Unit root and cointegration literature is based on the idea that integrated variables lead to spurious regression and spurious regression can be avoided by cointegration analysis but Granger et al. (2001) found that spurious regression can also occur in stationary series, implying that unit root and cointegration analysis are not much helpful to avoid spurious regression. In addition to that there are size and power problem associated with unit root testing. On the other hand there are methods to model time series without unit root and cointegration analysis. If we take the objective of forecasting, how do methods using with unit root and methods avoiding unit root perform? So far there is no answer to this question. The objective of this study is to evaluate performance of four methods of modelling time series of which two methods are based on unit root and cointegration analysis (i.e. Engel and Granger two-step process and Johansen and Juselius maximum likelihood approach) and two methods which do not require unit root and cointegration analysis root (i.e. ARDL bound test and vector auto regressive). The performance of four methods is compared on the basis of forecasting ability on real data. ARDL bound test proved to be the most efficient estimation method with least chance of spurious regression and optimal forecasting. Based on these estimation result, this study concludes that ARDL bound test is most powerful for testing long run relationship and also for the forecast. Supervisor:- Dr. Atiq ur Rehman