The Toolkit

A strategy with an annual Sharpe ratio of 1, measured on three years of monthly returns. Its standard error is about 0.59, so the 95% interval runs from −0.15 to 2.15. Three years cannot tell a good strategy from a useless one — or from an excellent one.

• Autocorrelated returns (smoothed or illiquid assets) make the usual √12 annualisation overstate the Sharpe ratio (Lo 2002).
• The ratio treats upside and downside volatility alike; skewed strategies (option selling) look better than they are.
• Comparing the best of many strategies by Sharpe needs a correction for the number tried.

sr_annual, months = 1.0, 36                         # a Sharpe of 1, from three years of monthly returns
sr_m = sr_annual / np.sqrt(12)
se_m = np.sqrt((1 + 0.5 * sr_m ** 2) / months)       # Lo (2002), i.i.d. returns
se_annual = se_m * np.sqrt(12)
ci = (sr_annual - 1.96 * se_annual, sr_annual + 1.96 * se_annual)

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 — risk-adjusted ratios
import numpy as np

returns = df['returns']
rf = 0.05 / 252  # daily risk-free rate

# Sharpe (annualised)
sharpe = (returns.mean() - rf) / returns.std() * np.sqrt(252)

# Sortino (downside only)
down = returns[returns < 0].std()
sortino = (returns.mean() - rf) / down * np.sqrt(252)

# Calmar
cum = (1 + returns).cumprod()
dd = (cum / cum.cummax() - 1).min()
calmar = returns.mean() * 252 / abs(dd)