Quantitative trading strategy design aims to discover trading programs from historical data that remain effective in future markets, which can be viewed as a black-box program optimization problem. LLM-based textual gradients offer a promising approach by providing explicit optimization directions for iterative strategy refinement. However, directly applying textual gradients faces two challenges: (1) optimization is myopic, underutilizing experience from previous evaluations; and (2) aggregate backtest feedback overlooks temporal robustness, potentially favoring strategies that perform well only in specific market periods. To address these challenges, we propose TradeGrad, an experience-guided textual-gradient framework for robust trading strategy optimization. TradeGrad leverages accumulated optimization experience to estimate textual gradients and employs multi-scale revisions for both strategy exploration and refinement. It further introduces the Cross-Period Robust Objective (CPRO), which emphasizes performance in unfavorable historical periods to promote temporal robustness. Experiments on cross-sectional and time-series strategy design in Chinese A-share and U.S. equity markets show that TradeGrad achieves the best in-sample and out-of-sample performance across all four settings. Notably, its Chinese cross-sectional strategy achieves 27.99% annualized return, 12.19% maximum drawdown, and a Sharpe ratio of 1.63, approximately 68% higher than the CSI 300 benchmark. Further analyses validate the proposed components and show consistent improvements in both in-sample and out-of-sample performance throughout optimization. The code is available at https://github.com/transcend-0/TradeGrad.
Figures & tables
Figure 1: (a) Trading Strategy as a Program. (b) Optimization for Generalizable Trading Strategy.
Figure 2: Overview of TradeGrad. TradeGrad iteratively samples archived strategies, estimates textual gradients from backtest feedback and optimization experience, updates strategies at varying revision scales, and evaluates and archives the resulting strategies.
CSI 300
S&P 500
Setting
Method
IS Score ↑
OOS Score ↑
AR ↑
MDD ↓
SR ↑
IS Score ↑
OOS Score ↑
AR ↑
MDD ↓
SR ↑
Market Index Reference
1.63
45.01
17.69
15.66
0.97
5.44
39.38
20.76
18.90
1.26
Cross-Sectional Strategies
Classical
Momentum
0.52
9.19
5.76
28.49
0.35
9.02
2.37
-15.35
31.13
-1.58
Reversal
0.28
14.51
17.80
28.59
0.74
5.12
26.76
0.56
11.59
0.11
Volatility
7.56
60.69
18.59
8.74
1.41
4.41
6.93
-9.38
23.79
-0.79
Table 1: Overall performance comparison. Results are averaged over three runs; standard deviations are reported in Appendix A.3 . Best and second-best results are bolded and underlined, respectively.
Method
IS Score ↑
OOS Score ↑
AR ↑
MDD ↓
SR ↑
Full Method
10.69±1.06
65.65±12.15
27.99±4.19
12.19±4.65
1.63±0.17
w/o Memory
4.56±0.42
35.62±7.07
11.78±1.73
16.09±5.97
0.80±0.08
w/o Sampling
8.32±0.85
46.72±10.68
12.98±2.03
11.35±4.38
0.94±0.10
w/o MSR (Diff)
5.79±0.54
46.88±12.10
14.87±1.14
13.04±5.14
0.96±0.10
w/o MSR (Rewrite)
7.42±0.77
55.74±10.01
15.98±2.39
8.86±3.48
0.98±0.11
Table 2: Ablation of experience-guided textual-gradient estimation. Results are mean ± standard deviation over three runs. Best and second-best means are bolded and underlined, respectively.
Objective
IS OV ↑
IS Score ↑
OOS Score ↑
AR ↑
MDD ↓
SR ↑
CPRO
10.69±1.06
10.69±1.06
65.65±12.15
27.99±4.19
12.19±4.65
1.63±0.17
CPRO w/o Freq Gate
10.99±1.07
8.10±0.79
40.24±8.68
10.43±1.51
12.66±4.81
0.73±0.07
Mean-Period Objective
37.72±4.18
4.40±0.40
23.37±5.06
6.08±0.93
18.70±7.13
0.44±0.05
Aggregate Objective
5.11±0.56
0.74±0.07
35.63±7.64
13.12±1.96
16.89±6.38
0.77±0.09
Table 3: Ablation of the CPRO. IS OV reports each variant’s own objective value and is not directly comparable across rows. Results are mean ± standard deviation over three runs. Best and second-best means are bolded and underlined, respectively.
