v1.2.2

Fixed: compute_bsm_vanilla_theta (and _vector wrapper) volatility-decay term was 4x too large and omitted the leading discfactor; both call and put branches affected. Reported by @gaoflow in ArturSepp/VanillaOptionPricers#1.

1.2.1

Changelog

Entries start at 1.2.0. For earlier releases see the git log.

[1.2.1] - 2026-07-21

Fixed

• The 1.2.0 release shipped without any of the changes to modules inside stochvolmodels/; only
  new files and renames reached the commit. import stochvolmodels therefore failed with
  ModuleNotFoundError: No module named 'qis' unless the research extra was installed, because
  pricers/hawkes_jd_pricer.py still imported qis for @qis.timer. 1.2.0 is yanked on PyPI.
  Everything listed under 1.2.0 below is in this release.

[1.2.0] - 2026-07-20

Added

• stochvolmodels.__version__, resolved from installed package metadata.
• __init__.py for stochvolmodels.examples, stochvolmodels.pricers.factor_hjm and
  stochvolmodels.pricers.rough_logsv. These shipped only through setuptools namespace-package
  discovery.
• .gitattributes storing *.py as LF and *.pdf as binary.
• AGENTS.md and CLAUDE.md at the repository root: layout, commands, conventions and
  constraints for coding agents.
• papers/local_path.py, resolving the output and resource directories for the reproduction code.
  get_output_path() and get_resource_path() read papers/settings.yaml when it exists and
  otherwise fall back to docs/figures and resources under the repository root, both of which are
  gitignored. papers/settings.yaml.example is the committed template; papers/settings.yaml is
  gitignored. yaml is imported only when that file exists, so PyYAML is not a dependency.
• papers/logsv_model_with_quadratic_drift/paper/ and papers/sv_for_factor_hjm/paper/, each
  holding the published article, its LaTeX source and a README recording which of the two the
  docstring equation references follow.
• NumPy-style docstrings across the package: 524 of 602 module members, up from 235. Every module
  outside pricers/rough_logsv/ now carries a header. Docstrings for pricers/logsv/,
  logsv_pricer.py and pricers/factor_hjm/ cite equation numbers from the published articles.
• This file.

Changed

• stochvolmodels.data.test_option_chain is now stochvolmodels.data.sample_option_chains.
  get_btc_test_chain_data, get_gld_test_chain_data, get_gld_test_chain_data_6m,
  get_qv_options_test_chain_data, get_spy_test_chain_data, get_sqqq_test_chain_data and
  get_vix_test_chain_data keep their names and stay exported from stochvolmodels. There is no
  compatibility shim: import them from the top-level package or from the new module path.
• stochvolmodels.pricers.factor_hjm.rate_core is now stochvolmodels.utils.rate_core. This
  removes the data to pricers import cycle: data/option_chain.py calls
  get_default_swap_term_structure and swap_rate.
• rough_logsv_mc_chain_pricer_fixed_randoms takes debug: bool = False. It printed per-slice
  path diagnostics on every call.
• stochvolmodels/tests/bsm_mgf_pricer.py moves to stochvolmodels/examples/run_bsm_mgf_pricer.py,
  stochvolmodels/tests/qv_pricer.py to stochvolmodels/examples/run_qvar_analytics.py, and
  stochvolmodels/pricers/rough_logsv/test_kernel_approx.py to
  stochvolmodels/examples/run_rough_kernel_approx.py. stochvolmodels/tests/ now holds the pytest
  suite and rough_logsv_perf.py.
• examples/{run_heston,run_heston_sv_pricer,run_hawkes_pricer,run_pricing_options_on_qvar, quick_run_lognormal_sv_pricer}.py execute under run_local_test behind
  if __name__ == '__main__':. They ran their demo, including plt.show(), on import.
• requires-python is >=3.10 and the 3.9 classifier is removed. CI tests 3.10, 3.11 and 3.12.
• CI installs .[dev] instead of .[dev,research], so the import check fails if a library module
  starts importing qis again.
• my_papers/ is now papers/, matching factorlasso and trendfollowing, and
  logsv_model_wtih_quadratic_drift is spelled logsv_model_with_quadratic_drift.
• The 23 hardcoded absolute paths across 9 modules in papers/ call lp.get_output_path() or
  lp.get_resource_path(). None of them ran on another machine without editing.
• The stochvolmodels.pricers.hawkes_jd_pricer module docstring cites Liu, Packham and Sepp (2025),
  arXiv:2510.21297, the bivariate Hawkes specification the pricer implements.
• papers/README.md links each directory to its local paper/ folder, publisher DOI and SSRN entry
  where those exist.

