Task #4 全天候 — Layer 2 Spec v0.1 (DRAFT for Kaite review)

**Author**: @RayDario

**Date**: 2026-05-18

**Status**: DRAFT — awaiting Kaite ack before passing to @JimSimons for implementation

**Supersedes**: v0+ funded baseline (Stage 1 inverse vol + Stage 4 cash zero redistribute)

**References**:

- framework-synthesis-v1.md (Qian + 林颖颖 + 桥水 synthesis, 38 deviation entries)

- brief-v0.md (Task #4 brief v0)

- ic-pack-v0.md (Task #4 IC pack v0, supersede after v0.1 baseline 跑出来)

- Kaite framework reset 8-point (#mini-cypress:1e7f3849:479adb40 + follow-ups)

---

§0 Spec 设计原则 (一句话总结)

**v0.1 = v0+ funded baseline + 5 个结构性 fix + 2 个 universe 扩展, 全部 align Qian Risk Parity Foundation + 林颖颖 4 regime framework + Kaite 8-point reset**.

不引入 Dynamic Risk Allocation overlay (deferred 到 v2).

不引入 中长期主观接管 (保留为 林颖颖 framework 的危机 trigger).

不追 Qian Sharpe 1.1 — 接受当前 macro regime (Japan-like low growth low inflation) 限制的 6-9% real return.

---

§1 5 个结构性 Fix (vs v0+ funded baseline)

Fix 1: Stage 1 从 "sleeve 内 inverse vol" → "sleeve 内 ERC" (deviation #1 + #15 + #22)

**问题**: v0+ Stage 1 用 60d 历史 vol inverse weighting, 当 vol 极度不均时 (城投 1% vs 沪深300 22%) 系统性 over-weight low-vol asset. 城投 拿 R1 sleeve 48%+ NAV → 整个 portfolio 实际 等于 重 信用 carry trade.

**Fix**: Stage 1 改用 **真正 Qian-style Equal Risk Contribution (ERC)** based on **covariance matrix** (vol + correlation):

- 每月重算 covariance matrix (60d rolling, robust shrinkage applied to handle 上市晚 资产)

- 每个 sleeve 内 用 numerical optimizer 求 weights such that 每资产 marginal risk contribution = 1/N (sleeve 内 N 个资产)

- Marginal risk contribution = w_i × (Σw)_i / σ_p (Qian Eq 1.3)

**Sub-Fix 1a: 加 regime-conditional covariance** (deviation #22):

- 全样本 cov 用作 baseline

- Stress-period cov (信用 stress, equity drawdown, commodity spike) 单独算

- 每月 ERC 时 用 max(baseline_cov, 1.3 × stress_cov) 取保守值

- 这 prevent post-2008 RORO 期间 信用 / 沪深300 实际 +0.5 correlation 被 全样本 0.2 cov 低估

**Validation**:

- 跑回测验证 城投 (or 任何低 vol 资产) 在 R1 sleeve 内不超 40% (vs v0+ 的 48%)

- 跑 stress sub-period 验证 sleeve 内 marginal risk 仍接近 equal

---

Fix 2: Stage 4 从 "cash zero redistribute" → "futures-based funded leverage" (deviation #2 + #26)

**问题**: v0+ Stage 4 把短债 / 货基 0 化然后 redistribute → 实际 把 cash buffer 的权重 inadvertently 转移给 R4 sleeve 内最低 vol 的资产 (城投), 造成 城投 53% NAV concentration.

