models, problems and technologies for complex systems research