【广发金工】2024精选深度报告系列之四：基于卷积神经网络的ETF轮动策略

Core Viewpoint - The domestic ETF market has reached a historical high, with index-based investment becoming a trend in the public fund industry. The report explores the effect of deep learning factors on ETF product rotation [1][47]. Group 1: ETF Market Overview - The global ETF market size surpassed $11 trillion in 2023, with a net inflow of nearly $1 trillion. The total assets of ETFs listed globally reached $11.61 trillion, a 21.83% increase from the end of 2022 [2]. - The domestic ETF market has also seen significant growth, exceeding 2 trillion yuan, with the number of ETFs listed reaching 889, an 18.06% increase from the end of 2022. The total market size grew by 28.13% [2][4]. - Equity ETFs dominate the global market, accounting for 74.2% of the total ETF assets, while bond ETFs and commodity ETFs account for 17.7% and 1.6%, respectively [2]. Group 2: Machine Learning Factors - The report discusses the use of convolutional neural networks (CNN) to predict future stock price movements based on price and volume data. This approach overcomes limitations of traditional time-series models [3][15]. - The CNN model processes standardized price-volume data charts, which include candlestick charts, moving averages, trading volumes, and MACD information, to effectively extract features and predict future price trends [15][16]. Group 3: Empirical Analysis - Under a weekly ETF rotation model, the average Information Coefficient (IC) of the ETF_fimage factor is 6.9%, with a win rate of 62%. The annualized return for long-short strategies is 20.4%, with a long annualized return of 14.4% and a short annualized return of -6.1% [4][19]. - The factor has shown stable performance over the years, achieving approximately 11% long-short returns since the beginning of 2024 [25][47]. Group 4: Fixed Holding Combinations - Backtesting results indicate that holding a smaller number of ETFs yields better returns. Holding five ETFs has achieved an annualized return of about 16% since 2020, outperforming both the equal-weighted configuration of all ETFs and the mixed equity fund index by 14.5% and 13.9%, respectively [2][44][47]. - The report highlights that stricter liquidity conditions can negatively impact the performance of the long positions in the ETF rotation strategy [44][41]. Group 5: Cost Impact - The backtested annualized returns for holding five ETFs under different trading fee scenarios are 19.9% (no trading fee), 16.2% (0.1% fee), and 12.5% (0.2% fee), with an annualized volatility of 25.6% [44][45].