Group 1 - The report explores the intrinsic relationship between ETF fund flows, holder structure, and index returns from a behavioral finance perspective, constructing an index rotation strategy based on the fund flow reversal effect [1][17][85] - The relationship between ETF fund flows and index returns is significantly influenced by the holder structure, with low institutional ownership (dominated by individual investors) exhibiting a pronounced "reversal effect" reflecting irrational behaviors such as "buy high, sell low" and "disposition effect" [3][84] - The historical returns of fund flow intensity groups display a notable "smile curve" characteristic, revealing the complex psychological interplay of "buying high" and "taking profits" among investors [4][62] Group 2 - The fund flow intensity factor demonstrates significant predictive power in both time series and cross-sectional dimensions, particularly the 10-day fund flow intensity factor showing a notable reversal predictive effect on future returns of low institutional ownership indices [5][84] - The index rotation strategy based on the fund flow reversal effect has shown excellent performance, with a monthly rebalancing yielding an annualized return of 16.75% since 2017, outperforming the Wind All A Index by 11.49% [6][73][85] - The portfolio's holding structure is relatively balanced, with the top five sectors being broad-based indices, pharmaceuticals, style strategies, TMT, and financial real estate, where the broad-based sector holds the highest weight at approximately 21% [7][74]

Summary of Key Points Core Insights - The report focuses on an ETF rotation strategy based on convolutional neural networks (CNN), utilizing the Wind industry thematic indices as the underlying assets. The strategy aims to measure the effectiveness of rotation based on these indices [4][9]. - The Wind industry thematic indices include over 1,000 indices across various categories, providing a broader selection compared to traditional ETFs, which have around 419 tracked indices as of January 2026 [4][45]. Section Summaries 1. Background Introduction - The report highlights the increasing acceptance of index-based investment strategies, particularly ETFs, which are favored for their transparency, low fees, and ease of trading. The strategy discussed has shown significant excess returns compared to the Wind mixed equity fund index since its implementation [9]. 2. Convolutional Neural Network Factor Logic - The methodology involves creating standardized price-volume data charts for stocks, which are then used to train a CNN model to predict future stock price movements. The model processes a large dataset of 115GB, significantly larger than traditional sequential data [13][19]. 3. Wind Industry Thematic Index Information - The Wind industry thematic indices are categorized into four subtypes: industry, theme, popular concepts, and thematic indices, with a total of 1,046 indices as of January 2026. This extensive categorization allows for a more nuanced investment approach compared to ETFs [27][45]. 4. Empirical Analysis - The empirical analysis indicates that the CNN-based rotation strategy achieved an average annualized return of approximately 30.7% since 2020, outperforming the Wind mixed equity fund index by about 21.7%. The strategy's IC (Information Coefficient) average is 3.7%, with a win rate of 59% [54][56]. - The analysis also shows that the strategy's performance is more stable across years compared to ETF rotation strategies, although the overall return in 2025 was lower than that of the ETF rotation [55][63]. 5. Parameter Adjustment Impact Measurement - The report examines the impact of various parameters on the strategy's performance, including the number of holdings (3, 5, or 10 indices), turnover frequency (weekly, bi-weekly, or monthly), and the price at which trades are executed. The findings suggest that a weekly turnover strategy yields higher returns [55][61]. 6. Conclusion - The report concludes that the CNN-based ETF rotation strategy, leveraging the diverse Wind industry thematic indices, presents a promising investment opportunity with significant potential for excess returns compared to traditional ETF strategies [4][9].