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.SVXY (short vol -0.5x), VIX, VIX3M, SHY (cash alternative). Note: SVXY avant Feb 2018 etait -1x, post-VIXplosion = -0.5x.
# ETFs
etf_tickers = ['SVXY', 'SHY', 'SPY']
symbols = {}
for ticker in etf_tickers:
symbols[ticker] = qb.add_equity(ticker, Resolution.DAILY).symbol
# VIX data
vix_symbol = qb.add_data(CBOE, 'VIX', Resolution.DAILY).symbol
vix3m_symbol = qb.add_data(CBOE, 'VIX3M', Resolution.DAILY).symbol
start = datetime(2012, 1, 1)
end = datetime(2026, 1, 1)
# ETF history
etf_history = qb.history(list(symbols.values()), start, end, Resolution.DAILY)
closes = etf_history['close'].unstack(level=0)
stm = {str(v): k for k, v in symbols.items()}
closes.columns = [stm.get(str(c), str(c)) for c in closes.columns]
# VIX history
vix_hist = qb.history(vix_symbol, start, end, Resolution.DAILY)
vix3m_hist = qb.history(vix3m_symbol, start, end, Resolution.DAILY)
vix = None
vix3m = None
if not vix_hist.empty and not vix3m_hist.empty:
vix = vix_hist['close'].droplevel(0)
vix3m = vix3m_hist['close'].droplevel(0)
vix.name = 'VIX'
vix3m.name = 'VIX3M'
# Align all data
all_data = pd.concat([closes, vix, vix3m], axis=1).dropna()
closes = all_data[etf_tickers]
vix = all_data['VIX']
vix3m = all_data['VIX3M']
print(f"Periode: {all_data.index[0].date()} a {all_data.index[-1].date()}")
print(f"Donnees: {len(all_data)} jours")
print(f"VIX range: [{vix.min():.1f}, {vix.max():.1f}]")
else:
# CBOE custom-data type needs QC Cloud alternative-data infra (empty in local
# Docker research). VIX/VIX3M are CBOE's PUBLIC 30-/93-day implied-vol indices;
# the identical series is available via yfinance (^VIX / ^VIX3M), provisioned as
# local CSVs -- regenerable via scripts/quantconnect/provision_vix_csv.py (the
# CSVs are gitignored like the LEAN data folder, NOT committed; same data, local
# pipe -- consistent with how sibling quantbooks use local LEAN equity/crypto
# data instead of QC Cloud).
import os
def _find(name):
for p in ['/Lean/Data/' + name, name, '../data/' + name, '../../data/' + name]:
if os.path.exists(p):
return p
raise FileNotFoundError(name + ' not found beside notebook or in LEAN data folder')
vix = pd.read_csv(_find('vix_daily.csv'), parse_dates=['date'], index_col='date')['close']
vix3m = pd.read_csv(_find('vix3m_daily.csv'), parse_dates=['date'], index_col='date')['close']
vix = vix.loc[start:end]; vix3m = vix3m.loc[start:end]
vix.name = 'VIX'; vix3m.name = 'VIX3M'
# Normalize all indices to naive date-only: closes comes from Lean qb.history
# (Timestamps with a time/tz component), vix/vix3m from CSV (midnight naive).
# Without this, concat axis=1 matches ZERO timestamps -> dropna -> empty.
def _norm(s):
idx = pd.to_datetime(s.index)
if getattr(idx, 'tz', None) is not None:
idx = idx.tz_convert('UTC').tz_localize(None)
return idx.normalize()
closes.index = _norm(closes); vix.index = _norm(vix); vix3m.index = _norm(vix3m)
all_data = pd.concat([closes, vix, vix3m], axis=1).dropna()
closes = all_data[etf_tickers]
vix = all_data['VIX']; vix3m = all_data['VIX3M']
print("[local VIX data] CBOE custom type unavailable locally (needs QC Cloud);")
print(" loaded genuine VIX/VIX3M series from local CSV (yfinance ^VIX/^VIX3M = CBOE public indices).")
print(f"Periode: {all_data.index[0].date()} a {all_data.index[-1].date()}")
print(f"Donnees: {len(all_data)} jours")
print(f"VIX range: [{vix.min():.1f}, {vix.max():.1f}]")[local VIX data] CBOE custom type unavailable locally (needs QC Cloud);
loaded genuine VIX/VIX3M series from local CSV (yfinance ^VIX/^VIX3M = CBOE public indices).