Figure 3: Effect of different hyperparameters: Memory window for summarizing optimization experience, sample size for textual-gradient estimation, and probability of large-scale revision.
Figure 4: Optimization trajectories of different methods. Points denote the strategies generated at each optimization step, and curves show their 10-step moving averages (MA(10)).
Appendix figures & tables2 assets
Supplementary material from the paper’s appendix.
Appendix
CSI 300
S&P 500
Method
IS Score
OOS Score
AR
MDD
SR
IS Score
OOS Score
AR
MDD
SR
Cross-Sectional Strategies
Island Algorithm
0.02
0.52
2.01
11.26
0.06
1.12
0.13
3.37
15.03
0.44
QuantEvolve
0.31
0.94
0.84
11.04
0.02
0.03
0.03
4.02
18.54
0.24
Vanilla TextGrad
0.01
0.97
1.21
11.50
0.04
0.10
0.74
2.23
10.09
0.27
Ours
1.06
12.15
4.19
4.65
0.17
2.67
8.39
0.94
1.50
0.11
Appendix
Table 4: Standard deviations of the main results over three independent runs.
Figure 5: Evolution of a final strategy along its ancestral lineage. Labels indicate program IDs and IS Scores.
Quantitative strategy optimization remains largely manual, requiring domain experts to identify weak signals, tune risk-control rules, and repeatedly validate iterative revisions. Large language models can accelerate this process, but directly relying on them to rewrite trading strategies often introduces hallucinated edits, strategy drift, and backtest overfitting. We propose EVOQUANT, a self-Evolving Verifier-guided framework for strategy Optimization in Quantitative trading. Our method utilizes LLMs to deeply diagnose performance bottlenecks, generates semantically controlled candidate edits, selects the best strategy through a multi-stage verification pipeline, and distills optimization experience into reusable knowledge for continual self-improvement. We evaluate our method using seven representative strategies: four from the A-share market and three from the Crypto market. Experimental results show that our method significantly improves the Sharpe ratio across all tested strategies: the average test Sharpe increases from -0.298 to 0.538, and the best-performing strategy achieves a 199% relative improvement. Ablation studies and stress tests under stricter conditions further validate the effectiveness and robustness of the framework. Overall, this work transforms quantitative strategy optimization from costly manual trial and error into an automated and verifiable iterative paradigm, offering a new path for applying large language models to financial strategy research.
Jie Mao, Changlun Li, Xiang Li +6
The Hong Kong University of Science and Technology (Guangzhou) · Paradoox AI Research
Large Language Models (LLMs) are evolving into autonomous trading agents, yet existing benchmarks often overlook the interplay between architectural reasoning and strategy consistency. We propose Strat-LLM, a framework grounded in Stratified Strategy Alignment. Operating in a live-forward setting throughout 2025, it integrates heterogeneous data including sequential prices, real-time news, and annual reports to eliminate look-ahead bias. Extensive stress tests on A-share and U.S. markets reveal: (1) reasoning-heavy models achieve peak utility in Free Mode via internal logic, whereas standard models require Strict Mode as a vital risk anchor; (2) alignment utility is regime-dependent, with Free and Guided modes capturing momentum in uptrending markets, while Strict Mode mitigates drawdowns in downtrends; (3) mid-scale models (35B) show optimal fidelity under strict constraints, whereas ultra-large models (122B) suffer an alignment tax under rigid rules but gain a performance premium in Guided Mode; (4) standard LLMs often fall into a high win-rate trap, optimizing for small gains at the expense of total returns, which can only be mitigated through deep reasoning or strict external guardrails. Project details are available at https://Strat-LLM.github.io.
Wenliang Huang, Zengyi Yu
School of Economics Zhejiang University of Technology Hangzhou, China · Faculty of Education East China Normal University Shanghai, China
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.