Paper: arXiv 2510.15903
Authors: Chi-Sheng Chen, Aidan Hung-Wen Tsai
Abstract
This study presents a comprehensive empirical comparison between quantum machine learning (QML) and classical machine learning (CML) approaches in Automated Market Makers (AMM) and Decentralized Finance (DeFi) trading strategies through extensive backtesting on 10 models across multiple cryptocurrency assets. Our analysis encompasses classical ML models (Random Forest, Gradient Boosting, Logistic Regression), pure quantum models (VQE Classifier, QNN, QSVM), hybrid quantum-classical models (QASA Hybrid, QASA Sequence, QuantumRWKV), and transformer models. The results demonstrate that hybrid quantum models achieve superior overall performance with 11.2% average return and 1.42 average Sharpe ratio, while classical ML models show 9.8% average return and 1.47 average Sharpe ratio. The QASA Sequence hybrid model achieves the highest individual return of 13.99% with the best Sharpe ratio of 1.76, demonstrating the potential of quantum-classical hybrid approaches in AMM and DeFi trading strategies.
Complexity vs Empirical Score
- Math Complexity: 8.5/10
- Empirical Rigor: 6.0/10 (capped from 9.0: transaction costs not modelled; see penalty flags)
- Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 9.0)
Why this score: The paper contains advanced mathematical formalisms, including quantum circuit equations and statistical formulas for feature engineering, while also presenting extensive backtesting results with specific performance metrics across multiple assets.
Research Flowchart
flowchart TD
A["Research Goal<br>QML vs CML in AMM DeFi Trading"] --> B["Data: Multi-Asset Backtesting<br>10 Models, Crypto Assets"]
B --> C["Computational Processes<br>Classical ML, QML, Hybrid Models"]
C --> D{"Backtesting & Evaluation<br>Returns & Sharpe Ratio"}
D --> E["Key Findings<br>Hybrid QML: 11.2% Avg Return<br>Best: QASA Sequence (13.99%)<br>CML: 9.8% Avg Return"]
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 and evaluates hybrid quantum‑classical machine learning models for AMM and DeFi trading, showing they achieve superior overall performance.
Source sentence
The results demonstrate that hybrid quantum models achieve superior overall performance with 11.2% average return and 1.42 average Sharpe ratio
Assumes 6
- The target prediction is a binary classification indicating whether rebalancing should occur. model
Source sentence
All models learn the target function: f(X) →y ∈{0, 1}
- The study uses daily frequency data over a one‑year period for three crypto assets. data
Source sentence
Time Period: January 1, 2024 - January 1, 2025 (1 years) • Assets: BTCUSDC, ETHUSDC, USDCUSDT • Frequency: Daily data • Total Samples: 252 trading days
- Performance metrics are based on average return and Sharpe ratio calculated on backtested equity curves. statistical
Source sentence
Classical machine learning methods achieved strong and balanced results, with a 9.8% average return and the highest Sharpe ratio (1.47).
- The backtesting assumes trades can be executed at the observed daily prices without additional slippage or market impact. execution
Source sentence
1} where X represents market features and y indicates whether rebalancing should occur.
Show 2 more
- Hybrid models use a moderate feature set (12 features) with quantum angle encoding. model
Source sentence
Hybrid models: Use moderate feature sets (12 features) with classical preprocessing and quantum angle encoding for optimal balance.
- Pure quantum models are trained with longer training times, limiting real‑time applicability. execution
Source sentence
Quantum model training time is longer, potentially limiting real-time applications.
Data
multiple cryptocurrency assets · crypto
- Crypto AMM Backtesting Dataset crypto-onchain unknown
Source sentence
Time Period: January 1, 2024 - January 1, 2025 (1 years) • Assets: BTCUSDC, ETHUSDC, USDCUSDT • Frequency: Daily data • Total Samples: 252 trading days
Costs
Not modeled
Finds 5
Sharpe 1.76 · total return 14.0% · max drawdown 10.1%
- Hybrid quantum models achieve the highest average return among quantum‑based methods (11.2%) with relatively low volatility (11.0%) and an average Sharpe of 1.42. performance
Source sentence
Hybrid quantum approaches provided a strong trade-off, delivering the highest average return among the quantum-based methods (11.2%) and relatively low volatility (11.0%), with performance close to classical models.
- The QASA Sequence hybrid model attains the highest individual return (13.99%) and best Sharpe ratio (1.76) across all tested models. performance
Source sentence
QASA Sequence achieved the highest return (13.99%), Sharpe ratio (1.76), and Calmar ratio (6.51), clearly outperforming both classical and purely quantum baselines.
- Classical machine learning models deliver strong results with a 9.8% average return and the highest Sharpe ratio (1.47). performance
Source sentence
Classical machine learning methods achieved strong and balanced results, with a 9.8% average return and the highest Sharpe ratio (1.47).
- Pure quantum models underperform, yielding only a 4.4% average return and a Sharpe ratio of 0.83, with higher volatility. negative
Source sentence
Pure quantum models underperformed in both returns (4.4%) and risk‑adjusted performance (Sharpe 0.83), while also exhibiting higher volatility.
- Transformer‑based models achieve the highest overall return (12.3%) but suffer from elevated volatility (15.4%) and a lower Sharpe (1.23) than classical methods. performance
Source sentence
Transformer‑based models yielded the best overall returns (12.3%) but at the cost of elevated volatility (15.4%), leading to a lower Sharpe ratio compared with classical methods.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 43 related papers). Coverage is partial, so this is neither replication nor refutation.
1 paper in the archive
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.
Cited References (25)
- Quantum adaptive self-attention for quantum transformer models (2025) — link
- Quantum-enhanced channel mixing in rwkv models for time series forecasting (2025) — link
- Quantum long short-term memory (2022)
- Linear-layer-enhanced quantum long short-term memory for carbon price forecasting (2023)
- Learning to program variational quantum circuits with fast weights (2024)
- Long short-term memory (1997)
- Attention is all you need (2017)
- A multi-source dataset of urban life in the city of Milan and the Province of Trentino (2015)
and 17 more
- Long Short-Term Memory (1997) — link
- Quantum Machine Learning (2017) — link
- RWKV: Reinventing RNNs for the Transformer Era (2023)
- Informer: Beyond efficient transformer for long sequence time-series forecasting (2021)
- Quantum long short-term memory (QLSTM) vs. classical LSTM in time series forecasting: a comparative study in solar power forecasting (2024)
- Differential privacy (2006)
- Quantum circuit learning (2018)
- R 'e nyi differential privacy (2017)
- Quantum machine learning for quantum anomaly detection (2018)
- Dynamic differential-privacy preserving sgd (2021) — link
- Learning differentially private recurrent language models (2017) — link
- Differentially private federated tensor completion for cloud–edge collaborative AIoT data prediction (2023)
- Quantum machine learning with differential privacy (2023)
- Differential privacy in quantum computation (2017)
- Quantum Reinforcement Learning Trading Agent for Sector Rotation in the Taiwan Stock Market — → 2506.20930
- Benchmarking Classical and Quantum Models for DeFi Yield Prediction on Curve Finance — → 2508.02685
- Quantum Reinforcement Learning Trading Agent for Sector Rotation in the Taiwan Stock Market — → 2506.20930
3 of 25 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.