contagionchannels
Two-Stage Detection and Attribution of Cross-Border Financial Contagion Channels
Implementation of a two-stage framework for the joint
detection-and-attribution of cross-border financial contagion. Stage
one detects directional information flows between equity markets via
Wavelet-Quantile Transfer Entropy, combining maximal-overlap discrete
wavelet decomposition (Percival and Walden, 2000, ISBN:9780521685085)
with the transfer-entropy estimator of Schreiber (2000)
contagionchannels
Two-Stage Detection and Attribution of Cross-Border Financial Contagion Channels
contagionchannels is an R package implementing the two-stage framework for the joint detection and attribution of cross-border financial contagion developed in Bhandari, Parida and Sahu (2026), arXiv:2604.26546. The framework first detects directional information flows between equity markets via Wavelet-Quantile Transfer Entropy and then attributes each significant directional link to one of five mutually exclusive transmission channels — Trade, Financial, Geopolitical, Behavioural, Monetary Policy — through a multi-method structural-identification architecture.
Highlights
- Stage 1: Wavelet-Quantile Transfer Entropy (WQTE) over a six-scale MODWT decomposition and three quantile levels, with absolute-thresholding to construct directed contagion networks.
- Stage 2: Five complementary identification strategies for channel attribution:
- Instrumental-variables / Two-Stage Least Squares with channel-specific external instruments — Stock & Watson (2018)
doi:10.1111/ecoj.12593 - LASSO instrument selection — Belloni, Chernozhukov & Hansen (2014)
doi:10.1093/restud/rdt044 - Local projections at h=1, 5, 22 days — Jordà (2005)
doi:10.1257/0002828053828518 - Heteroskedasticity-based identification — Rigobon (2003)
doi:10.1162/003465303772815727 - Cinelli-Hazlett (2020) robustness-value sensitivity —
doi:10.1111/rssb.12348
- Instrumental-variables / Two-Stage Least Squares with channel-specific external instruments — Stock & Watson (2018)
- Bundled datasets: 18 G20 equity markets × 5,036 daily log-returns (Jan 2006 – Mar 2026); 14 channel proxies; 8 crisis sub-period definitions.
- Three vignettes: full paper replication; methodology overview; using with custom datasets.
Installation
From source (the package is not yet on CRAN):
# install.packages("devtools")
devtools::install_local("path/to/contagionchannels", build_vignettes = TRUE)
Or directly from the source tarball:
install.packages("contagionchannels_0.1.3.tar.gz", repos = NULL, type = "source")
Quick start
library(contagionchannels)
# Load bundled data
d <- load_paper_data()
# Build the five channel composites
channels <- build_channel_composites(d$proxies)
# Run the full pipeline (Stage 1 detection + Stage 2 attribution)
res <- run_contagion_pipeline(d$returns, channels, d$periods,
scale = 5, tau = 0.50, n_cores = 4)
# Per-period channel-attribution shares
res$period_shares
#> Period Trade Financial Geopolitical Behavioral Monetary Dominant
#> 1 PreCrisis 8.8 35.9 9.4 13.8 32.1 Financial
#> 2 GFC 27.9 15.0 27.0 2.9 27.2 Trade
#> 3 ESDC 13.7 39.5 19.0 19.6 8.1 Financial
#> 4 CSC 15.9 21.6 15.6 21.7 25.3 Monetary
#> 5 PreCOVID 18.1 31.6 8.2 13.5 28.6 Financial
#> 6 COVID 18.7 18.3 27.5 8.2 27.2 Geopolit.
#> 7 RusUkr 22.6 23.4 6.1 20.3 27.6 Monetary
#> 8 MidEastTariffs 27.6 26.8 7.5 6.3 31.8 Monetary
Key functions
| Function | Purpose |
|---|---|
compute_wqte_matrix() |
Stage-1 pairwise Wavelet-Quantile Transfer Entropy |
build_channel_composites() |
Construct five channel composites from raw proxies |
iv_2sls_attribute() |
Stage-2 IV/2SLS attribution per significant link |
lasso_iv_attribute() |
LASSO IV variant per Belloni-Chernozhukov-Hansen |
local_projections() |
Jordà 2005 local projections at multiple horizons |
rigobon_id() |
Heteroskedasticity-based identification |
cinelli_hazlett_rv() |
Robustness-value sensitivity bound |
build_network() |
Construct directed contagion network from QTE matrix |
walktrap_communities() |
Pons-Latapy community detection |
run_contagion_pipeline() |
End-to-end wrapper |
Bundled datasets
| Object | Description |
|---|---|
g20_returns |
xts of daily log-returns: 5,036 days × 18 markets |
channel_proxies |
data.frame of 14 channel-proxy time series |
crisis_periods |
Named list of 8 crisis sub-period date ranges |
Vignettes
vignette("replication", package = "contagionchannels") # Reproduce the paper
vignette("methodology", package = "contagionchannels") # Methodology overview
vignette("custom_data", package = "contagionchannels") # Use custom datasets
Replication scripts
The package includes a standalone replication suite at
system.file("scripts", package = "contagionchannels"). To reproduce every
numerical result and figure:
master <- system.file("scripts", "99_replicate_paper.R",
package = "contagionchannels")
source(master)
Citation
If you use the package in published work, please cite the methodology paper:
Bhandari, A., Parida, I., & Sahu, H. K. (2026). What Drives Contagion? Identifying and Attributing Cross-Border Transmission Mechanisms. arXiv preprint arXiv:2604.26546.
@misc{bhandari2026contagion,
title = {What Drives Contagion? Identifying and Attributing
Cross-Border Transmission Mechanisms},
author = {Bhandari, Avishek and Parida, Ipsita and Sahu, Hitesh Kumar},
year = {2026},
eprint = {2604.26546},
archivePrefix = {arXiv},
primaryClass = {q-fin.ST},
url = {https://arxiv.org/abs/2604.26546}
}
@manual{contagionchannels2026,
title = {contagionchannels: Two-Stage Detection and Attribution of
Cross-Border Financial Contagion Channels},
author = {Bhandari, Avishek and Parida, Ipsita and Sahu, Hitesh Kumar},
year = {2026},
note = {R package version 0.1.3},
url = {https://github.com/avishekb9/contagionchannels}
}
License
GPL-3 © Bhandari, Parida & Sahu (2026), Indian Institute of Technology Bhubaneswar.