The Toolkit

One year of 250 trade returns has a maximum drawdown of 12.5%. Shuffle the same trades into 5,000 other orders: the final result is identical (True), but the maximum drawdown ranges from 12.6% to 26.8% (5th to 95th percentile). The order you happened to get was among the luckiest; plan for the distribution, not the history.

• Shuffling assumes trades are independent; strategies with streaks or regimes need block resampling.
• Simulated paths can only recombine what happened; they cannot produce a crash larger than any in the data.
• A thousand simulated paths from a wrong model are a thousand wrong answers.

rng = np.random.default_rng(12)
trades = rng.normal(0.002, 0.02, 250)                 # one year of trade returns
def mdd(rets):
    w = np.cumprod(1 + rets); return (1 - w / np.maximum.accumulate(w)).max()
actual = mdd(trades)
shuffled = [mdd(rng.permutation(trades)) for _ in range(5_000)]   # same trades, other orders
lo, hi = np.percentile(shuffled, [5, 95])
final_same = np.isclose(np.prod(1 + trades), np.prod(1 + rng.permutation(trades)))

Assumes import numpy as np and import pandas as pd (and SciPy where imported). Each snippet simulates or makes up its own data, as the comments say.

# Python — Monte Carlo equity simulation
import numpy as np

daily_returns = df['returns'].values
n_sims, n_days = 1000, 252

# Resample with replacement
paths = np.zeros((n_sims, n_days))
for i in range(n_sims):
    sampled = np.random.choice(daily_returns, size=n_days)
    paths[i] = np.cumprod(1 + sampled)

# Confidence bands
p5  = np.percentile(paths, 5, axis=0)
p95 = np.percentile(paths, 95, axis=0)