Core Viewpoint - The article outlines the agenda for the 2025 Quantitative Salon in Shanghai, focusing on various investment strategies and risk management techniques in the financial sector [1][2]. Group 1: Event Details - The event is scheduled for August 13, 2025, from 13:30 to 17:00 at the Jinling Zijinshan Hotel in Shanghai [1]. - The agenda includes multiple sessions led by experts from Guosen Securities, covering topics such as stock selection strategies, multi-strategy enhancement, and risk models [1][2]. Group 2: Session Summaries - The first session will discuss "Steady Stock Selection Strategies" led by Zhang Xinwei, the Chief Analyst of Financial Engineering at Guosen Securities [1]. - The second session will focus on "Multi-Strategy Enhanced Portfolio from a Heuristic Perspective," also presented by Zhang Xinwei [1]. - The third session will cover "Alpha Information Contained in Intraday Special Moments," presented by Neng Yu, Co-Chief Analyst of Financial Engineering [1]. - The fourth session will address "Comprehensive Guide to Risk Models," led by Zhang Yu, Co-Chief Analyst of Financial Engineering [2]. - The fifth session will explore "Expansion and Enhancement of Alpha Factors in Financial Statements," also by Zhang Yu [4]. - The sixth session will discuss "Contrarian Investment Ability and Performance of Fund Managers," presented by Chen Mengqi, an Analyst at Guosen Securities [4]. - The final session will focus on "Unified Improvement Framework for Selection Factors from the Perspective of Hidden Risks," led by Hu Zhichao, an Analyst at Guosen Securities [4]. Group 3: Participation and Benefits - Participation is limited, and interested attendees must register through a specific process to ensure a good experience [2]. - Attendees who successfully register and attend will receive a copy of the "Selected Research Report of Guosen Financial Engineering Team for 2025" [5].

Core Insights - The report emphasizes the development of a deep learning factor selection system for stock prediction and portfolio optimization, shifting from traditional logic-driven methods to data-driven approaches [7][10]. - The system integrates diverse data sources, including daily and minute market data, to enhance the performance of alpha signals [7][10]. - The report outlines the construction of multiple models that utilize different network architectures to extract unique alpha signals, demonstrating low correlation among them [8][54]. Data and Network - The input data consists of three categories: daily market data, minute market data, and manually crafted features, with neural networks independently extracting alpha features from each dataset [11]. - The report describes the use of Long Short-Term Memory (LSTM) networks combined with self-attention mechanisms to capture long-term dependencies in time series data [19]. - A Graph Attention Network (GAT) is employed to model the complex relationships between stocks, providing a global analysis perspective [20]. Alpha Models - The report presents various alpha models, including simple equal-weight, tree model weighting, and network weighting, with a focus on combining multiple signals to enhance robustness [3][3.1][3.2]. - The average Information Coefficient (IC) for the combined factors since 2019 is reported as 11.3% for 5-day IC and 12.4% for 10-day IC, indicating strong predictive power [31][32]. Risk Models - The report highlights the use of neural networks to identify high-dimensional non-linear risk patterns directly from raw price and volume data, enhancing risk control in portfolio construction [9]. Index Enhancement Strategies - The report details the performance of enhanced index strategies based on deep learning alpha signals, with annualized returns reported as follows: - CSI 300 enhanced portfolio: 18.2% annualized return, 14.2% excess return over the index [3][5.1]. - CSI 500 enhanced portfolio: 22.4% annualized return, 17.2% excess return over the index [3][5.2]. - CSI 1000 enhanced portfolio: 29.8% annualized return, 24.5% excess return over the index [3][5.3].

