drvarma

Exact maximum-likelihood estimation, forecasting and diagnostics of multivariate VARMA (vector ARMA) models — in pure Python, with an optional compiled C engine for speed.

import numpy as np
from drvarma import Model, datasets

series = datasets.simulate_varma(phi=[np.diag([0.5, 0.4, 0.3])], sigma=np.eye(3),
                                 n=300, mu=[100., 50., 75.], seed=1,
                                 names=["A", "B", "C"])
m = Model(series, p=2, q=0, include_mean=True).fit()
print(m.phi, m.sigma, m.loglik)
levels, lo, hi = m.forecast(12, bands=True)     # + 95% bands
print(m.diagnostics())                          # Hosking Q, Jarque-Bera

drvarma fits a stationary Gaussian VARMA(p, q), Φ(B)(wₜ − μ) = Θ(B) aₜ, aₜ ~ N(0, Σ), by exact maximum likelihood (no conditional/back-forecasting approximation), and gives forecasts (+ error bands), impulse responses, variance decompositions, residual diagnostics and volatility.

What this provides. drvarma is a pure-Python implementation of Mauricio's exact-likelihood algorithm for multivariate VARMA (1995 JASA / 1997 AS 311): an innovations-form factorisation that evaluates the exact Gaussian likelihood directly on the VARMA form, without a state-space / Kalman filter, maximised by a faithful factored-BFGS quasi-Newton. Exact ML for VARMA is also available in Python through state-space methods (e.g. statsmodels' Kalman-filter VARMAX); what drvarma adds is a faithful port of this specific, non-Kalman algorithm — which, as far as we know, is not otherwise available in Python — together with the surrounding forecasting/IRF/FEVD/diagnostics toolkit. It runs with no compiled code; the optional C engine is only an accelerator.

Install

pip install drvarma                       # pure-Python (numpy + scipy)

Optional extras: drvarma[plots] (matplotlib + pyfug charts), drvarma[forecast-report] (HTML forecast reports), drvarma[c-engine] (build the CFFI C engine — needs GSL dev headers, ~10–100× faster but optional).

Features

Exact-ML estimation of stationary VARMA(p, q) for general dimension m, with mean, diagonal AR/MA/covariance restrictions, and optional Hannan–Rissanen two-step start.
Forecasting in original units: level, period and annual variation, each with standard errors and 95 % bands; plus fixed-parameter recursive (out-of-sample) forecasting.
Structural analysis: orthogonalised impulse responses, accumulated responses, long-run gain, and the forecast-error variance decomposition.
Diagnostics: Hosking multivariate portmanteau, multivariate Jarque–Bera, per-series ACF/PACF and two-sided cross-correlation (CCF).
Volatility: exponential-weight and moving-window residual covariance.
Transforms: Box-Cox + regular/seasonal differencing and harmonic deseasonalisation (with re-seasonalised forecasts).
I/O & reports: .inp reader/writer, C-format .out/.forecast/ .recursive text reports, an HTML forecast report per series, and matplotlib charts.

Everything runs with no compiled code. The optional CFFI engine wraps the validated drvarma C core and is bit-compatible with the pure-Python path on well-conditioned problems — an accelerator only. The numerical methods are tabulated below; see docs/DEVELOPER_GUIDE.md for the complexity discussion and a pure-Python vs hybrid vs C performance study.

Command line

drvarma IPC3 3 0 -mean -deseason auto -forecast 24        # writes IPC3.out, .forecast
drvarma IPC3 3 0 -mean -forecast 24 -html                # + HTML report per series
drvarma IPC3 3 0 -mean -estwin 200 -forecast 12          # recursive (.recursive)
drvarma IPC3 3 0 -mean -volexp 0.05 20 -volmov 20        # volatility (.volexp/.volmov)

<file>.inp in, text reports out. Flags: -mean -diagar -diagma -diagcov -m {1,2} -twostep -deseason [auto|force] -scale S -forecast H -html -estwin N -volexp [α w] -volmov [w] (λ, d, D come from the .inp header).

Documentation

docs/USER_GUIDE.md — install, API, CLI, worked examples.
docs/INP_FORMAT.md — the .inp input format (a precise, assistant-friendly spec for preparing inputs).
docs/DEVELOPER_GUIDE.md — internals, algorithmic complexity from the literature, and the performance study.

