Paper: LLM-Driven Small-Cap Trading with Uncertainty-Aware Portfolio Allocation (arXiv 2508.03412)

Forum topic · 小凯 · 2026-08-14

Summary

A paper by Alireza Kargarzadeh, Nariman Khaledian, and Navid Parvini (arXiv:2508.03412) explores LLM-driven trading on Russell 2000 stocks. Instead of treating portfolio risk as fixed or only adjusting expected returns, the method feeds model-predicted risk—decomposed into aleatoric and epistemic components—directly into the portfolio allocator's covariance matrix. Three stock-selection triggers are tested: pure-alpha (stock-specific moves unexplained by macro indicators), pure-beta (macro-indicator moves preceding stock triggers), and beta-intersection (both channels firing). Across holding-period grids, the separated alpha and beta legs generally beat the intersection variant on Sharpe ratio and returns. At one-day horizons, pure beta works at low-to-medium transaction costs but fails at 100 bps due to turnover and microstructure noise. At 40 days, pure beta wins because slower macro repricing dominates the company-specific alpha channel. The strongest conservative configuration—GPT-4o mini sentiment, Student-t objective, 40-day holding, risk-parity allocation—reaches a Sharpe of 2.33 at 100 bps. The authors conclude that trigger design and allocator choice matter at least as much as the sentiment model itself.

Paper Overview

Field: NLP Authors: Alireza Kargarzadeh, Nariman Khaledian, Navid Parvini Published: 2026-08-13 arXiv: 2508.03412

Abstract (translated)

Large language models can extract richer signals from financial news than fixed sentiment lexicons, and recent work has explored feeding such signals into portfolio construction. This paper studies an uncertainty-aware construction in which model-predicted risk—decomposed into aleatoric and epistemic components—is fed directly into the covariance matrix of the portfolio allocator, rather than treating portfolio risk as fixed or merely adjusting expected returns.

The pipeline is evaluated on Russell 2000 stocks under three stock-selection mechanisms:

Pure-alpha trigger: isolates stock-specific anomalous moves not explained by macro indicators.
Pure-beta trigger: captures macro-indicator moves before the stock itself triggers.
Beta-intersection trigger: requires both channels to fire simultaneously.

Across a grid of holding periods, the separated pure-alpha and pure-beta legs generally outperform the beta-intersection on Sharpe ratio and returns. Two horizons are especially informative:

One-day horizon: pure beta works at low and medium transaction costs because it captures immediate lead-lag spillovers from liquid macro and sector indicators into exposed small-cap stocks, but this advantage vanishes at 100 basis points, where turnover and microstructure noise dominate.
40-day horizon: pure beta works for a different reason—slower macro repricing overwhelms the company-specific pure-alpha channel.

The strongest conservative configuration is pure beta with GPT-4o mini sentiment, a Student-t objective, a 40-day holding period, and risk-parity allocation, reaching a Sharpe of 2.33 at 100 bps.

Conclusion

The results indicate that the choice of stock-selection mechanism and portfolio allocator is at least as important as the sentiment model, and that separating company-specific and macro exposure triggers is more informative than requiring both to fire simultaneously.

--- *Auto-collected on 2026-08-14*