Reproduire l’analyse exploratoire de research.ipynb avec les données natives QuantConnect.
# Setup QuantBook
from AlgorithmImports import *
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import warnings
warnings.filterwarnings('ignore')
plt.style.use('seaborn-v0_8-darkgrid')
plt.rcParams['figure.figsize'] = (14, 6)
qb = QuantBook()
print("QuantBook initialise.")QuantBook initialise.Univers Antonacci: SPY (US equities), EFA (International), BND (US Aggregate Bonds). Candidats alternatifs: EEM (Emerging), IEF (7-10Y Treasuries), SHY (1-3Y), GLD (Gold), VNQ (REITs).
tickers = ['SPY', 'EFA', 'BND', 'EEM', 'IEF', 'SHY', 'GLD', 'VNQ']
symbols = {}
for ticker in tickers:
symbols[ticker] = qb.add_equity(ticker, Resolution.DAILY).symbol
start = datetime(2007, 1, 1)
end = datetime(2026, 1, 1)
history = qb.history(list(symbols.values()), start, end, Resolution.DAILY)
closes = history['close'].unstack(level=0)
symbol_to_ticker = {str(v): k for k, v in symbols.items()}
closes.columns = [symbol_to_ticker.get(str(c), str(c)) for c in closes.columns]
closes = closes.dropna()
print(f"Periode: {closes.index[0].date()} a {closes.index[-1].date()}")
print(f"Donnees: {len(closes)} jours de trading")Periode: 2007-04-10 a 2025-12-31
Donnees: 4714 jours de tradingdef backtest_dual_momentum(closes, risky_assets, refuge, lookback=252,
abs_threshold=0.0, rebal_freq='MS'):
"""Backtest Dual Momentum: relative + absolute.
- Choisit le meilleur actif risque sur lookback period
- Si son rendement > abs_threshold, investit 100% dedans
- Sinon, investit 100% dans le refuge
"""
all_assets = risky_assets + [refuge]
returns_df = closes[all_assets].pct_change()
# Momentum scores
mom = {}
for asset in all_assets:
mom[asset] = closes[asset] / closes[asset].shift(lookback) - 1
# first ACTUAL trading day of each month (month-start labels are not in
# closes.index -> would skip every rebalance -> 0% returns; fix #6891)
monthly = pd.Series(closes.index, index=closes.index).resample(rebal_freq).first().dropna().values
positions = pd.Series(index=closes.index, dtype=str)
current_pos = refuge
for date in monthly:
if date not in closes.index:
continue
if any(pd.isna(mom[a].get(date, np.nan)) for a in risky_assets):
continue
# Relative momentum: best risky asset
risky_scores = {a: mom[a][date] for a in risky_assets}
best_risky = max(risky_scores, key=risky_scores.get)
# Absolute momentum: is it positive enough?
if risky_scores[best_risky] > abs_threshold:
current_pos = best_risky
else:
current_pos = refuge
positions.loc[date:] = current_pos
# Calculer les rendements
port_ret = pd.Series(0.0, index=closes.index)
for asset in all_assets:
mask = positions == asset
port_ret[mask] = returns_df[asset][mask]
port_ret = port_ret[port_ret.index >= monthly[0]]
return port_ret
def calc_stats(returns, name=''):
total = (1 + returns).cumprod().iloc[-1] - 1
years = len(returns) / 252
cagr = (1 + total) ** (1 / years) - 1
vol = returns.std() * np.sqrt(252)
sharpe = (cagr - 0.03) / vol if vol > 0 else 0
cum = (1 + returns).cumprod()
dd = (cum / cum.cummax() - 1).min()
return {'name': name, 'CAGR': f'{cagr:.1%}', 'Vol': f'{vol:.1%}',
'Sharpe': f'{sharpe:.3f}', 'MaxDD': f'{dd:.1%}'}
print("Fonctions definies.")Fonctions definies.Source des donnees et reproductibilite. Les 8 ETF (SPY/EFA/BND/EEM/IEF/SHY/GLD/VNQ) sont servis localement au format LEAN daily via
DefaultDataProvider: donnees OHLCV journalieres converties depuis yfinance (source gratuite, sans cle API). Cette voie evite le Security Master payant –ApiDataProvider(telechargement cloud) exige un abonnement map/factor que l’organisation n’a pas, tandis queDefaultDataProviderlit directement le cache local. Les cours sont bruts (non ajustes des dividendes) : bénin pour un signal de momentum sur prix, documente ici en toute transparence.Note de debug (#6891). La re-execution honnete a revele un bug d’alignement mensuel latent dans
backtest_dual_momentum(les debuts de mois renvoyes parresample('MS')ne sont pas dans l’index des jours de bourse -> chaque rebalancement etait saute -> rendements systematiquement nuls). Le bug, masque jusque-la par les sorties fabriquees, est corrige dans la cellule precedente (premier jour de bourse reel de chaque mois). Voir #6891, #7265.
