Abstract
This thesis reports on research exploring the impact of social networks on the strategies and performance of traders in artificial financial markets.A Social Network Artificial stoCk marKet (SNACK) model is developed to do this.
The SNACK model investigates how traders may make decisions about trading strategies based on individual learning and social learning in continuous double auction markets with different communication networks. The case of no-network is also considered, for comparison. Two reinforcement learning methods for multi-armed bandit problems, Epsilon-greedy and SoftMax, are applied to help traders to make decisions by learning from the experiences of themselves and their neighbours. Moreover, a random learning strategy is applied for comparison with these two reinforcement learning methods.
There are four main statistical analyses of the simulation data from the SNACK model. First, I examine the influence of the proportion of rational traders on the performance of the financial markets and traders. This shows that the proportion of rational traders can impact market outcomes, which include market price, number of trades, buyers' profit, sellers' profit, and market efficiency. Second, the Tripartite Domination Conjecture about the convergence of proportions of traders choosing each trading strategy shows that the proportions of buyers and sellers selecting each trading strategy are stable after a fixed number of trading rounds with social networks and no-network. Third, the relationship between the proportion of rational traders and the financial markets is quantified. The findings show that there may be a strong sine and exponential mixture or polynomial quantitative relationship between the proportions of rational traders and the mean values of market outcomes with different social networks and no-network. The fourth study is how social networks impact the performance of traders and of the financial markets. These four studies are explored under SoftMax learning strategy. In addition, an-other small study of the influences of different machine learning methods for the decision making processes on the performance of market outcomes is given in the appendix.
In summary, this research contributes to an understanding of the relationship be-tween social networks and the performance of both traders and marketplaces, in artificial stock markets.
Date of Award | 1 Feb 2015 |
---|---|
Original language | English |
Awarding Institution |
|
Supervisor | Peter McBurney (Supervisor) & Katarzyna Musial-Gabrys (Supervisor) |