ICAIF 2025 Workshop | Rethinking Financial Time-Series

Call for Papers

Important Dates

(UTC-0)

Submission Deadline:

October 2nd, 2025
Author Notification:

October 15th, 2025
Camera-Ready Deadline:

October 31st, 2025
Workshop Date:

Nov 15th or 16th, 2025

Submit via OpenReview

Submission Guidelines

Authors must submit their paper (as a PDF) via the workshop submission site. At least one author of each accepted paper must attend the conference to present their work.

Format and Awards

Submissions are limited to 4 pages in length, excluding references and appendices. The paper format should be the same as the main ICAIF conference. We will be awarding a Best Paper Award to recognize outstanding contributions.

Review Process

The review process will be double-blind. There will be no rebuttal period.

Presentation and Proceedings

All accepted papers will be invited to a poster session. Participants are required to print and bring their own posters to the event. Detailed information regarding poster format and submission requirements will be provided by the organizers upon acceptance. Selected papers may also be given the opportunity for an oral presentation, subject to schedule constraints. The workshop is non-archival, and there will be no official proceedings. Only the names of the authors and the titles of the accepted papers will be posted on the website; the papers themselves will not be made public.

Accepted Papers

FinTSB: A Comprehensive and Practical Benchmark for Financial Time Series Forecasting Best Paper

Yifan Hu, Yuante Li, Peiyuan Liu, Naiqi Li, Tao Dai, Shu-Tao Xia, Dawei Cheng
Reasoning on Time-Series for Financial Analysis Best Paper

Kelvin J.L. Koa, Jan Chen, Yunshan Ma, Zheng Huanhuan, Tat-Seng Chua
MARCD: Tail-Aware, Regime-Expert Diffusion to Drive CVaR-Constrained Portfolio Decisions

Ali Atiah Alzahrani
Enhancing Portfolio Decisions Using Time Series Forecasting with LLMs

TAEKYUNG LIM, Jang Ho Kim
Robust Pricing of To-Be-Announced (TBA) securities

Matias Altamirano, Sophia Sosnina, Jan Szopinski, Miha Torkar, Saher Esmeir
A Neural Network Informer In Algorithmic Investment Strategies on High Frequency Bitcoin Data

Robert Ślepaczuk
LLM-guided Factor Generation for Explainable Coal Price Forecasting

Sangjin Jin*, Suhwan Park*, Kangmin Kim, Inwoo Tae, Juchan Kim, Hoyoung Lee, Junhyeong Lee, Yongjae Lee
Towards Causal Market Simulators

Dennis Thumm, Luis Ontaneda Mijares
Probabilistic Forecasting of High-Frequency Crypto Volatility via Decomposition-Integrated Recurrent Neural Networks

Yosep na, Jun Young Byun, Jungyoon Song, Namhyoung Kim, Jae Wook Song
Quantum-Inspired Image Encodings for Financial Time-Series Forecasting

Henry Woo, Gunnho Song, Taeyoung Park
Efficient Forward Curve Construction Using Neural Networks in Sparse Data Environments

Gagandeep Singh Kaler, Chenyu Zhao, Yury Krongauz, Dhagash Mehta

Invited Speakers

Agostino Capponi

Professor of Industrial Engineering and Operations Research at Columbia University.

Theis Ingerslev Jensen

Assistant Professor of Finance at the Yale School of Management.

Bo An

President's Chair Professor, College of Computing and Data Science at Nanyang Technological University.

Ying Chen

Department of Mathematics & Centre for Quantitative Finance, National University of Singapore

Savitha Ramasamy

Principal Scientist at the Agency for Science, Technology and Research (A*STAR).

Topics of Interest

Foundational Time-Series Principles

Principled time-series methods for heavy-tails, volatility clustering, and regime shifts.
Post-mortems of time-series model failures in live trading.
Robust benchmarks and backtesting protocols for financial time-series.
High-fidelity synthetic time-series generation (e.g., GANs, SDEs).

Frontiers in Time-Series Data & Models

Modeling non-stationary, multi-modal, and irregularly-sampled time-series.
Online learning and adaptation to distribution shifts in time-series.
Foundation models for financial time-series: Scaling laws and limitations.
Self-supervised representation learning for time-series.

New Paradigms for Time-Series Analysis

Causal discovery and inference from observational time-series data.
Interpretability of deep learning models for time-series forecasting (XAI).
Rigorous uncertainty quantification for probabilistic time-series forecasting.
Integrating market microstructure into time-series model design.

Schedule

09:00 - 09:05 am

Opening Remarks

09:05 - 09:35 am

Speaker Talk # 1 (Prof. Theis Ingerslev Jensen)

09:35 - 10:05 am

Speaker Talk # 2 (Prof. Agostino Capponi)

10:05 - 10:25 am

Best Papers Talks (# 1)

FinTSB: A Comprehensive and Practical Benchmark for Financial Time Series Forecasting

Yifan Hu, Yuante Li, Peiyuan Liu, Naiqi Li, Tao Dai, Shu-Tao Xia, Dawei Cheng

10:25 - 10:40 am

Coffee Break

10:40 - 11:00 am

Best Papers Talks (# 2)

Reasoning on Time-Series for Financial Analysis

Kelvin J.L. Koa, Jan Chen, Yunshan Ma, Zheng Huanhuan, Tat-Seng Chua

11:00 - 11:30 am

Speaker Talk # 3 (Prof. Bo An)

11:30 - 12:00 pm

Speaker Talk # 4 (Dr. Ramasamy Savitha)

12:00 - 12:30 pm

Speaker Talk # 5 (Prof. Chen Ying)

12:30 - 12:35 pm

Closing Remarks

12:35 pm onwards

Networking & Poster Session

Organizers

Yoontae Hwang

Pusan National University

Yongjae Lee

UNIST

Zhangyang (Atlas) Wang

UT Austin

Qingsong Wen

Squirrel AI Learning

Ming Jin

Griffith University

Stefan Zohren

University of Oxford

Dhagash Mehta

BlackRock, Inc.

Yaxuan Kong

University of Oxford

Changhee Lee

Korea University

Jaeho Kim

Korea University

Hao Ni

University College London

About the Workshop

Financial time-series analysis sits at the epicentre of today's algorithmic markets. However, its core foundations remain unsettled; the heavy-tailed noise, abrupt regime shifts, and market-microstructure frictions inherent to financial data continue to defy standard modelling assumptions. Ignoring these fundamentals has led to models that are often brittle in practice.

Simultaneously, the field is being reshaped by new frontiers in data and modelling. The proliferation of cross-modal signals from millisecond order-book updates and satellite feeds to generative-AI-curated news is overwhelming traditional pipelines. This data deluge has spurred the rise of large-scale foundation models, challenging us to reconcile immense scale with statistical soundness and avoid creating sophisticated yet unreliable black boxes.

This workshop provides a forum to confront these questions head-on. We solicit contributions that re-evaluate foundational principles, push the frontiers of model development, or chart future research paths. We welcome submissions on principled learning for volatile data, robustness and distribution shifts, novel benchmarks and evaluation protocols, case studies of model failures, and methods for causal inference or interpretability.