zeberg-salomon-rotator

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

<essential_principles>

<principle name="two_state_model"> **兩態切換模型核心**

Zeberg–Salomon 模型將市場簡化為兩種狀態：

RISK_ON: 持有股票（SPY），景氣擴張期
RISK_OFF: 持有長債（TLT），景氣收縮期

切換邏輯基於「領先指標先轉弱，同時指標後確認」的景氣循環規律。 </principle>

<principle name="leading_coincident"> **領先 vs 同時指標**

類型	作用	典型成分	領先時間
Leading	預警	殖利率曲線、新訂單、房市許可	6-12 月
Coincident	確認	就業、工業生產、實質收入	同步

合成方式：

各序列做 transform（yoy/mom/diff）
統一方向（direction +1/-1）
Rolling z-score 標準化
EMA 平滑
加權合成 </principle>

<principle name="iceberg_sinking"> **冰山事件 vs 下沉事件**

Iceberg Event: LeadingIndex < iceberg_threshold
  → 預警：景氣開始轉弱
  → 搭配「領先指標下降」+ 可選「市場亢奮」濾鏡

Sinking Event: CoincidentIndex < sinking_threshold
  → 確認：實體經濟收縮
  → 通常在 Iceberg 之後數月發生

狀態機邏輯：

RISK_ON → RISK_OFF：Iceberg 連續確認 + 斜率為負
RISK_OFF → RISK_ON：領先指標回升超過 (threshold + hysteresis) </principle>

<principle name="data_access"> **資料取得方式**

本 skill 使用無需 API key 的資料來源：

FRED CSV:

https://fred.stlouisfed.org/graph/fredgraph.csv?id={SERIES_ID}

Yahoo Finance:
```
yfinance
```
套件抓取 SPY, TLT, VIX

腳本位於

scripts/

目錄，可直接執行。 </principle>

</essential_principles>

<objective> 實作 Zeberg–Salomon 兩態輪動策略：

建構指標：從 FRED 數據合成 LeadingIndex 與 CoincidentIndex
偵測事件：識別「冰山」（領先轉弱）與「下沉」（同時確認）
切換訊號：產生 RISK_ON ↔ RISK_OFF 切換事件
回測績效：計算累積報酬、MaxDD、CAGR、與 benchmark 比較

輸出：切換事件清單、指標時間序列、回測摘要、診斷資訊。 </objective>

<quick_start>

最快的方式：執行預設回測

bash

cd skills/zeberg-salomon-rotator
pip install pandas numpy yfinance pandas-datareader  # 首次使用
python scripts/rotator.py --quick

輸出範例：

json

{
  "state": "RISK_ON",
  "latest_indices": {"LeadingIndex": 0.41, "CoincidentIndex": 0.22},
  "iceberg_event": false,
  "sinking_event": false,
  "last_switch": {"date": "2023-06-30", "action": "EXIT_LONG_BOND_ENTER_EQUITY"}
}

完整回測：

bash

python scripts/rotator.py --start 2000-01-01 --end 2026-01-01 --output result.json

</quick_start>

<intake> 需要進行什麼操作？

快速檢查 - 查看目前的景氣狀態與最新指標
完整回測 - 執行完整的歷史回測與績效分析
視覺化圖表 - 生成多面板回測結果圖表
監控模式 - 設定持續監控與切換警報
方法論學習 - 了解 Zeberg-Salomon 模型的邏輯

請選擇或直接提供分析參數。 </intake>

<routing> | Response | Action | |-------------------------------|---------------------------------------------| | 1, "快速", "quick", "check" | 執行 `python scripts/rotator.py --quick` | | 2, "回測", "backtest", "full" | 閱讀 `workflows/backtest.md` 並執行 | | 3, "視覺化", "chart", "plot" | 閱讀 `workflows/visualize.md` 並執行 | | 4, "監控", "monitor", "alert" | 閱讀 `workflows/monitor.md` 並執行 | | 5, "學習", "方法論", "why" | 閱讀 `references/methodology.md` | | 提供參數 (如日期範圍) | 閱讀 `workflows/backtest.md` 並使用參數執行 |

