Estimation of pair linear regression models with parameters in the form of linear operator matrices of two-dimensional vector space

The key problem in constructing a regression model is the choice of its structural specification, i.e. the composition of the variables and the mathematical form of the relationship between them. All currently known regression specifications are based on the fact that their unknown parameters are matrices of linear operators of a one-dimensional vector space. In this paper, for the first time, linear regression models with parameters in the form of matrices of linear operators of a two -dimensional vector space are considered. It is shown that such models can be used to predict the values of the explained variable, and for this, the researcher does not need to set the predicted values of the explanatory variable, since they are sequentially determined by the model. To estimate the proposed models, an optimization problem is formulated based on the least-squares method with restrictions. Using the method of Lagrange multipliers, it is proved that solving the formulated problem reduces to solving linear algebraic equations system. An example of estimating the proposed models for specific data is considered. As a result, the error sum of squares by the developed model turned out to be five times less than the error sum of squares by the classical pair linear regression model.

1. Kuhn M., Johnson K. Applied predictive modeling. Springer. 2018.

2. Harrell Jr., Frank E. Regression modeling strategies: with applications to linear models, logistic and ordinal regression, and survival analysis. Springer Series in Statistics. 2015.

3. Kleinbaum D.G., Kupper L.L., Nizam A., Rosenberg E.S. Applied regression analysis and other multivariable methods. Cengage Learning. 2013.

4. Noskov S.I. Tekhnologiya modelirovaniya ob"ektov s nestabil'nym funktsionirovaniem i neopredelennost'yu v dannykh. Irkutsk: RIC GP «Oblinformpechat'» Publ. 1996.

5. Kleyner G.B. Proizvodstvennye funktsii: Teoriya, metody, primenenie. Moscow: Finance and Statistics Publ. 1986.

6. Noskov S.I., Bazilevskiy M.P. Postroenie regressionnykh modeley s ispol'zovaniem apparata lineyno-bulevogo programmirovaniya. Irkutsk: IrGUPS. 2018.

7. Bazilevskiy M.P. MNK-otsenivanie parametrov spetsifitsirovannykh na osnove funktsiy Leont'eva dvukhfaktornykh modeley regressii. Yuzhno-Sibirskiy nauchnyy vestnik. 2019;26(2): 66-70.

8. Bazilevskiy M.P. Sintez modeli parnoy lineynoy regressii i prosteyshey EIV-modeli. Modelirovanie, optimizatsiya i informatsionnye tekhnologii. 2019;7(1): 170-182.

9. Bazilevskiy M.P. Issledovanie dvukhfaktornoy modeli polnosvyaznoy lineynoy regressii. Modelirovanie, optimizatsiya i informatsionnye tekhnologii. 2019;7(2): 80-96.

10. Akulich I.L. Matematicheskoe programmirovanie v primerakh i zadachakh. Moscow: High School Publ. 1986.