Agent-Based Models以及股票市場的實證現象

Abstract

我這次向 NSC 申請三年的研究經費，為的是進行以下三項研究計畫。所有的研究為單一作者研究。我計畫以模擬分析的方法來解釋顯著的股市現象，例如波動叢聚現象，波動不對稱性現象，以及成交量、波動性、order signs的長期記憶現象（然而，從股市不具持續回報的角度來看，股市在提供資訊方面算有效率）。近期的實證研究證明有這些結果；然而，試圖從理論的角度來解釋這些現象卻多為失敗。我以代理人基模型進行研究並可達成這個目標。在我第一個研究裡，我探討一個導致成交量、波動性、order signs的長期記憶現象的可能原因。我會探索以下的交易策略是否跟先前提到的長期記憶現象具有關連。我所探討的交易策略出自於金融實證研究的文獻。在股票投資人的行為方面，近期的實證研究得出兩個結論。其一，當委託單的同邊深度越厚（薄）時，投資人傾向更多（少）積極下單。其二，當買賣價差越大時，投資人會減少積極下單。從實證的證據來看，投資人的下單行為受到委託單的狀態的影響。我推測那些策略跟成交量、波動性、order signs的長期記憶現象有關。我將說明那些投資人的行為是否會造成所有項目的長期記憶現象發生。在我第二個研究裡，我探討一個導致波動不對稱現象發生的可能原因，並提出價格跟波動變化具有負相關的論證。我提供一個論點以解釋是什麼決定不對稱現象的幅度。我將探討投資人的從眾行為跟波動回饋效應共同作用時，是否會增強不對稱的幅度，因此可知從眾行為及波動回饋效應是決定不對稱的幅度的重要因素。在第三個研究裡，我探討一個造成波動叢聚現象發生的可能原因。近期的agent-based models顯示投資人的從眾行為造成股市的波動叢聚現象。在我所檢視的經濟體裡，投資人有從眾的行為，但是對於其他投資人的資訊卻掌握不多，因此他們的模仿行為是有限的。我推測當投資人掌握到越少其他投資人的資訊時，波動叢聚的現象越有可能消失；投資人需要掌握到其他投資人的詳細策略才能造成波動叢聚。若事實如此，由於現實中股票投資人無法觀察到其他投資人的詳細策略，因此我們的結果意謂著在現實中，除了從眾行為，另有其他的投資人行為（例如訂單分割行為）會造成波動叢聚現象的發生，如 Yamamoto 及 LeBaron（2009）已證明之。第一年裡，我將進行我的第一個研究，而第二個研究會在第二年完成。第三個研究會在第三年完成。在每年的研究完成後，我會撰寫那年的研究報告。我計畫將我的研究報告投至 Journal of Economic Dynamics and Control，或是 Journal of Economic Behavior and Organizations.

I am applying to NSC for three years’ research funds this time, in order to conduct the following three research projects. All of the projects will be done as single authored papers. I plan to conduct simulation analyses to explain well-known empirical properties in the stock market, e.g., volatility clustering, asymmetric volatility, and 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). Recent empirical research has demonstrated these results; however, theoretical attempts to explain these observations have foundered on the challenge of explaining the phenomena. My research projects achieve this goal in agent-based models. In my first research project, I examine a possible source of long-memories in volume, volatility, and order signs. I consider the following trading strategies which have been found in empirical studies in finance literature, and investigate whether that strategy is related to the previous mentioned long-memories. Recent empirical research has provided the following two results on the behaviors of stock investors. First, they tend to submit more (less) aggressive orders as the depth on the same side of the order book becomes thicker (thinner). Second, they are likely to place less aggressive orders as the bid-ask spread becomes wider. The empirical evidence suggests that investors’ order submissions are influenced by the state of the order book. I conjecture that those strategies are related to long-memories of volume, volatility and order signs. I will demonstrate whether those agents’ behaviors are critical for generating all long-memories. In my second project, I investigate a possible source of asymmetric volatility, suggesting a negative correlation between price and changes in volatility. I provide an explanation on what determines the magnitude of the asymmetry. I will examine whether agents’ herding behavior possibly intensifies the asymmetry when combined with the volatility feedback effect, and thus both the herding and volatility feedback effects are important determinants on the degree of the asymmetry. In the third project, I investigate a possible source of volatility clustering. Recent agent-based models have demonstrated that agents’ herding behavior causes volatility clustering in stock markets. I examine economies where agents herd on others, yet they have limited sets of information of other agents to imitate. I conjecture that volatility clustering tends to disappear as they observe less and less information of others, and agents need to observe the strategy details of others in order to generate the clustered volatility. If that is the case, since in reality stock investors may not be able to observe the strategy details of others, our result will imply that in reality, in addition to herding behavior, some other behaviors of agents (like order-splitting behavior) would also be important to generate clustered volatility as shown in Yamamoto and LeBaron (2009). In the first year, I will do my first research project, while the second project will be done in the second year. The third project will be done in the third year. I will write up papers for each project at the end of each project year. I plan to submit all papers to Journal of Economic Dynamics and Control, or Journal of Economic Behavior and Organizations. Keywords: agent-based models, boundedly rational, long-memory, asymmetric volatility, herding, volatility feedback effect, order aggressiveness