Among hedging methods for price fluctuations using currency futures trading, a GARCH error correction model was considered as a hedging model that yields time-varying hedge ratios. First, unit root tests and heteroskedasticity tests were conducted to determine whether the trends of futures and spot currency prices are non-stationary time series with heteroskedasticity, and a GARCH model that explains these heteroskedasticity characteristics was selected. Meanwhile, since the long-run equilibrium relationship between variables can be explained by utilizing the cointegration relationship when the linear combination of non-stationary time series data is stable, the optimal risk hedge ratio for currency futures was estimated using a GARCH error correction model that simultaneously considers the equilibrium relationship between variables represented by cointegration and the heteroskedasticity effect. In this study, hedging performance was compared across six models (no hedge, 1:1 naïve hedge model, OLS model, error correction model, GARCH model, and GARCH error correction model) for four currencies (Japanese yen, Australian dollar, Deutsche mark, and Canadian dollar). To examine whether the GARCH error correction model improves hedging efficiency compared to other hedging models when hedgers behave according to a mean-standard deviation expected utility function and transaction costs are considered, both within-sample and out-of-sample analyses were conducted. The results of both within-sample and out-of-sample analyses showed that the GARCH error correction model improved hedging performance compared to other models for the Japanese yen and Deutsche mark.