VAR-IRF

Impulse Response Function (IRF) utilities for VAR models with plotting helpers. Extracted from the workflow in Tutorial.ipynb.

Installation

pip install var-irf

Repository: https://github.com/TianzhuQin/Vector-Autoregression-Impulse-Response-Function

Paper: Hug fans or follow celebrities? How nationalism is reinforced on Chinese social media

Method summary (from the notebook)

• Fit a VAR(p) with statsmodels.tsa.api.VAR(df).fit(lags).
• Get IRF object via fit(...).irf(h):
  • IRF values: cumulative cum_effects (or non-cumulative irfs).
  • Error bands:
    • Asymptotic (asym): build ±z·stderr from covariance.
    • Monte Carlo (mc): direct lower/upper intervals.
• Plotting:
  • Central IRF curve, gradient shaded bands, style options.

Quick start

import pandas as pd
import matplotlib.pyplot as plt
from var_irf import compute_irf, plot_irf

# df is a DataFrame with multiple series columns and a time index
result = compute_irf(df, lags=2, horizon=7, stderr_type="asym", orth=False)

fig, ax = plt.subplots(figsize=(10, 8))
plot_irf(result, "Series1", "Series2", ax=ax, direction="single", band="shaded", layout="overlay", shaded=True)
ax.axhline(0, color='r', linestyle='dashdot', linewidth=0.5)
plt.show()

API

• compute_irf(df, lags=2, horizon=7, stderr_type='asym'|'mc', orth=False, cumulative=True, ...): Fit a VAR and return IRF arrays and error information.
• plot_irf(result, impulse, response, ax=None, direction='single'|'bi', band='shaded'|'interval', layout='overlay'|'side-by-side', shaded=True, ...): Plot using the result from compute_irf with three orthogonal switches:
  • direction: single A→B or bidirectional.
  • band: shaded (gradient) or interval (errorbar-like); set shaded=False for line-only.
  • layout: overlay in one axes or side-by-side two subplots.

License

MIT

Citation

If you use this package in your research, please cite:

@article{
  HongKimZhangQin2025,
  author = {Hong, Ji Yeon and Kim, Yong H. and Zhang, Han and Qin, Tianzhu},
  title = {Hug fans or follow celebrities? How nationalism is reinforced on Chinese social media},
  journal = {Science Advances},
  year = {2025},
  volume = {11},
  number = {45},
  pages = {eadu8241},
  doi = {10.1126/sciadv.adu8241},
  url = {https://www.science.org/doi/10.1126/sciadv.adu8241}
}