Theme Detector

# Theme Detector

## Overview

This skill detects and ranks trending market themes by analyzing cross-sector momentum, volume, and breadth signals. It identifies both bullish (upward momentum) and bearish (downward pressure) themes, assesses lifecycle maturity (Emerging/Accelerating/Trending/Mature/Exhausting), and provides a confidence score combining quantitative data with narrative analysis.

**3-Dimensional Scoring Model:**
1. **Theme Heat** (0-100): Direction-neutral strength of the theme (momentum, volume, uptrend ratio, breadth)
2. **Lifecycle Maturity**: Stage classification (Emerging / Accelerating / Trending / Mature / Exhausting) based on duration, extremity clustering, valuation, and ETF proliferation
3. **Confidence** (Low / Medium / High): Reliability of the detection, combining quantitative breadth with narrative confirmation. Script output is capped at Medium; Claude's WebSearch narrative confirmation step can elevate to High.

**Key Features:**
- Cross-sector theme detection using FINVIZ industry data
- Direction-aware scoring (bullish and bearish themes)
- Lifecycle maturity assessment to identify crowded vs. emerging trades
- ETF proliferation scoring (more ETFs = more mature/crowded theme)
- Integration with uptrend-dashboard for 3-point evaluation
- Dual-mode operation: FINVIZ Elite (fast) or public scraping (slower, limited)
- WebSearch-based narrative confirmation for top themes

---

## When to Use This Skill

**Explicit Triggers:**
- "What market themes are trending right now?"
- "Which sectors are hot/cold?"
- "Detect current market themes"
- "What are the strongest bullish/bearish narratives?"
- "Is AI/clean energy/defense still a strong theme?"
- "Where is sector rotation heading?"
- "Show me thematic investing opportunities"

**Implicit Triggers:**
- User wants to understand broad market narrative shifts
- User is looking for thematic ETF or sector allocation ideas
- User asks about crowded trades or late-cycle themes
- User wants to know which themes are emerging vs. exhausted

**When NOT to Use:**
- Individual stock analysis (use us-stock-analysis instead)
- Specific sector deep-dive with chart reading (use sector-analyst instead)
- Portfolio rebalancing (use portfolio-manager instead)
- Dividend/income investing (use value-dividend-screener instead)

---

## Prerequisites

**Required:**
- Python 3.10+ with core dependencies. The scripts use modern type-hint syntax such as `dict | None`, so Python 3.9 and older can fail at import time even when dependencies are installed.
  ```bash
  pip install requests beautifulsoup4 lxml pandas numpy yfinance
  ```

**Cron / mixed-Python fallback:** If the active `python3` is older than 3.10, or a newer Hermes venv lacks the data-science dependencies, run the detector through `uv` with an explicit modern interpreter and temporary dependencies instead of editing the environment mid-cron:
```bash
uv run --python 3.12 \
  --with requests --with beautifulsoup4 --with lxml \
  --with pandas --with numpy --with yfinance \
  --with finvizfinance --with PyYAML \
  python skills/theme-detector/scripts/theme_detector.py \
  --finviz-api-key "$FINVIZ_API_KEY" \
  --fmp-api-key "$FMP_API_KEY" \
  --output-dir reports/
```
Use this as a setup workaround, not as evidence that the detector is broken; still report FINVIZ/FMP/API-data caveats separately.

**Optional API Keys:**

FINVIZ Elite (recommended for full industry coverage and speed):
```bash
export FINVIZ_API_KEY=your_finviz_elite_api_key_here
```

FMP API (optional, for P/E ratio valuation data):
```bash
export FMP_API_KEY=your_fmp_api_key_here
```

**Optional Python packages:**
- `finvizfinance` - Required for FINVIZ Elite mode
- `PyYAML` - Required for `--themes-config` custom themes

Without FINVIZ Elite, the skill uses public FINVIZ scraping (limited to ~20 stocks per industry, slower rate limits).

