# K2J AI Quantitative Lab — Full System Knowledge Base & Whitepaper

URL: https://k2j.net/
Author: Justin (k2jysy@gmail.com | https://www.linkedin.com/in/justin-k2j)
Published by: K2J AI Quantitative Lab
License: Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0) | https://creativecommons.org/licenses/by-nc-nd/4.0/

---

## 1. Executive Summary

K2J AI Quantitative Lab is a production-grade, daily quantitative research and market intelligence system covering the entire US equities universe. The system generates daily research reports every trading morning, integrating macro regime detection, market breadth, 5-dimensional statistical factor attribution, unsupervised theme clustering, and Bayesian tail-risk portfolio management.

The entire platform is designed, engineered, and operated independently by Justin, an AI systems architect with over 20 years of experience across enterprise technology giants (Dell, Microsoft, IBM) and private investment funds, who actively trades US equities and options using this exact analytical framework.

---

## 2. Quantitative & Machine Learning Methodologies

### 2.1 Layer 1: Market Regime Classification (GBDT Ensemble)
- **Objective**: Determine the overarching macro market regime (Risk-On, Risk-Off, or Transitional / Compression) before analyzing individual stocks.
- **Model**: Gradient Boosted Decision Tree (GBDT) ensemble combined with hidden macro state indicators.
- **Inputs**: Cross-asset volatility surface (VIX, VVIX, term structure), interest rate differentials, credit spreads (HYG/LQD), foreign exchange dynamics, and liquidity momentum.
- **Output**: Calibrated regime probability distributions guiding directional exposure and hedging mandates.

### 2.2 Layer 2: Full-Market Breadth Analysis
- **Objective**: Dissect true market participation beneath headline capitalization-weighted index movements (e.g., S&P 500, Nasdaq 100).
- **Metrics**: Advance-decline volume ratios, percentage of stocks trading above key moving averages (20D, 50D, 200D), net new 52-week highs/lows, and sector-level participation dispersion.

### 2.3 Layer 3: 5-D Orthogonal Factor Attribution (Robust WLS + PCA)
- **Objective**: Explain the underlying structural forces driving cross-sectional daily returns while filtering noise.
- **Methodology**: Weighted Least Squares (WLS) regression with Huber loss robustness and Principal Component Analysis (PCA) orthogonalization.
- **Dimensions**:
  1. Market Beta (Systematic co-movement)
  2. Size & Capitalization (Large vs. Small Cap divergence)
  3. Value vs. Growth / Quality (Valuation multiples and profitability)
  4. Momentum & Reversal (Short-term cross-sectional trend persistence)
  5. Volatility & Liquidity / Residual Alpha (Idiosyncratic factor dynamics)

### 2.4 Layer 4: Unsupervised Theme Clustering (HDBSCAN + Semantic Vectors)
- **Objective**: Group moving assets into coherent, emerging narrative clusters rather than disjointed ticker lists.
- **Pipeline**: Dense text embeddings of corporate disclosures, real-time catalysts, and sector definitions, combined with high-dimensional price/volume co-movement matrices, clustered via HDBSCAN (Hierarchical Density-Based Spatial Clustering of Applications with Noise).

### 2.5 Layer 5: Portfolio Stance & Bayesian Tail-Risk Optimization (CVaR + Risk Parity)
- **Objective**: Convert quantitative insights into concrete, actionable portfolio positioning and risk management parameters.
- **Optimization**: Conditional Value-at-Risk (CVaR) minimization subject to Bayesian risk-parity constraints, dynamic volatility targeting, and asymmetric downside protection protocols.

---

## 3. Autonomous AI Systems Architecture

- **Multi-Agent Orchestration**: Powered by LangGraph, coordinating specialized autonomous agents for data ingestion, statistical computation, signal validation, natural language synthesis, and report publishing.
- **Agentic RAG**: Multi-stage retrieval-augmented generation utilizing Pinecone vector databases and hybrid sparse-dense indices to ground research narratives in historical regime precedents and market archives.
- **Deterministic Guardrails**: Strict constraint checks ensuring numerical consistency, mathematical verification of factor returns, and strict avoidance of LLM hallucination in financial statistics.
- **Autonomous Operations**: Completely automated pipeline executing pre-market calculations, data synthesis, multilingual HTML/Markdown generation, and CDN deployment daily.

