| name | l2-etf-strategy |
| description | Operate and reproduce the user's L2 ETF strategy package. Use when the user asks about L2 ETF, TRD_K20_ADV50M_NoKCB, quarterly ETF rebalancing, ETF picks/trades, live ETF holdings, NAV logs, or reproducing the ETF backtest/experiment evidence from the 2026-06-10 package. |
L2 ETF Strategy
Orientation
Use this skill for the user's locked L2 ETF strategy line only. Do not mix it with Turtle stock rules, VCP experiments, or single-stock live trading logic.
Core asset root:
SKILL_DIR="${CODEX_HOME:-$HOME/.codex}/skills/l2-etf-strategy"
PKG="$SKILL_DIR/assets/L2_ETF_STRATEGY_EXPERIMENT_PACKAGE_2026-06-10"
LIVE="$PKG/01_live_strategy/L2_ETF_LIVE"
BACKTEST="$PKG/02_backtest_reproduction/L2_ETF_BACKTEST"
SOURCE_DATA="$PKG/03_source_data/L2_ETF_DATA_2026-05-29"
Always start by reading references/handoff.md. For live rebalancing or holdings work, also read references/live-strategy.md and references/live-changelog.md. For performance reproduction, read references/backtest-reproduction.md.
Locked Strategy
- Strategy:
TRD_K20_ADV50M_NoKCB.
- Capital segment: 50,000 RMB.
- Pool: cleaned A-share equity ETF/LOF universe.
- Live exclusion: ETF codes prefixed
588 or 589.
- Liquidity filter: 60-day average traded value >= 50,000,000 RMB.
- Signal: 120-day momentum Top 20.
- Weighting: 20 equal-weight ETF positions, 100-share lots, floor sizing.
- Rebalance: quarterly.
- No stop loss, no take profit, no news-driven mid-quarter trading.
- Regime, REV, and ADAPT are not enabled for live operation.
The package cache is historical: source data is dated 2026-05-29. If a user asks for latest/current/future rebalance output, refresh or obtain up-to-date ETF pool and kline cache first, then run the same workflow with L2_ETF_DATA_ROOT pointing at the refreshed data. Do not present stale 2026-05-29 output as a current signal.
Live Rebalance Workflow
Use this when generating quarterly picks_YYYY-MM-DD.csv and trades_YYYY-MM-DD.csv.
- Inspect
current_holdings.json in LIVE and note confirmed live state. As of the package handoff, 2026-06-10 status is initial_build_completed_20_etfs; 159381 is intentionally above original equal-weight target and should be corrected only at the next quarterly rebalance unless the user explicitly requests interim action.
- Confirm data freshness. For bundled historical reproduction, use
SOURCE_DATA. For any current rebalance, use refreshed data with the same layout: all_etf_pool.csv plus kline_cache_all/*.json.
- Run from the live directory:
cd "$LIVE"
L2_ETF_DATA_ROOT="$SOURCE_DATA" python3 quarterly_picks.py --cutoff YYYY-MM-DD --capital 50000
If the cutoff should be the newest common date in the cache, omit --cutoff. Use --show-pool-size for a quick eligibility check. Do not use --commit unless the user confirms actual fills or explicitly asks to update holdings.
Generated files are written beside quarterly_picks.py. When preserving canonical assets matters, copy LIVE to a temporary working directory first and run there with L2_ETF_LIVE_HOME set to that copy.
Backtest Reproduction Workflow
Use this to verify the locked performance evidence:
cd "$BACKTEST"
python3 run_capital_sweep.py
diff ./baselines_capsweep/capsweep_summary.csv ./capsweep_summary.csv
Expected headline in stdout:
TRD_K20_ADV50M ret: 1w:+110.5% 2w:+117.8% 3w:+119.0% 5w:+120.8% 10w:+122.6% 20w:+123.3%
No diff output means the regenerated capsweep_summary.csv matches the reference. The 50,000 RMB line corresponds to about +120.83% total return, -20.62% max drawdown, and 5.86 Calmar for 2024-01 to 2026-05.
The scripts require pandas and numpy. If the default python3 lacks them, use a Python environment that has both installed.
Reporting Rules
- State the cutoff date, data root, eligible pool size, and whether the run used bundled or refreshed data.
- Separate target picks, trade diffs, and confirmed holdings.
- Mention that orders are operational outputs from the user's predefined strategy, not broker execution.
- If data is stale, say so plainly and avoid implying the list is current.
- Preserve evidence boundaries: the package supports the ETF strategy line only and excludes Turtle stock and later VCP experiments.
