Model selection is a fundamental issue in econometrics analysis. A major goal of econometric analysis is to develop an optimal model for observable phenomenon among various models. There are lots of procedures and criteria utilized for selecting model. Choice of model selection criteria depends on subjective judgement. General to specific methodology is theoretically superior wherever models could be nested in single large model. However general to specific is not feasible when the nested model becomes very large. In this situation, non-nested encompassing can be used for selecting models and is recommended by several authors. However, it is generally not known that how good non-nested encompassing is, in terms of selecting good model. The aim of this study is to examine the performance of non-nested hypothesis test for model selection and to select better model through encompassing on the basis of forecast performance of the models. On Forecast RMSE we found that Cox, Ericsson, Joint test have same power , and they chose the correct model for 13 out of 16 countries indicating 81 % power. Whereas, Sargan test chose the correct model for 11 out of 16 countries indicating 68 % power. So, Sargan is less suitable test for model selection than Cox, Ericsson, and Joint test. Supervisor:- Dr. Atiq-ur-Rehman