momentum-screen

name	momentum-screen
description	Deprecated shim — returns Minervini SEPA only. Use minervini-sepa, can-slim, darvas-box, livermore-pivot, or analyze_ticker momentum/full mode for all four master strategies.
license	Apache-2.0
compatibility	Prompt-first Agent Skill. Usable with the script absent. Script is Python 3.8+ stdlib-only, no network.
metadata	{"author":"stock-buddy","version":"0.2.0","prd_refs":["PRD-001:REQ-026","PRD-001:REQ-027","PRD-001:REQ-032"],"mode":["momentum"]}

Momentum Screen

Prompt-first skill. Score the 25-point checklist by reasoning over the data. scripts/screen.py is OPTIONAL — run it for exact indicator values; the screen is valid without it. Score only criteria you can evaluate; mark the rest "not counted".

Role & objective

You are a momentum screener (Minervini SEPA + Driehaus). Goal: check 25 criteria across 5 categories, report pass/fail with a beginner-friendly reason each, and map a category-weighted score (0..1) to a Momentum Grade A+..F.

When to use

"Is GP a momentum buy?", "does this qualify as a leader?", "trend-template / SEPA check", "relative-strength read". Pair with technical-analysis; size entries with risk-manager.

Inputs you need

Gather via dse-data-acquisition:

• ohlcv — daily bars, oldest-first, ≥30 (≥200 for the full MA stack).
• (optional) fundamentals.eps_history, earnings_surprise (Driehaus checks).
• (optional) market_index close series (relative strength).

Method — 25 criteria in 5 weighted categories

Each criterion is pass/fail; if its data is missing, mark it "not counted" (don't guess).

Trend — Minervini SEPA (weight 0.25): 1) close > 50-MA; 2) close > 150-MA; 3) close > 200-MA; 4) 50-MA > 150-MA; 5) 150-MA > 200-MA; 6) 200-MA rising (~1 month); 7) within 25% of 52-week high; 8) ≥30% above 52-week low.

Momentum power (weight 0.25): 9) RSI 40–70; 10) MACD > signal; 11) ROC positive and rising; 12) ADX > 25; 13) MFI 20–80; 14) Bollinger %B 0.5–1.0.

Volume (weight 0.20): 15) OBV trending up; 16) volume > 20-day avg; 17) Accumulation/ Distribution rising; 18) Volume-ROC positive.

Relative performance — Driehaus (weight 0.15): 19) earnings acceleration (each year's EPS growth beats the last); 20) positive recent earnings surprise; 21) institutional accumulation (relative volume > 1.2×); 22) relative strength vs index positive (stock ROC > index ROC).

Risk (weight 0.15): 23) ATR/price < 6%; 24) distance from 60-bar support < 8%; 25) no major resistance within 10%.

Scoring rubric

For each category, fraction = (criteria passed)/(criteria evaluable). score = 0.25·trend + 0.25·momentum_power + 0.20·volume + 0.15·relative_performance + 0.15·risk.

Grade: ≥0.9 A+ · ≥0.8 A · ≥0.7 B+ · ≥0.6 B · ≥0.5 C · ≥0.4 D · else F.

Confidence = clamp(0.5 + 0.45·(evaluable/25) − (0.1 if <200 bars), 0.1, 0.95).

Output (emit this Thinking Card)

{ "skill": "momentum-screen", "ticker": "..", "mode": "momentum", "as_of": "..",
  "score": 0.0, "confidence": 0.0, "rating": "B+",
  "key_metrics": { "overall_count": "18/25", "criteria_passed": 18, "criteria_evaluated": 23,
    "categories": { "trend": {"criteria_met": 0, "total": 0, "fraction": 0.0} } },
  "reasoning": ["✓/✗/? criterion — why"], "flags": ["..."],
  "disclaimer": "Educational analysis only. Not financial advice." }

DSE pitfalls

• Many DSE names lack earnings_surprise / clean eps_history and there's no index series — mark those Driehaus criteria "not counted" (flag missing:earnings_surprise, missing:market_index) instead of failing them.
• limited_history_<200_bars weakens the SEPA stack — flag and reduce confidence.
• A high grade on thin volume is suspect — volume criteria must genuinely pass.

Optional precision helper

python3 scripts/screen.py --input data.json --pretty

Returns each criterion's pass/fail, the per-category fractions, the 0..1 score and the grade. Use it to verify counts; trust the script if your tally differs.

Worked example

22 of 25 criteria evaluable, 18 pass — trend 7/8, momentum 5/6, volume 4/4, Driehaus 1/3, risk 1/3 → score ≈ 0.71 → B+, confidence ≈ 0.84.

References

Checklist detail: references/CHECKLIST.md. Output is educational analysis only, never financial advice.