sector-rotation-rs-screener

name

sector-rotation-rs-screener

description

Prescriptive sector-rotation screener that ranks sectors by relative strength, constituent breadth, and lifecycle stage, emitting a machine-readable FAVOR / ROTATING_IN / ROTATING_OUT / AVOID / NEUTRAL verdict per sector for both US and KR markets. Unlike the narrative sector-analyst skill, this produces a deterministic JSON snapshot (data/sector_rs_{market}.json) that the daily-briefing candidate discovery consumes to bias stock ranking toward sectors rotating in and away from sectors rotating out. Use when the user wants an actionable "which sectors to favor/avoid right now" answer, a sector leaderboard, sector relative-strength ranking, 섹터 로테이션, 섹터 상대강도, or which sectors are leading/lagging.

Sector Rotation RS Screener

Overview

This skill turns each sector's price structure — measured relative to the market benchmark — plus its constituent breadth into a prescriptive rotation verdict and a 0–100 score. It is the deterministic, machine-readable counterpart to sector-analyst (which writes a narrative cycle-phase report). Use this one when you want an actionable leaderboard and a JSON artifact that downstream automation (the daily briefing's candidate discovery) can read.

All scoring lives in mcp-market-data/sector_rs.py (pure, never-raise). The CLI command bin/stock-cli sector-rs fetches the data, builds the verdicts, and optionally persists the snapshot.

When to Use This Skill

• "Which sectors should I favor / avoid right now?"
• "Rank the sectors by relative strength."
• "Is tech rotating in or out?"
• "섹터 로테이션 점검해줘", "지금 어떤 섹터가 강해?"
• Before logging individual-stock predictions, to gate by the parent sector.

For a qualitative market-cycle narrative (Early/Mid/Late cycle, scenario probabilities), use sector-analyst instead — the two are complementary.

The Three Axes

1. Relative strength (RS)
   • rs_1m = sector_etf.return_1m − benchmark.return_1m
   • rs_3m = a 63-bar (~3-month) close-return spread vs the benchmark.
2. Breadth — fraction of the sector's constituent basket whose latest close is above its MA20 and whose 1-month return beats the benchmark's.
3. Stage (lifecycle) — from RSI14, distance to MA50, and overextension:
   • EARLY = RSI14 ∈ [45, 65] AND price within ±8% of MA50 AND overextension NONE AND rs_1m > 0 (constructive, not yet extended).
   • LATE = overextension ELEVATED/EXTREME OR price >15% above MA50 OR RSI14 > 70 (stretched / blow-off).
   • MID = otherwise.

Verdict Map (prescriptive)

Verdict	Condition	Action
FAVOR	rs_1m > 0 AND breadth ≥ 0.5 AND stage ∈ (EARLY, MID)	Overweight; prefer this sector's leaders.
ROTATING_IN	rs_1m > 0 AND rs_3m ≤ 0 AND stage == EARLY	Fresh leadership turning up — accumulate early.
ROTATING_OUT	(rs_1m < 0 AND rs_3m > 0) OR (stage == LATE with weak breadth)	Trim; momentum rolling over.
AVOID	rs_1m < 0 AND breadth < 0.4	Underweight; lagging with poor participation.
NEUTRAL	otherwise, or when the benchmark is missing	No edge; defer to bottom-up.

score = 0.5*RS + 0.3*breadth + 0.2*stage, normalized to 0–100. A missing benchmark floors every sector to NEUTRAL (score 50) — the screener never certifies rotation against an unknown market.

Workflow

Step 1 — Generate the snapshot

# US (S&P 500 sector ETFs vs SPY)
bin/stock-cli sector-rs --market US

# KR (KODEX sector proxies vs 069500 / KODEX 200)
bin/stock-cli sector-rs --market KR

# Persist for discovery to consume (atomic, per-market file)
bin/stock-cli sector-rs --market US --write   # → data/sector_rs_us.json
bin/stock-cli sector-rs --market KR --write   # → data/sector_rs_kr.json

The output is JSON: a sectors list ranked by score descending, each with verdict, stage, rs_1m, rs_3m, breadth_pct, score, the proxy etf, and the constituents basket. --write is the only side effect; without it the command is read-only.

• US sector ETFs and leader baskets are static inside sector_rs.py (US_SECTOR_ETFS / US_SECTOR_CONSTITUENTS), benchmark SPY.
• KR sectors come from data/kr_sector_map.csv (sector,proxy_etf,proxy_etf_name,constituents), benchmark 069500.

Step 2 — Interpret the leaderboard

Read top-to-bottom: FAVOR/ROTATING_IN sectors at the top are where new long ideas should concentrate; ROTATING_OUT/AVOID at the bottom are where to trim or stay out. Cross-check rs_1m vs rs_3m: a positive rs_1m with a negative rs_3m is the classic early rotation signature (ROTATING_IN).

Step 3 — Apply per stock

When evaluating an individual ticker, look up its sector's verdict/stage and let it gate the call: favor leaders in FAVOR/ROTATING_IN sectors; demote or skip names in ROTATING_OUT/AVOID sectors.

How discovery consumes the snapshot

scheduler/candidate_discovery.py reads the persisted file via _load_sector_verdicts(market) (never-raise: a missing or stale file → {}), which maps every constituent and proxy ETF ticker to its sector verdict. apply_sector_boost(candidates, verdicts) then:

1. stamps each candidate's sector_verdict / sector_stage in place, and
2. returns a bounded {ticker: multiplier} map the ranking sort folds in:

Sector state	Multiplier
FAVOR + EARLY	×1.30
ROTATING_IN (or FAVOR/MID)	×1.15
NEUTRAL / unknown	×1.00
ROTATING_OUT (non-LATE)	×0.80
AVOID / ROTATING_OUT+LATE	×0.60

When no snapshot exists, every multiplier is exactly 1.0 and candidate fields are left untouched — discovery ranking is byte-identical to the pre-sector behaviour (this no-op is covered by scheduler/tests/test_sector_boost.py).

Prerequisites

• bin/stock-cli available in the project root.
• Network access for the price batch (yfinance for US, PyKRX for KR).
• No API key required.

Notes

• Pure scoring lives in mcp-market-data/sector_rs.py; the CLI command is cmd_sector_rs in stock_cli.py.
• The snapshot files data/sector_rs_us.json / data/sector_rs_kr.json are gitignored — they are regenerated every run.
• Sector-cohort backtesting reuses scheduler/asof_backtest.py; this skill does not ship its own backtest harness.