Investor Learning, Information Dissemination Efficiency And Policy Design In Continuous Double Auction Markets

This thesis aims to study how the investors’ learning ability and orderaggressiveness affect the information dissemination efficiency, and then disposes theasset short-term price formation process and optimizes the policy design in continuousdouble auction (CDA) markets.Firstly, we introduce the logic of asset short-term price formation process andbuild a continuous double auction artificial stock market (CDA-ASM), which includesinformed traders, genetic algorithm (GA) learning traders and zero intelligence (ZI)traders with the help of the agent-based computational finance method. TheCDA-ASM reproduces a number of important stylized facts, including leptokurtosis,fat tails, short-term negative autocorrelation, volatility clustering, long memory andlong-term positive autocorrelation. Also the model passes the robustness test ofdifferent investor structures and different information durations. It proves that theagents can get more information by their leaning in short-lived information duration,hence uninformed traders learning ability could improve the informationdissemination efficiency. Under the four different investor structure markets, thereduction of the genetic algorithm learning traders and zero intelligence traders’ orderaggressiveness would give different impression on the information disseminationefficiency. Moreover the reduction of their order aggressiveness not only changestheir behaviors, but also affects the bid-ask spread and informed traders’ ordersubmission.Secondly, we also study how the combination of empirical analysis andagent-based modelling can provide scientific policy suggestions to meet the practice’sdemand. Based on the real investors’ structure and the statistical characteristics ofinvestors’ behaviors with different trading frequency in CSI300stock index futuresmarket, we establish a dynamic model with heterogeneous adaptively interactiveagents to study the policy design of the minimum tick size and position limits. Themodel includes informed traders and uninformed traders who are divided into threetypes: intelligent traders, simple traders, and liquidity traders. The model successfullyreproduces the key order-flow characteristics and stylized facts of the CSI300stockindex futures market. At last, it analyzes how the minimum tick size and position limitaffect the information dissemination efficiency, market liquidity and market volatility. This thesis focuses on the uninformed traders behaviors, builds up an agent-basedcomputational model to analyze the information dissemination process through theuninformed traders’ learning and order submission behaviors; it solves a part ofpuzzles of dynamic modeling in the continuous double auction market microstructuretheory. This model not only considers more comprehensive market quality indicatorsthan empirical studies, but also avoids the criticism that the parameters are chosenunrealistically under agent-based computational models. Finally we generate severalscientific policy design suggestions. The thesis makes some substantial progresses inthe agent-based computational finance, and also provides important guidingsignificance and practice value to the reform and innovations of Chinese financialmarkets.