Group 1 - The article highlights the increasing frequency of "black swan" events in the A-share market, leading to significant drawdowns in excess returns for public index-enhanced products in 2024, marking the largest historical drawdown [1][4][6] - The "black swan index" has shown a higher proportion of extreme events occurring in 2024 compared to previous years, indicating a substantial increase in the probability of extreme tail risks [1][10][14] Group 2 - The evolution of risk models has transitioned from single-factor to multi-factor approaches, and from explicit to implicit risks, reflecting a deeper understanding of market risks [18][19][21] - Implicit risks are defined as those that change with market conditions and are not fully captured by traditional explicit risk models, making them crucial for comprehensive risk management [46][52] Group 3 - A comprehensive risk control process is proposed, consisting of three stages: preemptive measures, in-process control, and post-event handling, aimed at effectively managing both explicit and implicit risks [60][63] - The introduction of a full-process risk control model has shown to significantly reduce drawdowns and volatility without adversely affecting long-term returns [3][61] Group 4 - The traditional multi-factor index-enhanced model has demonstrated an annualized excess return of 18.77% with a maximum drawdown of 9.68%, while the model incorporating full-process risk control has achieved an annualized excess return of 16.51% with a maximum drawdown of only 4.90% [3][5] - The performance metrics indicate that the full-process risk control model enhances the stability of excess returns while minimizing drawdowns and volatility [3][5][61]

Quantitative Models and Construction Methods Model Name: Black Swan Index - **Construction Idea**: Measure the extremity of market transactions based on the deviation of style factor returns[24][25] - **Construction Process**: 1. Calculate the daily return deviation of style factors: $$ \sigma_{s,t}=\frac{\bar{r}_{s,t}-\bar{r}_{s}}{\sigma_{s}} $$ where $\bar{r}_{s,t}$ is the daily return of style factor $s$ on day $t$, $\bar{r}_{s}$ is the average daily return of style factor $s$ over the entire sample period, and $\sigma_{s}$ is the standard deviation of daily returns of style factor $s$ over the entire sample period[25] 2. Calculate the Black Swan Index: $$ BlackSwan_{t}=\frac{1}{N}\times\sum_{s\in S}\left|\sigma_{s,t}\right| $$ where $BlackSwan_{t}$ is the Black Swan Index on day $t$, $S$ is the set of all style factors, and $N$ is the number of style factors[25] - **Evaluation**: The Black Swan Index effectively captures the extremity of market transactions, indicating higher probabilities of extreme tail risks[24][25] Model Name: Heuristic Style Classification for Cognitive Risk Control - **Construction Idea**: Address the discrepancy between individual and collective cognition in style classification to control cognitive risk[80][81] - **Construction Process**: 1. Calculate the value and growth factors for each stock based on predefined metrics[85] 2. Construct value and growth portfolios by selecting the top 10% and bottom 10% stocks based on factor scores[82] 3. Perform time-series regression to classify stocks into value, growth, or balanced styles: $$ r_{t,t}\sim\beta_{\mathit{Value}}\cdot r_{\mathit{Value},t}+\beta_{\mathit{Growth}}\cdot r_{\mathit{Growth},t}+\varepsilon_{t} $$ subject to $0\leq\beta_{\mathit{Value}}\leq1$, $0\leq\beta_{\mathit{Growth}}\leq1$, and $\beta_{\mathit{Value}}+\beta_{\mathit{Growth}}=1$[97] 4. Use weighted least squares (WLS) to estimate regression coefficients based on the most differentiated trading days[98] - **Evaluation**: The heuristic style classification method captures market consensus more accurately than traditional factor scoring methods, reducing cognitive risk[80][81] Model Name: Louvain Community Detection for Hidden Risk Control - **Construction Idea**: Cluster stocks based on excess return correlations to identify hidden risks[116][117] - **Construction Process**: 1. Calculate weighted correlation of excess returns between stocks: $$ Corr_{w}(X,Y)=\frac{Cov_{w}(X,Y)}{\sigma_{w,X}\cdot\sigma_{w,Y}}=\frac{\sum_{i=1}^{n}w_{i}(x_{i}-\overline{X_{w}})(y_{i}-\overline{Y_{w}})}{\sqrt{\sum_{i=1}^{n}w_{i}(x_{i}-\overline{X_{w}})^{2}}\cdot\sqrt{\sum_{i=1}^{n}w_{i}(y_{i}-\overline{Y_{w}})^{2}}} $$ where $w_{i}$ is the weight for day $i$, reflecting market volatility[118] 2. Use Louvain algorithm to cluster stocks based on weighted correlation matrix[117] 3. Ensure clusters have at least 20 stocks and remove clusters with fewer stocks[121] - **Evaluation**: The Louvain community detection method effectively identifies hidden risks by clustering stocks with similar return patterns, which traditional risk models may overlook[116][117] Model Name: Dynamic Style Factor Control - **Construction Idea**: Control style factors dynamically based on their volatility clustering effect[128][129] - **Construction Process**: 1. Identify style factors with high volatility or significant volatility increase: $$ \text{High volatility: Rolling 3-month volatility in top 3} $$ $$ \text{Volatility increase: Rolling 3-month volatility > historical mean + 1 standard deviation} $$ 2. Set the exposure of these style factors to zero in the portfolio[136] - **Evaluation**: Dynamic style factor control captures major market risks without significantly affecting portfolio returns, leveraging the predictability of volatility clustering[128][129] Model Name: Adaptive Stock Deviation Control under Target Tracking Error - **Construction Idea**: Adjust stock deviation based on tracking error to control portfolio risk[146][147] - **Construction Process**: 1. Calculate rolling 3-month tracking error for different stock deviation levels[153] 2. Set the maximum stock deviation that keeps tracking error within the target range[153] - **Evaluation**: Adaptive stock deviation control effectively reduces tracking error during high market volatility, maintaining portfolio stability[146][147] Model Backtest Results Traditional CSI 500 Enhanced Index - **Annualized Excess Return**: 18.77%[5][162] - **Maximum Drawdown**: 9.68%[5][162] - **Information Ratio (IR)**: 3.56[5][162] - **Return-to-Drawdown Ratio**: 1.94[5][162] - **Annualized Tracking Error**: 4.88%[5][162] CSI 500 Enhanced Index with Full-Process Risk Control - **Annualized Excess Return**: 16.51%[5][169] - **Maximum Drawdown**: 4.90%[5][169] - **Information Ratio (IR)**: 3.94[5][169] - **Return-to-Drawdown Ratio**: 3.37[5][169] - **Annualized Tracking Error**: 3.98%[5][169]

Core Viewpoint - The article emphasizes the importance of adopting appropriate risk models in the current volatile market environment, which is characterized by uncertainties from tariffs, rising inflation, and slowing economic growth [1][3]. Market Volatility Analysis - In April, the volatility levels in global financial markets were comparable to historical periods of significant market distress, such as the 2010 flash crash and the 2011 European debt crisis [1]. - The rapid transition from stable to volatile market conditions necessitates a shift in risk management approaches, requiring advanced probabilistic models to address the changing risks of extreme events [3]. - The FactSet fat-tail model provides real-time insights into market dynamics and demonstrates strong predictive capabilities across different volatility phases, highlighting the need for adaptive risk models [3]. Risk Models Comparison - The article compares the normal model (normal distribution with a half-life of 125 days) and the fat-tail short-term return model (fat-tail distribution with a half-life of 45 days) in terms of their daily value-at-risk estimates at a 99% confidence level [11]. - The fat-tail model reacts more quickly to recent market behavior, making it more sensitive to increases in market volatility [11]. - During periods of low volatility, the difference in risk estimates between the two models can turn negative, indicating that the fat-tail model may underestimate risk in stable conditions [11]. Industry Sector Perspective - The analysis focuses on the market from March to April 2025, examining the changes in daily value-at-risk for individual sectors within the S&P 500 index [20]. - Certain sectors, such as consumer services and transportation, showed earlier increases in the risk model difference, indicating a quicker response to volatility changes compared to others like durable goods and electronics [20][23]. - The fat-tail model is particularly valuable for sectors with non-normal return distributions, as it can more effectively capture changes in market conditions [28].