---

## Workflow

### Step 1: Verify Environment

Check that API keys are configured (see Prerequisites):

```bash
# Verify FINVIZ Elite API key (optional but recommended)
echo $FINVIZ_API_KEY

# Verify FMP API key (optional)
echo $FMP_API_KEY
```

### Step 2: Execute Theme Detection Script

Run the main detection script:

```bash
python3 skills/theme-detector/scripts/theme_detector.py \
  --output-dir reports/
```

**Script Options:**
```bash
# Full run (public FINVIZ mode, no API key required)
python3 skills/theme-detector/scripts/theme_detector.py \
  --output-dir reports/

# With FINVIZ Elite API key
python3 skills/theme-detector/scripts/theme_detector.py \
  --finviz-api-key $FINVIZ_API_KEY \
  --output-dir reports/

# With FMP API key for enhanced stock data
python3 skills/theme-detector/scripts/theme_detector.py \
  --fmp-api-key $FMP_API_KEY \
  --output-dir reports/

# Custom limits
python3 skills/theme-detector/scripts/theme_detector.py \
  --max-themes 5 \
  --max-stocks-per-theme 10 \
  --output-dir reports/

# Explicit FINVIZ mode
python3 skills/theme-detector/scripts/theme_detector.py \
  --finviz-mode public \
  --output-dir reports/
```

**Expected Execution Time:**
- FINVIZ Elite mode: ~2-3 minutes (14+ themes)
- Public FINVIZ mode: ~5-8 minutes (rate-limited scraping)

### Step 3: Read and Parse Detection Results

The script generates two output files:
- `theme_detector_YYYY-MM-DD_HHMMSS.json` - Structured data for programmatic use
- `theme_detector_YYYY-MM-DD_HHMMSS.md` - Human-readable report

Read the JSON output to understand quantitative results:

```bash
# Find the latest report
ls -lt reports/theme_detector_*.json | head -1

# Read the JSON output
cat reports/theme_detector_YYYY-MM-DD_HHMMSS.json
```

### Step 4: Perform Narrative Confirmation via WebSearch

For the top 5 themes (by Theme Heat score), execute WebSearch queries to confirm narrative strength:

**Search Pattern:**
```
"[theme name] stocks market [current month] [current year]"
"[theme name] sector momentum [current month] [current year]"
```

**Evaluate narrative signals:**
- **Strong narrative**: Multiple major outlets covering the theme, analyst upgrades, policy catalysts
- **Moderate narrative**: Some coverage, mixed sentiment, no clear catalyst
- **Weak narrative**: Little coverage, or predominantly contrarian/skeptical tone

Update Confidence levels based on findings:
- Quantitative High + Narrative Strong = **High** confidence
- Quantitative High + Narrative Weak = **Medium** confidence (possible momentum divergence)
- Quantitative Low + Narrative Strong = **Medium** confidence (narrative may lead price)
- Quantitative Low + Narrative Weak = **Low** confidence

### Step 5: Analyze Results and Provide Recommendations

Cross-reference detection results with knowledge bases:

**Reference Documents to Consult:**
1. `references/cross_sector_themes.md` - Theme definitions and constituent industries
2. `references/thematic_etf_catalog.md` - ETF exposure options by theme
3. `references/theme_detection_methodology.md` - Scoring model details
4. `references/finviz_industry_codes.md` - Industry classification reference

**Analysis Framework:**

For **Hot Bullish Themes** (Heat >= 70, Direction = Bullish):
- Identify lifecycle stage (Emerging = opportunity, Mature/Exhausting = caution)
- List top-performing industries within the theme
- Recommend proxy ETFs for exposure
- Flag if ETF proliferation is high (crowded trade warning)

For **Hot Bearish Themes** (Heat >= 70, Direction = Bearish):
- Identify industries under pressure
- Assess if bearish momentum is accelerating or decelerating
- Recommend hedging strategies or sectors to avoid
- Note potential mean-reversion opportunities if lifecycle is Mature/Exhausting