12 mois est le lookback canonique d’Antonacci. Tester la robustesse.
results_lb = []
for lb_months, lb_days in [(6, 126), (9, 189), (12, 252), (18, 378)]:
ret = backtest_dual_momentum(closes, ['SPY', 'EFA'], 'BND', lookback=lb_days)
results_lb.append(calc_stats(ret, f'{lb_months}m ({lb_days}d)'))
spy_ret = closes['SPY'].pct_change().dropna()
results_lb.append(calc_stats(spy_ret, 'SPY B&H'))
print("=== H1: Lookback ===")
print(pd.DataFrame(results_lb).set_index('name').to_string())=== H1: Lookback ===
CAGR Vol Sharpe MaxDD
name
6m (126d) 4.8% 14.5% 0.121 -35.9%
9m (189d) 5.5% 14.6% 0.170 -34.1%
12m (252d) 6.2% 14.9% 0.213 -34.1%
18m (378d) 7.6% 15.3% 0.303 -34.1%
SPY B&H 8.6% 19.9% 0.284 -56.5%Le lookback 12m est standard dans la litterature. Le COVID crash (Mars 2020) expose la latence du signal mensuel quel que soit le lookback.
Seuil de basculement risk-on/risk-off. Actuellement 0% (proxy T-bills).
results_thresh = []
for thresh in [-0.05, -0.02, 0.0, 0.02, 0.05]:
ret = backtest_dual_momentum(closes, ['SPY', 'EFA'], 'BND', abs_threshold=thresh)
results_thresh.append(calc_stats(ret, f'Seuil {thresh:+.0%}'))
print("=== H2: Absolute momentum threshold ===")
print(pd.DataFrame(results_thresh).set_index('name').to_string())=== H2: Absolute momentum threshold ===
CAGR Vol Sharpe MaxDD
name
Seuil -5% 7.0% 15.9% 0.253 -34.1%
Seuil -2% 6.6% 15.1% 0.239 -34.1%
Seuil +0% 6.2% 14.9% 0.213 -34.1%
Seuil +2% 4.6% 14.7% 0.108 -37.6%
Seuil +5% 4.6% 14.3% 0.114 -34.8%Un seuil positif (+2%, +5%) augmente le temps en refuge (plus conservateur). Un seuil negatif (-2%, -5%) maintient l’exposition plus longtemps.
BND vs IEF vs SHY vs GLD comme asset de refuge. Règle #3 du backlog: TLT risk-off detruit la valeur 2015-2026.
results_refuge = []
for refuge in ['BND', 'IEF', 'SHY', 'GLD']:
ret = backtest_dual_momentum(closes, ['SPY', 'EFA'], refuge)
results_refuge.append(calc_stats(ret, f'Refuge: {refuge}'))
print("=== H3: Refuge alternatives ===")
print(pd.DataFrame(results_refuge).set_index('name').to_string())=== H3: Refuge alternatives ===
CAGR Vol Sharpe MaxDD
name
Refuge: BND 6.2% 14.9% 0.213 -34.1%
Refuge: IEF 6.1% 15.1% 0.204 -34.1%
Refuge: SHY 5.8% 14.6% 0.192 -34.1%
Refuge: GLD 9.1% 17.5% 0.348 -34.1%research.ipynb (yfinance) montrait IEF/SHY legerement meilleur que BND, mais le QC cloud a rejete ce changement (Sharpe 0.350->0.324 avec SHY). Verifier avec les prix QC bruts.
Ajouter EEM (Emerging Markets) et VNQ (REITs) a l’univers risque.
results_univ = []
universes = {
'SPY+EFA (actuel)': ['SPY', 'EFA'],
'SPY+EFA+EEM': ['SPY', 'EFA', 'EEM'],
'SPY+EFA+VNQ': ['SPY', 'EFA', 'VNQ'],
'SPY+EFA+EEM+VNQ': ['SPY', 'EFA', 'EEM', 'VNQ'],
}
for name, risky in universes.items():
ret = backtest_dual_momentum(closes, risky, 'BND')
results_univ.append(calc_stats(ret, name))
print("=== H4: Univers elargi ===")
print(pd.DataFrame(results_univ).set_index('name').to_string())=== H4: Univers elargi ===
CAGR Vol Sharpe MaxDD
name
SPY+EFA (actuel) 6.2% 14.9% 0.213 -34.1%
SPY+EFA+EEM 5.3% 16.4% 0.143 -34.8%
SPY+EFA+VNQ 8.2% 16.0% 0.322 -34.1%
SPY+EFA+EEM+VNQ 6.7% 17.0% 0.219 -35.4%EEM et VNQ ajoutent de la diversification mais peuvent diluer le signal. Verifier si le gain en diversification compense la rotation accrue.
