基于元模型的异质交易行为主体下股票市场微观结构仿真研究

股票市场由大量具有不同的理性程度、行为偏好以及操作策略的异质类型交易者所构成的,这些交易者汇聚在一起相互作用,以自组织的方式涌现出诸多复杂的宏观金融市场现象。基于主体建模思想,从交易媒介、交易活动以及交易者行为三个层面进行设定,构建模拟真实市场交易的人工股票市场元模型。其中交易者(Agent)按其行为偏好分为风险厌恶型、损失厌恶型、过度乐观型、保守型、过度乐观的损失厌恶型以及保守的损失厌恶型6类;交易策略则分为基础交易者、趋势交易者、零信息交易者和自适应型交易者4类。不同的行为偏好和交易策略自由组合形成代表性异质交易者,并在交易环境的约束下相互作用推动市场的演化。以我国沪深300指数实际数据为参照,采用AnyLogic系统环境下优化算法对模型参数赋值并进行敏感性分析,发现股票市值和波动率之间不存在直接的联系,且所有交易者的总财富值在无风险利率改变的条件下差别不大,这意味着我国金融市场很大程度上是一个零和游戏过程,有效市场假说在市场运行的宏观层面成立。而在微观层面,不同类型交易者的财富分布均值明显不同,说明交易者的操作策略和行为偏好直接决定了其财富状况。

Microstructure simulation of stock market with meta model in a heterogeneous market environment

The stock market is a highly complex and evolving complex nonlinear system.The stock market consists of different types of participants,which have different levels of rationality,trading strategies,and risk preferences.This large number of heterogeneous traders have come together to interact,resulting in a number of complex macroscopic phenomena in a self-organizing manner,such as volatility clustering,leverage effect and sharp price bubble.In the past,most of the researches on such problems were based on regression models,such as the GARCH family model or the realized volatility model,which explored the asset pricing process and analyzed the dynamic of stock prices.The limitation lies in the inability to effectively describe the complex characteristics of financial markets,as well as the heterogeneity of trader??s behavior,the research results are insufficient for realistic interpretation.Therefore,in order to reveal the micro-operation mechanism of the stock market and analyze the dynamic characteristics of the stock complex system,we must describe the trading behavior of various heterogeneous traders from the perspective of trader??s behavior.The interaction process between the two can be used to effectively explain the complex dynamics of stock price movement.Based on the agent-based model,we set up stock market from three aspects:trading medium,trading activity and trader behavior,and establish the artificial stock market meta-model which simulates real market trading.Among them,the traders are classified into risk aversion,loss aversion,excessive optimism,conservative,excessively optimistic loss aversion and conservative loss aversion according to their behavioral preferences.Four investment strategies are included in the model:zero-intelligence,fundamental strategy,momentum,and adaptive trading strategy using the artificial neural network algorithm.Different behavioral preferences and trading strategies are freely combined to form representative heterogeneous traders,and interact under the constraints of the trading environment to promote the evolution of the market.The model maintains a high degree of freedom and flexibility while portraying the behavior of various types of market participants,so it can simulate the market operation process more realistically and observe the emergence phenomenon.The calibration is implemented using a scatter search heuristic approach using the HSI 300 index data.The simulation results show that the artificial stock market data and the real data have similar statistical characteristics,indicating that the meta-model constructed in this paper is effective.By performing a sensitivity test and performing an analysis of variance on the test results,the effects of various types of parameters on stock market value and fluctuations are observed.The results show that,overall,the sensitivity of volatility to market parameters is higher than the market value sensitivity to parameters.Transaction costs,risk-free interest rates,maximum buying positions,and maximum short-selling positions have a significant impact on volatility,but only transaction costs and risk-free interest rates have an impact on market capitalization.In addition,all trader behavior preference parameters and network structure parameters will have a significant impact on the volatility,but the impact on the market value is not significant,which shows that from a macro perspective,stock market is basically an effective market,market value.The market capital is basically synchronized with the risk-free rate.Finally,the distribution of traders??wealth in the process of artificial stock market trading is analyzed.The results show that the wealth levels of different types of traders are significantly different.Different trading strategies are the main reason for this phenomenon.The fundamental trader has a higher level of wealth,while the zero information traders have the lowest level of wealth.At the same time,overconfidence or conservative behavioral preferences have a negative impact on wealth levels.These all indicate that the market is not effective from a micro perspective,and the stock market is a zero-sum game for traders.The complexity of the real market is far from this.For further research,other types of traders??behavioral biases can also be considered into the model,such as optimism,pessimism,anchoring strategy or psychological account.Traders can also be influenced by other types of network structures,such as small world networks or distance-based networks.Whether the addition or improvement of these new elements can further enhance the ability of the META model to interpret the real world is a further direction that can be explored.