Integrate factor attribution into backtest CLI

This commit is contained in:
2026-04-07 17:10:42 +08:00
parent f6670d9e6d
commit 9c4a219c68
3 changed files with 488 additions and 0 deletions

View File

@@ -1,4 +1,5 @@
import http.client
import contextlib
import io
import socket
import ssl
@@ -18,9 +19,12 @@ from factor_attribution import (
KEN_FRENCH_DAILY_FF5_ZIP_URL,
_download_kf_zip_bytes,
_parse_kf_daily_csv,
attribute_strategies,
build_extension_factors,
build_proxy_core_factors,
export_attribution,
load_external_us_factors,
print_attribution_summary,
prepare_factor_models,
run_factor_regression,
)
@@ -573,3 +577,194 @@ class RegressionTests(unittest.TestCase):
list(prepared["factor_frame"].columns),
["MKT", "SMB_PROXY", "HML_PROXY", "RMW_PROXY", "CMA_PROXY", "MOM", "LOWVOL", "RECOVERY"],
)
class AttributionIntegrationTests(unittest.TestCase):
def test_attribute_strategies_exports_standard_model_summary_and_loadings(self):
dates = pd.date_range("2025-01-01", periods=320, freq="B")
angles = np.linspace(0.0, 24.0, len(dates))
factors = pd.DataFrame(
{
"MKT_RF": 0.010 * np.sin(angles),
"SMB": 0.006 * np.cos(angles * 0.7),
"HML": 0.004 * np.sin(angles * 1.3 + 0.4),
"RMW": 0.003 * np.cos(angles * 1.1 + 0.2),
"CMA": 0.002 * np.sin(angles * 0.5 + 0.8),
"RF": np.full(len(dates), 0.0001),
},
index=dates,
)
strategy_returns = (
0.0004
+ 1.10 * factors["MKT_RF"]
- 0.25 * factors["SMB"]
+ 0.35 * factors["HML"]
+ 0.10 * factors["RMW"]
- 0.05 * factors["CMA"]
+ factors["RF"]
)
benchmark_returns = 0.95 * factors["MKT_RF"] + factors["RF"]
results = pd.DataFrame(
{
"Strategy": 100_000.0 * (1.0 + strategy_returns).cumprod(),
"SPY (Benchmark)": 100_000.0 * (1.0 + benchmark_returns).cumprod(),
},
index=dates,
)
prices = self._make_price_frame(dates, benchmark="SPY")
with tempfile.TemporaryDirectory() as tmpdir:
summary, loadings = attribute_strategies(
results_df=results,
benchmark_label="SPY (Benchmark)",
benchmark="SPY",
price_data=prices,
market="us",
model_selection="ff5",
external_factors=factors,
)
export_attribution(summary, loadings, tmpdir)
self.assertTrue((Path(tmpdir) / "summary.csv").exists())
self.assertTrue((Path(tmpdir) / "loadings.csv").exists())
exported_summary = pd.read_csv(Path(tmpdir) / "summary.csv")
exported_loadings = pd.read_csv(Path(tmpdir) / "loadings.csv")
self.assertEqual(len(summary), 1)
self.assertListEqual(
list(summary.columns),
[
"strategy",
"market",
"model",
"factor_source",
"proxy_only",
"start_date",
"end_date",
"n_obs",
"alpha_daily",
"alpha_ann",
"alpha_t_stat",
"alpha_p_value",
"r_squared",
"adj_r_squared",
"residual_vol_ann",
"beta_mkt",
"beta_smb",
"beta_hml",
"beta_rmw",
"beta_cma",
"beta_mom",
"beta_lowvol",
"beta_recovery",
],
)
self.assertEqual(summary.loc[0, "strategy"], "Strategy")
self.assertEqual(summary.loc[0, "model"], "ff5")
self.assertEqual(summary.loc[0, "factor_source"], "external+local")
self.assertFalse(bool(summary.loc[0, "proxy_only"]))
self.assertAlmostEqual(summary.loc[0, "beta_mkt"], 1.10, places=3)
self.assertAlmostEqual(summary.loc[0, "beta_smb"], -0.25, places=3)
self.assertAlmostEqual(summary.loc[0, "beta_hml"], 0.35, places=3)