Fixed

• func_rhs_jac in pricers/logsv/affine_expansion.py and in
  pricers/factor_hjm/rate_affine_expansion.py returns 2 M A + L. Both returned
  2 M A + A0, adding the state instead of the linear matrix. Both are passed to
  solve_ivp(method='BDF', jac=...) when is_stiff_solver=True. Prices move by
  about 1e-5 relative, since BDF uses the Jacobian only to drive its Newton
  iteration; the gain is convergence robustness, not accuracy.
• simulate_vol_paths runs again. It was decorated @njit(cache=False, fastmath=False) with a brownians: np.ndarray = None default, which numba 0.60+
  cannot type, so every call raised TypingError. The decorator is removed; the
  body is vectorised across paths and loops only over time steps, and fastmath
  was already off, so results are unchanged where the function previously ran.
• simulate_vol_paths sizes its output array from nb_steps, not from
  nb_steps_per_year. The returned sigma_t now has nb_steps + 1 rows, matching
  grid_t. At ttm = 1.0 the old sizing indexed out of bounds; below 1.0 it
  returned trailing all-zero rows; above 1.0 it raised.
• LogSvParams.eta returns 2 (kappa2 theta - kappa1) / vartheta^2 - 1, the
  exponent of the generalized inverse Gaussian steady state in Eq. (3.38) of Sepp
  and Rakhmonov (2024). It returned kappa1 theta / vartheta^2 - 1, which is not
  that exponent. No caller in the repository reads the property.
• compute_analytic_vol_moments no longer branches on is_qvar to assign the same
  value to rhs[-1] twice. Behaviour is unchanged.
• import stochvolmodels no longer requires qis. pricers/hawkes_jd_pricer.py imported qis at
  module level for one @qis.timer decorator and is exported from __init__.py, so
  pip install stochvolmodels followed by import stochvolmodels raised ModuleNotFoundError. It
  now uses stochvolmodels.utils.funcs.timer.
• The ndarrays_regression baseline
  stochvolmodels/tests/test_rough_logsv_pricer_regression/test_rough_logsv_pricer_pricing_regression.npz
  is committed. The fixture changed from data_regression to ndarrays_regression without it, so
  pytest stochvolmodels/tests/ failed on every run and every Python version.
• [tool.pytest.ini_options] testpaths points at stochvolmodels/tests. It pointed at a tests
  directory that does not exist, which raised PytestConfigWarning.
• stochvolmodels.data.fetch_option_chain raises ImportError naming qis and
  option-chain-analytics rather than failing on a bare import of a package in no dependency group.
• MANIFEST.in includes *.npz so the regression baseline reaches the sdist, and drops
  stochvolmodels/templates, stochvolmodels/static and stochvolmodels/my_papers/figures, none of
  which exist.
• README dependency floors match pyproject.toml, and the project tree matches the repository.

Removed

• The superseded data_regression baseline
  test_rough_logsv_pricer_pricing_regression.yml. Its values are carried over unchanged into the
  .npz.
• Duplicate VariableType, compute_logsv_a_mgf_grid, solve_a_ode_grid and solve_ode_for_a
  entries in the stochvolmodels/__init__.py re-export of affine_expansion. Each named the same
  object twice, so the export list is unchanged.

1.2.0

Changelog

Entries start at 1.2.0. For earlier releases see the git log.

[1.2.0] - 2026-07-20

Added

• stochvolmodels.__version__, resolved from installed package metadata.
• __init__.py for stochvolmodels.examples, stochvolmodels.pricers.factor_hjm and
  stochvolmodels.pricers.rough_logsv. These shipped only through setuptools namespace-package
  discovery.
• .gitattributes storing *.py as LF and *.pdf as binary.
• AGENTS.md and CLAUDE.md at the repository root: layout, commands, conventions and
  constraints for coding agents.
• papers/local_path.py, resolving the output and resource directories for the reproduction code.
  get_output_path() and get_resource_path() read papers/settings.yaml when it exists and
  otherwise fall back to docs/figures and resources under the repository root, both of which are
  gitignored. papers/settings.yaml.example is the committed template; papers/settings.yaml is
  gitignored. yaml is imported only when that file exists, so PyYAML is not a dependency.
• papers/logsv_model_with_quadratic_drift/paper/ and papers/sv_for_factor_hjm/paper/, each
  holding the published article, its LaTeX source and a README recording which of the two the
  docstring equation references follow.
• NumPy-style docstrings across the package: 524 of 602 module members, up from 235. Every module
  outside pricers/rough_logsv/ now carries a header. Docstrings for pricers/logsv/,
  logsv_pricer.py and pricers/factor_hjm/ cite equation numbers from the published articles.
• This file.

