The most important feature that directed to the development of new time series econometrics was the spurious regression. It is a phenomenon known to econometricians since the times of Yule (1926) who attributed this problem to missing variable. A spurious regression occurs when two independent series come up with significant regression results. For a long time, missing variables were considered as root cause of spurious regression. However, Granger and Newbold (1974) challenged this wisdom and presented unit root as one of the causes of spurious regression. The extensive literature considers the nonstationarity as the only cause of spurious regression. The researchers frequently employed unit root and co-integration procedures for the treatment of spurious regression in case of nonstationarity but these procedures are equally unreliable because of uncertainty about various specification decisions like choice of the deterministic part, structural breaks, choice of autoregressive, lag length and distribution of error term. On the other hand Granger et al. (2001) show that unit root is not the only reason for spurious regression. They show the possibility of spurious regression in stationary time series. Whereas unit root and cointegration are unable to deal with this problem because they deal only nonstationary series. Such amount of conventional econometric literature is inadequate to deal with the problem of spurious regression in stationary time series. The objective of this study is to provide an alternative solution of spurious regression for both stationary and nonstationary time series. So, this study makes two contributions in this particular setup. First, spurious regression occurs due to missing variable and can be avoided by including missing lag values. Therefore, an alternative way to look at the problem of spurious regression takes us back to the missing variable (lag values) which further leads to ARDL model. Second, it significantly reduces the probability of spurious regression in both stationary and nonstationary time series case. This study mainly focusing on Monte Carlo simulations and real data is also used for performance comparison of ARDL model and conventional procedures. Our results indicate that conventional methods are significantly suffering in size and there is power problems but the performance of ARDL in both cases is far better than conventional methods. ARDL model significantly reduced the probability of spurious regression in stationary and nonstationary time series case. Supervisor:- Dr. Saud Ahmad Khan Co-Supervisor:- Dr. Attiq ur Rehman