Johansen (1988) proposed a maximum likelihood estimation method for multivariate co-integration test. The method can test the linear constraints of co-integrating vectors, and has solved the problem of test accuracy of symmetry and proportionality conditions. Compared to three-variable system, two-variable system uses more co-integration techniques. But the symmetry and proportionality conditions are often rejected; parameters in the PPP regression usually have a great gap between the theoretical value and the estimated value. In the two-step co-integration method this may be caused by the small amount sample bias, and the problem still exists no matter using the large amount of samples or Johansen’s method. It is argued that this may be because the time span of the sample data is too short to test mean reverting of the real exchange rate. This is called inefficient in econometric. The main reason of inefficient is few samples and information. Frankel (1986) was the first one who found the unit root test inefficient, he believed that when we use only ten or more than ten years of data to test whether the real exchange rate is generated by a random walk process, the tendency of rejecting the null hypothesis is still very low even if the real exchange rate is stationary. This view is proved by many scholars later, such as Froot & Rogoff (1995), Lothian & Taylor (1996) and so on. As the test value in traditional unit root test is low, the researchers expanded the sample in two ways to solve this problem: (1) using large-span or long-term data, (2) using cross time series data from different countries, that is panel data methods.