Periode: 2012-01-03 a 2025-12-31
Donnees: 3520 jours
VIX range: [9.1, 82.7]Le ratio VIX3M/VIX mesure la pente de la courbe de volatilite. - Ratio > 1 = contango (normal, premium a harvester) - Ratio < 1 = backwardation (stress, danger)
if vix is None or vix3m is None:
print("VIX data not available — skipping term structure analysis.")
ratio = None
else:
ratio = vix3m / vix
print(f"VIX3M/VIX ratio stats:")
print(f" Mean: {ratio.mean():.3f}")
print(f" Std: {ratio.std():.3f}")
print(f" Contango (>1.05): {(ratio > 1.05).mean():.0%} du temps")
print(f" Backwardation (<1.0): {(ratio < 1.0).mean():.0%} du temps")
print(f" Deep backwardation (<0.9): {(ratio < 0.9).mean():.0%} du temps")
# VIX distribution
print(f"\nVIX distribution:")
for threshold in [15, 18, 22, 25, 30]:
pct = (vix < threshold).mean()
print(f" VIX < {threshold}: {pct:.0%}")VIX3M/VIX ratio stats:
Mean: 1.130
Std: 0.088
Contango (>1.05): 83% du temps
Backwardation (<1.0): 7% du temps
Deep backwardation (<0.9): 1% du temps
VIX distribution:
VIX < 15: 41%
VIX < 18: 65%
VIX < 22: 82%
VIX < 25: 89%
VIX < 30: 96%def backtest_vix_strategy(closes, vix, vix3m, position_size=0.45,
vix_threshold=22, ratio_entry=1.05, ratio_exit=1.02,
trailing_stop=0.10, vix_spike=28, lockout=15,
cash_in='idle'):
"""Backtest VIX term structure strategy."""
svxy_ret = closes['SVXY'].pct_change()
shy_ret = closes['SHY'].pct_change() if 'SHY' in closes.columns else pd.Series(0.0, index=closes.index)
vix_sma10 = vix.rolling(10).mean()
n = len(closes)
port_ret = pd.Series(0.0, index=closes.index)
in_position = False
peak_price = 0
lockout_counter = 0
n_trades = 0
for i in range(11, n):
r = vix3m.iloc[i] / vix.iloc[i] if vix.iloc[i] > 0 else 0
if lockout_counter > 0:
lockout_counter -= 1
if cash_in == 'SHY':
port_ret.iloc[i] = shy_ret.iloc[i]
continue
if in_position:
port_ret.iloc[i] = position_size * svxy_ret.iloc[i]
if cash_in == 'SHY':
port_ret.iloc[i] += (1 - position_size) * shy_ret.iloc[i]
peak_price = max(peak_price, closes['SVXY'].iloc[i])
# Exit conditions
should_exit = False
if closes['SVXY'].iloc[i] < peak_price * (1 - trailing_stop):
should_exit = True
if vix.iloc[i] > vix_spike:
should_exit = True
if r < ratio_exit:
should_exit = True
if should_exit:
in_position = False
lockout_counter = lockout
n_trades += 1
else:
if cash_in == 'SHY':
port_ret.iloc[i] = shy_ret.iloc[i]
# Entry conditions
if (r > ratio_entry and
vix.iloc[i] < vix_threshold and
vix.iloc[i] < vix_sma10.iloc[i]):
in_position = True
peak_price = closes['SVXY'].iloc[i]
vals = (1 + port_ret).cumprod()
total = vals.iloc[-1] - 1
years = len(port_ret) / 252
cagr = (1 + total) ** (1 / years) - 1 if years > 0 else 0
vol = port_ret.std() * np.sqrt(252)
sharpe = (cagr - 0.03) / vol if vol > 0.001 else 0
max_dd = ((vals - vals.expanding().max()) / vals.expanding().max()).min()
return {'sharpe': sharpe, 'cagr': cagr, 'max_dd': max_dd, 'vol': vol,
'cum': vals, 'n_trades': n_trades}
print("Fonctions definies.")Fonctions definies.research.ipynb recommande 30% (securite). main.py utilise 45%.