**Fix**: Stage 4 改用 **真正 funded leverage on government bonds via 国债期货 (T / TF / TS 主连)** [Kaite ack 2026-05-18 msg befb1bd9]:

- 短债 / 货基 outside-regime cash buffer (Layer 1 已 fix ✓)

- Government bond sleeve (R4) 内 用 国债期货 leverage 把 effective duration / risk 推到 sleeve target

- Leverage 上限 = mathematical bound based on sleeve correlation (Qian Ch 5.2.5): 2-asset 200% 是 大致上限; 多资产 (我们 4 regime sleeve cross) ≤ 250% gross

- 当前 v0+ funded effective leverage 1.23x → v0.1 target effective leverage **1.5-1.8x** (跨 4 regime sleeve aggregate)

**重要 implementation note (Kaite 30Y 国债期货 clarification 2026-05-18, refined by JimSimons engineering 口径)**:

TL (30Y 国债期货 主连) 上线 2023-04, 历史短. 分两条线避免混淆:

- **Tradable backtest**: TL 主连 只从 2023-04 上市日起算; pre-2023 在 tradable backtest 不存在 TL 仓位 (因为彼时无 instrument)

- **Proxy diagnostic** (研究代理, 不作交易决策): pre-2023 用 30Y **现券** (国债指数 / 现券估值数据) 作 "风险/相关性代理", 用于研究 framework 在 pre-2023 regime 的 hypothetical 表现

- **实盘**: TL 流动性 evaluate (上线时间短 + 持仓量未必充足) → 如果不够, fallback 用 30Y 现券 + 多头持有 (无 leverage), 仍能 capture rate premium 但 effective leverage 降低

- IC pack v1 必须 transparent disclose: "post-2023 tradable backtest 是 真实 instrument; pre-2023 是 proxy diagnostic only, 不作 deployment 依据"

**路径 B (deferred to Task #8 portfolio-level leverage feasibility eval)** [Kaite clarification 2026-05-18 msg befb1bd9 + JimSimons engineering 重 framing]:

Kaite 原话意思: "ETF 融资融券" 是用于 **整体组合 vol target 未达成时, 在权益 ETF 侧 加 portfolio-level leverage** (e.g. 沪深300 ETF / 中证500 ETF 融资买入), 不是给 国债 加 leverage.

JimSimons 工程 framing 校准: 这不属于 Task #4 framework, 而属于 **Task #8 组合层 (4 strategy 跨 sleeve) portfolio-level leverage feasibility evaluation**.

**Task #4 v0.1 边界**: 主用 路径 A (国债期货 leverage on bond sleeve). **路径 B 不在 Task #4 scope**, 放 Task #8 评估 (届时考虑 cross-strategy leverage 整体框架).

- 当前 v0+ effective leverage 1.23x / vol 4.68% < target 6-8%

- 预期 路径 A 实施后 effective leverage ↑ to 1.5-1.8x, vol ↑ to 6.5-8% (达 target)

- 如 路径 A 实施后 vol 仍未达 target, escalate 给 Task #8 评估 portfolio-level leverage (路径 B 含 in scope)

**Implementation note (for JimSimons)**:

- 国债期货 main contract roll-over: 每季 IC date 自动 roll

- Margin: 期货保证金 7-10% 留作 cash collateral, 不计入 sleeve weight

- 期货 PnL 每日 mark-to-market, 计入 portfolio total return

---

Fix 3: Universe 显式 扩展 (deviation #4 + #26 + #35)

**v0+ universe** (11 资产):

- R1: 沪深300 / 中证500 / 城投债

- R2: 红利

- R3: 黄金 / 豆粕

- R4: 10Y国债

**v0.1 universe** (16-18 资产 target):

**Universe candidate ETF list** (JimSimons confirm 流动性 + 历史长度):

- 中证1000: 159845 / 512100

- 创业板: 159915

- 科创50: 588000

- 能化: 159981

- 有色: 512400

- 钢铁: 515210

- 30Y国债: 511130 (上市较晚, 用 国债期货 T 主连 + 部分指数模拟 PIT-safe)

- 农产品 扩展: 159825 (饲料豆粕 已有) + 162411 (有色基本金属) — TBD

**Data risks** (transparent disclosure 不 hack):

- 中证1000 / 创业板 / 科创50 上市晚 (科创50 2020-07 first IPO), 历史 < 5 年

- 30Y国债期货 (TL 主连) 2023-04 上线, 历史 < 3 年

- 能化 ETF 2019-12 上线, 5+ 年历史 OK

- Layer 2 回测: 标 "post-2020 sub-period" 单独跑 v0.1 整个 universe; 标 "pre-2020 sub-period" 用 reduced universe (skip 上市晚的) 验证 framework 稳健性

---

Fix 4: 3 Primary Risk Premium Check (deviation #33)

**问题**: 当前 v0+ funded risk contribution 拆 3 primary premium (equity / rate / inflation) ≈ 60% / 10-20% / 30% = **equity-biased**, 不是真 RP.

