mqqcause — Multivariate Quantile-on-Quantile Granger Causality

Author / maintainer: Dr Merwan Roudane  ·  merwanroudane920@gmail.com  ·  Repo: https://github.com/merwanroudane/qqcaus

mqqcause implements both bivariate and multivariate (conditional) Quantile-on-Quantile Granger causality tests. For each pair of quantile levels (theta, tau), the slope beta1(theta, tau) from the locally-weighted quantile regression of y_t on x_{t-1} is tested against zero with a paired bootstrap. The multivariate variant additionally conditions on a set of moderators Z (and optional x * Z interactions). A Sup-Wald summary across the grid is also provided (Troster 2018).

All 3D surfaces, heatmaps and contour plots default to MATLAB Parula.

Installation

# from CRAN (once accepted)
install.packages("mqqcause")

# development version
# install.packages("remotes")
remotes::install_github("merwanroudane/qqcaus")

# from a local source tarball
install.packages("mqqcause_1.0.0.tar.gz", repos = NULL, type = "source")

Quick start

library(mqqcause)

set.seed(1)
n <- 200
x <- rnorm(n)
z <- rnorm(n)
y <- 0.3 * c(0, x[-n]) + 0.2 * c(0, z[-n]) + rnorm(n, sd = 0.4)

# Bivariate
fit1 <- qq_causality(x, y, n_boot = 100)
print(fit1); sup_wald(fit1)

# Conditional (multivariate)
fit2 <- mqq_causality(x, y, moderators = list(Z = z), n_boot = 100)
print(fit2)

# Heatmap with significance stars (MATLAB Parula)
plot_qq_causality_heatmap(fit2, value = "t_value", show_stars = TRUE)

# Significance-only heatmap
plot_significance_heatmap(fit2)

References

License

GPL-3