Please rotate your device to landscape mode to view the charts.

Machine Learning in the Chinese Stock Market

Journal: Journal of Financial Economics

Date: 20220801

Author: Leippold, Markus; Wang, Qian; Zhou, Wenyu

Abstract:
We add to the emerging literature on empirical asset pricing in the Chinese stock market by building and analyzing a comprehensive set of return prediction factors using various machine learning algorithms. Contrasting previous studies for the US market, liquidity emerges as the most important predictor, leading us to closely examine the impact of transaction costs. The retail investors' dominating presence positively affects short-term predictability, particularly for small stocks. Another feature that distinguishes the Chinese market from the US market is the high predictability of large stocks and state-owned enterprises over longer horizons. The out-of-sample performance remains economically significant after transaction costs.

Link: Google Scholar

Background and Context

Market Size

China's stock market reached USD 10 trillion in 2020, making it the second largest globally after the US market at USD 39 trillion.

Market Structure

The Chinese market is dominated by retail investors who hold 99.8% of accounts, leading to high turnover of 224% of market capitalization compared to 108% in the US.

Methodology

The study analyzes 1,160 predictive signals using machine learning algorithms on data from 3,900+ A-share stocks from 2000-2020.

Superior Performance of Neural Networks vs Other Models

  • Neural networks (NN1-NN5) consistently outperform traditional models in predicting stock returns
  • GBRT achieves the highest R² of 2.71%, followed by neural networks ranging from 2.04-2.58%
  • Simple models like OLS perform significantly worse with R² of 0.81%

Higher Predictability for Small Stocks vs Large Stocks

  • Small stocks show significantly higher predictability (4.85% R²) compared to large stocks (0.57% R²)
  • This reflects the strong influence of retail investors who predominantly trade in small stocks
  • The difference is particularly pronounced at monthly horizons

Long-Only Portfolio Performance Across Different Models

  • Machine learning portfolios significantly outperform simple 1/N diversification strategy
  • Neural networks achieve highest Sharpe ratios around 1.70
  • Performance remains strong even after accounting for transaction costs

Superior Predictability for State-Owned Enterprises at Annual Horizon

  • State-owned enterprises (SOEs) show higher predictability at annual horizons
  • Neural networks achieve 7.61% R² for SOEs compared to 5.20% for non-SOEs
  • Reflects the importance of government signaling in SOE performance

Performance Resilience During Market Stress

  • Neural networks show lowest maximum drawdown during market stress periods
  • NN4 achieves 21.53% maximum drawdown compared to 47.24% for traditional OLS model
  • Demonstrates robustness of machine learning approaches during market turbulence

Contribution and Implications

  • The study demonstrates that machine learning methods can successfully predict stock returns in emerging markets with different characteristics than developed markets
  • Neural networks prove particularly effective for both short-term prediction of small stocks and long-term prediction of state-owned enterprises
  • The findings show that profitable trading strategies can be implemented even with Chinese market constraints like short-selling restrictions

Data Sources

  • Model Comparison Chart: Based on Table 1 showing monthly out-of-sample predictive R² values
  • Size Comparison Chart: Based on Table 3 showing average out-of-sample predictive R² for different size groups
  • Portfolio Performance Chart: Based on Table 6 showing Sharpe ratios for long-only portfolios
  • SOE Comparison Chart: Based on Table 3 showing annual out-of-sample predictive R² for SOE vs non-SOE
  • Stress Performance Chart: Based on Table 6 showing maximum drawdown statistics