**Fix**: Layer 2 加 **"RP Authenticity Audit"** as IC-level required monitoring:

- 每月跑 portfolio 后, 用 sector-level risk decomposition 算 each asset 属 哪个 primary premium:

- Equity premium: 股 ETF (沪深300 / 中证500 / 中证1000 / 创业板 / 科创50 / 红利 / 周期 ETF)

- Rate premium: 国债 / 国债期货 / 城投 (post-2008 RORO 视角 partially rate)

- Inflation premium: 商品 (黄金 / 豆粕 / 能化 / 有色) + TIPS proxy (无 China TIPS, 用 黄金 partial 代表)

- 信用 / 城投 partial classification: 用 conditional 拆分 (R1/R2 复苏期 80% rate / 20% equity; R3/R4 stress 期 40% rate / 60% equity)

**Target**: 每 primary premium 25-40% risk contribution. 任何 < 20% standalone 触发 IC alert (universe 或 Stage 1 ERC 失败).

**v0.1 expected outcome** (基于 universe 扩展 + Stage 1 ERC + Stage 4 国债 leverage):

- Equity: 35-40% (vs v0+ 60%)

- Rate: 30-35% (vs v0+ 10-20%)

- Inflation: 25-30% (vs v0+ 30%)

---

Fix 5: Benchmark 重设 (deviation #27 + #30)

**问题**: 当前 IC pack v0 用 沪深 300 / 60/40 (中国版) 作 benchmark, 都不适合 risk parity 框架.

**Fix**: Layer 2 IC pack v1 同时报 **3 个 benchmark**:

**Reframe "回测 < M2" 论点** (deviation #30, 回应 Kaite point 2):

- 不追 M2 5%+ nominal 作 absolute target

- 改追 **real return ≥ 0 in any macro regime** (Qian Ch 6 + Ch 6.5 全 4 decades 实证 RPF 都正 real return)

- v0+ funded 5.77% nominal (CPI ~0%) = +5.77% real ✓ pass

- v0.1 target: real return 5-7% in disinflationary regime, 3-5% in moderate inflation, 2-4% in high inflation (Qian 1970s precedent)

---

§2 2 个 Universe 扩展 不算 fix, 是 feature

Feature 1: 国债期货 / 国债 ETF leverage 作 Stage 4 主 implementation (vs cash zero redistribute)

See Fix 2 above.

Feature 2: 4 regime sleeve cross 加 sector-balanced 信号 (informational, 不改 Stage 2 ERC)

每月 IC date 计算并 disclose:

- Sleeve risk contribution: R1 / R2 / R3 / R4 各占 总 risk %

- Primary premium risk contribution (Fix 4)

- Asset-level risk contribution

- Sleeve correlation matrix (regime-conditional + 全样本)

这是 transparency 工具, 不影响 portfolio construction.

---

§3 v0+ → v0.1 升级 expected magnitude (per Qian 5.4.5 实证)

---

§4 Implementation phasing (for JimSimons engineering work split)

Phase A: Stage 1 ERC + Universe 扩展 (1.5-2 days)

- Stage 1 inverse vol → covariance-based ERC

- Universe 加 中证1000 / 创业板 / 科创50 / 能化 / 30Y 国债 (or 国债期货)

- 跑 baseline v0.1a: Stage 1 改, universe 扩, leverage 暂时维持 cash zero (不动 Stage 4)

- Verify: 城投 NAV % ↓ to < 30%, primary premium % balanced

Phase B: Stage 4 国债期货 leverage (1.5-2 days)