Changed

• stochvolmodels.data.test_option_chain is now stochvolmodels.data.sample_option_chains.
  get_btc_test_chain_data, get_gld_test_chain_data, get_gld_test_chain_data_6m,
  get_qv_options_test_chain_data, get_spy_test_chain_data, get_sqqq_test_chain_data and
  get_vix_test_chain_data keep their names and stay exported from stochvolmodels. There is no
  compatibility shim: import them from the top-level package or from the new module path.
• stochvolmodels.pricers.factor_hjm.rate_core is now stochvolmodels.utils.rate_core. This
  removes the data to pricers import cycle: data/option_chain.py calls
  get_default_swap_term_structure and swap_rate.
• rough_logsv_mc_chain_pricer_fixed_randoms takes debug: bool = False. It printed per-slice
  path diagnostics on every call.
• stochvolmodels/tests/bsm_mgf_pricer.py moves to stochvolmodels/examples/run_bsm_mgf_pricer.py,
  stochvolmodels/tests/qv_pricer.py to stochvolmodels/examples/run_qvar_analytics.py, and
  stochvolmodels/pricers/rough_logsv/test_kernel_approx.py to
  stochvolmodels/examples/run_rough_kernel_approx.py. stochvolmodels/tests/ now holds the pytest
  suite and rough_logsv_perf.py.
• examples/{run_heston,run_heston_sv_pricer,run_hawkes_pricer,run_pricing_options_on_qvar, quick_run_lognormal_sv_pricer}.py execute under run_local_test behind
  if __name__ == '__main__':. They ran their demo, including plt.show(), on import.
• requires-python is >=3.10 and the 3.9 classifier is removed. CI tests 3.10, 3.11 and 3.12.
• CI installs .[dev] instead of .[dev,research], so the import check fails if a library module
  starts importing qis again.
• my_papers/ is now papers/, matching factorlasso and trendfollowing, and
  logsv_model_wtih_quadratic_drift is spelled logsv_model_with_quadratic_drift.
• The 23 hardcoded absolute paths across 9 modules in papers/ call lp.get_output_path() or
  lp.get_resource_path(). None of them ran on another machine without editing.
• The stochvolmodels.pricers.hawkes_jd_pricer module docstring cites Liu, Packham and Sepp (2025),
  arXiv:2510.21297, the bivariate Hawkes specification the pricer implements.
• papers/README.md links each directory to its local paper/ folder, publisher DOI and SSRN entry
  where those exist.

Fixed

• func_rhs_jac in pricers/logsv/affine_expansion.py and in
  pricers/factor_hjm/rate_affine_expansion.py returns 2 M A + L. Both returned
  2 M A + A0, adding the state instead of the linear matrix. Both are passed to
  solve_ivp(method='BDF', jac=...) when is_stiff_solver=True. Prices move by
  about 1e-5 relative, since BDF uses the Jacobian only to drive its Newton
  iteration; the gain is convergence robustness, not accuracy.
• simulate_vol_paths runs again. It was decorated @njit(cache=False, fastmath=False) with a brownians: np.ndarray = None default, which numba 0.60+
  cannot type, so every call raised TypingError. The decorator is removed; the
  body is vectorised across paths and loops only over time steps, and fastmath
  was already off, so results are unchanged where the function previously ran.
• simulate_vol_paths sizes its output array from nb_steps, not from
  nb_steps_per_year. The returned sigma_t now has nb_steps + 1 rows, matching
  grid_t. At ttm = 1.0 the old sizing indexed out of bounds; below 1.0 it
  returned trailing all-zero rows; above 1.0 it raised.
• LogSvParams.eta returns 2 (kappa2 theta - kappa1) / vartheta^2 - 1, the
  exponent of the generalized inverse Gaussian steady state in Eq. (3.38) of Sepp
  and Rakhmonov (2024). It returned kappa1 theta / vartheta^2 - 1, which is not
  that exponent. No caller in the repository reads the property.
• compute_analytic_vol_moments no longer branches on is_qvar to assign the same
  value to rhs[-1] twice. Behaviour is unchanged.
• import stochvolmodels no longer requires qis. pricers/hawkes_jd_pricer.py imported qis at
  module level for one @qis.timer decorator and is exported from __init__.py, so
  pip install stochvolmodels followed by import stochvolmodels raised ModuleNotFoundError. It
  now uses stochvolmodels.utils.funcs.timer.
• The ndarrays_regression baseline
  stochvolmodels/tests/test_rough_logsv_pricer_regression/test_rough_logsv_pricer_pricing_regression.npz
  is committed. The fixture changed from data_regression to ndarrays_regression without it, so
  pytest stochvolmodels/tests/ failed on every run and every Python version.
• [tool.pytest.ini_options] testpaths points at stochvolmodels/tests. It pointed at a tests
  directory that does not exist, which raised PytestConfigWarning.
• stochvolmodels.data.fetch_option_chain raises ImportError naming qis and
  option-chain-analytics rather than failing on a bare import of a package in no dependency group.
• MANIFEST.in includes *.npz so the regression baseline reaches the sdist, and drops
  stochvolmodels/templates, stochvolmodels/static and stochvolmodels/my_papers/figures, none of
  which exist.
• README dependency floors match pyproject.toml, and the project tree matches the repository.

Removed

• The superseded data_regression baseline
  test_rough_logsv_pricer_pricing_regression.yml. Its values are carried over unchanged into the
  .npz.
• Duplicate VariableType, compute_logsv_a_mgf_grid, solve_a_ode_grid and solve_ode_for_a
  entries in the stochvolmodels/__init__.py re-export of affine_expansion. Each named the same
  object twice, so the export list is unchanged.