This paper investigated how to use prediction models to obtain higher arbitrage returns on Taiwan Stock Index Futures. The selection of arbitrage and exercise timing significantly influences the arbitrage performance and profits. The study examined Taiwan Stock Index Futures by utilizing: (1) absence of prediction models, (2) Neural Network, and (3) Markov GM(1,1) to arbitrage against the prediction of mis-pricing. Empirical results indicate that when no arbitrage boundaries ranged from 0.0148 to - 0.0146, the Grey theory showed the best performance of arbitrage, neural network the next, and the absence of prediction models the worse.