تقدير مَعْلَمَات انموذج انحدار kink بوجود مشكلة الأبعاد العالية مع تطبيق عملي

Abstract

Kink regression is one of the event topics that research has focused on recently because of its importance, which emerges from its association with the subject of continuity of regression, but there is a scarcity of them in Iraq, as this regression is characterized by the division of its parameters into two parameters, one of which is before and after the kink cut-off point, making the number of explanatory variables double according to the presence of the kink cut-off point in one or more explanatory variables, and this double number may cause the number of explanatory variables in the kink regression model to be larger of the sample size (n<p) or that the number of explanatory variables in the origin of the data is greater than the sample size of the kink, and in both cases the problem of high dimensions appears in the model, and in this case the kink regression model is called High Dimensions Kink Regression (HDKR). In this research, the method Penalized Least Squares method (PLS) was presented, to estimate parameters of the (HDKR) model by adopting different penalty functions, including the LASSO penalty, the SCAD penalty, the Minimax Concave Penalty (MCP) in addition to the Elastic-net penalty function, which was suggested to be used by the researcher. By applying the methods of the PLS on real data related to the market value and the variables affecting it (accounting data) of the Baghdad Soft Drinks Company, and by adopting the scale of the mean squares error for the model for the comparison, it was found that the method of PLS based on the penalty function Elastic - net is the most efficient in the selection and estimation.