Numerical methods

Algorithm	Reference	Used for
Exact Gaussian VARMA log-likelihood (innovations factorisation, not Kalman)	Mauricio (1995) JASA; Mauricio (1997) AS 311	Exact likelihood (`drvarma._as311`)
Factored-BFGS quasi-Newton + Dennis–Schnabel line search	Dennis & Schnabel (1983)	ML optimisation (`drvarma._qnewt.raxopt`)
Concentrated objective `f1ᵐ·f2` (σ² profiled, Σ = σ²·Q)	Mauricio (1995) JASA §3	Conditioning; covariance / std errors
Hannan–Rissanen two-step	Hannan & Rissanen (1982)	VARMA start (`-twostep`)
Companion-form Lyapunov / autocovariances	Mauricio (1997) AS 311	Stationary covariance, ψ-weights
Orthogonalised IRF & FEVD (Cholesky of Σ)	Lütkepohl (2005)	Structural analysis
Hosking multivariate portmanteau; multivariate Jarque–Bera	Hosking (1980); Jarque & Bera (1980)	Residual diagnostics
Exponential / moving-window conditional covariance	—	Volatility (`drvarma.volatility`)

Cross-references: Mauricio (2002, JTSA 23(4)) and Shea (1989, AS 242) are the other efficient exact-likelihood methods compared in the literature.

Authors and licence

drvarma is developed by David E. Guerrero and Arthur B. Treadway, based on the exact-likelihood algorithms designed and coded by José Alberto Mauricio.

Released under the GNU General Public License v2.0 or later (GPL-2.0-or-later) — see COPYING.

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

drvarma-0.1.2.tar.gz (205.3 kB view details)

Uploaded Aug 5, 2026 Source

Built Distributions

If you're not sure about the file name format, learn more about wheel file names.

The dropdown lists show the available interpreters, ABIs, and platforms. Enable javascript to be able to filter the list of wheel files.

drvarma-0.1.2-py3-none-any.whl (105.3 kB view details)

Uploaded Aug 5, 2026 Python 3

drvarma-0.1.2-cp313-cp313-win_amd64.whl (106.0 kB view details)

Uploaded Aug 5, 2026 CPython 3.13Windows x86-64

drvarma-0.1.2-cp313-cp313-musllinux_1_2_x86_64.whl (1.3 MB view details)

Uploaded Aug 5, 2026 CPython 3.13musllinux: musl 1.2+ x86-64

drvarma-0.1.2-cp313-cp313-musllinux_1_2_aarch64.whl (1.2 MB view details)

Uploaded Aug 5, 2026 CPython 3.13musllinux: musl 1.2+ ARM64

drvarma-0.1.2-cp313-cp313-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl (1.4 MB view details)

Uploaded Aug 5, 2026 CPython 3.13manylinux: glibc 2.24+ x86-64manylinux: glibc 2.28+ x86-64

drvarma-0.1.2-cp313-cp313-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl (1.3 MB view details)

Uploaded Aug 5, 2026 CPython 3.13manylinux: glibc 2.24+ ARM64manylinux: glibc 2.28+ ARM64

drvarma-0.1.2-cp313-cp313-macosx_14_0_arm64.whl (1.0 MB view details)

Uploaded Aug 5, 2026 CPython 3.13macOS 14.0+ ARM64

drvarma-0.1.2-cp312-cp312-win_amd64.whl (106.0 kB view details)

Uploaded Aug 5, 2026 CPython 3.12Windows x86-64

drvarma-0.1.2-cp312-cp312-musllinux_1_2_x86_64.whl (1.3 MB view details)

Uploaded Aug 5, 2026 CPython 3.12musllinux: musl 1.2+ x86-64

drvarma-0.1.2-cp312-cp312-musllinux_1_2_aarch64.whl (1.2 MB view details)

Uploaded Aug 5, 2026 CPython 3.12musllinux: musl 1.2+ ARM64

drvarma-0.1.2-cp312-cp312-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl (1.4 MB view details)

Uploaded Aug 5, 2026 CPython 3.12manylinux: glibc 2.24+ x86-64manylinux: glibc 2.28+ x86-64

drvarma-0.1.2-cp312-cp312-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl (1.3 MB view details)

