Sponsored
By:
Key Dates
Location
205A, COEX in Gangnam, Seoul, South Korea
Workshop Contact
Prof. Yongjae Lee
yongjaeleee@gmail.com
The finance sector is seeing a rapid increase in the application of machine learning and AI, with Large Language Models (LLMs), ESG (Environmental, Social, and Governance) investing, and AI Safety significantly reshaping the field. This workshop focuses on how these advancements intersect with core financial AI applications. We will foster interdisciplinary discussion on applying LLMs to finance, addressing challenges in multilingual and non-English markets like Korea. The event will also highlight the integration of ESG signals into algorithmic decision-making and explore AI Safety, emphasizing reliability, fairness, and explainability for AI systems in regulated financial environments. By bringing together experts from academia, industry, and regulatory bodies, the workshop aims to stimulate discussions on practical issues, ethical dilemmas, and cutting-edge research shaping financial AI's future. We welcome submissions that combine technical rigor with societal relevance in AI-driven financial decisions.
Yongjae Lee
UNIST
Keynote #1
LLMs in Financial Markets: Forecasting Signals, Agent-Based Trading, and Memorization Risks
Alejandro Lopez-Lira
University of Florida
Abstract: This talk synthesizes evidence on what large language models (LLMs) can—and cannot—do for investment practice today.
(i) Forecasting from text: Building on “Can ChatGPT Forecast Stock Price Movements?”, I show that off-the-shelf LLMs extract tradable signals from news headlines, with out-of-sample return predictability that is strongest for smaller firms and after negative news. I’ll discuss data/engineering choices practitioners care about (timing alignment, transaction costs, and model governance).
(ii) LLM trading agents: From “Can Large Language Models Trade?”, I present a framework that prompts LLMs to act as coherent trading agents (e.g., value, momentum, market making) and interact in simulated markets. We examine when these agents add liquidity or amplify volatility, how strategy design affects stability, and what this implies for deployment and risk controls.
(iii) Memorization and evaluation risk: Drawing on “The Memorization Problem,” I demonstrate that LLMs can recall pre–knowledge-cutoff economic data with high fidelity, confounding historical “forecasting” tests. I’ll offer practical protocols to reduce leakage—post-cutoff evaluations, time-locked inference, masked/perturbed holdouts, and auditable data pipelines.
Keynote #2
9:50 AM - 10:30 AM
How To Choose A Threshold For An Evaluation Metric For Large Language Models
Dhagash Mehta
BlackRock
Abstract: To ensure and monitor large language models (LLMs) reliably, various evaluation metrics have been proposed in the literature. However, there is little research on prescribing a methodology to identify a robust threshold on these metrics even though there are many serious implications of an incorrect choice of the thresholds during deployment of the LLMs. Translating the traditional model risk management (MRM) guidelines within regulated industries such as the financial industry, in this talk, I will discuss a step-by-step recipe for picking a threshold for a given LLM evaluation metric. I will emphasize that such a methodology should start with identifying the risks of the LLM application under consideration and risk tolerance of the stakeholders. We then propose concrete and statistically rigorous procedures to determine a threshold for the given LLM evaluation metric using available ground-truth data.
Coffee Break
10:30 AM - 11:00 AM
Keynote #3
11:00 AM - 11:45 AM
New Frontiers of Generative AI in Finance: Structured Agentic Workflows and Foundational Time-Series Models
Hao Ni
Abstract: Building accurate and trustworthy models for financial time-series data remains a core challenge in quantitative finance. This talk presents two complementary research directions advancing the frontiers of generative AI in finance: structured agentic workflows and domain-adapted foundation models.
In the first part, I will introduce TS-Agent, a modular framework that integrates the systematic optimization of AutoML with the reasoning and adaptability of agentic AI. TS-Agent automates time-series modelling through iterative stages of model selection, code refinement, and fine-tuning, guided by curated knowledge banks. Combining the strengths of both paradigms, it enables adaptive, transparent, and auditable model development, consistently outperforming strong AutoML and agentic baselines across diverse forecasting and generative tasks.
In the second part, I will present our empirical study on time-series foundation models (TSFMs) on financial data. Using a comprehensive dataset of daily excess returns across global markets, we systematically evaluate zero-shot inference, fine-tuning, and pre-training from scratch. Pre-training on financial data and synthetic data augmentation yields substantial gains in predictive accuracy and portfolio performance, underscoring the promise of TSFM in finance.
