Prediction and forecasting with linear regression models under multicollinearity

The Ordinary Least Squares (OLS) estimator is the widely used technique for estimating linear regression models. But the OLS estimator can be highly variable in certain directions, especially when the explanatory variables are collinear. Therefore, in the presence of multicollinearity, biased estimation techniques are often suggested as alternatives to the OLS. In many areas, the prediction/forecasting of the future values is very important because, forecasting provides information about the potential future events and their consequences. Thus, it increases the confidence of the policy maker (or the manager) to make important decisions. When a multiple linear regression model is used in predicting/forecasting unknown values of the response variable, its ability to produce an adequate prediction equation is of prime importance. In this study, some techniques are suggested to improve the prediction/forecasting performances of alternative biased estimators. Prediction equations formed by using these techniques are compared on real data sets.