Diffusion Models for Financial Time-Series Generation

3 posts · Sep 2021 – Jun 2025

What this trail covers

This post continues the exploration of diffusion models for financial time-series generation, building upon previous work. It details the application of Large Language Models (LLMs) to financial news for sentiment analysis and prediction, and the use of Transformers for modeling both tick-by-tick order book events and cross-sectional return data. The LLM section discusses prompt engineering and the impact of model size and examples on accuracy. The Transformer section covers data representation. This post further breaks down the application of AI to trading into three subproblems: Prediction, Optimization, and Execution, detailing the challenges and approaches for each within the context of financial markets.

All Hudson River Trading trails

Jun 2025Intern Spotlight: HRT AI Labs ProjectThis post details the exploration and application of diffusion models for generating stock return sequences. It covers techniques for conditioning generation on stock metadata (e.g., volume, volatility), modeling joint cross-sectional returns by treating stocks as channels, and using inpainting-like methods to generate returns conditioned on other stock trajectories. The post also addresses the challenge of defining evaluation metrics for synthetic time-series data and contrasts diffusion models with autoregressive models like transformers and GANs, noting the former's avoidance of mode collapse.
Dec 2022Intern Spotlight: 2022 HRT AI Labs Summer ProjectsThis post introduces the application of Large Language Models (LLMs) to financial news for sentiment analysis and prediction, detailing prompt engineering techniques and the impact of model size and examples on accuracy. It also explores the use of Transformer neural networks for modeling financial data at two scales: tick-by-tick order book events and minute-to-minute returns across a cross-section of instruments. For order book data, it discusses representation challenges and the integration of Variational Autoencoders (VAEs). For cross-sectional returns, it compares regression and classification objectives, highlighting the trade-offs in capturing correlations versus avoiding degenerate trajectories.
Sep 2021Applying Artificial Intelligence to TradingThis post decomposes the broad problem of applying AI to trading into three distinct subproblems: Prediction, Optimization, and Execution. It analyzes the challenges of applying AI techniques from board games to trading, highlighting differences in zero-sum nature, rule knowledge, state observation, and data availability. For each subproblem, it outlines specific AI tasks and considerations relevant to financial markets, such as supervised learning for price prediction, financial mathematics for portfolio optimization, and reinforcement learning or other AI for trade execution.
Advertisement