Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Leveraging Large Language Models for Institutional Portfolio Management: Persona-Based Ensembles

Yoshia Abe, Shuhei Matsuo, Ryoma Kondo, Ryohei Hisano

TL;DR

This work evaluates large language models for institutional portfolio management by predicting price movements of a stocks-and-bonds portfolio and adjusting positions accordingly. It examines how different investor personas within LLMs, combined via mode ensembles, affect predictive accuracy and downstream investment performance. The study finds that LLM-driven strategies outperform Buy-and-hold on the Sharpe ratio during rising CPI periods, while traditional approaches can outperform during CPI declines or sharp drawdowns; results vary across other metrics and market regimes. The findings suggest LLMs can augment institutional decision-making, particularly when paired with regime-aware ensembles and complementary strategies to handle diverse market conditions.

Abstract

Large language models (LLMs) have demonstrated promising performance in various financial applications, though their potential in complex investment strategies remains underexplored. To address this gap, we investigate how LLMs can predict price movements in stock and bond portfolios using economic indicators, enabling portfolio adjustments akin to those employed by institutional investors. Additionally, we explore the impact of incorporating different personas within LLMs, using an ensemble approach to leverage their diverse predictions. Our findings show that LLM-based strategies, especially when combined with the mode ensemble, outperform the buy-and-hold strategy in terms of Sharpe ratio during periods of rising consumer price index (CPI). However, traditional strategies are more effective during declining CPI trends or sharp market downturns. These results suggest that while LLMs can enhance portfolio management, they may require complementary strategies to optimize performance across varying market conditions.

Leveraging Large Language Models for Institutional Portfolio Management: Persona-Based Ensembles

TL;DR

This work evaluates large language models for institutional portfolio management by predicting price movements of a stocks-and-bonds portfolio and adjusting positions accordingly. It examines how different investor personas within LLMs, combined via mode ensembles, affect predictive accuracy and downstream investment performance. The study finds that LLM-driven strategies outperform Buy-and-hold on the Sharpe ratio during rising CPI periods, while traditional approaches can outperform during CPI declines or sharp drawdowns; results vary across other metrics and market regimes. The findings suggest LLMs can augment institutional decision-making, particularly when paired with regime-aware ensembles and complementary strategies to handle diverse market conditions.

Abstract

Large language models (LLMs) have demonstrated promising performance in various financial applications, though their potential in complex investment strategies remains underexplored. To address this gap, we investigate how LLMs can predict price movements in stock and bond portfolios using economic indicators, enabling portfolio adjustments akin to those employed by institutional investors. Additionally, we explore the impact of incorporating different personas within LLMs, using an ensemble approach to leverage their diverse predictions. Our findings show that LLM-based strategies, especially when combined with the mode ensemble, outperform the buy-and-hold strategy in terms of Sharpe ratio during periods of rising consumer price index (CPI). However, traditional strategies are more effective during declining CPI trends or sharp market downturns. These results suggest that while LLMs can enhance portfolio management, they may require complementary strategies to optimize performance across varying market conditions.

Paper Structure

This paper contains 15 sections, 5 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methods
  4. Task Definition
  5. Prompt Design
  6. Experiments
  7. Overview of the Experimental Procedure
  8. Experiment 1: Impact of Personas and Ensembles on Prediction Accuracy
  9. Experiment 2: Adaptive Investment Strategies with LLM Predictions
  10. Results
  11. Results of Experiment 1: Impact of Personas and Ensembles on Prediction Accuracy
  12. Results of Experiment 2: Adaptive Investment Strategies with LLM Predictions
  13. Results of Experiment 2: Comparison of Investment Strategies Across Different Time Periods
  14. Qualitative Analysis of the Reasoning of the LLM
  15. Conclusion

Figures (5)

  • Figure 1: Overview of our approach.
  • Figure 2: Accuracy of five trials for each persona condition (short, medium, or long term)
  • Figure 3: F1-score (Class 1) of five trials for each persona condition (short term, medium term, and long term)
  • Figure 4: Changes in the positions of the buy-and-hold, CM(D2), RG(S5), RG(S10), Pattern 1, Pattern 2, and Pattern 3 strategies. The background colors indicate the CPI trend: white for High, dark gray for Low, and light gray for periods outside the evaluation.
  • Figure 5: Changes in the portfolio values of the buy-and-hold, CM(D2), RG(S5), RG(S10), pattern 1, pattern 2, and pattern 3 strategies. The background colors indicate the CPI trend: white for High, dark gray for Low, and light gray for periods outside the evaluation.