results_pos = {}
if vix is None:
print("VIX data not available — skipping position size analysis.")
else:
print(f"{'Position':<15} {'Sharpe':>8} {'CAGR':>8} {'MaxDD':>8} {'Trades':>8}")
print("-" * 50)
for pos in [0.20, 0.30, 0.45, 0.60]:
r = backtest_vix_strategy(closes, vix, vix3m, position_size=pos)
name = f'{pos:.0%}'
results_pos[name] = r
print(f"{name:<15} {r['sharpe']:>8.3f} {r['cagr']:>7.1%} {r['max_dd']:>7.1%} {r['n_trades']:>7}")Position Sharpe CAGR MaxDD Trades
--------------------------------------------------
20% -0.156 2.1% -14.8% 68
30% -0.002 3.0% -21.7% 68
45% 0.088 4.2% -31.3% 68
60% 0.122 5.2% -40.1% 68Plus la position est petite, moins la MaxDD. Mais le CAGR baisse aussi. Post-VIXplosion (SVXY -0.5x), le premium est halve -> position plus petite = plus safe.
results_vix = {}
if vix is None:
print("VIX data not available — skipping VIX threshold analysis.")
else:
print(f"{'VIX Threshold':<15} {'Sharpe':>8} {'CAGR':>8} {'MaxDD':>8}")
print("-" * 40)
for vt in [18, 22, 25, 30]:
r = backtest_vix_strategy(closes, vix, vix3m, vix_threshold=vt)
name = f'VIX<{vt}'
results_vix[name] = r
print(f"{name:<15} {r['sharpe']:>8.3f} {r['cagr']:>7.1%} {r['max_dd']:>7.1%}")VIX Threshold Sharpe CAGR MaxDD
----------------------------------------
VIX<18 0.039 3.5% -23.7%
VIX<22 0.088 4.2% -31.3%
VIX<25 0.100 4.4% -34.5%
VIX<30 0.136 4.9% -30.1%VIX < 22 est le seuil actuel. VIX < 18 trop restrictif (trop peu d’entrees). VIX < 25-30 : plus d’entrees mais dans des periodes plus risquees.
results_ratio = {}
if vix is None:
print("VIX data not available — skipping contango depth analysis.")
else:
print(f"{'Ratio Entry':<15} {'Sharpe':>8} {'CAGR':>8} {'MaxDD':>8}")
print("-" * 40)
for re in [1.03, 1.05, 1.08, 1.10]:
r = backtest_vix_strategy(closes, vix, vix3m, ratio_entry=re)
name = f'Ratio>{re:.2f}'
results_ratio[name] = r
print(f"{name:<15} {r['sharpe']:>8.3f} {r['cagr']:>7.1%} {r['max_dd']:>7.1%}")Ratio Entry Sharpe CAGR MaxDD
----------------------------------------
Ratio>1.03 0.142 4.9% -30.3%
Ratio>1.05 0.088 4.2% -31.3%
Ratio>1.08 0.065 3.9% -33.1%
Ratio>1.10 0.083 4.1% -33.9%Un ratio plus eleve = contango plus profond = meilleur premium mais moins d’entrees.
results_stop = {}
if vix is None:
print("VIX data not available — skipping trailing stop analysis.")
else:
print(f"{'Trailing Stop':<15} {'Sharpe':>8} {'CAGR':>8} {'MaxDD':>8}")
print("-" * 40)
for ts in [0.05, 0.07, 0.10, 0.15]:
r = backtest_vix_strategy(closes, vix, vix3m, trailing_stop=ts)
name = f'{ts:.0%} stop'
results_stop[name] = r
print(f"{name:<15} {r['sharpe']:>8.3f} {r['cagr']:>7.1%} {r['max_dd']:>7.1%}")Trailing Stop Sharpe CAGR MaxDD
----------------------------------------
5% stop -0.290 -0.2% -40.8%
7% stop 0.042 3.5% -30.8%
10% stop 0.088 4.2% -31.3%
15% stop -0.115 1.3% -39.6%research.ipynb recommande 7% (vs 10% actuel). 5% = whipsaw.