- 实施 国债期货 (T / TF / TS 主连) 作 Stage 4 leverage

- 跑 baseline v0.1b: Phase A + Stage 4 真正 leverage

- Verify: effective leverage 1.5-1.8x, equity premium % 进一步降, rate premium % 升

Phase C: 3 Primary Premium Check + Benchmark v0.1 (0.5-1 day)

- 实施 RP Authenticity Audit 工具

- 建 "Naïve Chinese RP Benchmark"

- IC pack v1 数据准备

Phase D: Stress test on v0.1 (0.5 day)

- 2015 股灾 / 2018 单边熊 / 2024-01 小盘闪崩 / 2018-2019 城投违约潮 4 stress windows

- 通过红线: MDD ≤ 14%, no single asset > 30% NAV in any sub-period

**Total estimate**: ~5-6 days JimSimons engineering (vs Layer 2 之前估 2.5-3.5 days, 增 due to universe expansion + 国债期货 integration)

---

§5 Things explicitly NOT in v0.1 (deferred to v2 / v3)

Deferred to v2:

- Dynamic Risk Allocation overlay (Qian Ch 4.2.5 + Ch 6.5 1970s 实证 +1.2%/年; 林颖颖 "主观增强 capped 5%" 思路)

- Trigger 1 (inflation tilt): CPI yoy > 3% 持续 2Q → 商品 +5% / 国债 -5% NAV

- Trigger 2 (deleveraging): SHIBOR 3M > 4% AND 组合 60d vol > 1.5x target → leverage -50%

- Trigger 3 (rebound 加仓): 3-month rolling Sharpe < -1 (NNN stress) → tactical 加 5-10% leverage

- Regime-conditional cov 显式 explicit modeling (v0.1 用 conservative max approximation)

- Stock/bond correlation regime monitoring (deviation #31)

- Yield curve flat-curve monitor (deviation #24)

Deferred to v3+:

- Systematic factor enhancement (currency carry / value tilts / yield curve slope) — Qian Ch 8.3 提到可 +0.5-1% Sharpe

- 等权大盘 / 基本面权 宽基 ETF 替代 cap-weighted (deviation #13 + #35)

- 真正 inflation-linked bond proxy (中国 通胀挂钩国债 if/when 上线)

---

§6 Open questions for Kaite ack 前

1. **国债期货 vs ETF 融资融券**: primary path 选哪个? (我倾向 国债期货, 国债期货 透明度 + 流动性 更好; ETF 融资融券 operational simpler)

2. **R2 (过热) sleeve cyclical exposure**: 有色 / 钢铁 / PMI 强相关 ETF 加入合适吗? 还是 keep R2 仅 红利 + 沪深300/中证500 子集?

3. **30Y 国债 / 国债期货 (TL 主连)**: TL 主连 2023-04 才上线, 历史短. 是否 OK 用 (PIT-safe 模拟 pre-2023, 跑 sub-period 报告)?

4. **3 primary premium audit cadence**: 月度 IC report or 周度 alert?

5. **Real return benchmark "CPI + 5%"**: OK 作 absolute target? 或希望另设 (e.g. "CPI + 实际增长" or fixed nominal 8%)?

---

§7 Sign-off matrix

---

§8 Cross-reference

- 35+ deviation log entries in framework-synthesis-v1.md 直接 mapping 到 Fixes 1-5 above:

- Fix 1 ← deviation #1 / #15 / #22

- Fix 2 ← deviation #2 / #26

- Fix 3 ← deviation #4 / #26 / #35

- Fix 4 ← deviation #33

- Fix 5 ← deviation #27 / #30

- Synthesis sections directly informing:

- §A.1c-A.1e Qian Foundation/Dynamic + Spear & Shield + NNN + RORO

- §A.1f Leverage + Rebalancing math + Benchmarking

- §A.1g Japan analog + 1994 + 1970s

- §A.1h Style analysis + Stop-loss

**End of Layer 2 spec v0.1 draft. Awaiting @Kaite ack.**