Generator--critic trading systems pair an LLM-based trader that makes trading decisions with a red team agent that evaluates these decisions and provides feedback for revision. However, improving individual trading decisions does not ensure that both agents learn from experience and improve together over time. This raises an important question: how can the trader and red-Team select useful experience and learn from each other's evolving behavior? On one hand, experience that benefits decisions under one market condition may become ineffective or harmful under another. On the other hand, trading outcomes reflect both agents' actions, making it difficult to derive appropriate experience for each role. In this work, we propose ACE-Trader, an adversarial co-evolution trading framework. Specifically, we first develop separate context-dependent memory selectors that learn to identify useful experience for the current trading decision. Both selectors receive shared feedback from net-return differences between complete memory-assisted and memory-free trading pipelines, with online updates emphasizing recent outcomes. We further design a co-evolution mechanism that links role-specific memories through shared trading records. We use these records to evaluate the Red-Team's risk assessments and the effects of the Trader's revisions against realized market outcomes, then derive experience for their respective memories. By linking this experience to the corresponding records, we enable the Trader to improve its proposals based on prior critiques and the Red-Team to adjust its review focus as the Trader's behavior changes. Experiments on U.S. stock trading show that ACE-Trader outperforms all 22 active baselines in cumulative return and achieves the highest Sharpe ratio and lowest maximum drawdown across both active methods and passive benchmarks.
Figures & tables
Figure 1: Comparison of memory mechanisms in trading agents.
Figure 2: Overview of ACE-Trader.
Category
Model
Cum. Ret. (%) ↑
Ann. Ret. (%) ↑
Sharpe ↑
MDD. (%) ↓
Vol. (%) ↓
HR. (%) ↑
Turnover
Passive
NASDAQ Buy & Hold
+12.82
+28.29
1.340
12.84
20.10
57.38
0.000
Equal-Weight Buy & Hold
+8.68
+18.77
1.375
5.33
13.14
52.48
0.000
ML
LightGBM
+4.86
+10.30
0.502
13.13
26.48
50.74
1.341
XGBoost
-21.18
-38.84
-1.610
29.53
28.07
50.16
1.329
LSTM
+2.81
+5.90
0.399
9.25
18.60
47.52
1.044
GRU
+0.53
+1.09
0.148
11.16
17.42
49.59
1.028
Table 1: Portfolio performance on Stock.
Model
TS
RS
CE
Cum. Ret. (%) ↑
Sharpe ↑
MDD. (%) ↓
Vol. (%) ↓
HR. (%) ↑
Variant 1
×
×
×
+5.53
0.823
7.81
14.84
53.38
Variant 2
✓
×
×
+6.90
1.028
7.52
14.41
58.20
Variant 3
×
✓
×
+8.56
1.261
6.33
14.25
62.30
Variant 4
✓
✓
×
+10.07
1.529
6.36
13.57
55.74
ACE-Trader
✓
✓
✓
+12.44
2.023
5.21
12.35
57.38
Table 2: Ablation study of memory selection and co-evolution.
Figure 3: Impact of α and γ .
Appendix figures & tables2 assets
Supplementary material from the paper’s appendix.
Appendix
Phase
Decision, evidence, and outcome
Trader: Commit
Proposal: Buy additional AMZN worth 2% of NAV; sell JPM worth 1% and XOM worth 1.5% of NAV. Rationale: AMZN’s RSI of 23.35 suggests an oversold rebound.
RT: Challenge
Evidence: AMZN’s 10- and 20-day returns are -5.01% and -8.42%, and its price is 6.61% below the 20-day moving average. Challenge: Low RSI alone does not justify adding exposure while the downtrend persists. Risk condition: AMZN’s return at the end of the five-trading-day horizon is below -3%.
Trader: Revise
Accept the challenge and cancel the additional AMZN purchase. Retain the existing AMZN shares without selling them. Keep the JPM and XOM sell orders unchanged.
Risk evaluation
AMZN returns -4.42% over the evaluation window ending on June 25, satisfying the pre-specified risk condition. The warning is a hit .
Revision evaluation
The original proposal yields -0.0148%, whereas the revised plan yields +0.0745%. Canceling the purchase improves net return by 0.0893% , turning a small evaluated loss into a gain.
Appendix
Table 3: Case from the initial proposal to the revised decision.
Absolute predicted return
Position-size level qt,j
0.2%≤∣r^t,j∣≤0.6%
1
0.6%<∣r^t,j∣≤1.2%
2
1.2%<∣r^t,j∣≤1.8%
3
1.8%<∣r^t,j∣≤2.4%
4
∣r^t,j∣>2.4%
5
Appendix
Table 4: Mapping absolute predicted returns to position-size levels.
Large language model (LLM) trading agents can combine market data, news, and executable analysis, but their behavior is often controlled by static hand-written tool-use policies that are fixed before deployment. This limits their ability to adapt how they gather evidence, invoke tools, verify signals, and manage risk under changing market regimes. We introduce EvolveTrade, a self-evolving framework that treats the system prompt of a tool-using trading agent as a text-parameterized policy. After each update interval, a Policy Agent revises this policy using accumulated decision traces and realized portfolio feedback, while keeping the backbone LLM fixed. The updated policy is then used for the next batch of trading decisions, enabling the agent to refine its information-acquisition and portfolio-construction procedure over time. Experiments across multiple market regimes and two LLM backbones show that EvolveTrade often improves Sharpe Ratio and Cumulative Return over fixed-policy LLM baselines, achieving the improved SR and CR in most evaluated settings. Behavioral analyses further show that self-evolved policies increase code-mediated analysis and activate regime-relevant computations; case-level policy-to-return attributions trace how policy-induced allocation changes contribute to realized return differences. These results suggest that adapting the reusable procedure governing tool use is a key direction for building more robust LLM trading agents.
Recent deployments of large language models (LLMs) as autonomous trading agents raise questions about whether financial decision-making competence generalizes beyond specific market patterns and how it should be trained and evaluated in noisy markets lacking ground truth. We propose a structured framework for training and evaluating such models. Central to our approach is a curated, multiple-choice question (MCQ) dataset derived from classic textbooks and historical markets, verified by an AI committee, enriched with structured reasoning traces, and augmented to reduce shortcut learning. To evaluate whether performance on isolated MCQs generalizes to real-world trading, we introduce a two-stage protocol combining test-set evaluation with an MCQ-based chronological trading simulation. Extensive evaluations across market regimes provide statistically robust evidence that open models trained with our framework exhibit competitive, risk-aware behavior over time, outperform open-source baselines, and approach frontier-model performance at smaller scale. We release the dataset and evaluation framework to support further research.
Yuchen Pan, Soung Chang Liew
Department of Information Engineering, The Chinese University of Hong Kong
In financial trading, large language model (LLM)-based agents demonstrate significant potential, but their decisions can be sensitive to noisy and non-stationary market information. We propose ContestTrade, a multi-agent trading system with an internal competitive mechanism inspired by institutional investment workflows. The system consists of two specialized teams: (1) a Data Team that processes and condenses massive market data into diversified textual factors optimized for constrained LLM context windows, and (2) a Research Team that produces parallelized multipath trading decisions via tool-augmented deep research. The core design is a "Quantify-Predict-Allocate" contest mechanism within each team: agent outputs are scored only after market outcomes become observable, future utility is predicted from historical scores, and resources are allocated to agents with positive predicted utility. In a post-2024 A-share backtest, ContestTrade achieves higher backtested return and risk-adjusted performance than the evaluated baselines. We further describe the temporal protocol, implementation choices, and limitations to clarify the scope of these results.