路由後，閱讀對應文件並執行。 </routing>

<directory_structure>

zeberg-salomon-rotator/
├── SKILL.md                           # 本文件（路由器）
├── skill.yaml                         # 前端展示元數據
├── manifest.json                      # 技能元數據
├── workflows/
│   ├── backtest.md                    # 完整回測工作流
│   ├── visualize.md                   # 視覺化工作流
│   ├── monitor.md                     # 持續監控工作流
│   └── analyze.md                     # 深度分析工作流
├── references/
│   ├── data-sources.md                # FRED 系列代碼與資料來源
│   ├── methodology.md                 # Zeberg-Salomon 方法論解析
│   └── input-schema.md                # 完整輸入參數定義
├── templates/
│   ├── output-json.md                 # JSON 輸出模板
│   └── output-markdown.md             # Markdown 報告模板
└── scripts/
    ├── rotator.py                     # 主輪動腳本
    ├── visualize.py                   # 視覺化繪圖工具
    └── fetch_data.py                  # 數據抓取工具

</directory_structure>

<reference_index>

方法論: references/methodology.md

Zeberg-Salomon 模型概念
冰山/下沉事件定義
兩態切換邏輯

資料來源: references/data-sources.md

FRED 系列代碼（領先/同時）
Yahoo Finance 資產代碼
數據頻率與對齊

輸入參數: references/input-schema.md

完整參數定義
預設值與建議範圍

</reference_index>

<workflows_index>

Workflow	Purpose	使用時機
backtest.md	完整歷史回測	需要績效分析時
visualize.md	生成視覺化圖表	需要圖表展示時
monitor.md	持續監控狀態	日常監控或警報
analyze.md	深度指標分析	理解當前市場狀態
</workflows_index>

<templates_index>

Template	Purpose
output-json.md	JSON 輸出結構定義
output-markdown.md	Markdown 報告模板
</templates_index>

<scripts_index>

Script	Command	Purpose
rotator.py	`--quick`	快速檢查當前狀態
rotator.py	`--start DATE --end DATE`	完整回測
visualize.py	`-i result.json -o chart.png`	生成視覺化圖表
fetch_data.py	`--series T10Y3M,PAYEMS`	抓取 FRED 資料
</scripts_index>

<input_schema_summary>

核心參數

參數	類型	預設值	說明
start_date	string	2000-01-01	回測起始日
end_date	string	today	回測結束日
freq	string	M	頻率（M=月）
equity_proxy	string	SPY	風險資產代理
bond_proxy	string	TLT	長債代理

門檻參數

參數	類型	預設值	說明
iceberg_threshold	number	-0.3	領先指標門檻
sinking_threshold	number	-0.5	同時指標門檻
confirm_periods	int	2	連續確認期數
hysteresis	number	0.15	進出場間距

完整參數定義見

references/input-schema.md

。

</input_schema_summary>

<output_schema_summary>

json

{
  "skill": "zeberg-salomon-rotator",
  "as_of": "2026-01-14",
  "state": "RISK_ON",
  "latest_indices": {
    "LeadingIndex": 0.41,
    "CoincidentIndex": 0.22,
    "iceberg_event": false,
    "sinking_event": false
  },
  "switch_events": [...],
  "backtest_summary": {
    "cagr": 0.123,
    "max_drawdown": -0.27,
    "turnovers": 10
  }
}

完整輸出結構見

templates/output-json.md

。 </output_schema_summary>

<success_criteria> 執行成功時應產出：

當前狀態（RISK_ON 或 RISK_OFF）
LeadingIndex 與 CoincidentIndex 數值
冰山/下沉事件判定
切換事件清單（含日期、原因）
回測績效摘要（CAGR, MaxDD, 換手次數）
與 benchmark 比較（買入持有、60/40）
診斷資訊（各指標貢獻）
視覺化圖表（可選，輸出至
```
output/
```
目錄） </success_criteria>

