In the Beginning was the Word: LLM Risk Measures

Daniel Traian Pele (IDA, ASE), Min-Bin LIN, Rui REN, Andrei Theodor Ginavar, Vlad Bolovăneanu, Bruno Spilak, Alexandru Andrei, Mihai-Filip Toma, Stefan Lessmann, Wolfgang Karl Härdle

Value at Risk (VaR) and Expected Shortfall (ES) are essential measures in financial risk management, used to quantify potential losses in portfolios. Traditionally, these measures are estimated using parametric methods, historical simulations, or Monte Carlo simulations. Recently, machine learning, and more specifically, large language models (LLMs), have emerged as promising tools to enhance the accuracy and efficiency of these estimations. This paper addresses the research gap by investigating the use of LLMs to estimate market risk indicators such as VaR and ES, leveraging their zero-shot forecasting capabilities.

AI and National Security

Brent CHENG (IDP CO, LTD, Taipei)

In our research on efficient simulation and training of human security personal we discovered the need to connect our skills to work specific documents and information.  The Platograph architecture with its Q2 ecosystem has been trained to cover our AI applications to combine drones, robots, and other mobile elements in a survey system for modern multifunctional defence.  We present the various platforms of the Q2 eco system and show how the neuralink technology allows the control of the multi facet data analytics.

AFinCrime: A Platform to identify and predict crime risk

Monica Violeta Achim (Babes-Bolyai University)

We present the platform titled „Intelligent analysis and prediction of the economic and financial crime in a cyber-dominated and interconnected business world (FinCrime)” (https://fincrime.net/en/platform).  The FinCrime platform aims to be an innovative platform for Romania, providing the following information: It establishes a common database of worldwide economic and financial crime statistics, including various European, national, regional and county levels. It allows the calculation of an economic-financial crime score that provides a global assessment of the volume of economic and financial crime for the main types of crimes (corruption, shadow economy, money laundering, and cybercrime)

It makes interactive maps that allow us to easily visualize the level of economic and financial crime at the world, European, national, and county levels. it forecasts the crime-related risks associated with companies and industries. It provides an overview of geographical areas (regions, countries, counties of Romania) prone to a higher risk of crime, respectively, a higher risk of non-compliance with laws, regulations, and contracts (compliance risk).

Textual Data Analytics in Finance

Chuan-Ju WANG (Academia Sinica)

In this talk, we embark on a chronological journey through our lab's contributions to the field of text analytics in financial reports, emphasizing the evolution of methodologies in line with advancements in natural language processing (NLP) technologies from 2013 to 2023. This period marks a significant transition in how textual data is leveraged in financial analysis, mirroring the broader advancements in NLP techniques and computational power. Our exploration is structured around three pivotal studies:

  1. Textual Sentiment Analysis for Financial Risk Prediction (circa 2013-2016): An early exploration of sentiment extracted from financial texts to predict market trends and risk, highlighting the initial steps in financial NLP.
  2. Financial Keyword Expansion via Continuous Word Vector Representations (circa 2015-2017): A leap into more sophisticated NLP with the introduction of word vector representations, marking a shift towards understanding the nuanced context of financial terminology.
  3. A Multistage Pipeline for Uncovering Financial Signals in Financial Reports (circa 2020-2023): Showcasing the latest in NLP, this study integrates state-of-the-art models for deep insights into financial texts, reflecting the peak of NLP's application in finance during this period.

This talk emphasizes the timeline of advancements in NLP technologies and their impact on finance, illustrating the critical role of time-dependent evolution in the field's progress.

A safe Recomm System for knowledge platforms

Xiaorui ZUO (NUS, SG)

In the digital age, platforms like Coursera and   Quantinar serve as distributed knowledge proliferation environments, leveraging deep individual data sources that embody the distinct preferences and styles of contributors, whether they be students, researchers, or instructors. The nature of these platforms necessitates robust privacy-preserving mechanisms due to the sensitive personal data involved. One promising solution is the implementation of Encrypted Recommendation Systems utilizing Secure Multi-party Computation (SMPC). This approach allows multiple stakeholders to collaboratively compute recommendations based on encrypted graph data, ensuring that individual preferences and inputs remain confidential. By integrating SMPC into the quantinar.com recommendation system, it is guaranteed that this platform achieves user trust and security, hence promoting a safer and more personalized learning experience.