Skip to content

"Vibe-Trading: Your Personal Trading Agent"

feat(quantlib): add Corrado rank and Cowan sign non-parametric event study tests

santhreal10759fb28b23
diff --git a/agent/src/quantlib/eventstudy.py b/agent/src/quantlib/eventstudy.py
index a32d61513..afac3dfbe 100644
--- a/agent/src/quantlib/eventstudy.py
+++ b/agent/src/quantlib/eventstudy.py
@@ -58,8 +58,7 @@ from dataclasses import dataclass
import numpy as np
import pandas as pd
-from scipy.stats import norm, t as student_t
-
+from scipy.stats import norm, rankdata, t as student_t
__all__ = [
"DEFAULT_ESTIMATION_GAP",
"DEFAULT_ESTIMATION_WINDOW",
@@ -173,8 +172,12 @@ class EventStudyResult:
shared_event_dates: Event dates carried by more than one event. Non-empty
means the events overlap in calendar time, the independence every
statistic here assumes is violated, and the p-values are optimistic.
+ corrado_rank_z: Corrado non-parametric rank test z-statistic.
+ corrado_rank_p_value: Two-sided p-value of ``corrado_rank_z``.
+ cowan_sign_z: Cowan generalized sign test z-statistic.
+ cowan_sign_p_value: Two-sided p-value of ``cowan_sign_z``.
+ positive_car_fraction: Fraction of events with positive cumulative abnormal return.
"""
-
event_window: tuple[int, int]
events: tuple[EventOutcome, ...]
dropped: tuple[tuple[str, object, str], ...]
@@ -188,6 +191,11 @@ class EventStudyResult:
bmp_z: float
bmp_p_value: float
shared_event_dates: tuple[object, ...]
+ corrado_rank_z: float
+ corrado_rank_p_value: float
+ cowan_sign_z: float
+ cowan_sign_p_value: float
+ positive_car_fraction: float
def estimate_market_model(
@@ -411,15 +419,23 @@ def event_study(
car_se = float(np.sqrt(np.sum(daily_se**2)))
standardised = car / car_se if car_se > 0 else float("nan")
+ est_residuals = asset_estimation.to_numpy(dtype=float) - (
+ fit.alpha + fit.beta * market_estimation.to_numpy(dtype=float)
+ )
+
outcomes.append(
- EventOutcome(
- symbol=symbol,
- event_date=index[position],
- abnormal_returns=pd.Series(abnormal, index=relative_days, name=symbol),
- car=car,
- car_std_error=car_se,
- standardised_car=standardised,
- fit=fit,
+ (
+ EventOutcome(
+ symbol=symbol,
+ event_date=index[position],
+ abnormal_returns=pd.Series(abnormal, index=relative_days, name=symbol),
+ car=car,
+ car_std_error=car_se,
+ standardised_car=standardised,
+ fit=fit,
+ ),
+ est_residuals,
+ abnormal,
)
)
@@ -430,11 +446,12 @@ def event_study(
)
n_events = len(outcomes)
- ar_matrix = np.vstack([o.abnormal_returns.to_numpy() for o in outcomes])
+ event_outcomes = [o[0] for o in outcomes]
+ ar_matrix = np.vstack([o.abnormal_returns.to_numpy() for o in event_outcomes])
aar = pd.Series(ar_matrix.mean(axis=0), index=relative_days, name="aar")
caar = pd.Series(np.cumsum(aar.to_numpy()), index=relative_days, name="caar")
- cars = np.array([o.car for o in outcomes])
+ cars = np.array([o.car for o in event_outcomes])
if n_events > 1:
car_sd = float(cars.std(ddof=1))
t_stat = float(cars.mean() / (car_sd / np.sqrt(n_events))) if car_sd > 0 else float("nan")
@@ -442,7 +459,7 @@ def event_study(
else:
t_stat, t_p = float("nan"), float("nan")
- standardised = np.array([o.standardised_car for o in outcomes])
+ standardised = np.array([o.standardised_car for o in event_outcomes])
finite_scar = standardised[np.isfinite(standardised)]
# Patell: the standardised CARs are unit-variance under the null, up to the
@@ -453,7 +470,7 @@ def event_study(
(o.fit.observations - 2) / (o.fit.observations - 4)
if o.fit.observations > 4
else np.nan
- for o in outcomes
+ for o in event_outcomes
if np.isfinite(o.standardised_car)
]
)
@@ -476,14 +493,61 @@ def event_study(
bmp_z = float("nan")
bmp_p = float(2 * norm.sf(abs(bmp_z))) if np.isfinite(bmp_z) else float("nan")
+ # Cowan Generalized Sign Test: compares the proportion of positive CARs
+ # against the baseline proportion of positive abnormal returns in the estimation window.
+ est_pos_fractions = [
+ float(np.mean(est_res > 0)) for _, est_res, _ in outcomes if len(est_res) > 0
+ ]
