Deep Learning Enhanced Multi-Day Turnover Quantitative Trading Algorithm for Chinese A-Share Market
Yimin Du · 2025
Read Deep Learning Enhanced Multi-Day Turnover Quantitative Trading Algorithm for Chinese A-Share Market by Yimin Du (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-06356.txt.
Introduction
Yimin Du presents a deep learning trading system for Chinese A-shares in a 2025 paper. The design chains five parts. A neural network scores stocks. A mixture model reads opening signals. Position sizing follows company size and trading ease. Tuned rules set take-profit and stop-loss points. Timing models gauge market swings.
Training used market data from 2010 to 2020. Testing covered 2021 to 2024. The system claims yearly returns of fifteen point two percent. Its worst peak-to-trough loss stayed under five percent. Its risk score beat one point eight. It holds fifty to one hundred stocks at once. No position lasts beyond nine days. Readers get a full machine trading blueprint with stated results.
Catalog record
| CREATOR | Yimin Du |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 46 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2506-06356
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Deep Learning Enhanced Multi-Day Turnover Quantitative Trading Algorithm for Chinese A-Share Market Yimin Du Abstract This paper presents a sophisticated multi-day turnover quantitative trading algorithm that integrates advanced deep learning techniques with comprehensive cross-sectional stock prediction for the Chinese A-share market. Our framework combines five interconnected modules: initial stock selection through deep cross-sectional prediction networks, opening signal distribution analysis using mixture models for arbitrage identification, market capitalization and liquidity-based dynamic position sizing, grid-search optimized profit-taking and stop-loss mechanisms, and multi-granularity volatility-based market timing models. The algorithm employs a novel approach to balance capital efficiency with risk management through adaptive holding periods and sophisticated entry/exit timing. Trained on comprehensive A-share data from 2010-2020 and rigorously backtested on 2021-2024 data, our method achieves remarkable performance with 15.2% annualized returns, maximum drawdown constrained below 5%, and a Sharpe ratio of 1.87. The strategy demonstrates exceptional scalability by maintaining 50-100 daily positions with a 9-day maximum holding period, incorporating dynamic profit-taking and stop-loss mechanisms that enhance capital turnover efficiency while preserving risk-adjusted returns. Our approach exhibits robust performance across various market regimes while maintaining high capital capacity suitable for institutional deployment. I Introduction The evolution of quantitative trading in Chinese capital markets has been marked by increasing sophistication in both methodological approaches and technological implementation. The A-share market, characterized by its unique regulatory environment, high retail participation, and distinctive volatility patterns, presents both exceptional opportunities and significant challenges for algorithmic trading strategies. Traditional approaches often fall into two categories: high-frequency strategies with limited capital capacity, and long-term fundamental strategies that may miss intermediate-term market inefficiencies. This paper addresses these limitations by introducing a comprehensive multi-day turnover quantitative trading algorithm that bridges the gap between short-term tactical allocation and medium-term strategic positioning. Our approach is fundamentally built upon the integration of deep learning techniques with traditional quantitative methods, creating a robust framework that can adapt to the dynamic nature of Chinese equity markets while maintaining consistent risk-adjusted performance. The core innovation of our methodology lies in its multi-layered approach to market analysis and position management. Rather than relying on a single predictive model or trading signal, we construct a sophisticated ecosystem of interconnected components that work synergistically to identify opportunities, size positions appropriately, and manage risk dynamically. This holistic approach enables our algorithm to capture alpha from multiple sources while maintaining robust risk controls across varying market conditions. Our framework consists of five integrated modules, each designed to address specific aspects of the trading process. The cross-sectional prediction module employs deep neural networks to identify stocks with superior relative performance potential, moving beyond traditional factor-based approaches to capture complex non-linear relationships in market data. The opening signal distribution analysis component utilizes advanced statistical techniques to identify arbitrage opportunities in opening auction mechanisms, a particularly important feature given the unique characteristics of A-share market microstructure. The position sizing algorithm represents a significant advancement over conventional approaches by incorporating both market capitalization and liquidity dynamics in real-time, ensuring optimal capital allocation while