Predicting a regression model using elements of the theory of similarity

Authors: Noskov S. I., Vergasov A. S.

Annotation: The article discusses one of the main directions of practical application of mathematical models of regression type, associated with the prediction of future values of dependent variables. To improve the accuracy of these values, it is proposed, when estimating unknown parameters of the regression model, instead of the usual least squares method (LSM), to use the weighted least squares method (HMSC). At the same time, when calculating the weights of observations of the base period of the forecast, the theory of similarity developed by Professor Yu.A. Voronin is used, according to which the more similar the vector of values of independent variables of the lead-time observation is with the corresponding vector of observation of the base period, the greater the weight the latter must possess. This consideration is the basis of the weighting algorithm proposed in the paper. Considered in detail a numerical example.

Keywords: regression model, least squares method, weighted least squares method, theory of similarity, forecasting