Research and Development of Next-Generation Equity Investment Model and Quant Analysis System
R&D of Next-Generation Equity Investment Model and Quant Analysis System Integrated with LLM
We maintain a daily and monthly database of the Japanese stock market spanning over 45 years. We research and develop alpha analysis systems, portfolio analysis systems, and risk management systems. By designing the database from scratch using advanced programming, we have achieved high-speed daily calculations and big data processing.
In practical portfolio composition, there are various challenges. We are able to compose passive funds that can withstand scales of hundreds of billions of yen using our unique stratified sampling method. Furthermore, our alpha analysis system allows us to immediately implement and simulate alpha from cutting-edge research. By appropriately allocating the generated alpha to sectors or segments and optimizing it, we can now design active funds (Long/Short) exactly as intended.
In addition, by incorporating Large Language Models (LLMs) into the model, we are developing a system that automatically evaluates characteristics of the resulting portfolio from detailed perspectives and explains in what market phases it performs effectively.
Key Research Areas
- High-Precision Alpha Generation: Return prediction using multi-factor models and segment-specific approaches.
- Large-Scale Optimization Technology: Construction of practical-scale portfolios using optimization methods.
- Integration of LLM: Automatic evaluation and interpretation of portfolio characteristics by embedding Generative AI into the models.
System Configuration
In this research, we develop and operate the following integrated system groups.
| System | Overview | Key Functions |
|---|---|---|
| QuantDB | Financial Data Lake | Management of 45 years of daily/monthly data, Delta Lake integration |
| FactorAnalysis | Alpha Analysis | IC analysis, Quintile analysis, Factor return verification |
| MultiFactor Model | Multi-Factor | Segment-specific synthetic factors, Adaptive weight adjustment |
| TOPIX Index | Passive Fund | Large-scale index construction using Stratified Sampling |
| HighAlpha Long/Short | Active Fund | Optimization of Market Neutral and 130/30 type portfolios |
QuantDB: Financial Database Infrastructure
Data Structure
| Category | Content | Data Period |
|---|---|---|
| Price Data | Daily stock price, Adjusted close, Return | 1980 – Present |
| Fundamental | PBR, ROE, Dividend Yield, etc. | 1980 – Present |
| Market | Market Cap, Trading Volume, Floating Stock Ratio | 1990 – Present |
| Benchmark | TOPIX, Sector Indices, Segments | 1990 – Present |
FactorAnalysis: Alpha Analysis System
Analysis Modules
| Module | Analysis Content | Output |
|---|---|---|
| IC Analysis | Time-series evaluation of Information Coefficient | IC Mean, ICIR, Decay rate |
| Quintile Analysis | Return verification by factor strength | Quintile return, Spread |
| Factor Return | Cross-sectional regression | t-value, Return contribution |
| Segment Analysis | Validity verification by size/style | Interaction between segments |
| Synthetic Factor | Composite signal construction | Optimal weights, Correlation matrix |
HighAlpha Long/Short: Active Fund System
Investment Strategy Types
| Type | Long | Short | Benchmark |
|---|---|---|---|
| Market Neutral | 100% | 100% | Short-term interest rate |
| Long-Short (130/30) | 130% | 30% | TOPIX |
| Extension (120/20) | 120% | 20% | TOPIX |
Optimization Engine
Using high-speed optimization with Linear Programming (HiGHS solver), we construct portfolios that simultaneously satisfy the following constraints:
Objective Function: Maximize Alpha - Penalty Terms
Constraints:
├── Segment Constraints (Market weight ± tolerance)
├── Industry Constraints (Net / Total)
├── Factor Exposure Constraints
├── Turnover Constraints
└── Individual Stock Limits
Integration of LLM Advisor
Overview
Detailed information about the portfolio construction is embedded internally and sent to the LLM to automatically generate qualitative evaluations and recommendations.
Evaluation Items
| Evaluation Category | Content |
|---|---|
| Risk Analysis | Concentration risk, Sector bias, Factor tilt |
| Market Condition Suitability | Consistency with current market environment |
| Improvement Proposals | Specific proposals for portfolio adjustment |
| Scenario Analysis | Assumed impact during market fluctuations |
Future Research Direction
- Real-time data integration
- Machine learning-based factor discovery
- Integration of ESG factors
- Multi-asset expansion
- Sophistication of risk models (CVaR, MAD)
Conclusion
We focus on thinking quickly about what is needed for next-generation investment and analysis models, making hypotheses, and implementing them. Our goal is to freely perform research up to the point where it can be used in actual practice.