To identify price dynamics between WTI crude oil spot and futures prices and infer arbitrage behaviors on the basis of cost-of-carry hypothesis, I tried to apply bivariate 2-regime Threshold Error Correction Models to data along with conventional linear Vector Error Correction Models. After dividing the entire sample into 5 sub-samples by 4 structural break points as a result of Bai and Perron(2003) method, I employed estimation models which have been chosen by Hansen and Seo(2002) cointegration test and Brock, Dechert, and Scheinkman(1987) nonlinearity test between linear VECMs and nonlinear TVECMs. Estimation results showed the following things. First, onedirectional arbitrage was found in sub-sample 1 and 5, and bi-directional argitrage was found in sub-sample 5. Second, error correction effects were found in all the sub-samples. Third, with respect to price information leadership, while spot markets played it in sub-sample 1, and futures markets in sub-sample 2 and 3, information feedback between these two markets was found in sub-sample 4 and 5. Finally, I compared the forecasting performance of linear VECMs and nonlinear TVECMs with RMSE(Root Mean Square Error). Nonlinear TVECM filters were preferred for forecasting because of slightly lower RMSEs in the case of using nonlinear TVECM filters compared to those in the case of using linear VECM filters.