Duration / Treasury Timing久期 / 国债择时
Own long Treasuries when they're cheap and trending; step to T-bills before the rate shock.当长期国债便宜且趋势向上时持有；在利率冲击前退守短债。
As of截至 Jul 16, 2026 · Long Treasuries (TR)长期国债（总回报） · current income当前票息 ~4.61%
⚗ Experimental — display-only (Phase-0)实验性 — 仅展示（Phase-0）
A drawdown / Sharpe engine — not a yield-maximizer回撤 / 夏普引擎 — 并非收益最大化工具
Recommended allocation right now当前建议仓位
Long Treasuries (TR)长期国债（总回报）66%
T-bills短期国债34%
Long Treasuries (TR)长期国债（总回报）
T-bills短期国债 (cash carry现金套息)
Last change上次调整: 2026-07-15 — 100% → 66%
Current income blends the risk-asset yield (when invested) with the cash-leg yield (when de-risked).当前票息为风险资产收益（在场时）与现金腿收益（降险时）的加权。
Duration risk久期风险 score评分 (0–100)
32
Watch — trim关注 — 减仓
Higher = more risk-off. The score sets the risk-asset weight via graded bands (below), with a 5-day debounce to cut whipsaw.分数越高越偏避险。该分数通过分级区间（见下）设定风险资产权重，并有 5 日缓冲以抑制频繁换仓。
What is driving the score分数由什么驱动
Each leg adds points to the 0–100 score (renormalised weighted mean). The bar is the current score, split by contribution.各分项为 0–100 分数贡献分值（归一化加权平均）。下方进度条为当前分数，按贡献拆分。
Filled width = the score out of 100. Empty = headroom (calm).填充宽度 = 满分100中的得分。空白 = 余量（平静）。
| Signal leg信号分项 | Intensity强度 | Weight权重 | Pub lag发布滞后 | Points贡献分 |
|---|
| Real-yield value (expensive → step aside)实际收益率估值（昂贵 → 退避） |
1/100 |
1.0 |
1d |
0.5 |
| Curve carry (10y − 3m)曲线套息（10年 − 3月） |
22/100 |
0.5 |
1d |
4.3 |
| Trend (time-series momentum)趋势（时间序列动量） |
67/100 |
1.0 |
0d |
26.7 |
| Composite score综合分数 | | | | 32 |
The strategy in action · benchmark with the model on top策略实战 · 基准叠加模型
Green = fully invested, amber = trimmed, clear = defensive. ▲/▼ mark every allocation change. The strategy equity (blue) is rescaled onto the benchmark axis so you can read protection vs participation directly.绿色 = 满仓，琥珀 = 减仓，空白 = 防守。▲/▼ 标注每次仓位变动。策略净值（蓝）已缩放至基准轴，可直接对比保护与参与度。
Growth of $1 (log) — vs buy & hold1美元增长（对数）— 对比买入持有
Underwater (drawdown) — vs buy & hold水下回撤 — 对比买入持有
Honest framing: this does NOT reliably beat buy-and-hold on CAGR — sitting in the cash leg through stress gives up some carry. The edge is the higher Sharpe and the much shallower max drawdown: it times the LEFT TAIL that destroys the yield (the recession/credit/rate shock), not the yield itself.诚实说明：本策略在年化收益上并不稳定跑赢买入持有 —— 压力期退守现金腿会让出部分套息。优势在于更高的夏普与明显更浅的最大回撤：它择时的是吞噬收益的左尾（衰退/信用/利率冲击），而非收益本身。
Backtest scorecard · 24.0 years vs buy & hold回测记分卡 · 24.0 年对比买入持有
CAGR
3.5%
buy & hold买入持有 3.61%
Sharpe
0.48
buy & hold买入持有 0.32
Max drawdown最大回撤
-18.1%
buy & hold买入持有 -48.4%
Time invested在场时间
95.2%
1.84x turnover/yr换手/年
Sortino
0.68
buy & hold买入持有 0.48
Income yield now当前票息
4.61%
blended risk-asset + cash风险资产 + 现金加权
Deflated Sharpe贬值夏普
0.8144
multiple-testing haircut多重检验折减 · n=8
The rule-book · risk band → weight规则手册 · 风险区间 → 权重
| Risk band风险区间 | Score分数 | Long Treasuries (TR)长期国债（总回报） weight权重 | T-bills短期国债 |
|---|
| Calm — fully invested平静 — 满仓 |
< 25 |
100% |
0% |
| Watch — trim关注 — 减仓 ◀ |
25–50 |
66% |
34% |
| Stress — de-risk压力 — 降险 |
50–75 |
33% |
67% |
| Defensive — all cash防守 — 全现金 |
≥ 75 |
0% |
100% |
A new band must hold 5 trading days before it takes effect (hysteresis kills whipsaw). Bands and leg weights are pre-registered, not tuned per result.新区间须连续保持 5 个交易日才生效（滞后机制抑制频繁换仓）。区间与分项权重为预先登记，非按结果调参。
Independent bear episodes · the true sample size独立熊市事件 · 真实样本量
| Peak峰值 | Trough谷底 | Buy & hold drawdown买入持有回撤 |
|---|
| 2008-12-30 | 2009-06-10 | -26.6% |
| 2012-07-25 | 2013-08-21 | -20.7% |
| 2020-08-04 | 2023-10-19 | -48.4% (ongoing进行中) |
A handful of independent ≥20% bears govern a get-out / get-back-in switch — effective-N, not the day count, bounds confidence. Full leave-one-crisis-out validation is a fast-follow.少数几次独立的 ≥20% 熊市决定一次离场/再入场切换 — 决定置信度的是有效样本量，而非天数。完整的留一危机验证为后续跟进。
Phase-0 (2002→ TLT): the drawdown-control edge is robust — MaxDD −18% vs −48% buy & hold, beats the naive baselines, survives leave-one-crisis-out — but the deflated Sharpe is marginal (0.83 < 0.90), so the risk-adjusted edge may be noise at this sample depth. DISPLAY-ONLY / experimental, not a scored signal: a long-only single-ETF timer is far weaker than the leveraged multi-country bond-factor research; its real job is crisis convexity + standing aside in 2022-type regimes. See reports/duration-timing-phase0.md. Not investment advice.Phase-0（2002 起，TLT）：回撤控制优势稳健——最大回撤 −18% 对比买入持有 −48%，跑赢朴素基准，通过留一危机检验——但贬值夏普处于临界（0.83 < 0.90），在此样本深度下风险调整后优势可能是噪声。仅展示 / 实验性，非评分信号：仅做多的单一 ETF 择时远弱于带杠杆的多国债券因子研究；其真正作用是危机凸性 + 在 2022 式行情中退守。见 reports/duration-timing-phase0.md。非投资建议。