Uploaded Aug 5, 2026 CPython 3.12manylinux: glibc 2.24+ ARM64manylinux: glibc 2.28+ ARM64

drvarma-0.1.2-cp312-cp312-macosx_14_0_arm64.whl (1.0 MB view details)

Uploaded Aug 5, 2026 CPython 3.12macOS 14.0+ ARM64

drvarma-0.1.2-cp311-cp311-win_amd64.whl (106.0 kB view details)

Uploaded Aug 5, 2026 CPython 3.11Windows x86-64

drvarma-0.1.2-cp311-cp311-musllinux_1_2_x86_64.whl (1.3 MB view details)

Uploaded Aug 5, 2026 CPython 3.11musllinux: musl 1.2+ x86-64

drvarma-0.1.2-cp311-cp311-musllinux_1_2_aarch64.whl (1.2 MB view details)

Uploaded Aug 5, 2026 CPython 3.11musllinux: musl 1.2+ ARM64

drvarma-0.1.2-cp311-cp311-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl (1.4 MB view details)

Uploaded Aug 5, 2026 CPython 3.11manylinux: glibc 2.24+ x86-64manylinux: glibc 2.28+ x86-64

drvarma-0.1.2-cp311-cp311-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl (1.3 MB view details)

Uploaded Aug 5, 2026 CPython 3.11manylinux: glibc 2.24+ ARM64manylinux: glibc 2.28+ ARM64

drvarma-0.1.2-cp311-cp311-macosx_14_0_arm64.whl (1.0 MB view details)

Uploaded Aug 5, 2026 CPython 3.11macOS 14.0+ ARM64

drvarma-0.1.2-cp310-cp310-win_amd64.whl (106.0 kB view details)

Uploaded Aug 5, 2026 CPython 3.10Windows x86-64

drvarma-0.1.2-cp310-cp310-musllinux_1_2_x86_64.whl (1.3 MB view details)

Uploaded Aug 5, 2026 CPython 3.10musllinux: musl 1.2+ x86-64

drvarma-0.1.2-cp310-cp310-musllinux_1_2_aarch64.whl (1.2 MB view details)

Uploaded Aug 5, 2026 CPython 3.10musllinux: musl 1.2+ ARM64

drvarma-0.1.2-cp310-cp310-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl (1.4 MB view details)

Uploaded Aug 5, 2026 CPython 3.10manylinux: glibc 2.24+ x86-64manylinux: glibc 2.28+ x86-64

drvarma-0.1.2-cp310-cp310-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl (1.3 MB view details)

Uploaded Aug 5, 2026 CPython 3.10manylinux: glibc 2.24+ ARM64manylinux: glibc 2.28+ ARM64

drvarma-0.1.2-cp310-cp310-macosx_14_0_arm64.whl (1.0 MB view details)

Uploaded Aug 5, 2026 CPython 3.10macOS 14.0+ ARM64

File details

Details for the file drvarma-0.1.2.tar.gz.

File metadata

Download URL: drvarma-0.1.2.tar.gz
Upload date: Aug 5, 2026
Size: 205.3 kB
Tags: Source
Uploaded using Trusted Publishing? Yes
Uploaded via: twine/7.0.0 CPython/3.13.14

File hashes

Hashes for drvarma-0.1.2.tar.gz
Algorithm	Hash digest
SHA256	`014c36d046d181aa8f2319c46d47655734a5eb5c4f95554815b4aa5833c23913`
MD5	`a91558aef9d81370271dc0ffbe679b2c`
BLAKE2b-256	`549e1e6713b20acc0b5d456d6363ab83b0459ec3879e33c60c388189962fa528`

Algorithm	Hash digest
SHA256	`2ed136f1b198c384495393dd98ee5a1d6e367d841a85320b7e32dd9f3ded63de`
MD5	`36393b184c53132f878ab7bff85479ba`
BLAKE2b-256	`caa8725631091f9a5f8ea7365f322e79179643925475aee29f915cddd2c2dc79`

Algorithm	Hash digest
SHA256	`556572fa5de903ebad0491c1369f681511618fbefb0b97412b4e1b5b91d525a4`
MD5	`e8f87f81268ec26c28690301cd111a15`
BLAKE2b-256	`d31a48ff341269528944716c6640b68c5b665c6fb3ab7b7317a99459d5ba9fdb`

