DOI: 10.36347/sjpms.2021.v08i08.003

Pages: 143-152

Multicollinearity is a phenomenon when two or more predictors are highly correlated that leading the matrix X^T X to be singular, and hence identifying the least squares estimates will encounter numerical problems. In this work, we proposed two remedial measures for handling severe multicollinearity in the least-squares estimation, namely, the Ridge regression, and Least Absolute Shrinkage and Selection Operator (Lasso). A simulation study was conducted to compare the two proposed methods under different settings. These settings include different sample sizes, a variety of several explanatory variables used in the model along with the difference in the degree of correlation that exists among the explanatory variables, and finally the dependency of the error terms on the normal or non-normal distributions. This simulation study is novel in the field of Shrinkage Estimators, also may increase the effective capabilities of Ridge Regression, and several interesting results have been achieved.