A1505
Title: Cryptocurrency exchange simulation
Authors: Kirill Mansurov - Saint-Petersburg State University (Russia) [presenting]
Dmitry Grigoriev - Saint Petersburg State University (Russia)
Alexander Semenov - University of Florida (United States)
Abstract: The approach of applying state-of-the-art machine learning algorithms is considered to simulate some financial markets. The cryptocurrency market is chosen based on the assumption that such a market is more active today. As a rule, they have more volatility, attracting riskier traders. Considering classic trading strategies, an agent with a self-learning strategy is also introduced. Deep reinforcement learning algorithms are used to model the behaviour of such an agent, namely deep deterministic policy gradient (DDPG). Next, an agent-based model is developed with the following strategies. With this model, the main market statistics are evaluated, named stylized facts. Finally, a comparative analysis of results is conducted for the constructed model with outcomes of previously proposed models, as well as with the characteristics of a real market. As a result, it is concluded that the model with a self-learning agent gives a better approximation to the real market than a model with classical agents. In particular, unlike the model with classical agents, the model with a self-earning agent turns out to be not so heavy-tailed. Thus, it is demonstrated that for a complete understanding of market processes, simulation models should take into account self-learning agents that have a significant presence in modern stock markets.
