nonParQuantileCausality
0.1.0 (2025-09-15)
First public release (prepared for CRAN).
New features
- Introduces
np_quantile_causality()
— a nonparametric
causality-in-quantiles test for first-order lags,
supporting causality in mean and
variance.
- Returns an S3 object of class
np_quantile_causality
with fields for statistics, quantiles, bandwidth, type, and sample
size.
- Adds
plot()
method for
np_quantile_causality
objects to visualize test statistics
across quantiles with a reference critical-value line.
API changes
- Renames legacy
lrq.causality.test
→
np_quantile_causality
.
- Replaces dots with underscores in all function names.
- Deprecation shim:
lrq_causality_test()
calls
np_quantile_causality()
and warns.
- Replaces
do.causality.figure()
with the S3 plotting
interface plot.np_quantile_causality()
.
Data
- Bundles example dataset
gold_oil
(Gold, Oil) for
runnable examples and tests.
Implementation details
- Bandwidth: uses
KernSmooth::dpill()
as a
mean-regression proxy (Yu & Jones, 1998) with quantile-specific
rescaling.
- Internal local-linear quantile regression helper:
lprq2_()
(quantreg-backed).
- Kernel matrix uses a product Gaussian kernel with relative scaling
between lags.
Bug fixes
- Corrects a historical bug where
x2
lags were mistakenly
embedded from y2
in the variance case. Now uses
embed(x2, 2)
as intended.
Documentation
- Adds package-level documentation and function docs via
roxygen2.
- Includes a “References” section citing:
- Balcilar, M., Gupta, R., & Pierdzioch, C. (2016), Resources
Policy, 49, 74–80.
- Balcilar, M., Gupta, R., Kyei, C., & Wohar, M. E. (2016),
Open Economies Review, 27(2), 229–250.
- Provides
inst/CITATION
entries for standard package
citation.
- Examples demonstrate mean/variance tests and plotting using
gold_oil
.
Testing
testthat
suite covers:
- Object creation and basic structure for mean/variance runs.
- Plot method returns a
ggplot
object (skipped on
CRAN).
- Examples and tests are lightweight and CRAN-friendly (no network or
disk writes).
Licensing
- MIT license (
License: MIT + file LICENSE
).
Known limitations
- Current implementation supports first-order lags
only.
- No built-in bootstrap wrapper for small-sample critical values.
- O(n²) kernel matrix construction may be slow for very large n.