| |
Under the leadership of our Deputy Director for Industry Prof. Chen Kan, RMI has continued to make advances in the industry projects, which aim at developing models and tools for industrial applications, establish industry research program for training students on the use of analytic tools and methods for quantitative risk management, as well as establish industry collaborations. In 2024, the projects team worked on developing and finetuning new packages, collaborating with other researchers NUS and other universities on interdisciplinary project as well as further the training of students. Researchers at RMI continue to collaborate with Professor Thomas Nagler (University of Munich) and Professor Thibault Vatter (HES-SO University of Applied Sciences and Arts Western Switzerland), the original developers of the Vinecopula package (Pyvinecopulib). The team at RMI is developing a more comprehensive and efficient vine copula package called Torchvinecopulib, which now features GPU support. This package has been continuously updated since its release in April 2024, with over 1,000 downloads to date. Vine copulas have proven to be valuable tools for dependence modeling in various fields, including finance and climate science. Currently, Torchvinecopulib is the only package that enables vine copula conditional simulation and is the fastest for full vine copula simulations.
In addition, RMI researchers are advancing their work in NLP for Financial Text Analysis. Recent research focuses on using ChatGPT APIs for document embedding and sentiment analysis. We are exploring other components of the ChatGPT APIs for tokenization, clustering, and sentiment analysis of company news, including 8-K reports and other financial news sources.
RMI researchers are also involved in an interdisciplinary project based on the Complex System Approach. In collaboration with researchers from NUSCities and other universities, we are applying newly developed tail risk measures and vine copula dependence modeling tools to climate risk research. Additionally, we are exploring evolutionary reinforcement learning dynamics within an agent-based model inspired by spin glasses, a prototypical complex system model. The work of two recent Nobel laureates in physics, Giorgio Parisi and John Hopfield, is related to spin glasses. Our goal is to understand the emergent adaptive behaviors of agents, particularly concerning whether such systems generally self-organize into critical states, leading to optimal performance. The complex system approach is also helpful for understanding financial systems and the universal characteristics of tail risks.
|
|
