Haozhe Wu
d3c76819c3
Fix live option guards, grounding operand evidence, source-bound benchmarks, trade artifacts, attribution, drawdowns, valuation tolerances, and SEC point-in-time factor inputs. Add Brazil research data and market support, 13 fundamental and valuation factors, channel progress forwarding, and opt-in ZimaOS templates. Update seven README News sections with three visible dates. Validation: 17,635 backend tests passed; 1,084 final targeted tests and 103 frontend tests passed. Frontend build, package build, documentation checks, and CI gates passed.
121 lines
5.0 KiB
Python
121 lines
5.0 KiB
Python
"""Regression: after-close SEC filings must not leak into the IC one bar early.
|
|
|
|
SEC ``filed`` has no acceptance time. A filing on a trading date may be
|
|
published after its close. Same-day IC alignment can therefore include a
|
|
market reaction that preceded availability of the factor.
|
|
|
|
The known-answer check below plants a factor equal to the next day's return
|
|
(the strongest possible leak). Before the fix the bench reports a near-perfect
|
|
IC; with the filed-on-t value made first visible on bar t+1 the IC collapses
|
|
to ~0. These tests are offline: the loader is mocked, exactly like
|
|
``test_fundamental_factor_dependencies_are_loaded_at_filing_time``.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
from types import SimpleNamespace
|
|
from unittest.mock import Mock
|
|
|
|
import numpy as np
|
|
import pandas as pd
|
|
|
|
from src.factors.factor_analysis_core import compute_ic_series
|
|
from src.tools.alpha_bench_tool import _compute_forward_returns, _prepare_bench_panel
|
|
|
|
_DATES = pd.to_datetime(["2026-01-05", "2026-01-06", "2026-01-07", "2026-01-08", "2026-01-09", "2026-01-12"])
|
|
|
|
|
|
def _close_panel() -> pd.DataFrame:
|
|
"""Eight deterministic closes; cross-sectional IC needs >= 5 symbols.
|
|
|
|
The series is the seed-73 draw from ``default_rng(73).normal(0, 0.02)``
|
|
compounded from 100 — chosen so consecutive-day cross-sectional return
|
|
ranks are uncorrelated (|lag-IC| <= 0.12 every day), keeping the
|
|
known-answer test about the filing lag, not about the fixture.
|
|
"""
|
|
return pd.DataFrame(
|
|
{
|
|
"AAPL.US": [100.00, 97.81, 95.38, 98.83, 99.43, 96.99],
|
|
"MSFT.US": [100.00, 101.56, 103.82, 102.25, 99.69, 98.64],
|
|
"NVDA.US": [100.00, 99.24, 96.04, 92.59, 93.58, 91.74],
|
|
"AMZN.US": [100.00, 100.59, 97.40, 98.70, 99.25, 98.15],
|
|
"GOOG.US": [100.00, 100.67, 98.98, 99.41, 102.24, 104.99],
|
|
"META.US": [100.00, 96.11, 98.81, 99.49, 98.70, 100.09],
|
|
"TSLA.US": [100.00, 101.00, 102.87, 100.62, 101.13, 101.96],
|
|
"AMD.US": [100.00, 98.07, 97.74, 98.84, 100.68, 101.34],
|
|
},
|
|
index=_DATES,
|
|
)
|
|
|
|
|
|
def _leak_frame(close: pd.DataFrame) -> pd.DataFrame:
|
|
"""A fundamental whose value on bar t is exactly the t->t+1 return.
|
|
|
|
This is what a same-day ``filed`` value looks like when the filing is
|
|
accepted after the close: the value encodes information the market has
|
|
not priced in yet at close(t).
|
|
"""
|
|
return close.pct_change(fill_method=None).shift(-1)
|
|
|
|
|
|
def _registry(columns: list[str]) -> SimpleNamespace:
|
|
return SimpleNamespace(get=lambda _aid: SimpleNamespace(meta={"columns_required": columns}))
|
|
|
|
|
|
def _ic(factor: pd.DataFrame, close: pd.DataFrame) -> float:
|
|
"""Mean cross-sectional Spearman IC between the factor and forward returns."""
|
|
returns = _compute_forward_returns({"close": close})
|
|
ic_series = compute_ic_series(factor, returns)
|
|
assert not ic_series.empty
|
|
return float(ic_series.mean())
|
|
|
|
|
|
def test_unshifted_pit_panel_would_report_a_perfect_ic() -> None:
|
|
"""The known-answer sanity check: without the lag the leak is real.
|
|
|
|
Guards against someone removing the shift in ``_prepare_bench_panel``
|
|
without noticing it re-opens issue #1790.
|
|
"""
|
|
close = _close_panel()
|
|
|
|
ic = _ic(_leak_frame(close), close)
|
|
|
|
assert abs(ic) > 0.9
|
|
|
|
|
|
def test_pit_panel_shifts_filing_one_bar_later(monkeypatch) -> None:
|
|
"""A filed-on-t value must first appear on bar t+1 in the bench panel."""
|
|
close = _close_panel()
|
|
# What the loader hands back for a filing anchored on bar 2: NaN before
|
|
# the filing, forward-filled after it (fundamentals_loader L194 ffills).
|
|
frame = pd.DataFrame(np.nan, index=close.index, columns=close.columns)
|
|
frame.iloc[2] = 5.0
|
|
frame = frame.ffill()
|
|
loader = Mock(return_value={"net_income": frame})
|
|
monkeypatch.setattr("backtest.loaders.fundamentals_loader.load_fundamental_panel", loader)
|
|
|
|
result = _prepare_bench_panel(_registry(["close", "fund:net_income"]), ["aid"], {"close": close}, "sp500", "2026-2026")
|
|
|
|
column = result["fund:net_income"]
|
|
# Before the filing bar and on it: NaN. From t+1 on: the filed value.
|
|
assert column.iloc[2].isna().all()
|
|
assert (column.iloc[3:] == 5.0).all().all()
|
|
# The lag is recorded so downstream consumers can see the semantics.
|
|
assert result["_meta"]["fundamental_pit_lag_bars"] == 1
|
|
|
|
|
|
def test_after_close_filing_does_not_leak_into_the_ic(monkeypatch) -> None:
|
|
"""Known-answer: a factor equal to next-day returns must not produce IC.
|
|
|
|
This is the reproduce step from the issue, planted offline: the loader
|
|
returns a value anchored on the filed date (as for an after-close
|
|
filing); after the fix the bench must not report a signal.
|
|
"""
|
|
close = _close_panel()
|
|
loader = Mock(return_value={"net_income": _leak_frame(close)})
|
|
monkeypatch.setattr("backtest.loaders.fundamentals_loader.load_fundamental_panel", loader)
|
|
|
|
result = _prepare_bench_panel(_registry(["close", "fund:net_income"]), ["aid"], {"close": close}, "sp500", "2026-2026")
|
|
|
|
assert abs(_ic(result["fund:net_income"], close)) < 0.1
|