Quantitative Models and Construction Methods 1. Model Name: Daily Frequency "High/Low Volume" Signal - **Model Construction Idea**: Define "high volume at high price" and "high volume at low price" events using daily frequency data to identify event signals and construct a capital channel strategy [1][13] - **Model Construction Process**: 1. Define "low volume" events: - Closing price is in the bottom 10% percentile of the past 120 trading days - Trading volume exceeds the average of the past 120 trading days by 1.5 standard deviations 2. Define "high volume" events: - Closing price is in the top 90% percentile of the past 120 trading days - Trading volume exceeds the average of the past 120 trading days by 1.5 standard deviations [13] 3. Construct a capital channel strategy: - Set up 4 capital channels, each with a holding period of 20 trading days - At the beginning of each week, review the past 5 trading days and identify stocks that triggered high/low volume signals - Equally allocate funds to the identified stocks at the beginning of the week and hold for 20 trading days - Calculate the net value of the capital channel portfolio by summing the net values of the 4 channels [18] - **Model Evaluation**: The daily frequency "high/low volume" signals showed that the average excess return peaked around 20-25 trading days after the signal was triggered, but the returns were volatile and did not provide stable incremental returns [1][13][18] 2. Model Name: High-Frequency "High/Low Volume" Event Cluster - **Model Construction Idea**: Use high-frequency micro-level price and volume data to construct more stable "high/low volume" event clusters, which are less correlated and more effective [2][25] - **Model Construction Process**: 1. **Event Identification**: - Define "high/low price" using minute-level closing price data - Define "high/low volume" using minute-level trading volume data, considering factors such as "who's volume," "direction of volume," and "type of volume" [26][29][32] 2. **Signal Definition**: - Combine "high/low price" and "high/low volume" using two methods: - "Price first, volume second": Identify high/low price points first, then check if volume is high - "Volume first, price second": Identify high volume points first, then check if price is high/low [42][43] 3. **Signal Screening and Synthesis**: - Produce thousands of event signals by combining different identification methods - Evaluate the effectiveness and correlation of each signal - Retain effective and low-correlation signals to form "high volume event cluster" and "low volume event cluster" - Synthesize signals to construct comprehensive "high volume" and "low volume" signals [26][44][45] - **Model Evaluation**: The high-frequency "low volume" comprehensive signal provided stable positive excess returns, while the "high volume" comprehensive signal demonstrated strong negative selection effects [50][57] 3. Model Name: Combined "High/Low Volume" Signal - **Model Construction Idea**: Combine the positive selection effect of the "low volume" signal with the negative selection effect of the "high volume" signal to enhance the performance of the capital channel strategy [3][58] - **Model Construction Process**: 1. Use the "low volume" comprehensive signal to pre-screen the stock pool 2. Exclude stocks that triggered the "high volume" comprehensive signal in the past 5 trading days 3. Construct a capital channel strategy: - Set up 4 capital channels, each with a holding period of 20 trading days - At the beginning of each week, review the past 5 trading days and identify stocks that meet the combined signal criteria - Equally allocate funds to the identified stocks at the beginning of the week and hold for 20 trading days - Calculate the net value of the capital channel portfolio by summing the net values of the 4 channels [58] - **Model Evaluation**: The combination of the two signals improved the performance of the capital channel strategy, enhancing both returns and stability [58][60] --- Model Backtesting Results 1. Daily Frequency "High/Low Volume" Signal - **Low Volume Signal**: - Annualized excess return: 7.67% - IR: 2.22 - Maximum drawdown: 4.68% [50][51] - **High Volume Signal**: - Annualized excess return: -10.16% - IR: -0.44 - Maximum drawdown: 8.47% [57] 2. High-Frequency "High/Low Volume" Event Cluster - **Low Volume Comprehensive Signal**: - Annualized excess return: 7.67% - IR: 2.22 - Maximum drawdown: 4.68% [50][51] - **High Volume Comprehensive Signal**: - Annualized excess return: -10.16% - IR: -0.44 - Maximum drawdown: 8.47% [57] 3. Combined "High/Low Volume" Signal - **Combined Signal**: - Annualized excess return: 9.14% - IR: 2.42 - Maximum drawdown: 3.70% [60] --- Quantitative Factors and Construction Methods 1. Factor Name: Low Volume Signal - **Factor Construction Idea**: Identify stocks with low prices and high trading volumes as potential candidates for positive returns [13] - **Factor Construction Process**: 1. Define "low price" as the closing price in the bottom 10% percentile of the past 120 trading days 2. Define "high volume" as trading volume exceeding the average of the past 120 trading days by 1.5 standard deviations 3. Combine the two conditions to identify "low volume" events [13] - **Factor Evaluation**: The low volume signal showed positive returns, peaking around 20-25 trading days after the signal was triggered, but the returns were volatile [1][13] 2. Factor Name: High Volume Signal - **Factor Construction Idea**: Identify stocks with high prices and high trading volumes as potential candidates for negative returns [13] - **Factor Construction Process**: 1. Define "high price" as the closing price in the top 90% percentile of the past 120 trading days 2. Define "high volume" as trading volume exceeding the average of the past 120 trading days by 1.5 standard deviations 3. Combine the two conditions to identify "high volume" events [13] - **Factor Evaluation**: The high volume signal showed negative returns, with stocks underperforming after the signal was triggered [15][18] --- Factor Backtesting Results 1. Low Volume Signal - Annualized excess return: 7.67% - IR: 2.22 - Maximum drawdown: 4.68% [50][51] 2. High Volume Signal - Annualized excess return: -10.16% - IR: -0.44 - Maximum drawdown: 8.47% [57]

Group 1: Core Insights - The report highlights Shenghong Technology (300476.SZ) as a leading global AI PCB manufacturer, emphasizing its deep engagement with top GPU and ASIC clients [15][16][18] - The company has achieved significant advancements in HDI technology, being one of the first to mass-produce 6-layer 24-layer HDI products and is actively developing next-generation 10-layer 30-layer HDI technology [16] - Shenghong Technology is expanding its production capacity both domestically and internationally, with ongoing projects in Thailand and Vietnam, and is enhancing its collaboration with major clients to align with their R&D and production schedules [17] Group 2: Financial Projections - The revenue forecast for Shenghong Technology is projected to reach 370 billion and 599 billion yuan in 2026 and 2027, respectively, with expected net profits of 120 billion and 197 billion yuan [18] - The company is expected to benefit from the increasing value of PCBs in AI servers, which will further strengthen its competitive position in the market [16][18] Group 3: Industry Context - The report notes that the AI sector is experiencing rapid growth, with significant investments from major players like Google, which has increased its capital expenditure guidance for the year [15] - The demand for advanced PCB technology is rising in line with the growth of AI applications, positioning Shenghong Technology favorably within this expanding market [15][16]