Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Reasoning on Time-Series for Financial Technical Analysis

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

TL;DR

The paper tackles the challenge of interpretable stock forecasting by bridging time-series data and natural-language reasoning. It introduces Verbal Technical Analysis (VTA), which couples a reasoning LLM with a latent time-series backbone and uses textual annotations of price signals. Time-GRPO optimizes the reasoning traces with an inverse MSE reward, while joint conditional training gates forecasts on reasoning-derived attributes to preserve interpretability. Across US, Chinese, and European markets, VTA achieves state-of-the-art accuracy and generates reasoning traces that experts find highly informative, with investment portfolios built on its forecasts delivering strong risk-adjusted performance.

Abstract

While Large Language Models have been used to produce interpretable stock forecasts, they mainly focus on analyzing textual reports but not historical price data, also known as Technical Analysis. This task is challenging as it switches between domains: the stock price inputs and outputs lie in the time-series domain, while the reasoning step should be in natural language. In this work, we introduce Verbal Technical Analysis (VTA), a novel framework that combine verbal and latent reasoning to produce stock time-series forecasts that are both accurate and interpretable. To reason over time-series, we convert stock price data into textual annotations and optimize the reasoning trace using an inverse Mean Squared Error (MSE) reward objective. To produce time-series outputs from textual reasoning, we condition the outputs of a time-series backbone model on the reasoning-based attributes. Experiments on stock datasets across U.S., Chinese, and European markets show that VTA achieves state-of-the-art forecasting accuracy, while the reasoning traces also perform well on evaluation by industry experts.

Reasoning on Time-Series for Financial Technical Analysis

TL;DR

The paper tackles the challenge of interpretable stock forecasting by bridging time-series data and natural-language reasoning. It introduces Verbal Technical Analysis (VTA), which couples a reasoning LLM with a latent time-series backbone and uses textual annotations of price signals. Time-GRPO optimizes the reasoning traces with an inverse MSE reward, while joint conditional training gates forecasts on reasoning-derived attributes to preserve interpretability. Across US, Chinese, and European markets, VTA achieves state-of-the-art accuracy and generates reasoning traces that experts find highly informative, with investment portfolios built on its forecasts delivering strong risk-adjusted performance.

Abstract

While Large Language Models have been used to produce interpretable stock forecasts, they mainly focus on analyzing textual reports but not historical price data, also known as Technical Analysis. This task is challenging as it switches between domains: the stock price inputs and outputs lie in the time-series domain, while the reasoning step should be in natural language. In this work, we introduce Verbal Technical Analysis (VTA), a novel framework that combine verbal and latent reasoning to produce stock time-series forecasts that are both accurate and interpretable. To reason over time-series, we convert stock price data into textual annotations and optimize the reasoning trace using an inverse Mean Squared Error (MSE) reward objective. To produce time-series outputs from textual reasoning, we condition the outputs of a time-series backbone model on the reasoning-based attributes. Experiments on stock datasets across U.S., Chinese, and European markets show that VTA achieves state-of-the-art forecasting accuracy, while the reasoning traces also perform well on evaluation by industry experts.

Paper Structure

This paper contains 12 sections, 10 equations, 4 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Verbal Technical Analysis
  4. Problem Formulation
  5. Time-Series Reasoning
  6. Time-Series Forecasting
  7. Joint Conditional Training
  8. Experiments
  9. Results
  10. Conclusion
  11. List of Technical Indicators
  12. Evaluation Metrics

Figures (4)

  • Figure 1: The Verbal Technical Analysis (VTA) framework. We first teach an LLM to reason over time-series data. The reasoning outputs are used to condition a time-series forecasting model, to produce forecasts with similar attributes. This results in forecasts with interpretable reasoning traces.
  • Figure 2: Joint conditional training component.
  • Figure 3: Correctness reward over steps.
  • Figure 4: An example of the reasoning trace and forecasts produced by our VTA framework.