For **Emerging Themes** (Heat 40-69, Lifecycle = Emerging):
- These may represent early rotation signals
- Recommend monitoring with watchlist
- Identify catalyst events that could accelerate the theme

For **Exhausted Themes** (Heat >= 60, Lifecycle = Exhausting):
- Warn about crowded trade risk
- High ETF count confirms excessive retail participation
- Consider contrarian positioning or reducing exposure

### Step 6: Generate Final Report

Present the final report to the user using the report template structure:

```markdown
# Theme Detection Report
**Date:** YYYY-MM-DD
**Mode:** FINVIZ Elite / Public
**Themes Analyzed:** N
**Data Quality:** [note any limitations]

## Theme Dashboard
[Top themes table with Heat, Direction, Lifecycle, Confidence]

## Bullish Themes Detail
[Detailed analysis of bullish themes sorted by Heat]

## Bearish Themes Detail
[Detailed analysis of bearish themes sorted by Heat]

## All Themes Summary
[Complete theme ranking table]

## Industry Rankings
[Top performing and worst performing industries]

## Sector Uptrend Ratios
[Sector-level aggregation if uptrend data available]

## Methodology Notes
[Brief explanation of scoring model]
```

Save the report to `reports/` directory.

---

## Output

The skill generates two output files in the `reports/` directory:

**JSON Output** (`theme_detector_YYYY-MM-DD_HHMMSS.json`):
```json
{
  "report_type": "theme_detector",
  "generated_at": "2026-04-18 10:30:00",
  "metadata": {
    "generated_at": "2026-04-18 10:30:00",
    "data_mode": "full",
    "finviz_mode": "elite",
    "fmp_available": true,
    "max_themes": 14,
    "max_stocks_per_theme": 5,
    "data_sources": {
      "finviz_industries": 152,
      "yfinance_stocks": 68,
      "etf_volume": 24
    }
  },
  "summary": {
    "total_themes": 14,
    "bullish_count": 8,
    "bearish_count": 6,
    "top_bullish": "AI & Machine Learning",
    "top_bearish": "Regional Banks"
  },
  "themes": {
    "all": [
      {
        "name": "AI & Machine Learning",
        "direction": "bullish",
        "heat": 85.3,
        "maturity": 42.1,
        "stage": "Accelerating",
        "confidence": "Medium",
        "heat_label": "Hot",
        "industries": ["Software - Infrastructure", "Semiconductors"],
        "representative_stocks": [{"symbol": "NVDA"}, {"symbol": "MSFT"}],
        "proxy_etfs": ["BOTZ", "ROBO"],
        "theme_origin": "seed"
      }
    ],
    "bullish": [...],
    "bearish": [...]
  },
  "industry_rankings": {
    "top": [...],
    "bottom": [...]
  },
  "sector_uptrend": {...},
  "data_quality": {...}
}
```

**Markdown Report** (`theme_detector_YYYY-MM-DD_HHMMSS.md`):
- Theme Dashboard with sortable rankings
- Bullish/Bearish theme detail sections
- Industry performance rankings
- Sector uptrend ratio summary
- Methodology notes

**Key Output Fields (per theme):**
| Field | Description |
|-------|-------------|
| `heat` | 0-100 direction-neutral theme strength |
| `direction` | `"bullish"` (LEAD) or `"bearish"` (LAG) |
| `stage` | Emerging / Accelerating / Trending / Mature / Exhausting |
| `confidence` | Low / Medium / High (script caps at Medium; WebSearch can elevate) |
| `representative_stocks` | Top stocks for the theme (list of objects with `symbol` and metrics) |
| `proxy_etfs` | Thematic ETF tickers (length = ETF count; higher = more crowded) |
| `theme_origin` | `"seed"` (from YAML config) or `"discovered"` (auto-clustered) |