Invited Talk #1
11:45 AM - 12:15 PM
Agents, Tools, and Truth in Finance
Nicole Cho
JP Morgan AI Research
Abstract: Large language models are rapidly shifting from single-shot predictors to agentic systems that plan, act, and reflect while invoking external tools. Yet many barriers keep LLM agents from becoming trustworthy systems in finance - including the difficulty in knowing when an agent will hallucinate, and how to orchestrate tools so the agent actually follows evidence. This 30-minute talk will focus on recent work presented at AAAI, NeuRIPs, EMNLP, COLING, ICAIF to drive closer to a unified playbook for building trustworthy LLM agents in finance. We will dive deeper into inverse reinforcement learning, hallucination prediction methods during inference, and grounding general models with inductive biases through the use of domain-specific tools.
Invited Talk #2
12:15 PM - 12:45 PM
Harnessing AI in Economic Statistics
Dr. Sojung Kim
Bank of Korea
Abstract: International organizations and central banks are increasingly exploring the use of artificial intelligence in economic statistics, leveraging high-frequency and alternative data for real-time analysis and policy decisions. This talk presents recent research by the Bank of Korea, focusing on the use of language models to analyze web-based and textual data sources.
The first part of the talk focuses on the News Sentiment Index, an experimental statistic derived from economic news. We introduce a methodology to filter economically meaningful articles and classify sentiment at both sentence and article levels, primarily through fine-tuning encoder-based models. We also discuss experiments using small LLMs with various prompting strategies for sentiment analysis.
The second part highlights an application of LLMs in official statistical research, measuring data intensity in labor markets to classify occupational text data. This study provides a systematic comparison between traditional rule-based NLP pipelines and LLM-based approaches.
Through these examples, the talk demonstrates how LLMs can complement conventional econometric methods, enabling more responsive, explainable, and scalable approaches to economic measurement while addressing remaining challenges.
Lunch Break
12:45 PM - 1:45 PM
Invited Talk #3
1:45 PM - 2:05 PM
AI Transformation in KFTC
Seonkyu Lim
KFTC
Abstract: This presentation introduces the AI-driven initiatives led by the Korea Financial Telecommunications and Clearings Institute (KFTC). Building on its nationwide payment infrastructure, KFTC is advancing graph analytics for anti-money laundering, fraudulent activity monitoring in Onnuri gift certificates, and representation learning tailored for financial fraud analysis. KFTC also contributes to the national World Best LLM Project by developing a financial domain benchmark and promoting a synthetic data ecosystem to accelerate innovation in financial AI.
Invited Talk #4
2:05 PM - 2:25 PM
Toward Reliable and Explainable Fiscal–Economic Generative AI
Yeonhee Lee
ETRI
Abstract: Recent advances in generative artificial intelligence have expanded efforts to apply such models to economic forecasting and narrative generation. While statistical generative AI shows promise across policy, finance, and industrial analysis, it still struggles to fully reflect the structural principles and real-world dynamics of economic systems. This talk introduces a new approach — Structural Generative AI — grounded in the concept of a Fiscal–Economic Digital Twin. The proposed framework embeds economic logic, policy mechanisms, and inter-sectoral interactions directly into the generative process, aiming to bridge the gap between statistical generative modeling and structurally consistent economic simulation. We explore the complementary potential of these two paradigms by comparing results from generative AI–based statistical models and digital-twin–based structural simulations under fiscal and macroeconomic scenarios. By combining structural consistency with generative flexibility, this research envisions a path toward sustainable, explainable, and trustworthy AI, ultimately evolving into an “AI Policy Assistant” that supports fiscal and economic decision-making.
Invited Talk #5
2:25 PM - 2:45 PM
Financial AI Data Services and Models at Korea Credit Information Services(KCIS)
Jongin Choi
K Credit
Abstract: As a comprehensive credit information agency in Korea, KCIS provides various financial AI services and models for financial institutions, the public sector, and academia. We provide models, datasets and services such as synthetic data generation models, persona databases and open datasets for AI learning. In this presentation, we introduce and share how to support research, innovation, and responsible AI in finance.
Invited Talk #6
2:45 PM - 3:15 PM
Foundation Model Challenges and Opportunities in Financial Services
Eric Jiawei He
RBC Borealis
Abstract: Financial services are at the core of our economy. Opportunities for machine learning abound in this space, from capital markets to insurance services to wealth management to lending to tools that assist clients in managing their money. Modern machine learning methods have transformed industries, yet particular challenges exist in realizing the full potential of machine learning in financial services. These include explainability, data imbalance, partial observations, distribution shift, and self-supervised learning in low-signal settings. I will describe the ATOM foundation model, which specializes in learning from asynchronous event sequences, to maximally utilize the richness of transactional data in financial services.