---

---

## 4. Institutional 8-Pillar Data Architecture (100+ Granular Metrics)

Every post-close trading session, K2J filters the broader US market for optimal liquidity and tradability, ingesting and computing across 8 comprehensive data pillars for over 1,500 liquid US equities:

1. **Asset Metadata, Share Structure & Wall Street Consensus** (20+ metrics): CUSIP/ISIN identifiers, Free Float %, institutional ownership, upcoming earnings proximity alerts, analyst ratings, and 1-year target price upside dispersion.
2. **Multi-Period Price, Volume & Microstructure** (50+ metrics): Candlestick intent (`BULLISH_DRIVE` vs `FADING_GAP_DUMP`), VWAP deviation %, relative volume (10D/30D/60D/90D averages), turnover, multi-period highs/lows (ATH, 52W, 6M, 3M), and extended-hours pre/post-market quotes.
3. **Full-Period Technical Indicators & Pivot Systems** (50+ metrics): Multi-timeframe moving averages (SMA/EMA/Hull), Ichimoku cloud, momentum oscillators (RSI, Stochastic, ADX, ATR, CCI), institutional money channels (VWAP, Bollinger Bands, CMF, MFI), and 5 classic pivot point systems (Camarilla, Classic, DeMark, Fibonacci, Woodie).
4. **Financial Statements Deep-Dive (TTM & FQ)** (30+ metrics): Income statement (revenue, EBITDA, R&D expense), Cash Flow statement (Free Cash Flow, Operating Cash Flow, CapEx), and Balance Sheet (cash, net debt, total equity, goodwill).
5. **Advanced Enterprise Valuation Multiples** (15+ metrics): EV/EBITDA, EV/EBIT, EV/Revenue, EV/FCF, P/FCF, PEG, Earnings Yield %, Buyback Yield %.
6. **Multi-Period Growth Rates** (8+ metrics): YoY quarterly and TTM growth across revenue, gross profit, EBITDA, net income, free cash flow, assets, and liabilities.
7. **Capital Ratios & 4 Classic Forensic Quant Models** (20+ metrics): Return on capital (ROE, ROA, ROIC), liquidity ratios, and four classic forensic accounting risk models:
   - **Piotroski F-Score (0-9)**: Fundamental operational health verification (Longs must have F ≥ 3).
   - **Altman Z-Score**: Corporate bankruptcy and default risk barrier (Z < 1.81 distressed warning).
   - **Sloan Ratio %**: Accrual earnings quality audit (Sloan > 10% flags aggressive earnings manipulation).
   - **Zmijewski Score**: Financial distress probability cross-validation.
8. **Shareholder Yield & Dividend Policy** (8+ metrics): Dividend yield, dividend growth persistence, payout ratios, and Total Shareholder Yield (Buybacks + Dividends).

---

## 5. Intelligence & Regulatory Surveillance Engine

### 5.1 NLP-Powered Causal News Intelligence (70+ Wire Feeds)
- **Real-Time Surveillance Across 70+ Sources**: Ingests and parses wire services, financial media, and regulatory feeds 24/7.
- **Causal Event–Ticker Relevance Scoring**: Employs deep NLP models to score the true causal impact between breaking news and targeted liquid equities.
- **Bullish/Bearish Sentiment Polarity Decomposition**: Dissects sentiment magnitude, directional conviction, and institutional narrative weight.
- **Event-Driven Tail-Risk Signal Extraction**: Filters out background noise and isolates asymmetric tail-risk catalysts (earnings surprises, regulatory actions, dilutive financing, management turnover).
- **Alpha Overlay**: Synthesizes unstructured news signals into a forward-looking catalytic layer that resonates with price/volume and factor data.

### 5.2 SEC EDGAR & 13F Smart Money Regulatory Intelligence Pipeline
- **Real-Time SEC EDGAR Ingestion**: Automated ingestion and structured parsing of mandatory regulatory submissions:
  - **Form 13F**: Quarterly institutional investment manager holdings and interim portfolio amendments.
  - **Form 13D / 13G**: Beneficial ownership disclosures triggered when institutional entities cross the 5% equity ownership threshold.
  - **Form 4**: Statements of changes in beneficial ownership by corporate insiders, officers, and directors.
  - **Form 8-K**: Unscheduled material corporate events, executive departures, financing terms, and strategic transactions.
- **Smart Money Flow & Divergence Forensics**: Cross-references disclosed institutional positioning against intraday and multi-day price-volume microstructure. Isolates subtle divergences between declared institutional allocations and actual market order flow to detect stealth accumulation or liquidation before broad consensus discovery.
- **Triple-Engine Resonance**: Unifies "Price/Volume Multi-Factor Models + Causal News Sentiment + Smart Money Regulatory Filings" into an institutional-grade multi-layer decision framework.

---

## 6. Daily 20 Focus Equities Selection & Sector Divergence Forensics

### 6.1 Three-Layer Screening Funnel
- **Layer 1 (2 Extreme Movers)**: Single largest daily gainer and single largest decliner automatically qualify, ensuring market extremes and black-swan anomalies are tracked.
- **Layer 2 (18 5-D Resonance Leaders)**: Sorted by 5-D factor resonance tier (`5/5★` → `4/5` → `3/5`), composite attribution score, and an Industry Alignment filter (≥50%) to discard isolated noise.
- **Layer 3 (Diversity Cap)**: Maximum 3 tickers per thematic cluster and 3 tickers per sub-industry to prevent sector clustering and preserve macro market breadth.

