Research Framework研究框架

Systematic process.
Repeatable output. 系統化流程
可重複的研究分析輸出

Rules-based methodology with zero discretionary judgment. The same process runs every week, regardless of market sentiment. 規則驅動的方法論，零主觀判斷。無論市場情緒如何，每週相同的流程準時運行。

Core Process核心流程

  • ▶Evaluate ETF relative strength across universe評估 ETF 宇宙中的相對強弱
  • ▶Rank ETFs by multi-period momentum metrics基於多週期動量指標對 ETF 進行排名
  • ▶Apply trend strength and volatility filters應用趨勢強度與波動率過濾器
  • ▶Monitor risk environment and market regime監測風險環境與市場狀態
  • ▶Publish standardised research analysis weekly每週發佈標準化研究分析輸出

Research Analysis Labels研究分析輸出標籤

  • ▶Trend Above Filter — research observation: price above trend threshold, strong relative reading趨勢在過濾線上方 — 研究觀察：價格在趨勢閾值上方，相對強度讀數強
  • ▶Trend Recovering — research observation: trend stabilising, relative strength improving趨勢企穩 — 研究觀察：趨勢趨於穩定，相對強度讀數改善
  • ▶Neutral Reading — research observation: mixed indicators, no clear directional trend中性讀數 — 研究觀察：指標混合，無明確方向性趨勢
  • ▶Below Filter Threshold — research observation: below trend threshold, elevated risk reading低於過濾線閾值 — 研究觀察：在趨勢閾值下方，風險讀數偏高

Market Regime Analysis市場狀態研判

⚔ Offensive (Risk-On) ⚔ 進攻（Risk-On）
≥80% momentum filters pass · Strong trend · Aggressive allocation to leading sectors ≥80% 動量過濾器通過 · 趨勢強勢 · 積極配置領先板塊
⚖ Neutral ⚖ 中性
50–79% filters pass · Borderline trend · Balanced allocation, reduced concentration 50–79% 過濾器通過 · 趨勢邊際 · 均衡配置，降低集中度
🛡 Defensive (Risk-Off) 🛡 防守（Risk-Off）
<50% filters pass · Trend deteriorating · Defensive sectors / cash equivalents priority <50% 過濾器通過 · 趨勢惡化 · 優先配置防禦性板塊 / 現金等價物
⚡

Open Source on GitHubGitHub 開源

Simplified versions of both ETF rotation researchs are open-sourced. The open-source versions demonstrate the core framework logic — the production models used in the weekly publication include additional signal layers and refinements not included in the public versions. 兩個 ETF 輪動模型的簡化版本已公開開源。開源版本展示了核心框架邏輯——每週發佈中使用的正式版本包含開源版本未涵蓋的額外信號層與優化。

⚡ GitHub Official SiteGitHub 官網 → 🇺🇸 US ETF Rotation — Simplified美股 ETF 輪動模型簡化版 → 🇨🇳 A-Share ETF Rotation — SimplifiedA 股 ETF 輪動模型簡化版 →
US ETF Rotation — Backtest美股 ETF 輪動 — 回測

🇺🇸 US Model Performance
Jan 2019 – Apr 2026 · Hypothetical 🇺🇸 美股模型業績
2019年1月 – 2026年4月 · 假設性結果

Hypothetical results based on historical data with 0.15% cost per trade. Does not represent actual trading returns. 基於歷史數據的假設性模擬，含每筆 0.15% 交易成本，不代表實際交易收益。

16.81%CAGR年化收益率
-23.58%Max Drawdown最大回撤
0.937Sharpe Ratio夏普比率
DD -30%vs SPYvs 標普500 回撤縮小
α
Alpha Rotation Lab阿爾法輪動實驗室

US ETF Rotation  ·  Model V6.2

Backtest 2019.01.02 – 2026.05.06  ·  Benchmark: S&P 500 (SPY)  ·  Cost: 0.15% one-way  ·  Weekly rebalance
01 · Full Period Overview (2019.01 – 2026.05)
Total Return
+212.16%
SPY: +227.04%
Nearly identical
CAGR
16.81%
SPY: 17.56%
Alpha: −0.75%/yr
Max Drawdown
−23.58%
SPY: −33.72%
30% smaller
Sharpe Ratio
0.937
SPY: 0.824
+0.113
Calmar Ratio
0.713
SPY: 0.521
+0.192
02 · Performance Analysis

📈 Why S&P 500 Leads in Raw Return

The 2019–2026 period was dominated by mega-cap tech concentration. By 2024, the top 10 S&P 500 holdings represented over 35% of the index — Nvidia alone rose 200%+ in both 2023 and 2024.

SPY's market-cap weighting automatically captures this concentration in full. The rotation research, diversifying across a broad ETF universe with explicit regime-based risk control, intentionally reduces mega-cap exposure — most visible in 2021 (SPY +30.51% vs model +22.01%) and 2024 (SPY +25.59% vs model +5.16%).

This is the model's design: accept a moderate return lag to achieve substantially lower drawdown. The same mega-cap concentration that drove 2024 returns caused the S&P 500 to fall 34% in 2022. The rotation research lost only 17.78% that year.

🛡️ The Risk-Control Trade-Off

Giving up 0.75% annual return, the model achieves dramatic improvements in every risk metric — the strongest case for systematic risk management.

Annual return given up
−0.75%/yr
Max drawdown reduced
−10.14 pp (30%)
Sharpe ratio improved
+0.113
Calmar ratio improved
+0.192

2020 COVID crash: model −21.80% vs SPY −33.72% — 11.9 pp protected at the worst moment.
2022 bear market: model −17.78% vs SPY −18.65% — defensive cash in T-bills turned the tide.

03 · Annual Breakdown (Model vs S&P 500)
Year Ann. ReturnMax DD SharpeCalmar Annual AlphaBenchmark
2019Model +27.29-7.95+2.076+3.43-3.80%
+31.09-6.62+2.072+4.70 S&P 500
2020Model +30.60-21.80+1.256+1.40+13.36%
+17.24-33.72+0.584+0.51 S&P 500
2021Model +22.01-5.45+1.302+4.04-8.50%
+30.51-5.11+1.978+5.96 S&P 500
2022Model -17.78-23.58-1.135-0.76+0.87%
-18.65-24.50-0.818-0.76 S&P 500
2023Model +21.97-11.30+1.364+1.96-4.74%
+26.71-9.97+1.743+2.70 S&P 500
2024Model +5.16-12.40+0.282+0.42-20.43%
+25.59-8.41+1.723+3.04 S&P 500
2025Model +21.73-9.82+1.287+2.23+3.72%
+18.01-18.76+0.852+0.97 S&P 500
2026 YTDModel +17.99-7.53+3.029+8.28+10.28%
+7.71-8.88+1.485+2.74 S&P 500
Full PeriodModel +212.16%
CAGR 16.81%		 -23.58%0.9370.713 -0.75%/yr
+227.04%
CAGR 17.56%		 -33.72%0.8240.521 S&P 500
ℹ  2024 underperformance note: 2024 was an extraordinary mega-cap concentration year. Nvidia's weighting in SPY exceeded most sector ETFs. The model's diversification was a material headwind (−20.43% alpha). The same structure, however, limited 2025 drawdown to −9.82% while SPY fell −18.76% — a 8.9 pp gap that illustrates the model's primary purpose: drawdown control over maximum return capture.
04 · NAV Curve · Drawdown · Annual Return Bars
US ETF Rotation Model V6.2 vs S&P 500
US ETF Rotation Model V6.2
S&P 500 (SPY)
Equity Curve
Hypothetical backtest · Log scale · 0.15% cost/trade · Weekly rebalance · Not investment advice
⚠ All results are hypothetical backtests. Past performance does not guarantee future results. 0.15% cost/trade included. Excludes taxes, slippage, market impact.
Alpha Rotation Lab  ·  Research publication only  ·  Not investment advice
α
Alpha Rotation Lab阿爾法輪動實驗室

US ETF 輪動  ·  模型 V6.2

回測區間：2019.01.02 – 2026.05.06  ·  基準：標普500（SPY） ·  單邊手續費：0.15%  ·  每週調倉
01 · 總區間概覽（2019.01 – 2026.05）
累計收益
+212.16%
標普500：+227.04%
幾乎相同
年化收益 CAGR
16.81%
標普500：17.56%
年均超額：−0.75%
最大回撤
−23.58%
標普500：−33.72%
縮小 30%
夏普比率
0.937
標普500：0.824
+0.113
卡瑪比率
0.713
標普500：0.521
+0.192
02 · 業績分析

📈 標普500原始收益領先的原因

2019–2026年市場結構上被科技巨頭高度集中化所主導。2024年標普500前十大持倉佔指數超35%——英偉達在2023、2024年分別上漲超200%。

SPY的市值加權機制自動完整捕獲了這種集中度。輪動模型在廣泛ETF宇宙中分散配置，並進行明確的Regime風控，有意識地降低了對科技巨頭的暴露——在2021年（SPY +30.51% vs 模型 +22.01%）和2024年（SPY +25.59% vs 模型 +5.16%）尤為明顯。

這是模型的設計初衷：以適度讓出收益換取大幅降低回撤。2024年推高收益的同一集中度，導致了2022年標普500暴跌34%，而模型當年僅虧損17.78%。

🛡️ 風控能力的代價與收益

讓出0.75%的年化收益，模型在每項風控指標上均取得了顯著改善——這是系統化風險管理最有力的論據。

每年讓出的收益
−0.75%/年
最大回撤縮小
−10.14 pp（縮小30%）
夏普比率提升
+0.113
卡瑪比率提升
+0.192

2020年新冠崩盤：模型 −21.80% vs SPY −33.72%——最關鍵時刻保護了11.9個百分點。
2022年熊市：模型 −17.78% vs SPY −18.65%——防守期短債收益扭轉超額為正。

03 · 逐年業績分拆（模型 vs 標普500）
年份 年度收益率最大回撤 夏普卡瑪 年度超額比較基準
2019模型 +27.29-7.95+2.076+3.43-3.80%
+31.09-6.62+2.072+4.70 標普500
2020模型 +30.60-21.80+1.256+1.40+13.36%
+17.24-33.72+0.584+0.51 標普500
2021模型 +22.01-5.45+1.302+4.04-8.50%
+30.51-5.11+1.978+5.96 標普500
2022模型 -17.78-23.58-1.135-0.76+0.87%
-18.65-24.50-0.818-0.76 標普500
2023模型 +21.97-11.30+1.364+1.96-4.74%
+26.71-9.97+1.743+2.70 標普500
2024模型 +5.16-12.40+0.282+0.42-20.43%
+25.59-8.41+1.723+3.04 標普500
2025模型 +21.73-9.82+1.287+2.23+3.72%
+18.01-18.76+0.852+0.97 標普500
2026 年初至今模型 +17.99-7.53+3.029+8.28+10.28%
+7.71-8.88+1.485+2.74 標普500
全區間模型 +212.16%
年化16.81%		 -23.58%0.9370.713 -0.75%/年
+227.04%
年化17.56%		 -33.72%0.8240.521 標普500
ℹ  2024年跑輸說明：2024年是科技巨頭集中化的極端年份，英偉達在SPY中的權重超過了大多數行業ETF，模型分散結構構成顯著逆風（超額 −20.43%）。然而正是這一結構，在2025年SPY下跌−18.76%時將模型損失控制在−9.82%——差距8.9個百分點，體現了模型的核心定位：以回撤控制優先，而非追求最高收益。
04 · 淨值曲線 · 回撤對比 · 年度收益柱圖
US ETF 輪動模型 V6.2 vs 標普500
US ETF 輪動模型 V6.2
標普500（SPY）
Equity Curve
假設性回測結果 · 對數軸 · 每次交易0.15%手續費 · 每週調倉 · 不構成投資建議
⚠ 以上所有數據均為假設性回測結果，歷史業績不代表未來收益。已含每次交易0.15%手續費，未扣除稅費、滑點及市場衝擊成本。
阿爾法輪動實驗室  ·  純屬研究出版物  ·  不構成投資建議

⚠ Historical simulations for research and informational purposes only · Does not represent actual investment results · 0.15% cost/trade included · Past hypothetical performance does not guarantee future results歷史模擬僅供研究與資訊參考 · 不代表實際投資結果 · 已含每筆 0.15% 交易成本 · 假設性歷史業績不保證未來收益

A-Share ETF Rotation — BacktestA 股 ETF 輪動 — 回測

🇨🇳 A-Share Model Performance
Jan 2019 – Apr 2026 · Hypothetical 🇨🇳 A 股模型業績
2019年1月 – 2026年4月 · 假設性結果

Backtest includes 0.15% cost per trade. Substantially improved risk-adjusted returns vs CSI 300 on both return and risk metrics. 回測含每筆 0.15% 交易成本。在回報與風險指標上均風險調整後收益明顯優於滬深300。

28.18%CAGR年化收益率
-13.78%Max Drawdown最大回撤
1.339Sharpe Ratio夏普比率
+19%/yrAlpha vs CSI 300vs 滬深300超額
α
Alpha Rotation Lab 阿爾法輪動實驗室

A-Share ETF Rotation  ·  Model V3.6 (Pure Momentum)

Backtest 2019.01.02 – 2026.04.30  ·  Benchmark: CSI 300 (510300)  ·  Cost: 0.15% one-way  ·  Rebalance every 5 trading days
01 · Full Period Overview (2019.01 – 2026.04)
Total Return
+475.2%
CSI300: +85.4%
Alpha +389.8%
CAGR (Ann. Return)
28.18%
CSI300: 9.15%
+19.03%/yr
Max Drawdown
-13.78%
CSI300: -42.13%
DD 67% smaller
Sharpe Ratio
1.339
CSI300: 0.449
+0.890
Calmar Ratio
2.045
CSI300: 0.217
9.4× better
02 · Annual Breakdown (Model vs CSI 300)
Year Ann. Return Max DD Sharpe Calmar Annual Alpha Benchmark
2019Model +17.8%-10.4%+1.005+1.78-23.7%
+40.5%-13.4%+1.774+3.14 CSI 300
2020Model +39.1%-9.0%+1.782+4.56+12.3%
+27.4%-16.1%+1.106+1.78 CSI 300
2021Model +26.7%-6.5%+1.338+4.30+32.8%
-4.9%-17.2%-0.297-0.29 CSI 300
2022Model +5.5%-5.7%+0.390+1.01+26.3%
-19.8%-27.4%-1.147-0.75 CSI 300
2023Model +7.9%-7.8%+0.576+1.06+18.7%
-10.0%-19.9%-0.907-0.52 CSI 300
2024Model +22.1%-13.8%+0.920+1.67+3.2%
+19.0%-12.4%+0.832+1.60 CSI 300
2025Model +61.0%-8.8%+2.255+7.26+38.6%
+24.3%-10.1%+1.477+2.51 CSI 300
2026 YTDModel +14.6%-9.9%+1.854+5.67+12.4%
+2.2%-7.6%+0.404+0.96 CSI 300
Full PeriodModel +475.2%
CAGR 28.18%		 -13.78%1.3392.045 +19.03%/yr
+85.4%
CAGR 9.15%		 -42.13%0.4490.217 CSI 300
ℹ  2019 note: CSI 300 surged +40.5% in 2019 led by large-cap financials and consumer blue-chips. The model, in its early data-accumulation phase, rotated cautiously — but successfully capped max drawdown at -10.4% vs CSI300's -13.4%. An honest presentation including the year the benchmark outperformed.
03 · NAV Curve · Drawdown · Annual Return Bars
Model vs CSI 300 · 2019.01 – 2026.04
ETF Rotation Model V3.6
CSI 300 Benchmark
Equity Curve
Hypothetical backtest · Log scale · 0.15% cost/trade · 5-day rebalance · Not investment advice
⚠ All results are hypothetical backtests. Past performance does not guarantee future results. 0.15% per-trade cost included. Results exclude taxes, slippage, market impact.
Alpha Rotation Lab · Research publication only · Not investment advice
α
Alpha Rotation Lab 阿爾法輪動實驗室

A股ETF輪動  ·  模型 V3.6（純輪動）

回測區間：2019.01.02 – 2026.04.30  ·  基準：滬深300（510300） ·  單邊手續費：0.15%  ·  每5個交易日調倉
01 · 總區間概覽（2019.01 – 2026.04）
累計收益
+475.2%
滬深300：+85.4%
超額 +389.8%
年化收益 CAGR
28.18%
滬深300：9.15%
+19.03%/yr
最大回撤
-13.78%
滬深300：-42.13%
回撤縮小 67%
夏普比率
1.339
滬深300：0.449
+0.890
卡瑪比率
2.045
滬深300：0.217
高出 9.4 倍
02 · 逐年業績分拆（模型 vs 滬深300）
年份 年度收益率 最大回撤 夏普 卡瑪 年度超額 比較基準
2019模型 +17.8%-10.4%+1.005+1.78-23.7%
+40.5%-13.4%+1.774+3.14 滬深300
2020模型 +39.1%-9.0%+1.782+4.56+12.3%
+27.4%-16.1%+1.106+1.78 滬深300
2021模型 +26.7%-6.5%+1.338+4.30+32.8%
-4.9%-17.2%-0.297-0.29 滬深300
2022模型 +5.5%-5.7%+0.390+1.01+26.3%
-19.8%-27.4%-1.147-0.75 滬深300
2023模型 +7.9%-7.8%+0.576+1.06+18.7%
-10.0%-19.9%-0.907-0.52 滬深300
2024模型 +22.1%-13.8%+0.920+1.67+3.2%
+19.0%-12.4%+0.832+1.60 滬深300
2025模型 +61.0%-8.8%+2.255+7.26+38.6%
+24.3%-10.1%+1.477+2.51 滬深300
2026 年初至今模型 +14.6%-9.9%+1.854+5.67+12.4%
+2.2%-7.6%+0.404+0.96 滬深300
全區間模型 +475.2%
年化28.18%		 -13.78%1.3392.045 年均+19.03%
+85.4%
年化9.15%		 -42.13%0.4490.217 滬深300
ℹ  2019年說明：2019年A股牛市由大盤金融股與消費藍籌主導，滬深300大漲40.5%。模型在早期數據積累階段切換較為謹慎，但成功將最大回撤控制在-10.4%，優於滬深300當年的-13.4%。這是包含基準跑贏年份的誠實回測呈現。
03 · 淨值曲線 · 回撤對比 · 年度收益柱圖
模型 vs 滬深300 · 2019.01 – 2026.04
ETF輪動模型 V3.6
滬深300 基準
Equity Curve
假設性回測結果 · 對數軸 · 每次交易0.15%手續費 · 每5個交易日調倉 · 不構成投資建議
⚠ 以上所有數據均為假設性回測結果。歷史業績不代表未來收益。已含每次交易0.15%手續費，未扣除稅費、滑點及市場衝擊成本。
阿爾法輪動實驗室 · 純屬研究出版物 · 不構成投資建議

⚠ Historical simulations for research and informational purposes only · Does not represent actual investment results · 0.15% cost/trade included · Past hypothetical performance does not guarantee future results歷史模擬僅供研究與資訊參考 · 不代表實際投資結果 · 已含每筆 0.15% 交易成本 · 假設性歷史業績不保證未來收益