Temporal Relational Reasoning of Large Language Models for Detecting Stock Portfolio Crashes

Paper: arXiv 2410.17266

Abstract

Stock portfolios are often exposed to rare consequential events (e.g., 2007 global financial crisis, 2020 COVID-19 stock market crash), as they do not have enough historical information to learn from. Large Language Models (LLMs) now present a possible tool to tackle this problem, as they can generalize across their large corpus of training data and perform zero-shot reasoning on new events, allowing them to detect possible portfolio crash events without requiring specific training data. However, detecting portfolio crashes is a complex problem that requires more than reasoning abilities. Investors need to dynamically process the impact of each new piece of information found in news articles, analyze the relational network of impacts across different events and portfolio stocks, as well as understand the temporal context between impacts across time-steps, in order to obtain the aggregated impact on the target portfolio. In this work, we propose an algorithmic framework named Temporal Relational Reasoning (TRR). It seeks to emulate the spectrum of human cognitive capabilities used for complex problem-solving, which include brainstorming, memory, attention and reasoning. Through extensive experiments, we show that TRR is able to outperform state-of-the-art techniques on detecting stock portfolio crashes, and demonstrate how each of the proposed components help to contribute to its performance through an ablation study. Additionally, we further explore the possible applications of TRR by extending it to other related complex problems, such as the detection of possible global crisis events in Macroeconomics.

Complexity vs Empirical Score

• Math Complexity: 5.0/10
• Empirical Rigor: 7.5/10
• Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper involves complex graph algorithms (PageRank, temporal graph reasoning) and advanced LLM prompting techniques, indicating moderate-to-high mathematical and algorithmic complexity. It demonstrates strong empirical rigor through extensive experiments, ablation studies, and comparisons against state-of-the-art models across multiple time periods and portfolios, though it lacks explicit code/data disclosure or statistical metrics.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Detect Stock Portfolio Crashes<br>using LLMs without specific training"] --> B["Core Methodology:<br>Temporal Relational Reasoning TRR Framework"];
  B --> C{"Inputs"};
  C --> C1["Portfolio Constituents"];
  C --> C2["News Feeds / Events"];
  C --> C3["Historical Context"];
  C --> D["Computational Process:<br>Dynamic Impact Processing"];
  D --> E["Temporal Analysis:<br>Impacts across time-steps"];
  E --> F["Relational Analysis:<br>Network of impacts across events/stocks"];
  F --> G["LLM Zero-Shot Reasoning:<br>Emulating human cognition<br>Attention & Memory"];
  G --> H["Key Outcomes"];
  H --> H1["High Accuracy<br>Portfolio Crash Detection"];
  H --> H2["State-of-the-art<br>Performance"];
  H --> H3["Generalizability<br>e.g., Macro-Economic Crises"];

Cited References (86)

1. Diffusion Variational Autoencoder for Tackling Stochasticity in Multi-Step Regression Stock Price Prediction — → 2309.00073
2. Learning to Generate Explainable Stock Predictions using Self-Reflective Large Language Models — → 2402.03659
3. Portfolio Selection (1952) — link
4. Building diversified portfolios that outperform out of sample (2016)
5. The Black Swan: The Impact of the Highly Improbable (2007) — link
6. Global governance and systemic risk in the 21st century: Lessons from the financial crisis (2010)
7. COVID-19 pandemic and connectedness across financial markets (2021)
8. Large language models are zero-shot reasoners (2022)

3 of 86 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.