Simulation Experiment Research On Artificial Financial Market Based On Multiple Competitive Market-makers

In recent years, the OTC (over the counter) market in our country develops at a sharplyhigh speed. In OTC market the listed securities exhibit low price, low liquidity, high risk.Auction mechanism is unable to solve the problems caused by market low liquidity.Especially when block trade happens, auction mechanism can easily lead to drasticfluctuation in price, which will cause unbearably high risk. Faced with such situation, thediscussion about whether to introduce market-makers to China’s financial market hasattracted wide attention. All kinds of complex economic phenomena in financial market areemerging from market participants’ micro-level interaction in macro mode. Agent-basedcomputational economics theory provides a unified, open research method for complexfinancial system.According to complex adaptive theory and agent-based modeling computationaleconomics principles, after introducing multi-market makers quotation mechanism, anartificial financial market simulation framework based on multiple competitive marketmakers has been built in the paper through agent modeling techniques to research themarket emerging characteristics in specific market structure.Firstly, the market was divided into3parts: market-makers, investors and marketenvironment, afterwards, the mathematical models of the decision behavior of market-makerand investor were designed and the market environment structure was defined. Secondly,market-makers agent class, investor agent class and market environment agent class weredeveloped on ANYLOGIC simulation software, in which rich methods were defined toimitate the behavior of market participants. Furthermore, a usefully class was developed toimport date to MS EXCEL for detailed statistical analysis. Lastly, two groups of simulationexperiments were carried out to verify the model’ reasonableness and figure out themarket-makers’ quotes characteristics.The first group experiments results show3important features. The first is that the dailyreturn time series replicate the typical characteristics in real market, e.g., fat tails and nonnonlinearity. The second is closing transaction price immediately follows the up and down ofthe real value in a stable amplitude and frequency. The third is market spread in our model is smaller than in the model with a single market-maker. All above feature can correspond tothe real market. Therefore, the conclusion that the simulation model in the paper isreasonable can be made. The Second group experiments results show that the quotes of eachmarket-maker converge to the fundamental value with some deviation, in which thedeviation of the market-maker with dynamic learning ability is smaller while the convergingspeeds slower. The convergence of market-makers’ quotes to fundamental value proves thatquoting decision methods of market-makers are effective.