Coffee Break
3:15 PM - 4:00 PM
Decision by Supervised Learning with Deep Ensembles: A Practical Framework for Robust Portfolio Optimization
Juhyeong Kim
Mirae Assets
Abstract: We propose Decision by Supervised Learning (DSL), a practical framework for robust portfolio optimization. DSL reframes portfolio construction as a supervised learning problem: models are trained to predict optimal portfolio weights, using cross-entropy loss and portfolios constructed by maximizing the Sharpe or Sortino ratio. To further enhance stability and reliability, DSL employs Deep Ensemble methods, substantially reducing variance in portfolio allocations. Through comprehensive backtesting across diverse market universes and neural architectures, shows superior performance compared to both traditional strategies and leading machine learning-based methods, including Prediction-Focused Learning and End-to-End Learning. We show that increasing the ensemble size leads to higher median returns and more stable risk-adjusted performance. The code is available at https://github.com/DSLwDE/DSLwDE.
NMIXX: Domain-Adapted Neural Embeddings for Cross-Lingual eXploration of Finance
Hanwool Lee
Shinhan Securities
Abstract: General-purpose sentence embedding models often struggle to capture specialized financial semantics—especially in low-resource languages like Korean—due to domain-specific jargon, temporal meaning shifts, and misaligned bilingual vocabularies. To address these gaps, we introduce NMIXX (Neural eMbeddings for Cross-lingual eXploration of Finance), a suite of cross-lingual embedding models fine-tuned with 18.8,K high-confidence triplets that pair in-domain paraphrases, hard negatives derived from a semantic-shift typology, and exact Korean-English translations. Concurrently, we release KorFinSTS, a 1,921-pair Korean financial STS benchmark spanning news, disclosures, research reports, and regulations, designed to expose nuances that general benchmarks miss. When evaluated against seven open-license baselines, NMIXX’s multilingual bge-m3 variant achieves Spearman’s rho gains of +0.10 on English FinSTS and +0.22 on KorFinSTS—outperforming its pre-adaptation checkpoint and surpassing other models by the largest margin—while revealing a modest trade-off in general STS performance. Our analysis further shows that models with richer Korean token coverage adapt more effectively, underscoring the importance of tokenizer design in low-resource, cross-lingual settings. By making both models and benchmark publicly available, we provide the community with robust tools for domain-adapted, multilingual representation learning in finance.
Invited Talk #7
4:30 PM - 5:00 PM
Engineering an AI-native Hedge Fund: From Data Infrastructure to Intelligent Agents
Joo Lee
Arrowpoint Investment Partners
Jin Kim
LinqAlpha
Invited Talk #8
5:00 PM - 5:20 PM
From Forecasting to Understanding: Explainable Financial AI in Real World Markets
Wonbin Ahn
LG AI Research
Abstract: Forecasting has always been at the heart of financial AI, but the next challenge lies in understanding why markets move. This talk introduces EXAONE Business Intelligence (EXAONE-BI), LG AI Research’s agentic framework for explainable forecasting powered by large language models. By combining time-series modeling with unstructured data such as news, filings, and analyst reports, EXAONE-BI integrates multi-modal reasoning and LLM-based explainability to generate both forecast signals and natural language interpretations that reveal the underlying drivers of market movements. We highlight its deployment in real financial ecosystems, including AI-managed ETFs (LQAI) and insight generation with the London Stock Exchange Group (LSEG), demonstrating how explainable AI can operate reliably in live market environments. The talk concludes with reflections on how large language models can connect data, reasoning, and decision-making to advance responsible and transparent financial intelligence.