---

## Resources

### Scripts Directory (`scripts/`)

**Main Scripts:**
- `theme_detector.py` - Main orchestrator script
  - Coordinates industry data collection, theme classification, and scoring
  - Generates JSON + Markdown output
  - Usage: `python3 theme_detector.py [options]`

- `theme_classifier.py` - Maps industries to cross-sector themes
  - Reads theme definitions from `cross_sector_themes.md`
  - Calculates theme-level aggregated scores
  - Determines direction (bullish/bearish) from constituent industries
  - Display mapping: "bullish" → "LEAD", "bearish" → "LAG" (see report_generator.py::_direction_label())

- `finviz_industry_scanner.py` - FINVIZ industry data collection
  - Elite mode: CSV export with full stock data per industry
  - Public mode: Web scraping with rate limiting
  - Extracts: performance, volume, change%, avg volume, market cap

- `calculators/lifecycle_calculator.py` - Lifecycle maturity assessment
  - Duration scoring, extremity clustering, valuation analysis
  - ETF proliferation scoring from thematic_etf_catalog.md
  - Stage classification: Emerging / Accelerating / Trending / Mature / Exhausting

- `report_generator.py` - Report output generation
  - Markdown report from template
  - JSON structured output
  - Theme dashboard formatting

### References Directory (`references/`)

**Knowledge Bases:**
- `cross_sector_themes.md` - Theme definitions with industries, ETFs, stocks, and matching criteria
- `thematic_etf_catalog.md` - Comprehensive thematic ETF catalog with counts per theme
- `finviz_industry_codes.md` - Complete FINVIZ industry-to-filter-code mapping
- `theme_detection_methodology.md` - Technical documentation of the 3D scoring model

### Assets Directory (`assets/`)

- `report_template.md` - Markdown template for report generation with placeholder format

---

## Important Notes

### FINVIZ Mode Differences

| Feature | Elite Mode | Public Mode |
|---------|-----------|-------------|
| Industry coverage | All ~145 industries | All ~145 industries |
| Stocks per industry | Full universe | ~20 stocks (page 1) |
| Rate limiting | 0.5s between requests | 2.0s between requests |
| Data freshness | Real-time | 15-min delayed |
| API key required | Yes ($39.50/mo) | No |
| Execution time | ~2-3 minutes | ~5-8 minutes |

### Direction Detection Logic

Theme direction is determined by majority vote of constituent industries' relative rank:

1. **Industry ranking**: All ~145 industries are ranked by multi-timeframe momentum score
2. **Rank-based direction**: Industries in the top half of the ranked list are classified as "bullish"; bottom half as "bearish"
3. **Theme majority vote**: `_majority_direction()` counts bullish vs. bearish industries within each theme; the majority wins

Display mapping: "bullish" → **LEAD**, "bearish" → **LAG** (see `report_generator.py::_direction_label()`)

A LEAD theme indicates relative outperformance of its constituent industries. A LAG theme may still have positive absolute returns — it indicates relative underperformance, not a short signal.

### Known Limitations

1. **Survivorship bias**: Only analyzes currently listed stocks and ETFs
2. **Lag**: FINVIZ data may lag intraday moves by 15 minutes (public mode)
3. **Theme boundaries**: Some stocks fit multiple themes; classification uses primary industry
4. **ETF proliferation**: Catalog is static and may not capture very new ETFs
5. **Narrative scoring**: WebSearch-based and inherently subjective
6. **Public mode limitation**: ~20 stocks per industry may miss small-cap signals

### Disclaimer

**This analysis is for educational and informational purposes only.**
- Not investment advice
- Past thematic trends do not guarantee future performance
- Theme detection identifies momentum, not fundamental value
- Conduct your own research before making investment decisions

---

**Version:** 1.0
**Last Updated:** 2026-02-16
**API Requirements:** FINVIZ Elite (recommended) or public mode (free); FMP API optional
**Execution Time:** ~2-8 minutes depending on mode
**Output Formats:** JSON + Markdown
**Themes Covered:** 14+ cross-sector themes