Abstract
This paper presents results of an artificial stock market and tries to make it more consistent with the statistical features of real stock data. Based on the SFI-ASM, a novel model is proposed to make agents more close to the real world. Agents are divided into four kinds in terms of different learning speeds, strategy-sizes, utility functions, and level of intelligence; and a crucial parameter has been found to ensure system stability. So, some parameters are appended to make the model which contains zero-intelligent and less-intelligent agents run steadily. Moreover, considering real stock markets change violently due to the financial crisis; the real stock markets are divided into two segments, before the financial crisis and after it. The optimal modified model before the financial crisis fails to replicate the statistical features of the real market after the financial crisis. Then, the optimal model after the financial crisis is shown. The experiments indicate that the optimal model after the financial crisis is able to replicate several of real market phenomena, including the first-order autocorrelation, kurtosis, standard deviation of yield series and first-order autocorrelation of yield square. We point out that there is a structural change in stock markets after the financial crisis, which can benefit people forecast the financial crisis.
Highlights
Advances in computing had an immense influence to finance and economics
On the view of agents about financial markets, heterogeneity is appending to agents by four different modes: evolving speed, size of strategy pool, aversion level of risk and intelligent level
The optimal of all the three experimental data possess a lower level of first-order autocorrelation about yield series than the original model
Summary
Advances in computing had an immense influence to finance and economics. Today, there is a huge amount of financial data available for analysis, and we have such strong processing power to analyze these data. Using the modified structure of ASM anterior mentioned, the below experiments try to better simulate the real market data by appending heterogeneous to agents from different aspects. Comparing with real data, when using parameters vector (2) and (5), Table 4 shows that the yield series presents no first-order auto-correlation, and the square of yield series exhibits large positive first-order auto-correlation which the original model does not show, well fitting the statistical features of real financial market data at daily frequency. They have little larger standard deviation value. The kurtoses are larger than the value of the original model, though still smaller than the value of real data
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call