Abstract:
"Algorithmic trading has been continuously dominating the financial market and making the
percentage of manual traders becoming lesser by each year. Financial institutions and hedge funds
have to carry out continuous extensive research to find the alpha trading strategy because they
have to adapt to continuously changing market conditions while beating competitor strategies.
In recent years many reinforcement learning algorithms have been developed to solve challenging
tasks that have complicated problem space and imperfect information and a dynamically changing
environment. Reinforcement learning appears to be promising in these exact problem sets that the
algorithmic trading domain has. But there is much research to be conducted in order to deploy a
strategy that has been trained in a simulated environment to be deployed in a real-world
environment. This paper is focused on reducing the gap between the real-world environment and
simulated one by incorporating factors such as slippage, bid-ask spread, and exchange
commission. This paper attempts to cater to the issue by introducing an action scheme that has a
limit orders to reduce the effect of mentioned issues and has a realistic transaction cost so that the
actions that the agents learn will behave better in the real work environment."
