From 8f17c56dbf21d95e955f021539309c280608f153 Mon Sep 17 00:00:00 2001
From: zengbin93 <zeng_bin8888@163.com>
Date: Fri, 29 Nov 2024 22:36:47 +0800
Subject: [PATCH] =?UTF-8?q?0.9.61=20=E6=96=B0=E5=A2=9E=20limit=5Fleverage?=
 =?UTF-8?q?=20=E5=87=BD=E6=95=B0?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 czsc/__init__.py |  1 +
 czsc/eda.py      | 38 ++++++++++++++++++++++++++++++++++++++
 test/test_eda.py | 34 ++++++++++++++++++++++++++++++++++
 3 files changed, 73 insertions(+)

diff --git a/czsc/__init__.py b/czsc/__init__.py
index 233c747dc..81dd9fa4a 100644
--- a/czsc/__init__.py
+++ b/czsc/__init__.py
@@ -218,6 +218,7 @@
     sma_long_bear,
     dif_long_bear,
     tsf_type,
+    limit_leverage,
 )
 
 
diff --git a/czsc/eda.py b/czsc/eda.py
index 2f045932a..6b3c08038 100644
--- a/czsc/eda.py
+++ b/czsc/eda.py
@@ -511,3 +511,41 @@ def tsf_type(df: pd.DataFrame, factor, n=5, **kwargs):
 
     sorted_layers = sorted(layer_returns.items(), key=lambda x: x[1], reverse=True)
     return "->".join([f"{x[0]}" for x in sorted_layers])
+
+
+def limit_leverage(df: pd.DataFrame, leverage: float = 1.0, **kwargs):
+    """限制杠杆比例
+
+    原理描述：
+
+    1. 计算滚动窗口内权重的绝对均值 abs_mean，初始窗口内权重的绝对均值设为 leverage
+    2. 用 leverage 除以 abs_mean，得到调整比例 adjust_ratio
+    3. 将原始权重乘以 adjust_ratio，再限制在 -leverage 和 leverage 之间
+
+    :param df: DataFrame, columns=['dt', 'symbol', 'weight']
+    :param leverage: float, 杠杆倍数
+    :param kwargs:
+
+        - copy: bool, 是否复制 DataFrame
+        - window: int, 滚动窗口，默认为 300
+        - min_periods: int, 最小样本数，小于该值的窗口不计算均值，默认为 50
+        - weight: str, 权重列名，默认为 'weight'
+
+    :return: DataFrame
+    """
+    window = kwargs.get("window", 300)
+    min_periods = kwargs.get("min_periods", 50)
+    weight = kwargs.get("weight", "weight")
+
+    assert weight in df.columns, f"数据中不包含权重列 {weight}"
+    assert df['symbol'].nunique() == 1, "数据中包含多个品种，必须单品种"
+    assert df['dt'].is_monotonic_increasing, "数据未按日期排序，必须升序排列"
+    assert df['dt'].is_unique, "数据中存在重复dt，必须唯一"
+
+    if kwargs.get("copy", False):
+        df = df.copy()
+
+    abs_mean = df[weight].abs().rolling(window=window, min_periods=min_periods).mean().fillna(leverage)
+    adjust_ratio = leverage / abs_mean
+    df[weight] = (df[weight] * adjust_ratio).clip(-leverage, leverage)
+    return df
diff --git a/test/test_eda.py b/test/test_eda.py
index fd6c40542..b257338d4 100644
--- a/test/test_eda.py
+++ b/test/test_eda.py
@@ -69,5 +69,39 @@ def test_weights_simple_ensemble_only_long():
     pd.testing.assert_series_equal(result["weight"], expected)
 
 
+def test_limit_leverage():
+    from czsc.eda import limit_leverage
+
+    data = {
+        "dt": pd.date_range(start="2023-01-01", periods=10, freq="D"),
+        "symbol": ["TEST"] * 10,
+        "weight": [0.1, 0.2, -0.3, 3, -0.5, 0.6, -0.7, 0.8, -0.9, 1.0],
+        "price": [100 + i for i in range(10)],
+    }
+    df = pd.DataFrame(data)
+
+    # Test with leverage = 1.0
+    df_result = limit_leverage(df, leverage=1.0, copy=True, window=3, min_periods=2)
+    assert df_result["weight"].max() <= 1.0
+    assert df_result["weight"].min() >= -1.0
+
+    # Test with leverage = 2.0
+    df_result = limit_leverage(df, leverage=2.0, copy=True, window=3, min_periods=2)
+    assert df_result["weight"].max() <= 2.0
+    assert df_result["weight"].min() >= -2.0
+
+    # Test with different window and min_periods
+    df_result = limit_leverage(df, leverage=1.0, window=5, min_periods=2, copy=True)
+    assert df_result["weight"].max() <= 1.0
+    assert df_result["weight"].min() >= -1.0
+
+    df1 = df.copy()
+    df1.rename(columns={"weight": "weight1"}, inplace=True)
+    # Test with leverage = 1.0
+    df_result = limit_leverage(df1, leverage=1.0, copy=True, window=3, min_periods=2, weight="weight1")
+    assert df_result["weight1"].max() <= 1.0
+    assert df_result["weight1"].min() >= -1.0
+
+
 if __name__ == "__main__":
     pytest.main()