Algorithm	Hash digest
SHA256	`afe01089c7f8c3570e3d8e3eab655e7d8e3cdabfe0c1c08e6cace78367fc7a0d`
MD5	`28284d6e1ca93e143194dc10ebdbe8a3`
BLAKE2b-256	`7ba0287a74e8b9147d7303df0b1668b12e56734975911f0e133621d4c42762fb`

Algorithm	Hash digest
SHA256	`982f350cf16941eeb2da6b6e31213eb06c505b2019e95b2032faf0b61040c9de`
MD5	`bfbd06894eb69e793763ce24347721b7`
BLAKE2b-256	`7921053d8a75558272e8e3075fa2fdadc29b7c7fc98844fe5a3d9ddf1d36ef3e`

Algorithm	Hash digest
SHA256	`b577a231c0b274d335e98960bcd374ec39f1b37e9d682320f70d9ececd5d6b85`
MD5	`1163d82ad35bca0f44704ddb662a5444`
BLAKE2b-256	`88b9f6b40498b8a149ae8cf6ce5678025a2056a89b7237cf3f2b5d8243b8894f`

Algorithm	Hash digest
SHA256	`237b10c4636abf275d7b0f1a3cb763c2db171fcc001d12518115b30e0de30bd0`
MD5	`e28252097e1c6b149b61885836b412a5`
BLAKE2b-256	`00680befe7b0742e9d204f81d472cefafbf2fe68f652202c83e4fc8eb98b61e6`

Algorithm	Hash digest
SHA256	`cea85287514012eae84512b8d600f97ca869d2fc56a0a0d3b7efef537be1f0a4`
MD5	`7d8bcc661612ca74c8242d762f6d4120`
BLAKE2b-256	`781df4598a4a566276078d3c0169df2030addcffd86b6be38d59c8c0c5ec5cae`

Algorithm	Hash digest
SHA256	`45873861c353f52e0cf1a1f86e1caa86186bd167bc40d23c9e8ea2d71ff4bbcf`
MD5	`691419defbb9ae914007d9867ba63251`
BLAKE2b-256	`26cd674b337e840b410c25a2dd12c36ae656339d0b13604dfee881b9dcf770df`

Algorithm	Hash digest
SHA256	`6ceec304b35357008c0aaadee57a78f970168e7575a724d7a932162e5432e033`
MD5	`88979a476acced1726328d2896523ea5`
BLAKE2b-256	`be63bb03fd0da80389a8af393b8f24ad4087382eed4c352446ade1e77f4d5d7f`

Algorithm	Hash digest
SHA256	`854f7f7eaf69a424f5700babeb9f355980d5eb61dca67a03ed1fd454f913d4de`
MD5	`dc3545f81247c2bd81966d53ab99d85a`
BLAKE2b-256	`731a5291b0a4d565046b8270894d01dfa46f04e189ed20dbf0cdf89dae60b3c0`

Algorithm	Hash digest
SHA256	`4128e403e1a3f3693fb61e4fc11cfb21ba2303bd8f924d7c837d267eeca8b301`
MD5	`934fa237fad3cc02f821532888060ca4`
BLAKE2b-256	`75ae4e1d9ee5fb0dfa925cd32ce8a9349badf89fe85aa09c760d23079e933526`

Algorithm	Hash digest
SHA256	`f9bf27d107d4e3f1a8f707024304ea2bd4b4c68926b53cb3bd9bbab9303c65d2`
MD5	`dee5776f4c6ed8593c5285a8fba9ce8a`
BLAKE2b-256	`ac76a24c7c0bf35e043aac6a5318dfa2041e7dafa0d28bc92d6df70c83792861`

Algorithm	Hash digest
SHA256	`d2eb0db4c23b51f5cc113c28de3ed6384f68f81a36ea0cd10143ebffdfaa01c9`
MD5	`cf02995b0fad257ede7d3730f790b376`
BLAKE2b-256	`9901a51d89b3ad3e7262f38aee8943b3ce369377fbf1c4122344e9a84138f230`

Algorithm	Hash digest
SHA256	`ce658af0001d17694a10b20c20f07505ce8a71646c6a4a2769e1c6fe8ce4f932`
MD5	`83ea7e00d6b179929c06a491701d5d9a`
BLAKE2b-256	`5c7c0653026645401dfcc544c04eb1dec1541b92e6f55985700176b6f899f845`

