feat(evaluate): add Sortino, Calmar, and max drawdown duration to risk_analysis

[JKDasondee]

Summary

Extends risk_analysis() in qlib/contrib/evaluate.py with three standard risk metrics that every quant platform should expose:

Metric	Formula	Edge case
sortino_ratio	annualized_return / downside_std * sqrt(N)	inf when no negative returns
calmar_ratio	annualized_return / abs(max_drawdown)	inf when no drawdown
max_drawdown_duration	longest period below cumulative peak (in observations)	0 when no drawdown

These are the three most commonly used risk-adjusted performance metrics after Sharpe/IR and max drawdown, which risk_analysis already computes.

Implementation

19 lines added, no new dependencies. All three metrics:

• Respect the existing mode parameter ("sum" vs "product" accumulation)
• Use the same annualization scaler N
• Are appended to the existing output pd.Series, so existing code that reads risk_analysis() output is fully backwards compatible

Downside deviation uses r.clip(upper=0).std(ddof=1) (standard Sortino convention — only negative returns contribute to risk).

Max drawdown duration counts contiguous periods where the cumulative curve is below its running maximum, using cummax() group labeling.

Verification

r = pd.Series(np.random.randn(252) * 0.01)
res = risk_analysis(r, freq='day')

# Manual Sortino matches
ds = r.clip(upper=0).std(ddof=1)
assert abs(res.loc['sortino_ratio', 'risk'] - r.mean() / ds * np.sqrt(238)) < 1e-10

# Manual Calmar matches
ann_ret = r.mean() * 238
max_dd = (r.cumsum() - r.cumsum().cummax()).min()
assert abs(res.loc['calmar_ratio', 'risk'] - ann_ret / abs(max_dd)) < 1e-10

# Edge cases
r_pos = pd.Series(np.abs(np.random.randn(100)) * 0.01)
assert risk_analysis(r_pos, freq='day').loc['calmar_ratio', 'risk'] == np.inf

# Product mode
res_prod = risk_analysis(r, freq='day', mode='product')
assert 'sortino_ratio' in res_prod.index

Existing tests/ops/ and tests/misc/ suites pass (18 tests, 0 failures).

[JKDasondee]

@microsoft-github-policy-service agree