Agent Skills › AlphaGBM/skills › alphagbm-bps-backtest

alphagbm-bps-backtest

GitHub

对牛市看跌价差策略进行历史回测，对比信号触发与无信号控制组的收益表现。支持自定义参数并输出权益曲线、KPI及交易明细，用于验证策略Alpha有效性。

skills/alphagbm-bps-backtest/SKILL.md AlphaGBM/skills

触发场景

backtest BPS on QQQ bull put spread backtest does FearScore work on SPY

安装

npx skills add AlphaGBM/skills --skill alphagbm-bps-backtest -g -y

更多选项

不安装直接使用

npx skills use AlphaGBM/skills@alphagbm-bps-backtest

指定 Agent (Claude Code)

npx skills add AlphaGBM/skills --skill alphagbm-bps-backtest -a claude-code -g -y

安装 repo 全部 skill

npx skills add AlphaGBM/skills --all -g -y

预览 repo 内 skill

npx skills add AlphaGBM/skills --list

SKILL.md

Frontmatter

{
    "name": "alphagbm-bps-backtest",
    "globs": [
        "mock-data\/bps-backtest\/**"
    ],
    "description": "Full walk-forward Bull Put Spread backtest over ~8 years of daily history. Runs\nboth the signal (FearScore ≥ 60 entry) version AND a no-signal control in the\nsame request, so you can quantify whether the fear-entry rule actually delivers\nalpha for this ticker under your parameters. Returns equity curve, 4 KPIs\n(annualized return \/ win rate \/ max drawdown \/ Sharpe), trade ledger, and a\nplain-language takeaway.\nTriggers: \"backtest BPS on QQQ\", \"bull put spread backtest\", \"does FearScore\nwork on SPY\", \"what DTE for BPS\", \"optimal bull put spread delta\", \"BPS strategy\nbacktest\", \"credit spread backtest\", \"backtest short put spread\"\n"
}

AlphaGBM BPS Backtest

Backtests the Bull Put Spread (short put + long put at lower strike) as a mechanical strategy over 2018–present on any ticker, with two passes per call:

With Signal — only enters when the per-ticker FearScore is ≥ your threshold
No Signal (Control) — enters unconditionally every Monday

The side-by-side comparison shows whether the signal is doing work, or whether you're paying 1 credit for noise.

Parameters

All optional except ticker:

Param	Default	Range	Meaning
`ticker`	required	US / HK / CN	Underlying
`dte_target`	14	7–45	Days to expiry on entry
`short_delta`	0.25	0.15–0.35	Absolute delta of the short put leg
`spread_width`	5.0	2–10	Dollar width of the spread
`take_profit_pct`	0.50	0.20–0.80	Close when realized % of max profit hits this
`fear_threshold`	60	40–80	FearScore ≥ X is entry signal
`start_date`	2018-01-01	YYYY-MM-DD	Backtest start
`end_date`	2026-04-20	YYYY-MM-DD	Backtest end
`include_control`	true	bool	Run no-signal control pass alongside

What's Returned

Per pass (with_signal and no_signal):

total_trades, win_rate_pct, annual_return_pct, sharpe, max_drawdown_pct, roc_pct, avg_holding_days, avg_pnl_per_trade, total_pnl, final_capital
exit_reasons — count by take_profit / stop_loss / expiry_otm / expiry_itm / close_early
trades[] — full ledger (entry/exit date, strikes, credit, pnl, reason)
equity_curve[] — per-day cumulative capital
pnl_histogram — bucket counts for the P&L distribution

Plus:

summary — one-paragraph zh/en takeaway comparing signal vs control, with ⚠️ flags when drawdown or win rate look problematic

Methodology Notes

IV is proxied by 20-day historical volatility (HV20) for BS pricing. Historical option-chain IV is unaffordable to source at scale; HV20 is a reasonable proxy but will under-estimate IV around events. Live results typically outperform backtest because of this.
FearScore is reconstructed from the same 6 indicators the live version uses, but computed from cheap historical price + volume data only.
Entries filtered by max_positions (3) and min_entry_spacing_days (3) and a risk_per_trade cap (0.5% of capital).

How to Use

Example Queries:

backtest BPS on QQQ — Default params, signal vs control comparison
does FearScore work on SPY — Same call, reads the comparison summary
backtest bull put spread IWM DTE 21 delta 0.30 — Custom params
what DTE works best for BPS on QQQ — Run a few with different DTEs, compare
bps fear threshold 70 vs 60 on NVDA — Run two calls with different thresholds

Mock Data

Mock data in mock-data/bps-backtest/ — examples for QQQ with signal ON and OFF.

API Endpoint

POST /api/options/bps-backtest
Content-Type: application/json

Request body:

{
  "ticker": "QQQ",
  "dte_target": 14,
  "short_delta": 0.25,
  "spread_width": 5.0,
  "take_profit_pct": 0.50,
  "fear_threshold": 60,
  "start_date": "2018-01-01",
  "end_date": "2026-04-20",
  "include_control": true
}

Response:

{
  "success": true,
  "ticker": "QQQ",
  "period": {"start": "2018-01-01", "end": "2026-04-20"},
  "with_signal": {
    "total_trades": 28, "win_rate_pct": 100, "annual_return_pct": 10.8,
    "sharpe": 16.3, "max_drawdown_pct": 0.0, "trades": [...], "equity_curve": [...],
    "pnl_histogram": {...}, "exit_reasons": {"take_profit": 20, "expiry_otm": 8}
  },
  "no_signal": {
    "total_trades": 185, "win_rate_pct": 82, "annual_return_pct": 3.5,
    "sharpe": 2.1, "max_drawdown_pct": -8.2, ...
  },
  "summary": {
    "zh": "QQQ · 2018-2026 · 使用 FearScore ≥ 60 触发 BPS 入场，共交易 28 笔，年化 +10.8%，胜率 100%，最大回撤 0.0%。 同参数无信号对照组年化 +3.5%、胜率 82%；信号版本高出无信号组 7.3 个百分点。",
    "en": "QQQ · 2018-2026 · BPS entry on FearScore ≥ 60 over 28 trades: annualized +10.8%, win rate 100%, max drawdown 0.0%. The no-signal control under the same params: annualized +3.5%, win rate 82%. Signal version outperforms by 7.3 pp."
  }
}

Pricing: 1 option-analysis credit per call; 30-min cache per parameter hash (cache hits free). Expect ~5-10s compute for a fresh hash.

Related Skills

Skill	Relevance
alphagbm-fear-score	The live version of the entry signal being backtested
alphagbm-options-strategy	Build a custom BPS after deciding params
alphagbm-pnl-simulator	Forward-simulate a specific BPS at various future prices