Utiliser SHY (1-3Y bonds) pour le cash idle au lieu de 0% return. research.ipynb recommande 70% SHY allocation.
results_cash = {}
if vix is None:
print("VIX data not available — skipping cash allocation analysis.")
else:
print(f"{'Cash Policy':<20} {'Sharpe':>8} {'CAGR':>8} {'MaxDD':>8}")
print("-" * 45)
for name, cash in [('Idle (actuel)', 'idle'), ('SHY allocation', 'SHY')]:
r = backtest_vix_strategy(closes, vix, vix3m, cash_in=cash)
results_cash[name] = r
print(f"{name:<20} {r['sharpe']:>8.3f} {r['cagr']:>7.1%} {r['max_dd']:>7.1%}")Cash Policy Sharpe CAGR MaxDD
---------------------------------------------
Idle (actuel) 0.088 4.2% -31.3%
SHY allocation 0.086 4.2% -32.6%SHY ajoute ~2-3%/an de rendement pendant les periodes hors-position. Comme la stratégie est 60-70% du temps en cash, c’est un gain significatif.
if not any([results_pos, results_vix, results_stop, results_cash]):
print("No backtest results available — skipping visualization.")
print("Run this notebook on QC Cloud with CBOE data access.")
else:
fig, axes = plt.subplots(2, 2, figsize=(16, 10))
ax = axes[0, 0]
if results_pos:
for name, r in results_pos.items():
ax.plot(r['cum'].values, label=f"{name} (S={r['sharpe']:.2f})", linewidth=1.5)
ax.set_title('H1: Position size', fontweight='bold')
ax.legend(fontsize=8)
ax.grid(True, alpha=0.3)
ax = axes[0, 1]
if results_vix:
for name, r in results_vix.items():
ax.plot(r['cum'].values, label=f"{name} (S={r['sharpe']:.2f})", linewidth=1.5)
ax.set_title('H2: VIX threshold', fontweight='bold')
ax.legend(fontsize=8)
ax.grid(True, alpha=0.3)
ax = axes[1, 0]
if results_stop:
for name, r in results_stop.items():
ax.plot(r['cum'].values, label=f"{name} (S={r['sharpe']:.2f})", linewidth=1.5)
ax.set_title('H4: Trailing stop', fontweight='bold')
ax.legend(fontsize=8)
ax.grid(True, alpha=0.3)
ax = axes[1, 1]
if results_cash:
for name, r in results_cash.items():
ax.plot(r['cum'].values, label=f"{name} (S={r['sharpe']:.2f})", linewidth=1.5)
ax.set_title('H5: Cash allocation', fontweight='bold')
ax.legend(fontsize=8)
ax.grid(True, alpha=0.3)
plt.tight_layout()
plt.savefig('vix_termstructure_quantbook_analysis.png', dpi=150, bbox_inches='tight')
plt.show()
print("Graphique sauvegarde.")
Graphique sauvegarde.| Hypothese | Résultat QuantBook | Coherent avec research.ipynb? |
|---|---|---|
| H1 Position size | 60% Sharpe 0.122 (MaxDD -40%) ; 30% Sharpe -0.002 (MaxDD -22%) | Partiellement : 60% maximise le Sharpe mais 30% offre un meilleur ratio Sharpe/MaxDD |
| H2 VIX threshold | VIX<30 Sharpe 0.136 > VIX<22 Sharpe 0.088 | Non : seuil plus large (VIX<30) perf mieux sur 2012-2026 |
| H3 Contango depth | Ratio>1.03 Sharpe 0.142 > Ratio>1.05 Sharpe 0.088 | Non : seuil plus permissif (1.03) capture plus d’entrees |
| H4 Trailing stop | 10% Sharpe 0.088 ; 7% Sharpe 0.042 | Partiellement : 10% legerement meilleur |
| H5 SHY cash | SHY Sharpe 0.086 vs Idle 0.088 (~equivalent) | Non : SHY n’ajoute pas de CAGR net sur 2012-2026 |
Post-VIXplosion 2018, SVXY est passe de -1x a -0.5x. Le premium est halve. MaxDD 35% est structural (tail events VIX). Sharpe 0.051 reflete le fait que le risk-free rate (~3-5%) mange presque tout le CAGR.