| Abstract: | 此研究專題是與美國布蘭迪斯大學Blake LeBaron教授共同研究，主要在探討委託單驅動市場中，交易員委託單分割行為與長期記憶性的關聯。例如：交易量、波動性、委託單訊號的長期記憶性（但報酬無持續性）。計畫的第一部份，主要在研究NYSE TAQ (New York Stock Exchange Trades and Quotes 資料庫中有關中長期記憶性的統計實例。相較於利用日頻率資料的研究，使用極高頻率資料來做長期記憶性的實證研究是較新的研究方向，因此，在將這些現象理論化之前，提出一些實證資料是很值得的。接著，將建立一個一般自動化委託單切割系統的理論模型，並檢視委託單切割對於引起長期記憶性的重要性。LeBaron and Yamamoto (2007) 證明股票交易員的模仿行為種類是引起長期記憶性的關鍵，故我們將比較這些結果並證明哪個因素（模仿與委託單切割）的重要性較高，以證明這些因素與實證現象的關聯性。 因此，此研究計畫預計需要兩年以上的時間來完成，主要原因包括： 1) 一般而言，整理極高頻率資料會耗費大量的時間。 2) 如研究計畫表C012-2中所陳述，交易量與波動性的自我相關函數有季節性（交易量與波動性在開盤與收盤期間有較大的傾向）存在，故此研究要將季節性因素移除，以便分析長期記憶性。此次研究主題有所進展時，我們計劃在下列兩個研討會發表研究結果： 1) ESHIA/WEHIA 2008, 19th to 21st June 2008, Warsaw University of Technology, Warsaw, POLAND 2) Econophysics Colloquium 2008, August 29-30, Kiel, Germany.
This is a research project with Professor Blake LeBaron at Brandeis University. This project asks how traders’ order splitting behavior is related to the long-memory properties in an order-driven market, i.e., long-memories in volume, volatility, and order signs (but yet, the market is informationally efficient in a sense that there is no persistence in returns). We first show a statistical example of the long-memory properties with our NYSE TAQ (New York Stock Exchange Trades and Quotes) dataset that National Chengchi University library has. Empirical research on long-memories with ultra-high frequency data is relatively new to those with daily frequency. So, we think that it would be worthwhile to provide some evidence on them before theorizing those phenomena. Then we set up a theoretical model on a simple automated order-splitting system, and examine how important order-splitting is to generate the long-memories. LeBaron and Yamamoto (2007) show that some form of imitative behavior of stock traders is key for generating the long-memories. We compare these results and show which factor (imitation and order-splitting) is more important and how these are related to those empirical phenomena. We estimate that this project takes more than two years to complete mainly because: 1) It normally takes extremely huge amount of time to clean such ultra-high frequency data. 2) As I describe the details in the proposal (a form C012-2), the autocorrelation functions for volume and volatility show hourly seasonality (volume and volatility tend to be higher around at the opening and closing sessions) so that we need to remove to analyze the long-memories. I am applying to NSC for two years’ research funds this time. As we make some progress on this research, we plan to present our results at the following two conferences: 1) ESHIA/WEHIA 2008, 19th to 21st June 2008, Warsaw University of Technology, Warsaw, POLAND 2) Econophysics Colloquium 2008, August 29-30, Kiel, Germany. |