### 6.2 5-D Factor Attribution Formula
$$\text{Narrative Score} = \Phi\left(0.20 D_1 + 0.25 D_2 + 0.25 D_3 + 0.15 D_4 + 0.15 D_5\right) \times 100$$
- **D1 (20%)**: Index Alignment & Systematic Beta co-movement.
- **D2 (25%)**: Sector Clustering, residual Alpha, and sub-industry win rate.
- **D3 (25%)**: Capital Flow, Candlestick Intent, VWAP deviation, and Chaikin Money Flow.
- **D4 (15%)**: Price Action, ATH/52W breakout badges, and moving average momentum.
- **D5 (15%)**: Fundamentals, cash flow multiples, and forensic accounting red lines.

### 6.3 Sector Divergence (⚠️) Forensic Alerts
When a stock collapses while its sub-industry surges (e.g. gold mining up +2.81% across 80% of peers, but SA drops -9.35%), the system tags it as **Sector Divergence**:
- D2 Sector dimension receives a punitive `-2.0` score.
- Composite Z-score is reduced by `-0.50`, classifying the move as an *Isolated Divergence Drop* rather than a resonant short.
- Targeted AI news reasoning extracts the company-specific black swan (e.g., dilutive offerings) and displays an orange `⚠️ Divergent` warning banner.

---

## 7. Top-Down Reading Blueprint (Macro → Sector → Equities)

Readers are guided through a disciplined 4-step 5-minute workflow:
1. **Step 1 (1 Min) · Macro Regime**: Check Section 01 for macro climate (Risk-On vs Risk-Off) and full-market advance/decline breadth to calibrate overall portfolio risk budget.
2. **Step 2 (2 Mins) · Thematic Clusters**: Scan Section 02 to identify sub-industries where institutional smart money is accumulating with high internal win rates (>70%).
3. **Step 3 (2 Mins) · 20 Focus Tickers**: Review Section 03 summary and Section 06 master table for `5/5★` golden resonance badges and `⚠️ Divergent` warnings.
4. **Step 4 (On-Demand) · Equities Deep-Dive**: Explore Section 04/05 cards for AI causal news intelligence ("Why It Moved"), 6-pillar business moat, analyst targets, and 3-day earnings proximity warnings.

---

## 8. Frequently Asked Questions (FAQ)

### Q1: What makes K2J AI Quantitative Lab unique?
A1: Unlike purely discretionary commentary or black-box academic models, K2J combines rigorous multi-factor statistical mechanics with production-grade autonomous agentic AI. It is built to institutional rigor by a veteran practitioner who actively trades the signals daily.

### Q2: What technology stack does K2J use?
A2: The AI architecture leverages LangGraph for multi-agent workflows, Pinecone for vector retrieval, and Python (uv, NumPy, SciPy, LightGBM, scikit-learn) for quantitative modeling, backed by deterministic guardrail engines.

### Q3: How can I access the daily reports and reading guide?
A3: Daily reports are published every trading morning in both English and Simplified Chinese under `https://k2j.net/reports/`. The comprehensive methodology guide is available at `https://k2j.net/guide.html` (`https://k2j.net/guide.zh.html` and `https://k2j.net/guide.en.html`).

### Q4: What collaboration opportunities are available with Justin?
A4: Justin is open to strategic advisory, technical consulting, and collaborative partnerships with hedge funds, proprietary trading firms, asset managers, and AI startups seeking proven expertise in building production agentic AI systems and quantitative finance models. Contact via email at `k2jysy@gmail.com` or LinkedIn at `https://www.linkedin.com/in/justin-k2j`.

---

### Q5: How does K2J track smart money through 13F and SEC filings?
A5: K2J continuously parses SEC EDGAR filings (Form 13F institutional holdings, 13D/13G beneficial ownership, Form 4 insider trades, 8-K material notices) and cross-references them against price-volume microstructure. This reveals divergences between disclosed institutional stakes and real-time market liquidity, pinpointing smart money accumulation and distribution before broad price discovery.

## 9. Contact & Key Resources

- **Website**: https://k2j.net/
- **Methodology Guide**: https://k2j.net/guide.html
  - Chinese Guide: https://k2j.net/guide.zh.html
  - English Guide: https://k2j.net/guide.en.html
- **Privacy Policy**: https://k2j.net/privacy.html
  - Chinese Privacy Policy: https://k2j.net/privacy.zh.html
  - English Privacy Policy: https://k2j.net/privacy.en.html
- **Author**: Justin
- **Email**: k2jysy@gmail.com
- **LinkedIn**: https://www.linkedin.com/in/justin-k2j
- **Reports Directory**: https://k2j.net/reports/
- **LLM Summary**: https://k2j.net/llms.txt
- **LLM Full Whitepaper**: https://k2j.net/llms-full.txt