Abstract: Most of the recent works on ETF are based on daily data,thus the strategy of instantaneous ar bitrage could not be carefully studied.In this paper,two ETFfunds in Chinese stock market are examined based on high frequency data.We investigate the extent and properties of the premiums and discounts of ETF from their market value.The cost of arbitrage is carefully studied.In the last part of our work,the Autoregressive Conditional Duration (ACD) is used to study the dynamic structure of ETF transactions; we introduce a statistical method to forecast the time cost to realize the arbitrage chance.

Key words: exchange-traded fund, high frequency data, cost of arbitrage