arch: ARCH for Python

8.0.0 · active · verified Sun Apr 12

ARCH is a Python library providing tools for financial econometrics, focusing on Autoregressive Conditional Heteroskedasticity (ARCH) models, unit root tests, cointegration analysis, and bootstrapping. It is actively developed with regular releases, with the current version being 8.0.0, requiring Python 3.10 or newer for optimal compatibility and performance.

Warnings

breaking Version 8.0.0 and later require Python 3.10+. Older Python versions are no longer supported. The build system has also moved from setuptools to Meson.
Fix: Upgrade your Python environment to 3.10 or newer. If you encounter build issues, ensure your environment is compatible with Meson or consider `--no-binary` installation if performance is not critical and a compiler is unavailable.
gotcha When using `arch` for financial econometrics, proper handling and preparation of time-series data (e.g., ensuring correct datetime indices, handling missing values, and calculating returns appropriately) is crucial for valid model results. External libraries like `pandas_datareader` (used in quickstart) are not core dependencies and must be installed separately.
Fix: Always preprocess your time-series data carefully using libraries like Pandas. Validate data types and structures before feeding them into `arch` models. Install any external data acquisition libraries required for your specific workflow separately.
gotcha Mixing `pip` installations with system-managed Python packages (e.g., via `pacman` on Arch Linux) can lead to conflicts, especially during Python version updates. This can break virtual environments and lead to 'module not found' errors.
Fix: Always use virtual environments (e.g., `venv`, `conda`, `pyenv`) for your Python projects to isolate dependencies. Install all project-specific packages within the activated virtual environment using `pip`.

Install

pip install arch Standard installation
pip install 'arch[all]' With all optional dependencies for full functionality

Imports

arch_model
```
from arch import arch_model
```
univariate
```
from arch import univariate
```
unitroot
```
from arch import unitroot
```

Quickstart

This example demonstrates how to fetch historical financial data, calculate returns, and then use `arch_model` to specify and fit a GARCH(1,1,1) model with a Student's T distribution. It prints a summary of the fitted model. Note that `pandas_datareader` is used for data fetching and needs to be installed separately.

import datetime as dt
import pandas_datareader.data as web
from arch import arch_model

# Fetch financial data (example: FTSE 100 index)
# Requires 'pandas_datareader' to be installed (pip install pandas_datareader)
st = dt.datetime(1990,1,1)
en = dt.datetime(2014,1,1)
data = web.DataReader('^FTSE', 'yahoo', start=st, end=en)
returns = 100 * data['Adj Close'].pct_change().dropna()

# Initialize and fit an ARCH model
am = arch_model(returns, vol='Garch', p=1, o=1, q=1, dist='StudentsT')
res = am.fit(update_freq=5)
print(res.summary())

view raw JSON →