respecting market impact constraints. Our grid-search optimization framework for profit-taking and stop-loss parameters goes beyond static rule-based approaches, dynamically adapting to market conditions to optimize the trade-off between profit maximization and risk control. Finally, the multi-granularity market timing component integrates analysis across multiple time horizons and volatility regimes, enabling the algorithm to make nuanced decisions about entry and exit timing that account for broader market dynamics while maintaining focus on individual security selection. The key contributions of this research include: (1) A novel integration architecture that combines cross-sectional prediction with opening signal distribution analysis for enhanced stock selection and precise entry timing; (2) An adaptive position sizing algorithm that dynamically balances market impact, liquidity constraints, and capital efficiency; (3) A comprehensive grid-search optimization framework that simultaneously optimizes profit-taking, stop-loss, and holding period parameters; (4) A multi-granularity volatility analysis system that incorporates market regime identification for robust timing decisions; (5) Extensive empirical validation demonstrating superior risk-adjusted performance with institutional-scale capital capacity. II Literature Review and Related Work The landscape of quantitative trading research has evolved dramatically over the past two decades, with particular acceleration in applications to Asian markets. Traditional factor-based models, pioneered by [ 1 ] and extended by [ 2 ] , established the foundational framework for systematic equity strategies. However, these linear models often fail to capture the complex, non-linear relationships present in modern financial markets, particularly in emerging market contexts such as the Chinese A-share market. The integration of machine learning techniques into quantitative finance has opened new avenues for alpha generation. Early applications focused on traditional supervised learning approaches, with [ 3 ] demonstrating the effectiveness of support vector machines for stock return prediction. The advent of deep learning has further revolutionized the field, with [ 4 ] showing that LSTM networks can effectively capture temporal dependencies in financial time series. Recent research has increasingly focused on the application of sophisticated neural network architectures to financial prediction tasks. [ 5 ] provides a comprehensive survey of deep learning applications in financial forecasting, highlighting the superior performance of neural networks over traditional econometric models in many contexts. [ 6 ] introduces the Informer architecture, which addresses the computational challenges of applying transformer models to long sequences commonly encountered in financial data. Cross-sectional analysis has emerged as a particularly powerful approach for equity selection, as it naturally controls for market-wide factors while focusing on relative performance differences. [ 7 ] demonstrates that cross-sectional models can significantly improve the information ratio of equity strategies by reducing exposure to systematic risk factors. This approach is particularly relevant in the Chinese market context, where sector rotation and style factors play significant roles in return generation. The Chinese A-share market presents unique characteristics that distinguish it from developed markets. [ 8 ] documents numerous market anomalies specific to Chinese equities, including calendar effects, momentum patterns, and liquidity premiums that differ substantially from those observed in Western markets. [ 9 ] specifically examines the application of machine learning techniques to Chinese equity selection, finding that ensemble methods combining multiple algorithms tend to outperform single-model approaches. Volatility modeling has been a central focus of financial econometrics, with [ 10 ] introducing the DCC-GARCH framework for multivariate volatility modeling. More recent work has explored the application of neural networks to volatility prediction, with [ 11 ] demonstrating that deep learning models can capture volatility clustering and regime changes more effectively than traditional GARCH models. Position sizing and portfolio construction represent critical components of any systematic trading strategy. The Kelly criterion, introduced by [ 12 ] and adapted for financial applications by [ 13 ] , provides a theoretical framework for optimal position sizing under uncertainty. However, practical implementations must account for transaction costs, market impact, and liquidity constraints. [ 14 ] develops a framework for optimal execution that balances market impact with timing risk, while [ 15 ] extends this work to high-frequency trading contexts. Risk management in quantitative strategies has evolved from simple volatility targeting to sophisticated multi-dimensional approaches. [ 16 ] proposes modifications to the traditional Sharpe ratio that better capture tail risk characteristics. [ 17 ] examines the risk characteristics of systematic trading strategies, emphasizing the importance of drawdown control and regime-aware risk management. The integration of