Artificial Intelligence (AI) is often hailed as a solution to complex modeling issues, and one area of interest is its potential application in managing risks associated with derivative contracts in investment banking. However, concerns have been raised about the practicality of implementing AI-based solutions.
A recent study published in The Journal of Finance and Data Science conducted by researchers from Switzerland and the US delves into the use of reinforcement learning (RL) agents for derivative contract hedging. RL agents are trained using simulated market data, but the lack of sufficient training data has been a challenge in this field.
To overcome this data scarcity, researchers typically rely on accurate market simulators to train their AI agents. However, this approach presents its own set of challenges, such as model selection, calibration, and similarity to traditional Monte Carlo methods. The success of AI systems heavily relies on the availability of market data, which is rare in realistic derivative markets.
In collaboration with the investment bank UBS, the study utilized Deep Contextual Bandits, a well-known RL technique. This method addresses the low training data requirement and adaptability to changing markets, while also incorporating end-of-day reporting obligations. The researchers aimed to create a model that closely aligns with real-world operational practices in investment firms.
Oleg Szehr, the senior author of the study and a former staff member at various investment banks, explains that the availability of data and meeting the operational demands of reporting end-of-day risk figures are crucial factors in real-world scenarios. The newly developed model encompasses these practical considerations and demonstrates superior efficiency, adaptability, and accuracy compared to benchmark systems.
Despite its simplicity, the new method proves to be successful in managing risk under realistic conditions. The study emphasizes that in risk management, less complexity often leads to better outcomes. By mirroring the operational processes in investment firms, this model offers practical applicability rather than aiming for ideal agent training.
This research brings forth promising insights into the use of reinforcement learning agents for derivative contract hedging. It highlights the importance of practicality and real-world considerations in developing effective AI solutions for risk management in the investment banking sector. With further advancements and refinements, these methods have the potential to revolutionize and optimize derivative contract hedging strategies.