+ base_p = float(np.mean(est_pos_fractions)) if est_pos_fractions else 0.5
+ pos_cars_count = int(np.sum(cars > 0))
+ pos_car_frac = float(pos_cars_count / n_events) if n_events > 0 else 0.0
+
+ if n_events > 1 and 0.0 < base_p < 1.0:
+ denom = np.sqrt(n_events * base_p * (1.0 - base_p))
+ cowan_z = float((pos_cars_count - n_events * base_p) / denom)
+ cowan_p = float(2 * norm.sf(abs(cowan_z)))
+ else:
+ cowan_z, cowan_p = float("nan"), float("nan")
+
+ # Corrado Rank Test: non-parametric rank test across combined estimation + event period
+ L2 = end - start + 1
+ rank_dev_est_list = []
+ rank_dev_evt_list = []
+ for _, est_res, evt_abn in outcomes:
+ combined = np.concatenate([est_res, evt_abn])
+ ranks = rankdata(combined)
+ M = len(combined)
+ mean_rank = (M + 1.0) / 2.0
+ ranks_est = ranks[: len(est_res)] - mean_rank
+ ranks_evt = ranks[len(est_res) :] - mean_rank
+ rank_dev_est_list.append(ranks_est)
+ rank_dev_evt_list.append(ranks_evt)
+
+ if n_events > 1 and all(len(r) == len(rank_dev_est_list[0]) for r in rank_dev_est_list):
+ est_rank_matrix = np.vstack(rank_dev_est_list) # (n_events, T_est)
+ mean_est_ranks = est_rank_matrix.mean(axis=0) # mean rank dev per estimation day
+ s_k = float(np.sqrt(np.mean(mean_est_ranks**2)))
+
+ evt_rank_matrix = np.vstack(rank_dev_evt_list) # (n_events, L2)
+ mean_evt_ranks = evt_rank_matrix.mean(axis=0) # mean rank dev per event day
+ cum_evt_rank_dev = float(np.sum(mean_evt_ranks))
+
+ if s_k > 0:
+ corrado_z = float(cum_evt_rank_dev / (np.sqrt(L2) * s_k))
+ corrado_p = float(2 * norm.sf(abs(corrado_z)))
+ else:
+ corrado_z, corrado_p = float("nan"), float("nan")
+ else:
+ corrado_z, corrado_p = float("nan"), float("nan")
+
seen: dict[object, int] = {}
- for outcome in outcomes:
+ for outcome in event_outcomes:
seen[outcome.event_date] = seen.get(outcome.event_date, 0) + 1
shared = tuple(sorted((d for d, c in seen.items() if c > 1), key=str))
return EventStudyResult(
event_window=(start, end),
- events=tuple(outcomes),
+ events=tuple(event_outcomes),
dropped=tuple(dropped),
aar=aar,
caar=caar,
@@ -494,5 +558,11 @@ def event_study(
patell_p_value=patell_p,
bmp_z=bmp_z,
bmp_p_value=bmp_p,
+ corrado_rank_z=corrado_z,
+ corrado_rank_p_value=corrado_p,
+ cowan_sign_z=cowan_z,
+ cowan_sign_p_value=cowan_p,
+ positive_car_fraction=pos_car_frac,
shared_event_dates=shared,
)
+
diff --git a/agent/tests/quantlib/test_eventstudy.py b/agent/tests/quantlib/test_eventstudy.py
index e7488b6db..c0e7e50c8 100644
--- a/agent/tests/quantlib/test_eventstudy.py
+++ b/agent/tests/quantlib/test_eventstudy.py
@@ -381,3 +381,36 @@ def test_post_event_only_window_is_allowed():
result = event_study(returns, market, events, event_window=(1, 5))
assert result.event_window == (1, 5)
assert list(result.aar.index) == [1, 2, 3, 4, 5]
+
+# --- non-parametric tests (Corrado rank and Cowan sign) ---
+
+
+def test_corrado_rank_and_cowan_sign_detect_positive_reaction():
+ returns, market = _panel(seed=801)
+ event_day = returns.index[250]
+ # Inject positive abnormal returns in the event window for all assets
+ returns_injected = returns.copy()
+ for col in returns_injected.columns:
+ returns_injected.loc[event_day, col] += 0.05
+
+ events = [(s, event_day) for s in returns.columns]
+ result = event_study(returns_injected, market, events, event_window=(0, 2))
+
+ assert result.positive_car_fraction == pytest.approx(1.0)
+ assert result.cowan_sign_z > 2.0
+ assert result.cowan_sign_p_value < 0.05
+ assert result.corrado_rank_z > 2.0
+ assert result.corrado_rank_p_value < 0.05
+
+
+def test_corrado_rank_and_cowan_sign_single_event_safe_nan():
+ returns, market = _panel(seed=802)
+ event_day = returns.index[250]
+ events = [("S00", event_day)]
+ result = event_study(returns, market, events, event_window=(0, 2))
+
+ assert np.isnan(result.corrado_rank_z)
+ assert np.isnan(result.corrado_rank_p_value)
+ assert np.isnan(result.cowan_sign_z)
+ assert np.isnan(result.cowan_sign_p_value)
+ assert 0.0 <= result.positive_car_fraction <= 1.0