Authors: Kirill Mansurov, Alexander Semenov, Dmitry Grigoriev, Andrei Radionov & Rustam Ibragimov
Journal: Computational Economics
Abstract: In this paper, we consider the approach of applying state-of-the-art machine learning algorithms to simulate some financial markets. In this case, we choose the cryptocurrency market based on the assumption that such markets more active today. As a rule, they have more volatility, attracting riskier traders. Considering classic trading strategies, we also introduce an agent with a self-learning strategy. To model the behavior of such agent, we use deep reinforcement learning algorithms, namely Deep Deterministic policy gradient. Next, we develop an agent-based model with following strategies. With this model, we will be able to evaluate the main market statistics, named stylized-facts. Finally, we conduct a comparative analysis of results for constructed model with outcomes of previously proposed models, as well as with the characteristics of real market. As a result, we conclude that our 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-learning agent turns out to be not so heavy-tailed. Thus, we demonstrate that for a complete understanding of market processes simulation models should take into account self-learning agents that have a significant presence at modern stock markets.
Journal: Computational Economics
Abstract: In this paper, we consider the approach of applying state-of-the-art machine learning algorithms to simulate some financial markets. In this case, we choose the cryptocurrency market based on the assumption that such markets more active today. As a rule, they have more volatility, attracting riskier traders. Considering classic trading strategies, we also introduce an agent with a self-learning strategy. To model the behavior of such agent, we use deep reinforcement learning algorithms, namely Deep Deterministic policy gradient. Next, we develop an agent-based model with following strategies. With this model, we will be able to evaluate the main market statistics, named stylized-facts. Finally, we conduct a comparative analysis of results for constructed model with outcomes of previously proposed models, as well as with the characteristics of real market. As a result, we conclude that our 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-learning agent turns out to be not so heavy-tailed. Thus, we demonstrate that for a complete understanding of market processes simulation models should take into account self-learning agents that have a significant presence at modern stock markets.