The Toolkit

Simulated stock-like returns — 7% a year on average with 16% volatility. The median maximum drawdown is 13.7% over one year, 25.7% over five and 38.0% over twenty. The same strategy has a deeper worst drawdown the longer you watch it, so drawdowns from histories of different lengths cannot be compared directly.

• A backtest’s maximum drawdown is one draw from a distribution; expect worse live.
• Limits set from a short history will be breached by a perfectly normal strategy.
• Drawdown depends on the order of returns (see Monte Carlo); the same trades can produce very different worst cases.

rng = np.random.default_rng(10)
def median_mdd(years, paths=2_000, mu=0.07, sigma=0.16):
    r = rng.normal(mu / 252, sigma / np.sqrt(252), (paths, 252 * years))
    wealth = np.exp(np.cumsum(r, axis=1))
    return np.median((1 - wealth / np.maximum.accumulate(wealth, axis=1)).max(axis=1))
table = {y: round(median_mdd(y), 3) for y in (1, 5, 20)}

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 — drawdown analysis
import numpy as np

cum_returns = (1 + returns).cumprod()
running_max = cum_returns.cummax()
drawdown = (cum_returns - running_max) / running_max
max_dd = drawdown.min()
print(f"Max Drawdown: {max_dd:.1%}")

# Recovery time
underwater = drawdown < 0
recovery_periods = underwater.astype(int).groupby(
    (~underwater).cumsum()
).sum()