This paper proposes a four-regime bivariate Markov regime-switching model to estimate the daily time-varying minimum variance hedge ratios for West Texas Intermediate (WTI) crude oil, and evaluates its in- and out-of-sample hedging performances with two-regime model, CC-GARCH, TVC-GARCH, and OLS models. Empirical results reveal that the four-regime Markov switching model outperforms the other models for both in- and out-of-sample hedging performance. Based on Hansen's SPA test (2005), the four-regime model significantly outperforms the other models for only in-sample hedging. (C) 2011 Elsevier Ltd. All rights reserved.