Files
stack/tests/market/test_abnormal.py
kert 82246dcc58
All checks were successful
CI / lint (push) Successful in 31s
CI / test (push) Successful in 4m1s
CI / notebooks-smoke (push) Successful in 1m25s
Deploy / changes (push) Successful in 13s
Infra CI / notebooks (push) Successful in 1m6s
Infra CI / zotero (push) Successful in 16s
Infra CI / docs (push) Successful in 1m16s
Infra CI / api (push) Successful in 1m13s
Infra CI / bcda (push) Successful in 2m23s
Infra CI / mc (push) Successful in 15s
Deploy / notebooks (push) Has been skipped
Deploy / zotero (push) Has been skipped
Deploy / docs (push) Has been skipped
Infra CI / llm (push) Successful in 57s
Deploy / api (push) Successful in 1m17s
Deploy / bcda (push) Successful in 1m48s
Deploy / mc (push) Has been skipped
Deploy / llm (push) Successful in 1m25s
Deploy / report (push) Successful in 13s
Renovate / renovate (push) Successful in 2m2s
LLM Golden / llm-golden (push) Successful in 23m11s
feat(market): market-model abnormal returns with CLM intervals; shared bootstrap helpers in pfs.event_study
2026-11-16 22:27:37 -05:00

225 lines
7.3 KiB
Python

"""market.abnormal — market-model CARs, CLM intervals, pooled bootstrap."""
from __future__ import annotations
from datetime import date, timedelta
import pytest
from market.abnormal import Car, car_from_arrays, event_car, fit, pooled
_ASSOC = "association, not a causal estimate"
def _weekdays(d0: date, n_before: int, n_after: int) -> list[date]:
back: list[date] = []
day = d0
while len(back) < n_before:
day -= timedelta(days=1)
if day.weekday() < 5:
back.append(day)
fwd: list[date] = []
day = d0
while len(fwd) < n_after:
day += timedelta(days=1)
if day.weekday() < 5:
fwd.append(day)
return [*reversed(back), d0, *fwd]
def _market(cal: list[date]) -> tuple[dict[date, float], dict[str, dict[date, float]]]:
"""Firm equals SPY. SPY varies, so the market model is identified."""
spy: dict[date, float] = {}
firm: dict[date, float] = {}
for i, day in enumerate(cal):
spy[day] = ((i % 5) - 2) * 0.01
firm[day] = spy[day]
return firm, {"SPY": spy}
def _car(value: float | None, **kw: object) -> Car:
return Car(
event_id="e",
unit=kw.get("unit", "U"),
ticker="T",
model=kw.get("model", "market"),
window=kw.get("window", (-1, 1)),
car=value,
se=None if value is None else 0.01,
ci_low=None if value is None else value - 0.02,
ci_high=None if value is None else value + 0.02,
n_est=200,
n_members=1,
caveats=(_ASSOC,),
)
def test_fit_recovers_alpha_beta():
x = [i - 65 for i in range(130)]
y = [0.001 + 1.5 * xi for xi in x]
fitted = fit(y, x)
assert fitted.alpha == pytest.approx(0.001, abs=1e-12)
assert fitted.betas[0] == pytest.approx(1.5, abs=1e-12)
assert fitted.sigma2 == pytest.approx(0.0, abs=1e-12)
assert fit([1.0, 2.0], [1.0, 2.0]) is None
def test_car_variance_clm():
fitted = fit([-0.9, -0.2, 1.1], [-1.0, 0.0, 1.0])
assert fitted.alpha == pytest.approx(0.0, abs=1e-9)
assert fitted.betas[0] == pytest.approx(1.0, abs=1e-9)
assert fitted.sigma2 == pytest.approx(0.06)
car, se = car_from_arrays(fitted, [1.5, 2.0], [1.0, 2.0])
assert car == pytest.approx(0.5)
assert se == pytest.approx(0.6855654600, abs=1e-9)
d0 = date(2026, 1, 8)
cal = [
date(2026, 1, 5),
date(2026, 1, 6),
date(2026, 1, 7),
d0,
date(2026, 1, 9),
date(2026, 1, 10),
]
xs = {
date(2026, 1, 5): -1.0,
date(2026, 1, 6): 0.0,
date(2026, 1, 7): 1.0,
date(2026, 1, 9): 1.0,
date(2026, 1, 10): 2.0,
}
ys = {
date(2026, 1, 5): -0.9,
date(2026, 1, 6): -0.2,
date(2026, 1, 7): 1.1,
date(2026, 1, 9): 1.5,
date(2026, 1, 10): 2.0,
}
out = event_car(
ys,
{"SPY": xs},
cal,
d0,
est=(-3, -1),
window=(1, 2),
min_est=3,
)
assert out.car == pytest.approx(0.5)
assert out.se == pytest.approx(0.6855654600, abs=1e-9)
assert out.ci_low == pytest.approx(0.5 - 1.96 * 0.6855654600)
assert out.ci_high == pytest.approx(0.5 + 1.96 * 0.6855654600)
assert out.n_est == 3
assert _ASSOC in out.caveats
def test_car_window_sum():
d0 = date(2026, 11, 9)
cal = _weekdays(d0, 250, 2)
firm, factors = _market(cal)
firm[d0] = factors["SPY"][d0] + 0.02
day = event_car(firm, factors, cal, d0, window=(0, 0))
three = event_car(firm, factors, cal, d0, window=(-1, 1))
assert day.car == pytest.approx(0.02)
assert three.car == pytest.approx(0.02)
assert _ASSOC in day.caveats and _ASSOC in three.caveats
def test_short_estimation_window_caveat():
d0 = date(2026, 11, 9)
