Currently, algorithm trading has gained immense popularity in recent years. In addition, machine learning is also gaining increasing importance in the financial market, where people usually use it to predict price movements of the asset and make an optimal trading strategy. However, a number of limitations of traditional machine learning have been exposed. For instance, the financial market is very unpredictable as it constantly changes based on the events. As a result, methods that rely on historical data tend to have low prediction accuracy. Moreover, it is also difficult for classical machine learning algorithms to determine the thresholds for optimal policy and convert the predictions into actions. Therefore, our group is trying to leverage a different machine learning paradigm, one of multi-agent reinforcement learning, where the agents themselves learn the improved techniques by accounting for the changes in the environment.
Environment working well RL implemented
We are five student at UC Berkeley, in a Master of Engineering program within the Fintech concentration of the IEOR department: