From 921f84cb6afd5b1d8643e2ec14d10c27c7fdbb51 Mon Sep 17 00:00:00 2001 From: aszerW Date: Sat, 6 Jun 2026 16:40:01 +0800 Subject: [PATCH] =?UTF-8?q?feat:=20=E6=96=B0=E5=A2=9E=20standardized=5Fslo?= =?UTF-8?q?pe=20(t-statistic)=20=E5=9B=A0=E5=AD=90=E5=B9=B6=E5=AE=9E?= =?UTF-8?q?=E9=AA=8C=E9=AA=8C=E8=AF=81?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit - simple_rotation.py: 新增 standardized_slope_score 函数 (slope/SE) - config_loader.py: FactorType 枚举新增 STANDARDIZED_SLOPE - 对比实验结果: standardized_slope 年化 13.73% vs slope_r2 19.84% - 结论: t-statistic 过度惩罚高波动资产的有效趋势信号，不适合本场景 - 文档更新: 动量因子对比调研报告新增 3.3 节详细分析 --- docs/动量因子对比调研报告.md | 36 ++++- rotation/config_loader.py | 1 + .../output/std_slope_test_results.json | 25 ++++ rotation/experiments/std_slope_test.py | 129 ++++++++++++++++++ rotation/simple_rotation.py | 35 +++++ 5 files changed, 224 insertions(+), 2 deletions(-) create mode 100644 rotation/experiments/output/std_slope_test_results.json create mode 100644 rotation/experiments/std_slope_test.py diff --git a/docs/动量因子对比调研报告.md b/docs/动量因子对比调研报告.md index 1a0f7cd..adaba69 100644 --- a/docs/动量因子对比调研报告.md +++ b/docs/动量因子对比调研报告.md @@ -86,9 +86,12 @@ $$\text{Score} = \frac{\text{prices}[-1]}{\text{prices}[0]} - 1$$ | vol_adjusted_momentum | 13.16% | 0.85 | -18.61% | 0.71 | 393 | 55.9% | | **slope_r2（当前默认）** | **19.84%** | **1.14** | **-15.35%** | **1.29** | 394 | 54.1% | | momentum | 9.27% | 0.57 | -17.42% | 0.53 | 729 | 53.3% | +| standardized_slope | 13.73% | 1.01 | -13.52% | 1.02 | 335 | 54.5% | **结论**：`slope_r2` 全面胜出，年化 +1.48%，夏普 +0.12，回撤改善 +1.01%。 +> **注**：`standardized_slope`（t-statistic）回撤更小但收益大幅落后（年化 -6.11%），说明统计显著性过滤在高波动资产上过度惩罚趋势信号，不适合本场景（详见 3.3）。 + ### 3.2 数值尺度分析（2024-06-03 截面） | 因子 | 最大值 | 最小正值 | max/min 比值 | @@ -100,6 +103,31 @@ $$\text{Score} = \frac{\text{prices}[-1]}{\text{prices}[0]} - 1$$ `slope_r2` 的跨资产数值差距仅 31 倍，远小于其他因子的 2000~3000 倍，这是其跨市场可比的根本原因。 +### 3.3 standardized_slope（t-statistic）实验 + +**公式**： +$$\text{Score} = \frac{\hat{\beta}}{\text{SE}(\hat{\beta})}, \quad \text{SE}(\hat{\beta}) = \sqrt{\frac{\text{MSE}}{S_{xx}}}$$ + +**学术动机**：t-statistic 同时考虑了斜率大小和估计的统计显著性，理论上比 `slope × R²` 更严格。 + +**实验结果**： + +| 指标 | slope_r2 | standardized_slope | Δ | +|------|---------|-------------------|---| +| 年化收益 | 19.84% | 13.73% | **-6.11%** | +| 夏普比率 | 1.14 | 1.01 | **-0.12** | +| 最大回撤 | -15.35% | -13.52% | +1.83% | +| Calmar | 1.29 | 1.02 | -0.27 | +| 调仓次数 | 394 | 335 | -59 | + +**失败原因分析**： + +- **绝对度量 vs 相对度量**：`SE(β)` 是绝对度量（量纲同斜率），而 `R²` 是相对度量（无量纲）。在跨资产比较中，SE 对高波动资产（如 CL=F、HSTECH）惩罚过重，即使趋势方向正确，score 也会被压低。 +- **过度过滤**：调仓次数减少 59 次，说明 t-statistic 把大量"方向对但波动大"的有效信号过滤掉了，反而错失趋势行情。 +- **数学等价性**：`slope / SE(slope) = slope × √(Sxx / MSE)`，而 `slope × R² = slope × (1 - SS_res/SS_tot)`。前者惩罚的是残差方差绝对值，后者惩罚的是偏离趋势线的比例——后者更适合作为趋势质量因子。 + +**结论**：t-statistic 不适合本场景，保持 `slope_r2` 为默认因子。 + --- ## 4. slope_r2 胜出的原因分析 @@ -245,18 +273,22 @@ factor: | 负价格排除 | 窗口内出现非正价格时返回 None | 低（实际影响极小） | | 多窗口融合 | 结合 5/25/60 天信号 | 中 | | 截面 rank | 动量值转截面百分位排名 | 低（slope_r2 已天然可比） | +| ~~标准化斜率~~ | slope/SE(slope)，已验证不适合（详见 3.3） | **已排除** | --- ## 附录：实验代码 -对比实验脚本：`rotation/experiments/factor_comparison.py` +对比实验脚本：`rotation/experiments/factor_comparison.py`、`rotation/experiments/std_slope_test.py` 运行方式： ```bash cd /Users/aszer/code/etf set -a && source .env && set +a python rotation/experiments/factor_comparison.py +python rotation/experiments/std_slope_test.py ``` -结果输出：`rotation/experiments/output/factor_comparison_results.json` +结果输出： +- `rotation/experiments/output/factor_comparison_results.json` +- `rotation/experiments/output/std_slope_test_results.json` diff --git a/rotation/config_loader.py b/rotation/config_loader.py index 419b9b6..b34aad5 100644 --- a/rotation/config_loader.py +++ b/rotation/config_loader.py @@ -34,6 +34,7 @@ class FactorType(str, Enum): SLOPE_R2 = "slope_r2" WEIGHTED_MOMENTUM = "weighted_momentum" VOL_ADJUSTED_MOMENTUM = "vol_adjusted_momentum" + STANDARDIZED_SLOPE = "standardized_slope" class PremiumMode(str, Enum): diff --git a/rotation/experiments/output/std_slope_test_results.json b/rotation/experiments/output/std_slope_test_results.json new file mode 100644 index 0000000..5998495 --- /dev/null +++ b/rotation/experiments/output/std_slope_test_results.json @@ -0,0 +1,25 @@ +{ + "timestamp": "2026-06-06T16:36:39.736366", + "results": [ + { + "factor_type": "slope_r2", + "annual_return": 0.198416094188119, + "total_return": 2.0421974188211456, + "sharpe_ratio": 1.1350010914615083, + "max_drawdown": -0.15352659557851117, + "win_rate": 0.541343669250646, + "rebalance_count": 394, + "calmar_ratio": 1.2923890707043786 + }, + { + "factor_type": "standardized_slope", + "annual_return": 0.13732579856023497, + "total_return": 1.2055386092808908, + "sharpe_ratio": 1.0139515617271433, + "max_drawdown": -0.13523854511100616, + "win_rate": 0.5452196382428941, + "rebalance_count": 335, + "calmar_ratio": 1.0154338650087928 + } + ] +} \ No newline at end of file diff --git a/rotation/experiments/std_slope_test.py b/rotation/experiments/std_slope_test.py new file mode 100644 index 0000000..a7f2c79 --- /dev/null +++ b/rotation/experiments/std_slope_test.py @@ -0,0 +1,129 @@ +""" +slope_r2 vs standardized_slope 对比实验 + +测试两种信号质量优化的回测表现： +1. slope_r2: slope × R² (当前默认) +2. standardized_slope: slope / SE(slope) (t-statistic) + +运行方式: + cd /Users/aszer/code/etf + set -a && source .env && set +a + python3 rotation/experiments/std_slope_test.py +""" + +import os +import sys +import json +import yaml +from pathlib import Path +from datetime import datetime + +PROJECT_ROOT = Path(__file__).parent.parent.parent +sys.path.insert(0, str(PROJECT_ROOT)) + +from rotation.simple_rotation import SimpleRotationStrategy + +FACTOR_TYPES = [ + ("slope_r2", "slope_r2 (slope×R², 当前默认)"), + ("standardized_slope", "standardized_slope (t-statistic)"), +] + + +def run_factor_experiment(factor_type: str): + """Run backtest with a specific factor type""" + print(f"\n{'='*60}") + print(f" Testing: {factor_type}") + print(f"{'='*60}") + + config_path = PROJECT_ROOT / "rotation" / "config_simple.yaml" + with open(config_path, 'r', encoding='utf-8') as f: + config = yaml.safe_load(f) + + original_type = config['factor']['type'] + config['factor']['type'] = factor_type + + with open(config_path, 'w', encoding='utf-8') as f: + yaml.dump(config, f, allow_unicode=True, default_flow_style=False) + + try: + strategy = SimpleRotationStrategy() + result = strategy.run() + if result: + metrics = result.get('metrics', {}) + return { + 'factor_type': factor_type, + 'annual_return': metrics.get('annual_return', 0), + 'total_return': metrics.get('total_return', 0), + 'sharpe_ratio': metrics.get('sharpe_ratio', 0), + 'max_drawdown': metrics.get('max_drawdown', 0), + 'win_rate': metrics.get('win_rate', 0), + 'rebalance_count': metrics.get('rebalance_count', 0), + 'calmar_ratio': metrics.get('calmar_ratio', 0), + } + finally: + config['factor']['type'] = original_type + with open(config_path, 'w', encoding='utf-8') as f: + yaml.dump(config, f, allow_unicode=True, default_flow_style=False) + + return None + + +def main(): + if 'FLASK_API_URL' not in os.environ: + os.environ['FLASK_API_URL'] = 'https://k3s.tokenpluse.xyz' + + print("="*60) + print(" slope_r2 vs standardized_slope 对比实验") + print("="*60) + + results = [] + for factor_type, description in FACTOR_TYPES: + print(f"\n>>> {description}") + result = run_factor_experiment(factor_type) + if result: + results.append(result) + print(f" ✓ {factor_type}: 年化={result['annual_return']:.2%}, " + f"夏普={result['sharpe_ratio']:.2f}, 回撤={result['max_drawdown']:.2%}") + else: + print(f" ✗ {factor_type}: 运行失败") + + print(f"\n{'='*60}") + print(" 对比结果汇总") + print(f"{'='*60}") + print(f"{'因子类型':<25} {'年化收益':>10} {'夏普比率':>8} {'最大回撤':>10} {'Calmar':>8} {'调仓次数':>8} {'胜率':>6}") + print("-"*80) + for r in results: + print(f"{r['factor_type']:<25} " + f"{r['annual_return']:>9.2%} " + f"{r['sharpe_ratio']:>8.2f} " + f"{r['max_drawdown']:>9.2%} " + f"{r['calmar_ratio']:>8.2f} " + f"{r['rebalance_count']:>8d} " + f"{r['win_rate']:>5.1%}") + + if len(results) >= 2: + base = results[0] + new = results[1] + print(f"\n{'='*60}") + print(" 变化对比 (standardized_slope vs slope_r2)") + print(f"{'='*60}") + print(f" 年化收益: {base['annual_return']:.2%} → {new['annual_return']:.2%} " + f"(Δ={new['annual_return']-base['annual_return']:+.2%})") + print(f" 夏普比率: {base['sharpe_ratio']:.2f} → {new['sharpe_ratio']:.2f} " + f"(Δ={new['sharpe_ratio']-base['sharpe_ratio']:+.2f})") + print(f" 最大回撤: {base['max_drawdown']:.2%} → {new['max_drawdown']:.2%} " + f"(Δ={new['max_drawdown']-base['max_drawdown']:+.2%})") + print(f" 调仓次数: {base['rebalance_count']} → {new['rebalance_count']} " + f"(Δ={new['rebalance_count']-base['rebalance_count']:+d})") + + output_dir = PROJECT_ROOT / "rotation" / "experiments" / "output" + output_dir.mkdir(exist_ok=True) + output_path = output_dir / "std_slope_test_results.json" + with open(output_path, 'w', encoding='utf-8') as f: + json.dump({'timestamp': datetime.now().isoformat(), 'results': results}, + f, ensure_ascii=False, indent=2) + print(f"\n结果已保存: {output_path}") + + +if __name__ == "__main__": + main() diff --git a/rotation/simple_rotation.py b/rotation/simple_rotation.py index 289b2e7..3f007f5 100644 --- a/rotation/simple_rotation.py +++ b/rotation/simple_rotation.py @@ -123,6 +123,39 @@ def slope_r2_score(prices: np.ndarray) -> float: return 10000 * slope * r2 +def standardized_slope_score(prices: np.ndarray) -> float: + """Standardized slope (t-statistic): slope / SE(slope) + + Academic basis: + - Uses normalized prices (p/p[0]) for cross-asset comparability, + consistent with slope_r2_score. + - Divides slope by its standard error, yielding a statistical significance + measure rather than raw magnitude. + - Equivalent to the t-value for H0: slope=0, penalizing noisy trends. + + Formula: + SE(slope) = sqrt(MSE / Sxx) + MSE = SS_res / (n - 2) + Sxx = sum((xi - x_bar)^2) = n*(n-1)*(n+1)/12 for x = 0..n-1 + """ + n = len(prices) + if n < 5: + return 0.0 + prices = np.clip(prices, 0.01, None) + y = prices / prices[0] # normalize + x = np.arange(n) + slope, intercept = np.polyfit(x, y, 1) + y_pred = slope * x + intercept + ss_res = np.sum((y - y_pred) ** 2) + # Standard error of slope + mse = ss_res / (n - 2) # unbiased MSE + sxx = n * (n - 1) * (n + 1) / 12 # sum of squared deviations of x + se_slope = math.sqrt(mse / sxx) if sxx > 0 else 1e-9 + if se_slope < 1e-12: + se_slope = 1e-12 + return slope / se_slope + + def momentum_score(prices: np.ndarray) -> float: """Simple price return: (last / first) - 1""" if len(prices) < 5: @@ -409,6 +442,8 @@ class SimpleRotationStrategy: return vol_adjusted_momentum_score(prices) elif ft == FactorType.SLOPE_R2: return slope_r2_score(prices) + elif ft == FactorType.STANDARDIZED_SLOPE: + return standardized_slope_score(prices) elif ft == FactorType.MOMENTUM: return momentum_score(prices) return weighted_momentum_score(prices)