# 101 trading days fall in the default window; one has no firm return.
cal = _weekdays(d0, 131, 2)
firm, factors = _market(cal)
del firm[cal[cal.index(d0) - 131]]
out = event_car(firm, factors, cal, d0)
assert out.car is None and out.n_est == 100
assert "short estimation window (n=100)" in out.caveats
assert _ASSOC in out.caveats
def test_missing_window_return():
d0 = date(2026, 11, 9)
cal = _weekdays(d0, 250, 2)
firm, factors = _market(cal)
del firm[date(2026, 11, 10)]
out = event_car(firm, factors, cal, d0, window=(-1, 1))
assert out.car is None
assert "missing returns in event window" in out.caveats
assert _ASSOC in out.caveats
def test_offsets_count_trading_days():
d0 = date(2026, 11, 9)
cal = _weekdays(d0, 250, 3)
firm, factors = _market(cal)
shocks = {
date(2026, 11, 5): 0.50, # trading offset -2, outside (-1, 1)
date(2026, 11, 6): 0.01, # Friday
date(2026, 11, 9): 0.02,
date(2026, 11, 10): 0.04, # Tuesday
date(2026, 11, 11): 0.50, # trading offset +2, outside
}
for day, shock in shocks.items():
firm[day] = factors["SPY"][day] + shock
out = event_car(firm, factors, cal, d0, window=(-1, 1))
assert out.window == (-1, 1)
assert out.car == pytest.approx(0.07)
assert _ASSOC in out.caveats
def test_two_factor_and_sector_models():
x1 = [((i % 5) - 2) * 0.01 for i in range(130)]
x2 = [((i % 7) - 3) * 0.01 for i in range(130)]
y = [0.2 * a + 0.3 * b for a, b in zip(x1, x2, strict=True)]
assert len(fit(y, [x1, x2]).betas) == 2
assert len(fit(x2, x2).betas) == 1
d0 = date(2026, 11, 9)
cal = _weekdays(d0, 250, 2)
spy: dict[date, float] = {}
xlv: dict[date, float] = {}
firm: dict[date, float] = {}
for i, day in enumerate(cal):
spy[day] = ((i % 5) - 2) * 0.01
xlv[day] = ((i % 7) - 3) * 0.01
firm[day] = 0.2 * spy[day] + 0.3 * xlv[day]
factors = {"SPY": spy, "XLV": xlv}
sector = event_car(firm, factors, cal, d0, model="sector")
two = event_car(firm, factors, cal, d0, model="two-factor")
note = "sector index absorbs sector-wide effects"
assert note in sector.caveats and note in two.caveats
assert sector.model == "sector" and two.model == "two-factor"
assert _ASSOC in sector.caveats and _ASSOC in two.caveats
def test_estimation_overlap_caveat():
d0 = date(2026, 11, 9)
cal = _weekdays(d0, 250, 2)
inside = cal[cal.index(d0) - 40]
outside = cal[cal.index(d0) - 10]
assert inside == date(2026, 9, 14)
assert outside == date(2026, 10, 26)
firm, factors = _market(cal)
out = event_car(firm, factors, cal, d0, other_events=(inside, outside))
assert f"estimation window overlaps event on {inside}" in out.caveats
assert f"estimation window overlaps event on {outside}" not in out.caveats
assert out.car is not None
assert _ASSOC in out.caveats
def test_pooled_bootstrap_deterministic():
cars = [_car(0.01), _car(0.03), _car(-0.02), _car(None)]
first = pooled(cars)
second = pooled(cars)
assert (first.ci_low, first.ci_high) == (second.ci_low, second.ci_high)
assert first.ci_low is not None and first.ci_high is not None
assert first.mean_car == pytest.approx((0.01 + 0.03 - 0.02) / 3)
assert first.n_events == 3
assert first.unit == "U" and first.model == "market"
assert "fewer than 5 events" in first.caveats
assert _ASSOC in first.caveats
one = pooled([_car(0.05)])
assert one.ci_low is None and one.ci_high is None
assert one.mean_car == pytest.approx(0.05)
assert one.n_events == 1