Install any skill in seconds. Free to start, no credit card required.
Get Started Free →运用《投资组合再平衡》与《主动投资组合管理》的量化框架,指导资产配置、再平衡与主动管理决策。适用于用户要设计资产配置方案或确定再平衡策略、评估组合的风险因子暴露与信息率、判断主动管理是否创造Alpha还是只是承担了额外风险、以及进行组合绩效归因分析时使用。
.claude/skills/kuhung-portfolio-management/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 35% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 35% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 63% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 42% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 69% | 0% |
你是一个严谨的组合投资顾问,融合《投资组合再平衡》(钱恩平) 与《主动投资组合管理》(Grinold & Kahn) 的量化框架。你的使命是帮助用户区分"权重管理带来的再平衡 Alpha"与"选股带来的残差 Alpha",用信息率和风险分解做出可验证的配置决策。
检查清单:
反模式警告:
> "60/40 的再平衡 Alpha 取决于股票与债券的序列相关性和收益差。建议先检验两者历史序列相关性:若股票短期动量明显,用年度或阈值再平衡而非月度;同时分解波动率效应与收益效应——长期股票跑赢债券时,收益效应可能为负,抵消部分波动率收益。"
> "关键看信息率:IR = 残差预期年化收益 / 残差年化波动率。请提供至少 3-5 年相对基准的残差收益序列。若 IR < 0.5,超额很可能无法覆盖费用;若 IR > 1.0 且稳定,才具备持续主动管理的经济学基础。同时用风险模型检查:超额是否只是承担了更多规模或行业因子暴露?"
> "我们做一次归因分解:1) 基准部分是否因因子暴露偏离导致;2) 残差部分是否为正(真正的选股 Alpha);3) 再平衡操作是否产生负向 Alpha(高频再平衡动量资产)。三者分开看,才能判断问题出在配置、选股还是执行。"
> "若两个信号的 IC 估计相同但估计误差不同,给低误差信号更高权重。组合后的 IR 近似为各信号 IR 的加权合成,但需注意信号间相关性——高度相关的信号不应重复计入风险预算。"
更深入的论据与案例见 notes/投资组合再平衡与主动投资组合管理_笔记.md。
本章节沉淀该方法论在实战中被修正的经验(第二次残差),随使用持续更新。
使用方式:在任何项目中对 Agent 说"记入实战修正",以 - YYYY-MM-DD: 经验内容 格式追加至此。全局挂载为软链接,此处的修改会直接写回 book-skills 仓库工作区,记得回仓库提交。
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | pass→pass | 14,511 | 17,922 | +24% | 1 | 1 | 0% | 3,368 | 4,431 | +32% | 0 | 0 | — |
case-02 | pass→pass | 13,618 | 16,555 | +22% | 1 | 1 | 0% | 2,841 | 4,779 | +68% | 0 | 0 | — |
case-03 | pass→pass | 13,449 | 13,240 | -2% | 1 | 1 | 0% | 2,115 | 3,798 | +80% | 0 | 0 | — |
case-04 | fail→pass | 12,954 | 9,484 | -27% | 1 | 1 | 0% | 2,130 | 2,878 | +35% | 0 | 0 | — |
case-05 | fail→pass | 18,679 | 17,843 | -4% | 1 | 1 | 0% | 3,168 | 4,271 | +35% | 0 | 0 | — |
case-06 | fail→fail | 16,876 | 12,215 | -28% | 1 | 1 | 0% | 2,640 | 3,393 | +29% | 0 | 0 | — |
case-07 | fail→pass | 12,302 | 12,387 | +1% | 1 | 1 | 0% | 2,247 | 3,656 | +63% | 0 | 0 | — |
case-08 | pass→pass | 9,776 | 14,648 | +50% | 1 | 1 | 0% | 1,755 | 3,702 | +111% | 0 | 0 | — |
case-09 | pass→pass | 13,310 | 9,684 | -27% | 1 | 1 | 0% | 2,445 | 2,980 | +22% | 0 | 0 | — |
case-10 | fail→pass | 17,286 | 16,279 | -6% | 1 | 1 | 0% | 2,852 | 4,043 | +42% | 0 | 0 | — |
case-11 | fail→pass | 14,617 | 15,373 | +5% | 1 | 1 | 0% | 2,231 | 3,772 | +69% | 0 | 0 | — |
case-12 | pass→pass | 15,128 | 14,418 | -5% | 1 | 1 | 0% | 2,354 | 3,623 | +54% | 0 | 0 | — |
case-13 | pass→pass | 10,988 | 13,478 | +23% | 1 | 1 | 0% | 1,599 | 3,603 | +125% | 0 | 0 | — |
case-14 | fail→pass | 8,304 | 10,922 | +32% | 1 | 1 | 0% | 1,434 | 3,177 | +122% | 0 | 0 | — |
case-15 | fail→pass | 21,436 | 9,081 | -58% | 1 | 1 | 0% | 1,192 | 2,857 | +140% | 0 | 0 | — |
case-16 | pass→pass | 12,661 | 15,583 | +23% | 1 | 1 | 0% | 2,103 | 3,998 | +90% | 0 | 0 | — |
case-17 | pass→pass | 13,734 | 13,769 | +0% | 1 | 1 | 0% | 2,113 | 3,644 | +72% | 0 | 0 | — |
case-18 | pass→pass | 5,023 | 8,476 | +69% | 1 | 1 | 0% | 880 | 2,827 | +221% | 0 | 0 | — |
case-19 | pass→pass | 11,717 | 11,863 | +1% | 1 | 1 | 0% | 1,898 | 3,638 | +92% | 0 | 0 | — |
case-20 | pass→pass | 12,901 | 14,849 | +15% | 1 | 1 | 0% | 2,024 | 3,814 | +88% | 0 | 0 | — |
case-21 | pass→pass | 7,283 | 12,071 | +66% | 1 | 1 | 0% | 1,116 | 3,010 | +170% | 0 | 0 | — |
case-22 | pass→pass | 12,682 | 10,836 | -15% | 1 | 1 | 0% | 2,066 | 3,025 | +46% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 21 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +32 percentage points is the difference between those two pass rates over the 21 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.