<essential_principles>

<principle name="two_state_model"> **Core of Two-State Switching Model**

The Zeberg–Salomon model simplifies the market into two states:

RISK_ON: Hold equities (SPY), business cycle expansion phase
RISK_OFF: Hold long-term bonds (TLT), business cycle contraction phase

The switching logic is based on the business cycle rule that "leading indicators weaken first, followed by confirmation from coincident indicators." </principle>

<principle name="leading_coincident"> **Leading vs Coincident Indicators**

Type	Function	Typical Components	Lead Time
Leading	Early Warning	Yield curve, new orders, housing permits	6-12 months
Coincident	Confirmation	Employment, industrial production, real income	Synchronous

Synthesis Method:

Transform each series (yoy/mom/diff)
Unify direction (+1/-1)
Rolling z-score standardization
EMA smoothing
Weighted synthesis </principle>

<principle name="iceberg_sinking"> **Iceberg Event vs Sinking Event**

Iceberg Event: LeadingIndex < iceberg_threshold
  → 預警：景氣開始轉弱
  → 搭配「領先指標下降」+ 可選「市場亢奮」濾鏡

Sinking Event: CoincidentIndex < sinking_threshold
  → 確認：實體經濟收縮
  → 通常在 Iceberg 之後數月發生

State Machine Logic:

RISK_ON → RISK_OFF: Iceberg Event confirmed consecutively + negative slope
RISK_OFF → RISK_ON: Leading indicator rebounds above (threshold + hysteresis) </principle>

<principle name="data_access"> **Data Acquisition Methods**

This skill uses data sources that do not require an API key:

FRED CSV:

https://fred.stlouisfed.org/graph/fredgraph.csv?id={SERIES_ID}

Yahoo Finance: Fetch SPY, TLT, VIX using the
```
yfinance
```
package

Scripts are located in the

scripts/

directory and can be executed directly. </principle>

</essential_principles>

<objective> Implement the Zeberg–Salomon two-state rotation strategy:

Construct Indicators: Synthesize LeadingIndex and CoincidentIndex from FRED data
Detect Events: Identify "Iceberg" (leading indicator weakening) and "Sinking" (coincident indicator confirmation) events
Generate Switch Signals: Produce RISK_ON ↔ RISK_OFF switch events
Backtest Performance: Calculate cumulative returns, MaxDD, CAGR, and compare with benchmarks

Outputs: List of switch events, indicator time series, backtest summary, diagnostic information. </objective>

<quick_start>

Fastest Way: Run Default Backtest

bash

cd skills/zeberg-salomon-rotator
pip install pandas numpy yfinance pandas-datareader  # First-time use
python scripts/rotator.py --quick

Sample Output:

json

{
  "state": "RISK_ON",
  "latest_indices": {"LeadingIndex": 0.41, "CoincidentIndex": 0.22},
  "iceberg_event": false,
  "sinking_event": false,
  "last_switch": {"date": "2023-06-30", "action": "EXIT_LONG_BOND_ENTER_EQUITY"}
}

Full Backtest:

bash

python scripts/rotator.py --start 2000-01-01 --end 2026-01-01 --output result.json

</quick_start>

<intake> What operation do you need to perform?

Quick Check - View current business cycle state and latest indicators
Full Backtest - Perform complete historical backtest and performance analysis
Visualization Charts - Generate multi-panel backtest result charts
Monitoring Mode - Set up continuous monitoring and switch alerts
Methodology Learning - Understand the logic of the Zeberg-Salomon model

Please select or provide analysis parameters directly. </intake>

<routing> | Response | Action | |-------------------------------|---------------------------------------------| | 1, "quick", "check" | Execute `python scripts/rotator.py --quick` | | 2, "backtest", "full" | Read `workflows/backtest.md` and execute | | 3, "chart", "plot" | Read `workflows/visualize.md` and execute | | 4, "monitor", "alert" | Read `workflows/monitor.md` and execute | | 5, "learn", "methodology", "why" | Read `references/methodology.md` | | Provide parameters (e.g., date range) | Read `workflows/backtest.md` and execute with parameters |

