Ridge regression parameter selection: Turkey's foreign direct investment

The problem that arises when there is a linear relationship between one or more independent variables in the multiple linear regression model is called the multicollinearity problem. In the case of multicollinearity, the variance of the least squares estimators is high and the parameter estimates are calculated incorrectly. In this case, least squares estimators are incapable of interpreting the results, making the least squares method unreliable. Biased regression estimators are used as an alternative to the least squares estimator to solve the multicollinearity problem and estimate the model parameters. In this study, Ridge regression method, which is one of the biased regression estimators, is discussed. The solution of the Ridge regression parameter depends on the tuning parameter k. In this study, analyzes were made on the selection of the Ridge regression adjustment parameter k. In this study, the factors affecting Turkey's foreign direct investments between 1974-2019 were taken as a data set. A simulation study was carried out with the multiple linear regression model established with the factors affecting foreign direct investments. With the simulation study, it was aimed to choose the one that gives the best result among the model selection criteria for the selection of the tuning parameter k. After the simulation, analysis was made with real data and multicollinearity was determined according to the results. By reanalyzing with Ridge regression, one of the biased estimators included in the model, it was determined that Ridge regression solved the connection problem by giving better results than the least squares method. Keywords: Multiple Linear Connection, Ridge Regression, Parameter Estimators, LASSO.