Liquidity Cost of Market Orders in Taiwan Stock Market: A Study based on An Order-Driven Agent-Based Artificial Stock Market

Abstract

This thesis construct an order-driven artificial stock market base on Daniels et al.\r\n(2003) model. We also use autoregressive conditional duration (ACD) model initiated\r\nby Engle and Russell (1998) to model duration or order size. We analyzed the\r\ntransaction cost of ten securities, including stocks, Exchange-Traded Funds (ETFs)\r\nand Real Estate Investment Trusts (REITs), in Taiwan stock market and compared this\r\nresult with the simulated cost of our models. We find that for those frequently traded\r\nsecurities, for example, TSMC (2330.TW) or China Steel (2002.TW), it is better not\r\nto incorporate ACD model of duration in the model, and for those not frequently\r\ntraded securities, for example, President Chain Store (2912.TW) or Gallop No.1 Real\r\nEstate Investment Trust Fund (01008T.TW), it is better to incorporate ACD model of\r\nduration in the model. Our empirical estimates show that the liquidity costs of market\r\norder of these ten securities are generally smaller than 3%, and largely lied between\r\n-1% and 1%. We, however, find that simulation costs of market orders in our model,\r\nwith a range from 0% to 10%, are generally larger than those of real data. One\r\npossible reason for this departure is that investors in stock markets generally do not\r\nplace their orders blindly. They tend to wait for the appearance of opposite order size,\r\nand then place their orders. They also tend to split up a large order, and then reduce\r\nmarket impact. These behavior do not exist in our simulation. Regardless of these\r\ndifferences, our models may still be a simulation tool for transaction cost assessment\r\nwhen one would like to liquidate their asset in a short span of time.