merton 1.0.0

Production-grade Merton structural credit-risk model and extensions.

merton

A production-grade Python package for the Merton structural credit-risk model and its industry-standard descendants (KMV, Black-Cox, Geske, Longstaff-Schwartz, CreditGrades, Leland-Toft, jump-diffusion, Vasicek portfolio, climate overlays).

🚀 New in 1.0 — first stable release. API frozen per semver, cold import merton down to ~500 ms (from ~1.1 s), conda-forge feedstock in review. See the announcement post and the migration guide.

Highlights

• Vectorized core — single-firm and panel-scale (10 000+ firms × decades) with NumPy by default and optional Numba / JAX / CuPy / MLX backends.
• Multiple calibration methods — Vassalou-Xing iterative MLE, Duan transformed-data MLE (with survivorship-bias correction), Jones-Mason-Rosenfeld, Bharath-Shumway naive, KMV iterative (Crosbie-Bohn).
• Full extensions library — Black-Cox first-passage, Geske compound options, Longstaff-Schwartz stochastic rates, CreditGrades, Leland-Toft endogenous default, Zhou jump-diffusion, hybrid structural/reduced-form, climate overlays.
• Portfolio engine — Vasicek single-factor (Basel IRB closed-form), Gaussian / t / Clayton / factor copulas, Monte Carlo loss distribution with VaR / ES / economic capital.
• Backtesting harness — AUC, Brier, KS, accuracy ratio, reliability / calibration curves, rolling-window and walk-forward validation.
• Excel integration — =MERTON_DD(equity, σE, debt, rf, T) and friends, available in Excel Web, Excel Mac (M365), and Excel Windows via xlwings Server.
• Climate stress — composable Scenario framework with packaged NGFS Phase V (2024) scenarios (net_zero_2050, delayed_transition, current_policies, fragmented_world); ClimateOverlay wraps any structural model with carbon-price paths and sectoral PD multipliers.
• OpenTelemetry observability — opt-in via merton.obs.enable(...); pipes spans to any OTLP-compatible backend (Datadog, Honeycomb, Tempo, …).
• Modern Python — type hints, Pydantic settings, structured logging, Apache-2.0 license, cross-platform wheels for Python 3.11-3.14 (incl. free-threaded).

Quickstart

uv pip install merton          # or: pip install merton

from merton import Firm, fit

firm = Firm(
    equity=100_000_000,        # $100M market cap
    debt_short=20_000_000,
    debt_long=30_000_000,
    equity_vol=0.30,
    rf=0.045,
    horizon=1.0,
)

result = fit(firm, method="vassalou_xing")
print(result.summary())
# MertonResult
#   distance_to_default : 3.142
#   probability_of_default : 0.000839
#   asset_value : 148.6M
#   asset_vol : 0.207
#   implied_spread (LGD=0.6) : 5.0 bps

Excel integration

merton excel install
merton excel server start --port 8000

Then in any Excel workbook (Web / Mac / Windows):

=MERTON_DD(B2, B3, B4+B5, B6, B7)        # distance to default
=MERTON_PD(B2, B3, B4+B5, B6, B7)        # probability of default
=MERTON_SPREAD(B2, B3, B4+B5, B6, B7, 0.6)

See the Excel docs for the full reference.

Performance

• Single-firm fit: <50 ms
• 10 000-firm × 10-year daily panel: <60 s on 8 cores (Numba backend)
• 100 000-sim portfolio VaR: <30 s with GPU (CuPy)

Documentation

Full docs at https://merton.readthedocs.io including:

• Theory deep-dives (Merton 1974, KMV, Black-Cox, Geske, …)
• Executable tutorials (AAPL time series, panel backtest, portfolio VaR, climate stress test, GPU acceleration, Excel integration)
• API reference
• Performance benchmarks and backend-selection guide

Installation matrix

Goal	Command
Minimum (CPU NumPy)	pip install merton
Numba JIT (default fast path)	pip install merton (Numba is a hard dep)
GPU acceleration	pip install "merton[gpu]" (requires CUDA 12)
JAX autodiff calibration	pip install "merton[jax]"
Apple Silicon GPU	pip install "merton[mlx]"
Excel integration	pip install "merton[excel]"
Visualization (matplotlib + plotly)	pip install "merton[viz]"
Bayesian MCMC calibration	pip install "merton[mcmc]"
OpenTelemetry tracing	pip install "merton[obs]"
Everything	pip install "merton[all]"

License

Apache-2.0 — see LICENSE.

Citing

If you use merton in academic work, please cite via the CITATION.cff file or the package's Zenodo DOI (assigned on first tagged release).