After routing, read the corresponding document and execute. </routing>

<directory_structure>

zeberg-salomon-rotator/
├── SKILL.md                           # 本文件（路由器）
├── skill.yaml                         # 前端展示元數據
├── manifest.json                      # 技能元數據
├── workflows/
│   ├── backtest.md                    # 完整回測工作流
│   ├── visualize.md                   # 視覺化工作流
│   ├── monitor.md                     # 持續監控工作流
│   └── analyze.md                     # 深度分析工作流
├── references/
│   ├── data-sources.md                # FRED 系列代碼與資料來源
│   ├── methodology.md                 # Zeberg-Salomon 方法論解析
│   └── input-schema.md                # 完整輸入參數定義
├── templates/
│   ├── output-json.md                 # JSON 輸出模板
│   └── output-markdown.md             # Markdown 報告模板
└── scripts/
    ├── rotator.py                     # 主輪動腳本
    ├── visualize.py                   # 視覺化繪圖工具
    └── fetch_data.py                  # 數據抓取工具

</directory_structure>

<reference_index>

Methodology: references/methodology.md

Zeberg-Salomon model concepts
Definitions of Iceberg/Sinking events
Two-state switching logic

Data Sources: references/data-sources.md

FRED series codes (leading/coincident)
Yahoo Finance asset codes
Data frequency and alignment

Input Parameters: references/input-schema.md

Complete parameter definitions
Default values and recommended ranges

</reference_index>

<workflows_index>

Workflow	Purpose	Usage Scenario
backtest.md	Complete historical backtest	When performance analysis is needed
visualize.md	Generate visualization charts	When chart presentation is needed
monitor.md	Continuous state monitoring	For daily monitoring or alerts
analyze.md	In-depth indicator analysis	To understand current market conditions
</workflows_index>

<templates_index>

Template	Purpose
output-json.md	JSON output structure definition
output-markdown.md	Markdown report template
</templates_index>

<scripts_index>

Script	Command	Purpose
rotator.py	`--quick`	Quick check of current state
rotator.py	`--start DATE --end DATE`	Complete backtest
visualize.py	`-i result.json -o chart.png`	Generate visualization charts
fetch_data.py	`--series T10Y3M,PAYEMS`	Fetch FRED data
</scripts_index>

<input_schema_summary>

Core Parameters

Parameter	Type	Default Value	Description
start_date	string	2000-01-01	Backtest start date
end_date	string	today	Backtest end date
freq	string	M	Frequency (M=Monthly)
equity_proxy	string	SPY	Risk asset proxy
bond_proxy	string	TLT	Long-term bond proxy

Threshold Parameters

Parameter	Type	Default Value	Description
iceberg_threshold	number	-0.3	Leading indicator threshold
sinking_threshold	number	-0.5	Coincident indicator threshold
confirm_periods	int	2	Number of consecutive confirmation periods
hysteresis	number	0.15	Entry/exit gap

Complete parameter definitions can be found in

references/input-schema.md

</input_schema_summary>

<output_schema_summary>

json

{
  "skill": "zeberg-salomon-rotator",
  "as_of": "2026-01-14",
  "state": "RISK_ON",
  "latest_indices": {
    "LeadingIndex": 0.41,
    "CoincidentIndex": 0.22,
    "iceberg_event": false,
    "sinking_event": false
  },
  "switch_events": [...],
  "backtest_summary": {
    "cagr": 0.123,
    "max_drawdown": -0.27,
    "turnovers": 10
  }
}

Complete output structure is available in

templates/output-json.md

. </output_schema_summary>

<success_criteria> When execution is successful, the following should be produced:

Current state (RISK_ON or RISK_OFF)
Values of LeadingIndex and CoincidentIndex
Iceberg/Sinking event determination
List of switch events (including date, reason)
Backtest performance summary (CAGR, MaxDD, number of turnovers)
Comparison with benchmarks (buy-and-hold, 60/40)
Diagnostic information (contribution of each indicator)
Visualization charts (optional, output to
```
output/
```
directory) </success_criteria>