Algorithm	Hash digest
SHA256	`45474f388ea5dd02b6796b9902ca43e3b38ebb9c6bc8de0451956a2176beae46`
MD5	`6627f2e7c4fb7d6a18aa62b3ef669fc0`
BLAKE2b-256	`e023a4cb3172c08a4a170f05c5af4bef8bc65a0b2c2b3fce7b95be6d8088b99a`

Algorithm	Hash digest
SHA256	`c2b9e02ace9825d84bcfd69c4ba9761f57ff97a28107fda773bb7096c2612b1b`
MD5	`e10caf8ac4a5dc9ac69462260bdeecde`
BLAKE2b-256	`e9369c4287bc188b6d7adf9c936d5f64d383fe506bf91720edffced796eed2a1`

Algorithm	Hash digest
SHA256	`3ba11a129559f56c9ccd574639754f032e88509d6f093233d39a709a1ebe2ee0`
MD5	`cd6e16733910c229d4998d63d874b018`
BLAKE2b-256	`22c32e73dce2c447fbccecee98df0072e719daa45c6a9856005a5f7d434bd325`

Algorithm	Hash digest
SHA256	`424d93ce0d60215d078a0f32b3ba8b439e7c839f5652c2f19a20ad8d15610682`
MD5	`a1939a9fb01906da6793219f2e797359`
BLAKE2b-256	`1be3fc70e2f95503d129faa676a09fe7a597d90d9249e320b01d2d66f1f890bd`

Algorithm	Hash digest
SHA256	`fd98f8d2dae90ecd5add95c1b4983a943ca0ef6c78eae29decc1cdcf91cb86e5`
MD5	`27aa43a7f7a377d2b9ab768d3b89d657`
BLAKE2b-256	`26cc513f1a0b04996d8aaf24c52007a34e0dc2dc55646cc5ef4c88ddf7e8e741`

Algorithm	Hash digest
SHA256	`d58096607b4e21e4664d0d6cd48852d3c2a50bb28a867b422a5426d014264ee6`
MD5	`fa24a8c7c9b1d435172e5e3f90179a99`
BLAKE2b-256	`93e5c1289f1e5951d2312f2291e1ac2b91ac7dc2b6e80e3fab9c16b8a72e6546`

drvarma 0.1.2

drvarma

Install

Features

Command line

Documentation

Numerical methods

Authors and licence

Project links

Key dates

1 maintainer

Credits

GitHub Statistics

License expression

Requires

Provides Extra

Tags

Classifiers

Download files

Source Distribution

Built Distributions

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Algorithm	Hash digest
SHA256	`57ab22b340162b58baab8cf63ccda5e79b11c86d7478d6225f457a274df2dbd1`
MD5	`cd21d9ee498584b23ca49afa5d567fe2`
BLAKE2b-256	`e905a1ed08aaa0a44f98ca0b1d4e5d53dbd2ec9a79186bfaef8b4810263b32bf`

Algorithm	Hash digest
SHA256	`8dc5a4e3cefd555060552853bbf9f73f90685dceb46c9156b71536e284222759`
MD5	`25169d410d8b925aca433c1d404ea7bc`
BLAKE2b-256	`51b8950a8569b7b7e306357101e66f2c349851e91003f3f6b66715272f56522f`

Algorithm	Hash digest
SHA256	`0c9e91b597dea4af3919cfb0858d7bdf7fc0f9441f7edc53140ba313353fa989`
MD5	`3bea7f788059d4553bdc5b8826ba425c`
BLAKE2b-256	`097d2ef634030f6c3e1db98dcbeb6f11e4a11fb9ad3a988aa213013d7e161c48`

Algorithm	Hash digest
SHA256	`23707d4ae6fb50fe227c47b6b1df7724543f2f705c9b119b4a647dfdb7591690`
MD5	`de1fa73619ac03bfd221782a2a9a7e59`
BLAKE2b-256	`6327b46a32c15c9aa39ddfd28ecabd680363fd9aaefc81699cc94478249ea65d`

Algorithm	Hash digest
SHA256	`f2460a09560f8ee23365553c783580d2d216fd8a1b105d07ef97a612efb7f990`
MD5	`7035d89d1e899f9b3d5febe6cf99b69d`
BLAKE2b-256	`a22a9c9b757a670439f6c79c7068cc574114487496f503b0c97aedc27ca91d16`