Assessing independence of two series was, is and will be the most fundamental goal of econometric/economic practitioners. Most of the economic (especially macroeconomic) data are time dependent and economists are interested to validate different economic theories using sophisticated data analysis tools. At the end of 19th century Karl Pearson developed coefficient of correlation to assess correlation between two series, however Yule (1926) criticized the use of this coefficient for time series because of its autocorrelated behavior. Haugh (1976) was considered the first to propose a measure of correlation for time series. His idea of pre-whitening the series’ first and then apply correlation coefficient became very famous. Since then different versions of Haugh test were developed. The latest version was developed by Rehman and Malik (2014) that is also based on the same idea, however, the pre-whitening process became more refined and modified. For the last four decades, several tests of independence for time series are developed. Every test was developed on a particular property of underlying assumptions, and it works in its own domain and fails to work in other situations/domains. Researchers working in this area, presented few results to compare proposed test and one or two previously developed tests and concluded superiority of his/her proposed test. These studies include Hong (1996b), Duchesne and Roy (2003a) and so on. These comparisons are ad hoc in nature and no comprehensive study available in the literature to unfold the strengths and weaknesses of these tests. We organized a comprehensive Monte Carlo simulation study to compare tests of independence and selected a broad data generating process in a common framework. For this purpose, standard stringency criteria of Zaman (1996) has been used. In presented study, we have selected eleven available tests of independence for time series. To use stringency criteria, tests of independence should maintain a stable size and then based on its powers we can decide about the appropriateness of the test. For checking the size stability of the tests, we have used twenty-one different specifications of stochastic part in our data generating process, similarly two specifications of deterministic part have been used for each stochastic part case. So, we have tested size stability at total forty-two different combinations of data generating process. Simulated critical values were used, as past studies suggested that asymptotic critical values are not appropriate. All eleven tests of independence have their sizes around nominal size of 5%. Following the stable size, power analysis has been carried out for the same combinations of the data generating process and results suggests that Atiq test performs well in small sample size in almost all cases. PhamRoy test remains on second position in small samples but in many situations, it supersedes Atiq test in medium and large sample sizes. Haugh test remains at third place in almost all cases of the simulation study however the difference between the shortcomings of Haugh and PhamRoy test are very large. The remaining tests have not shown any considerable performance. Kim Lee, LiHui and Bouhaddioui tests considered worst in all sample sizes and with and without deterministic part cases. Another important contribution of this study is to compare three important techniques used to check the dependence of two or more-time series, these include cointegration, Granger causality and Tests of independence for time series. Using renowned Keynesian function of income and consumption, we applied these tests on real economic data of income and consumption of 100 countries from 1970 to 2014. The results depict that three selected tests of independence, i.e. Atiq, PhamRoy and Haugh tests have appreciable power gains and lower size distortions. Again, it is observed that Atiq test shows better empirical power gain with least size distortion. PhamRoy and Haugh tests also shows good performance, have good power gains and small size distortions. However, five cointegration tests which are considered relatively better by Khan (2017) and famous Granger causality test have shown very poor performance both in terms of real empirical size and power.