market-top-detector

Market Top Detector Skill

Purpose

Detect the probability of a market top formation using a quantitative 6-component scoring system (0-100). Integrates three proven market top detection methodologies:

1. O'Neil - Distribution Day accumulation (institutional selling)
2. Minervini - Leading stock deterioration pattern
3. Monty - Defensive sector rotation signal

Unlike the Bubble Detector (macro/multi-month evaluation), this skill focuses on tactical 2-8 week timing signals that precede 10-20% market corrections.

When to Use This Skill

English:

• User asks "Is the market topping?" or "Are we near a top?"
• User notices distribution days accumulating
• User observes defensive sectors outperforming growth
• User sees leading stocks breaking down while indices hold
• User asks about reducing equity exposure timing
• User wants to assess correction probability for the next 2-8 weeks

Japanese:

• 「天井が近い？」「今は利確すべき？」
• ディストリビューションデーの蓄積を懸念
• ディフェンシブセクターがグロースをアウトパフォーム
• 先導株が崩れ始めているが指数はまだ持ちこたえている
• エクスポージャー縮小のタイミング判断
• 今後2〜8週間の調整確率を評価したい

Prerequisites

Required:

• FMP API Key: Set $FMP_API_KEY environment variable or pass --api-key. Free tier sufficient (~33 API calls per execution).
• WebSearch Access: Required to collect S&P 500 breadth (50DMA %) and CBOE Put/Call ratio data.

Optional:

• Margin Debt Data: Enhances sentiment scoring but typically 1-2 months lagged.
• VIX Term Structure: Auto-detected from FMP API if VIX3M quote available; manual override via --vix-term.

Data Freshness: All manually collected data should be from the most recent 3 business days for accurate analysis.

Difference from Bubble Detector

Execution Workflow

Phase 1: Data Collection via WebSearch

Before running the Python script, collect the following data using WebSearch. Data Freshness Requirement: All data must be from the most recent 3 business days. Stale data degrades analysis quality.

1. S&P 500 Breadth (200DMA above %)
   AUTO-FETCHED from TraderMonty CSV (no WebSearch needed)
   The script fetches this automatically from GitHub Pages CSV data.
   Override: --breadth-200dma [VALUE] to use a manual value instead.
   Disable: --no-auto-breadth to skip auto-fetch entirely.

2. [REQUIRED] S&P 500 Breadth (50DMA above %)
   Valid range: 20-100
   Primary search: "S&P 500 percent stocks above 50 day moving average"
   Fallback: "market breadth 50dma site:barchart.com"
   Direct fallback when search snippets are poor: fetch `https://www.barchart.com/stocks/quotes/$S5FI/overview` and extract the embedded `lastPrice` / `tradeTime` for “S&P 500 Stocks Above 50-Day Average”.
   Record the data date

3. [REQUIRED] CBOE Equity Put/Call Ratio
   Valid range: 0.30-1.50
   Primary search: "CBOE equity put call ratio today"
   Fallback: "CBOE total put call ratio current"
   Fallback: "put call ratio site:cboe.com"
   Direct fallback when Cboe CSV endpoints are stale: fetch `https://ycharts.com/indicators/cboe_equity_put_call_ratio` and parse the “Last Value” / “Latest Period” table fields. Treat this as a secondary source and cite it in freshness notes.
   Record the data date

4. [OPTIONAL] VIX Term Structure
   Values: steep_contango / contango / flat / backwardation
   Primary search: "VIX VIX3M ratio term structure today"
   Fallback: "VIX futures term structure contango backwardation"
   Note: Auto-detected from FMP API if VIX3M quote available.
   CLI --vix-term overrides auto-detection.

5. [OPTIONAL] Margin Debt YoY %
   Primary search: "FINRA margin debt latest year over year percent"
   Fallback: "NYSE margin debt monthly"
   Note: Typically 1-2 months lagged. Record the reporting month.

Phase 2: Execute Python Script

Run the script with collected data as CLI arguments:

python3 skills/market-top-detector/scripts/market_top_detector.py \
  --api-key $FMP_API_KEY \
  --breadth-50dma [VALUE] --breadth-50dma-date [YYYY-MM-DD] \
  --put-call [VALUE] --put-call-date [YYYY-MM-DD] \
  --vix-term [steep_contango|contango|flat|backwardation] \
  --margin-debt-yoy [VALUE] --margin-debt-date [YYYY-MM-DD] \
  --output-dir reports/ \
  --context "Consumer Confidence=[VALUE]" "Gold Price=[VALUE]"
# 200DMA breadth is auto-fetched from TraderMonty CSV.
# Override with --breadth-200dma [VALUE] if needed.
# Disable with --no-auto-breadth to skip auto-fetch.

The script will:

1. Fetch S&P 500, QQQ, VIX quotes and history from FMP API
2. Fetch Leading ETF (ARKK, WCLD, IGV, XBI, SOXX, SMH, KWEB, TAN) data
3. Fetch Sector ETF (XLU, XLP, XLV, VNQ, XLK, XLC, XLY) data
4. Calculate all 6 components
5. Generate composite score and reports

Phase 3: Present Results

Present the generated Markdown report to the user, highlighting:

• Composite score and risk zone
• Data freshness warnings (if any data older than 3 days)
• Strongest warning signal (highest component score)
• Historical comparison (closest past top pattern)
• What-if scenarios (sensitivity to key changes)
• Recommended actions based on risk zone
• Follow-Through Day status (if applicable)
• Delta vs previous run (if prior report exists)

6-Component Scoring System

Risk Zone Mapping

Exchange Calendar and Replay

Install requirements.txt before running the detector. Freshness uses XNYS sessions rather than weekdays. --as-of YYYY-MM-DD is accepted for the live evaluation date, but historical live replay fails closed because the current quote endpoints are not point-in-time sources.

API Requirements

Required: FMP API key (free tier sufficient: ~33 calls per execution) Optional: WebSearch data for breadth and sentiment (improves accuracy)

Output Files

• JSON: market_top_YYYY-MM-DD_HHMMSS.json
• Markdown: market_top_YYYY-MM-DD_HHMMSS.md

Reference Documents

references/market_top_methodology.md

• Full methodology with O'Neil, Minervini, and Monty frameworks
• Component scoring details and thresholds
• Historical validation notes

references/distribution_day_guide.md

• Detailed O'Neil Distribution Day rules
• Stalling day identification
• Follow-Through Day (FTD) mechanics

references/historical_tops.md

• Analysis of 2000, 2007, 2018, 2022 market tops
• Component score patterns during historical tops
• Lessons learned and calibration data

When to Load References

• First use: Load market_top_methodology.md for full framework understanding
• Distribution day questions: Load distribution_day_guide.md
• Historical context: Load historical_tops.md
• Regular execution: References not needed - script handles scoring