self.assertTrue(np.isnan(summary.loc[0, "beta_mom"]))
self.assertListEqual(
list(loadings.columns),
["strategy", "market", "model", "factor_source", "proxy_only", "factor", "beta", "t_stat", "p_value"],
)
self.assertEqual(set(loadings["factor"]), {"MKT_RF", "SMB", "HML", "RMW", "CMA"})
self.assertEqual(len(loadings), 5)
pd.testing.assert_frame_equal(summary, exported_summary, check_dtype=False)
pd.testing.assert_frame_equal(loadings, exported_loadings, check_dtype=False)
def test_attribute_strategies_uses_proxy_model_for_cn_runs(self):
dates = pd.date_range("2025-01-01", periods=320, freq="B")
prices = self._make_price_frame(dates, benchmark="000300.SS")
returns = prices["000300.SS"].pct_change().fillna(0.0) * 0.7 + 0.0002
results = pd.DataFrame(
{
"Strategy": 100_000.0 * (1.0 + returns).cumprod(),
"CSI 300 (Benchmark)": 100_000.0 * (1.0 + prices["000300.SS"].pct_change().fillna(0.0)).cumprod(),
},
index=dates,
)
summary, loadings = attribute_strategies(
results_df=results,
benchmark_label="CSI 300 (Benchmark)",
benchmark="000300.SS",
price_data=prices,
market="cn",
model_selection="ff5",
external_factors=None,
)
self.assertEqual(len(summary), 1)
self.assertEqual(summary.loc[0, "model"], "proxy")
self.assertEqual(summary.loc[0, "factor_source"], "proxy_only")
self.assertTrue(bool(summary.loc[0, "proxy_only"]))
self.assertEqual(
set(loadings["factor"]),
{"MKT", "SMB_PROXY", "HML_PROXY", "RMW_PROXY", "CMA_PROXY", "MOM", "LOWVOL", "RECOVERY"},
)
def test_print_attribution_summary_prints_compact_table_and_interpretation(self):
summary = pd.DataFrame(
[
{
"strategy": "Strategy",
"market": "us",
"model": "ff5",
"factor_source": "external+local",
"proxy_only": False,
"start_date": "2025-01-02",
"end_date": "2026-03-24",
"n_obs": 319,
"alpha_daily": 0.0004,
"alpha_ann": 0.1008,
"alpha_t_stat": 2.1,
"alpha_p_value": 0.04,
"r_squared": 0.82,
"adj_r_squared": 0.81,
"residual_vol_ann": 0.12,
"beta_mkt": 1.05,
"beta_smb": -0.20,
"beta_hml": 0.30,
"beta_rmw": 0.05,
"beta_cma": np.nan,
"beta_mom": np.nan,
"beta_lowvol": np.nan,
"beta_recovery": np.nan,
}
]
)
buffer = io.StringIO()
with contextlib.redirect_stdout(buffer):
print_attribution_summary(summary)
output = buffer.getvalue()
self.assertIn("Factor attribution", output)
self.assertIn("Strategy", output)
self.assertIn("ff5", output)
self.assertIn("alpha_ann", output)
self.assertIn("Interpretation", output)
def _make_price_frame(self, dates: pd.DatetimeIndex, benchmark: str) -> pd.DataFrame:
steps = np.arange(len(dates), dtype=float)
data = {}
for symbol, base, drift, amplitude, frequency, phase in (
("AAA", 45.0, 0.0005, 0.030, 19.0, 0.1),
("BBB", 60.0, 0.0002, 0.025, 23.0, 0.8),
("CCC", 75.0, -0.0001, 0.035, 17.0, 1.4),
("DDD", 90.0, 0.0007, 0.020, 29.0, 0.5),
("EEE", 55.0, -0.0002, 0.028, 31.0, 1.9),
("FFF", 70.0, 0.0004, 0.032, 21.0, 2.5),
):
log_path = drift * steps + amplitude * np.sin(steps / frequency + phase)
data[symbol] = base * np.exp(log_path)
benchmark_path = 0.0004 * steps + 0.018 * np.sin(steps / 27.0 + 0.3)
data[benchmark] = 250.0 * np.exp(benchmark_path)
return pd.DataFrame(data, index=dates)