armagarch package

The package provides a flexible framework for modelling time-series data. The main focus of the package is implementation of the ARMA-GARCH type models.

Full documentation is coming soon.

Installation

The latest stable version can be installed by using pip

pip install armagarch

The master branch can be installed with

git clone https://github.com/iankhr/armagarch
cd armagarch
python setup.py install

Example: Modelling conditional volatility of the US excess market returns

The code requires: NumPy, Pandas, SciPy, Shutil, Matplotlib, Pandas_datareader and Statsmodels

import armagarch as ag
import pandas_datareader as web
import matplotlib.pyplot as plt
import numpy as np

# load data from KennethFrench library
ff = web.DataReader('F-F_Research_Data_Factors_daily', 'famafrench')
ff = ff[0]

# define mean, vol and distribution
meanMdl = ag.ARMA(order = {'AR':1,'MA':0})
volMdl = ag.garch(order = {'p':1,'q':1})
distMdl = ag.normalDist()

# create a model
model = ag.empModel(ff['Mkt-RF'].to_frame(), meanMdl, volMdl, distMdl)
# fit model
model.fit()

# get the conditional mean
Ey = model.Ey

# get conditional variance
ht = model.ht
cvol = np.sqrt(ht)

# get standardized residuals
stres = model.stres

# make a prediction of mean and variance over next 3 days.
pred = model.predict(nsteps = 3)

# pred is a list of two-arrays with first array being prediction of mean
# and second array being prediction of variance

Authors

• Ian Khrashchevskyi - iankhr

License

This project is licensed under the MIT License - see the LICENSE.md file for details

Acknowledgments

• Special thanks to Kevin Sheppard for his Python for Econometrics, which was an inspiration to write current code