This paper investigates that the volatility is statistically predictable using prediction models and this predictability is significant enough to generate abnormal returns. To forecast future option prices, GARCH(1,1) model and Implied Volatility Regression model are used. We compare the performance of two different methods in the KOSPI 200 index option market over the period from January 14, 1999 to October 30, 2000 by straddle trading strategy. After pricing according to the results of two prediction models, we buy(sell) a straddle when the model price is higher(lower) than the market price on each day. The model prices using two prediction models are not unbiased estimates by the Wald statistic test. Using the IVR model exhibits larger profits than the GARCH model. Two prediction models earn abnormal returns. But it is difficult to reject the efficient market hypothesis accounting for transaction costs and the high standard deviation of returns.