Panel Discussion
5:20 PM - 6:00 PM
From Models to Markets: Deploying AI Responsibly in Financial Systems
Bhaskarjit Sarma
Domyn
Hye-young Yoon
KFTC
Eric Jiawei He
RBC Borealis
Sojung Kim
Bank of Korea
Joo Lee
Arrowpoint
Jin Kim
Moderator
Yongjae Lee
UNIST
Signed Bridge Pruning Framework : Improving Structural Reliability in Financial Knowledge Graphs
Jeongseon Kim, Yeonhee Lee, Sungsu Lim
A Double-Edged Sword: Benchmarking the Trade-off Between Bias Mitigation and Helpfulness of LLM Guardrails in Finance
Sungjun Lim, Hoyoon Byun, Jihee Kim, Kyungwoo Song
Multilingual Conversational AI for Financial Assistance: Bridging Language Barriers in Indian FinTech
Bharatdeep Hazarika, Arya Suneesh, Prasanna Devadiga, Pawan Kumar Rajpoot, Anshuman B Suresh, Ahmed Ifthaquar Hussain
LLM-Enhanced Black-Litterman Portfolio Optimization
Youngbin Lee, Yejin Kim, Juhyeong Kim, Suin Kim, Yongjae Lee
FPQ-VAE: A Dynamic Factor Model Fusing Financial Priors and Vector-Quantized Factors for Stock Rank Prediction
Namhyoung Kim, Janghyuk Youn, Jae Wook Song
FinNuE: Exposing the Risks of Using BERTScore for Numerical Semantic Evaluation in Finance
Yu-Shiang Huang, Yun-Yu Lee, Chou Tzu Hsin, Che Lin, Chuan-Ju Wang
Interpreting LLMs as Credit Risk Classifiers: Do Their Feature Explanations Align with Classical ML
Saeed Bark AlMarri, Kristof Juhasz, Mathieu Ravaut, Gautier Marti, Hamdan Ali Al Ahbabi, Ibrahim M. Elfadel
Decision by Supervised Learning with Deep Ensembles: A Practical Framework for Robust Portfolio Optimization
Juhyeong Kim, Sungyoon Choi, Youngbin Lee, Yejin Kim, Yongmin Choi, Yongjae Lee
NMIXX: Domain-Adapted Neural Embeddings for Cross-Lingual eXploration of Finance
Hanwool Lee, Yewon hwang, Jonghyun Choi, Heejae ahn, Sung Bum Jung, Youngjae Yu
Modeling Corporate Health Rating Using Online Business Reviews
Makanjuola Adekunmi Ogunleye, Nazanin Zaker, Xue Han, Jeremy Krohn
FINCH: Financial Intelligence using Natural language for Contextualized SQL Handling Avinash
Kumar Singh, Bhaskarjit Sarmah, Stefano Pasquali
Verified LRM Agents for Finance: Economics-Informed World Models and Abstract Knowledge Management for Safe and Responsible Decisioning
David Scott Lewis, Anar Batkhuu, Haley Yi, Enrique Zueco, Zijin Wu
VAP: Preventing Plausibility Traps in Algorithmic Trading with Physics-Informed Learning and Formal Verification
David Scott Lewis, Enrique Zueco
UIAGR:Up-to-date Infomation Awareness Generative Retrieval for Alipay Fund Search
Jingyuan Wen, Hong Liu, chenglei shen, Gang Yang, yedan shen, Kaixin Wu, mingjie zhong, Xu Jia, Linjian Mo
Towards Macroeconomic Policy Analysis with LLM-Agent Simulations
Taewan You
Earnings-Call Tone and Market Reactions: Evidence from Korea with a Text-Based Trading Strategy
Eunchong Kim, Jaehee Jang, NakYoung Lee
Collaborative Multi-Agent LLMs for Autonomous Portfolio Management: An Empirical Study of Distributed AI Decision-Making in Financial Markets
Quang-Vinh Dang, Ngoc-Son-An Nguyen
SLRL: Strategy-Level Reinforcement Learning for Robust Portfolio Management
Junyoung Park, SeungEun Ock, Yosep na, Young Hoon Choi, Jae Wook Song
GuruAgents: Emulating Wise Investors with Prompt-Guided LLM Agents
Yejin Kim, Youngbin Lee, Juhyeong Kim, Yongjae Lee
Knowledge Graph Construction for Stock Markets with LLM-Based Explainable Reasoning
Cheonsol Lee
Algorithmic Echo Chambers: Interpretation Cascades, Model Risk Personas, and Systemic Risk in LLM-Driven Financial Workflows
Anar Batkhuu, David Scott Lewis, Enrique Zueco, Haley Yi
FinAI Data Assistant: LLM-based Financial Database Query Processing with the OpenAI Function Calling API
Juhyeong Kim, Yejin Kim, Youngbin Lee, Hyunwoo Byun
Your AI, Not Your View: The Bias of LLMs in Investment Analysis
Hoyoung Lee, Junhyuk Seo, Suhwan Park, Junhyeong Lee, Wonbin Ahn, CHANYEOL CHOI, Alejandro Lopez-Lira, Yongjae Lee
Contact