ret_base = backtest_dual_momentum(closes, ['SPY', 'EFA'], 'BND')
regimes = {
'GFC (2008-2009)': ('2008-01-01', '2009-12-31'),
'Bull (2010-2019)': ('2010-01-01', '2019-12-31'),
'COVID (2020)': ('2020-01-01', '2020-12-31'),
'Rate Hikes (2022)': ('2022-01-01', '2022-12-31'),
'Recovery (2023-2025)': ('2023-01-01', '2025-12-31'),
}
print("=== Performance par regime ===")
regime_stats = []
for regime, (s, e) in regimes.items():
mask = (ret_base.index >= s) & (ret_base.index <= e)
if mask.sum() > 50:
regime_stats.append(calc_stats(ret_base[mask], regime))
print(pd.DataFrame(regime_stats).set_index('name').to_string())=== Performance par regime ===
CAGR Vol Sharpe MaxDD
name
GFC (2008-2009) 3.4% 9.3% 0.048 -9.8%
Bull (2010-2019) 6.3% 14.0% 0.237 -25.3%
COVID (2020) 0.8% 30.9% -0.070 -34.1%
Rate Hikes (2022) -18.3% 14.5% -1.468 -21.5%
Recovery (2023-2025) 14.3% 15.1% 0.748 -19.0%fig, axes = plt.subplots(2, 2, figsize=(16, 10))
# H1: Lookback
ax = axes[0, 0]
for lb_m, lb_d in [(6, 126), (9, 189), (12, 252), (18, 378)]:
ret = backtest_dual_momentum(closes, ['SPY', 'EFA'], 'BND', lookback=lb_d)
cum = (1 + ret).cumprod()
ax.plot(cum.index, cum, label=f'{lb_m}m', linewidth=1.5)
ax.set_title('H1: Lookback period', fontweight='bold')
ax.legend(fontsize=8)
ax.grid(True, alpha=0.3)
# H2: Threshold
ax = axes[0, 1]
for thresh in [-0.05, 0.0, 0.05]:
ret = backtest_dual_momentum(closes, ['SPY', 'EFA'], 'BND', abs_threshold=thresh)
cum = (1 + ret).cumprod()
ax.plot(cum.index, cum, label=f'Seuil {thresh:+.0%}', linewidth=1.5)
ax.set_title('H2: Abs. momentum threshold', fontweight='bold')
ax.legend(fontsize=8)
ax.grid(True, alpha=0.3)
# H3: Refuge
ax = axes[1, 0]
for refuge in ['BND', 'IEF', 'SHY', 'GLD']:
ret = backtest_dual_momentum(closes, ['SPY', 'EFA'], refuge)
cum = (1 + ret).cumprod()
ax.plot(cum.index, cum, label=f'Refuge: {refuge}', linewidth=1.5)
ax.set_title('H3: Refuge alternatives', fontweight='bold')
ax.legend(fontsize=8)
ax.grid(True, alpha=0.3)
# H4: Universe
ax = axes[1, 1]
for name, risky in universes.items():
ret = backtest_dual_momentum(closes, risky, 'BND')
cum = (1 + ret).cumprod()
ax.plot(cum.index, cum, label=name, linewidth=1.5)
spy_cum = (1 + spy_ret).cumprod()
ax.plot(spy_cum.index, spy_cum, label='SPY B&H', linestyle='--', alpha=0.5)
ax.set_title('H4: Univers elargi', fontweight='bold')
ax.legend(fontsize=7)
ax.grid(True, alpha=0.3)
plt.tight_layout()
plt.savefig('dualmomentum_quantbook_analysis.png', dpi=150, bbox_inches='tight')
plt.show()
Lecture des 4 panneaux. Chaque panneau trace la richesse cumulee
(1+r).cumprod()de la strategie Dual Momentum sur 2007-2025. H1 (lookback) : le rendement croit avec la fenetre (6m -> 18m), signe d’un momentum plus stable sur periode longue. H2 (seuil absolu) : un seuil plus permissif (-5%) maintient une exposition au marche plus longtemps -> rendement legerement superieur. H3 (refuge) : GLD offre le meilleur couple rendement/MaxDD sur la periode (or 2007-2025), BND/IEF/SHY proches. H4 (univers) : ajouter VNQ ameliore le rendement (8.2%), EEM le penalise. En pointille contre SPY B&H (pointille) : la strategie reduit le drawdown max (~-34% vs ~-56%) au prix d’un rendement legerement inferieur – l’arbitrage attendu d’une strategie de refuge dynamique.
| Hypothese | Résultat QuantBook | Coherent avec yfinance? |
|---|---|---|
| H1 Lookback | (a remplir) | (a verifier) |
| H2 Threshold | (a remplir) | (a verifier) |
| H3 Refuge | (a remplir) | (a verifier) |
| H4 Univers | (a remplir) | (a verifier) |