alphagbm-take-profit

AlphaGBM Take-Profit Strategy Lab

Answer one question mechanically for any ticker: Can you just hold it, or do you need to actively take profits? Most retail losses come from poor exits, not poor entries. This skill quantifies the exit decision with 10 years of daily data.

The Core Metric: Rollercoaster Rate

A "rollercoaster event" happens when an entry's paper profit exceeds +50% and then falls more than 50% from that peak before exit. Example: enter at 100, peak at 190, fall back to 90 — you didn't lose money, but the 90 of peak profit you "touched" evaporated, and the journey was brutal.

Rollercoaster rate varies up to 97 percentage points across instruments:

• Broad-index ETFs (SPY, VTI): 0% — hold forever
• Blue chips (AAPL, MSFT): 0% — hold forever
• Sector ETFs (SOXX, XLK): 0% — hold forever
• Large-cap mega-caps (META, AMZN): ~47% — tiered exit preferred
• HK tech (腾讯, 阿里): ~49% — tiered exit preferred
• High growth (NVDA, TSLA, AMD): ~85% — tiered exit mandatory
• Crypto-related (COIN, MSTR): ~90% — tiered exit mandatory
• Leveraged ETFs (TQQQ, SOXL): ~97% — structurally un-holdable

Whether you can hold is an instrument property, not an attitude problem.

Strategy Universe (15 total)

• A family (sell all at trigger): A_+50%, A_+100%, A_+200%
• B family (tiered): B_50/100/200 (default), B_30/60/100, B2_20/40/80, B3_40/80/150, B5 back-weighted, B6 front-weighted
• C_10x (conviction hold)
• D (-20% / -30% trailing stop) — loses to hold on every tested ticker
• E (never sell / long-hold)
• F (peak-pullback after +50% activation)
• G (HV-aware: picks A_+100% or A_+200% based on entry-day vol)

How to Use

Input:

• ticker (required) — any US / HK / CN stock, ETF, or leveraged ETF

Output:

• Profile: color (green/amber/red) + special_flag (no_hold for leveraged ETFs, reverse_alpha for declining stocks where active selling beats hold)
• Headline numbers: rollercoaster_rate, max_drawdown, hold_cagr
• strategy_results: 15 strategies, each with {cagr, rc, mdd} medians
• Provenance: sample_size (typically ~120 entry points), period, computed_at

The caller is expected to:

1. Display the headline profile
2. Recommend a strategy matching user's personality + position size (front-end logic)
3. Generate concrete GTC limit-sell orders at entry × 1.5 / 2.0 / 3.0 etc.

Example Queries

• should I hold TQQQ long-term → no_hold flag + rollercoaster 97% → tiered exit
• take-profit strategy for NVDA → high-growth profile, B_50/100/200 default
• is AAPL holdable → blue-chip profile, 0% rollercoaster, hold recommended
• when should I sell COIN → crypto profile, mandatory tiered exit
• rollercoaster rate for SPY → 0%, long-hold optimal
• backtest sell strategies for MSFT → full 15-strategy comparison

Mock Data

Mock data in mock-data/take-profit/ — sample responses for TQQQ (no_hold), AAPL (hold-optimal), and PYPL (reverse_alpha).

API Endpoint

POST /api/stock/take-profit-analyze
Content-Type: application/json

Request body:

{"ticker": "TQQQ"}

Also available for reading the cached library (no quota):

GET /api/stock/take-profit-library

Returns list of already-cached tickers with their headline numbers — useful for agents to know which queries are instant vs first-time.

Response shape:

{
  "success": true,
  "ticker": "TQQQ",
  "color": "red",
  "special_flag": "no_hold",
  "rollercoaster_rate": 97,
  "max_drawdown": -82,
  "hold_cagr": 37.0,
  "strategy_results": {
    "A_50": {"cagr": 6.0, "rc": 21, "mdd": -32},
    "A_100": {"cagr": 11.5, "rc": 44, "mdd": -50},
    "A_200": {"cagr": 17.6, "rc": 64, "mdd": -62},
    "B_50_100_200": {"cagr": 12.0, "rc": 36, "mdd": -42},
    "B6_front": {"cagr": 11.0, "rc": 29, "mdd": -37},
    "E_hold": {"cagr": 37.0, "rc": 97, "mdd": -82},
    "...": "..."
  },
  "sample_size": 120,
  "period": {"start": "2014-04-20", "end": "2026-04-20"},
  "computed_at": "2026-04-24T08:00:00"
}

Pricing: 1 stock-analysis credit per first-time ticker compute; DB-cached for 30 days globally — once computed, all users get instant reads (including cache hits within 5 min in-process). Cache hits do not deduct credits.

First-time compute takes ~30s (10 years of daily data × 15 strategies × ~120 entry points = ~1800 simulations). Subsequent reads are <100 ms.

Related Skills

---

Powered by AlphaGBM — Real-data options & research intelligence. 10K+ users.