regime-aware-multi-factor/Regression.py at main · TorresYue/regime-aware-multi-factor · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
import numpy as np
import pandas as pd


def align_ff5_clock(ff5_df: pd.DataFrame, shift_months: int = 1) -> pd.DataFrame:
    ff5 = ff5_df.copy()
    ff5.columns = [column.strip() for column in ff5.columns]

    required = ["Mkt-RF", "SMB", "HML", "RMW", "CMA", "RF"]
    if ff5[required].abs().max().max() > 1:
        ff5[required] = ff5[required] / 100.0

    ff5.index = pd.to_datetime(ff5.index) + pd.DateOffset(months=shift_months)
    return ff5


class FF5Regression:
    def __init__(self, price_df: pd.DataFrame, ff5_df: pd.DataFrame):
        self.price_df = price_df.copy()
        self.ff5_df = ff5_df.copy()
        self.returns = None
        self.ff5_clean = None

    def compute_returns(self) -> pd.DataFrame:
        df = self.price_df.copy()

        if "Date" in df.columns:
            df["Date"] = pd.to_datetime(df["Date"])
            df = df.set_index("Date")

        df = df.select_dtypes(include=[np.number])
        returns = df.pct_change().iloc[1:]
        returns = returns.loc[:, returns.isna().mean() < 0.2]

        self.returns = returns
        return returns

    def prepare_ff5(self, shift_months: int = 1) -> pd.DataFrame:
        # Price data are stamped at the first day of each month, so the return
        # indexed by 2024-04-01 represents the realized move over March. Shift
        # FF5 monthly factors forward so they align with the month-start return
        # labels used throughout the backtest.
        ff5 = align_ff5_clock(self.ff5_df, shift_months=shift_months)
        self.ff5_clean = ff5
        return ff5

    def run_single_regression(self, ticker: str, ridge_alpha: float = 0.0) -> dict:
        if self.returns is None:
            self.compute_returns()
        if self.ff5_clean is None:
            self.prepare_ff5()

        common_idx = self.returns.index.intersection(self.ff5_clean.index)
        y_stock = self.returns.loc[common_idx, ticker]
        risk_free = self.ff5_clean.loc[common_idx, "RF"]
        y = (y_stock - risk_free).values

        factor_values = self.ff5_clean.loc[
            common_idx, ["Mkt-RF", "SMB", "HML", "RMW", "CMA"]
        ].values
        x_mat = np.column_stack((np.ones(len(factor_values)), factor_values))

        valid_mask = ~np.isnan(y)
        if valid_mask.sum() < 6:
            raise ValueError(f"[WARNING] {ticker} does not have enough valid observations.")

        x_valid = x_mat[valid_mask]
        y_valid = y[valid_mask]

        if ridge_alpha > 0:
            penalty = np.eye(x_valid.shape[1]) * ridge_alpha
            penalty[0, 0] = 0.0
            betas = np.linalg.solve(x_valid.T @ x_valid + penalty, x_valid.T @ y_valid)
        else:
            betas, _, _, _ = np.linalg.lstsq(x_valid, y_valid, rcond=None)

        return {
            "ticker": ticker,
            "alpha": betas[0],
            "beta_mkt": betas[1],
            "beta_smb": betas[2],
            "beta_hml": betas[3],
            "beta_rmw": betas[4],
            "beta_cma": betas[5],
        }

    def run_all_regression(self, ridge_alpha: float = 0.0) -> pd.DataFrame:
        if self.returns is None:
            self.compute_returns()
        if self.ff5_clean is None:
            self.prepare_ff5()

        common_idx = self.returns.index.intersection(self.ff5_clean.index)
        all_returns = self.returns.loc[common_idx]
        factor_values = self.ff5_clean.loc[
            common_idx, ["Mkt-RF", "SMB", "HML", "RMW", "CMA"]
        ]
        risk_free = self.ff5_clean.loc[common_idx, "RF"]

        excess_returns = all_returns.sub(risk_free, axis=0)
        x_mat = np.column_stack((np.ones(len(factor_values)), factor_values.values))

        penalty = np.eye(x_mat.shape[1]) * ridge_alpha
        penalty[0, 0] = 0.0

        results = []
        for ticker in excess_returns.columns:
            y = excess_returns[ticker].values
            valid_mask = ~np.isnan(y)
            if valid_mask.sum() < 6:
                continue

            x_valid = x_mat[valid_mask]
            y_valid = y[valid_mask]

            if ridge_alpha > 0:
                betas = np.linalg.solve(x_valid.T @ x_valid + penalty, x_valid.T @ y_valid)
            else:
                betas, _, _, _ = np.linalg.lstsq(x_valid, y_valid, rcond=None)

            results.append(
                {
                    "ticker": ticker,
                    "alpha": betas[0],
                    "beta_mkt": betas[1],
                    "beta_smb": betas[2],
                    "beta_hml": betas[3],
                    "beta_rmw": betas[4],
                    "beta_cma": betas[5],
                }
            )

        return pd.DataFrame(results)