Paper: arXiv 2608.12283
Authors: Alireza Kargarzadeh, Nariman Khaledian, Navid Parvini, Arman Khaledian
Abstract
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. We study an uncertainty-aware construction that feeds model-predicted risk – decomposed into aleatoric and epistemic components – directly into the covariance matrix of portfolio allocators, rather than treating portfolio risk as fixed or adjusting only expected returns. We evaluate the pipeline on Russell 2000 equities under three stock-selection regimes: a pure-alpha trigger that isolates abnormal stock moves not explained by macro indicators, a pure-beta trigger that captures macro-indicator moves before the stock itself fires, and a beta trigger in which both channels agree. Across the full holding-period grid, the separated pure-alpha and pure-beta legs usually dominate the beta intersection on Sharpe and return. Two horizons are especially informative. At one day, pure beta can work under low and moderate transaction costs because it captures immediate lead-lag spillovers from liquid macro and sector indicators into exposed small-cap stocks, but this advantage disappears at 100 bps when turnover and microstructure noise dominate. At 40 days, pure beta works for a different reason: slower macro repricing overtakes the firm-specific pure-alpha channel. The strongest conservative row is pure beta with GPT-4o mini sentiment, a Student-t target, a 40-day holding period, and risk parity allocation, reaching Sharpe 2.33 at 100 bps. The results suggest that stock-selection regime and allocator choice matter at least as much as the sentiment model, and that separating firm-specific and macro-exposure triggers is more informative than requiring both to fire simultaneously.
Complexity vs Empirical Score
- Math Complexity: 6.0/10
- Empirical Rigor: 8.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper presents a sophisticated approach to portfolio construction, integrating LLM sentiment with macroeconomic and technical signals, and rigorously evaluates it across various dimensions. The methodology for decomposing risk into aleatoric and epistemic components and feeding it into the covariance matrix is particularly noteworthy. The empirical evaluation is extensive, covering different stock-selection regimes, holding periods, and transaction costs.
Research Flowchart
flowchart TD
A["Research Goal: LLM-Driven Small-Cap Trading"] --> B("Methodology: Uncertainty-Aware Portfolio Construction");
B --> C{"Inputs: Financial News, Macro Indicators, Technical Signals"};
C --> D["LLM Processing: Extract Ricker Signals, Decompose Risk (Aleatoric & Epistemic)"];
D --> E("Portfolio Allocation: Covariance Matrix Integration, Risk Parity");
E --> F{"Stock Selection Regimes: Pure-Alpha, Pure-Beta, Beta Intersection"};
F --> G["Outcomes: Sharpe Ratio, Return Across Holding Periods (1-day, 40-day)"];
G --> H{"Key Findings: Pure-Alpha & Pure-Beta Dominate; Regime & Allocator Matter; Pure-Beta Strongest (Sharpe 2.33)"};
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes an uncertainty‑aware portfolio construction that feeds model‑predicted risk, split into aleatoric and epistemic components, directly into the covariance matrix used by allocators.
Source sentence
We study an uncertainty-aware construction that feeds model-predicted risk -- decomposed into aleatoric and epistemic components -- directly into the covariance matrix of portfolio allocators, rather than treating portfolio risk as fixed or adjusting only expected returns.
Assumes 6
- The aleatoric risk is modeled with either a Gaussian or Student‑t distribution and the model is trained by maximizing likelihood under that distribution. model
Source sentence
The model is trained by maximizing the likelihood of realized returns under its own predicted aleatoric distribution, Gaussian or Student-t (full training objective in Appendix E), so risk is learned jointly with return rather than fit afterward from residuals.
- All data used for universe selection, standardization, model fitting, and early stopping are confined to an in‑sample period ending 31 Dec 2024. data
Source sentence
All universe selection, standard‑ization, model fitting, and early stopping use only the in‑sample period ending December 31, 2024.
- Transaction costs are represented by a set of fixed basis‑point scenarios plus a constant 5 bps one‑way slippage. execution
Source sentence
Transaction-cost scenarios are {0, 1, 2, 5, 10, 20, 50, 100} basis points, with a fixed 5 bps one-way slippage component in addition to the reported transaction-cost setting.
- Holding periods are limited to the discrete set {1,2,3,5,10,20,40,60} trading days. other
Source sentence
Holding periods are H ∈{1, 2, 3, 5, 10, 20, 40, 60} trading days.
Show 2 more
- The predictive covariance matrix is constructed by summing aleatoric and epistemic covariances, with epistemic covariance zero when S=1. model
Source sentence
We combine both into a single predictive covariance, ˆΣt = ˆΣA t + ˆΣE t (6) extending it from a single-signal quantity to a full N × N cross-asset covariance matrix. With S = 1, epistemic covariance is exactly zero, recovering a model with no explicit account of its own uncertainty.
- Backtesting is performed on a single out‑of‑sample calendar year (2025), so annualized metrics are summaries of one test year. data
Source sentence
Because the out-of-sample period is one calendar year, annualized metrics should be read as annualized summaries of a single test year rather than as estimates over many independent years.
Data
2025 – 2026-01 · Russell 2000 · equities · daily
- Russell 2000 equities daily OHLCV market-prices unknown
Source sentence
We evaluate the pipeline on Russell 2000 equities with daily OHLCV data from October 2, 2023 through December 31, 2025
- Scored news articles text-news unknown
Source sentence
scored news from October 1, 2023 through December 31, 2025
- Macro‑indicator panel macro unknown
Source sentence
Macro-beta selection additionally uses a macro‑indicator panel spanning January 1, 2022 through January 1, 2026
Costs
Modeled
Costs are modeled as explicit basis‑point scenarios (0–100 bps) plus a fixed 5 bps one‑way slippage.
Source sentence
Transaction-cost scenarios are {0, 1, 2, 5, 10, 20, 50, 100} basis points, with a fixed 5 bps one-way slippage component in addition to the reported transaction-cost setting.
Finds 5
Sharpe 2.33
- The pure‑beta leg with GPT‑4o mini sentiment, Student‑t target, 40‑day holding, and risk‑parity allocation achieves a Sharpe of 2.33 at 100 bps transaction costs. performance
Source sentence
The strongest conservative row is pure beta with GPT-4o mini sentiment, a Student-t target, a 40-day holding period, and risk parity allocation, reaching Sharpe 2.33 at 100 bps.
- Separated pure‑alpha and pure‑beta legs usually dominate the beta intersection on Sharpe and return across the holding‑period grid. performance
Source sentence
Across the full holding-period grid, the separated pure-alpha and pure-beta legs usually dominate the beta intersection on Sharpe and return.
- Pure‑alpha leads at 5, 10, and 60 trading days; pure‑beta leads at 20 and 40 days, with pure‑beta’s one‑day edge disappearing under 100 bps costs. performance
Source sentence
Pure alpha leads at 5, 10, and 60 trading days; pure beta leads at 20 and 40 days. The one-day result is cost-sensitive: pure beta is strongest at low costs, but the edge disappears under the 100 bps stress case.
- Transaction‑cost robustness shows pure‑alpha remains strongest from zero to 100 bps at a 60‑day horizon, with its lead widening as costs rise. performance
Source sentence
pure alpha remains the strongest filtered regime from zero costs through 100 bps; if anything, its lead over pure beta and the beta intersection widens rather than narrows as costs rise.
- The way the tradable set is defined (pure‑alpha, pure‑beta, or their intersection) matters at least as much as the sentiment model or allocator choice. other
Source sentence
The central result is that the way the tradable set is defined matters at least as much as the sentiment backend or allocator.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 29 related papers). Coverage is partial, so this is neither replication nor refutation.
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.