Announcing `SignalY`: A Unified Framework for Signal Extraction, Decomposition, and Sparse Regression

Package development
, , , , , ,
1

Hello R Community,

I'm excited to announce the first stable release of SignalY, an R package now available on CRAN. This package is designed for econometricians, data scientists, and researchers in signal processing who need to extract meaningful latent structures from complex panel data.

What does it do?
Drawing from the wiki documentation and the underlying Bayesian architecture, SignalY offers a comprehensive toolkit for Signal Extraction, Sparse Regression, and Spectral Decomposition. Unlike standard time-series libraries, it integrates three distinct analytical layers into a single orchestrator:

  • Bayesian Variable Selection: Implements "Horseshoe" priors (Piironen & Vehtari, 2017) to solve the column selection problem, identifying which variables in a high-dimensional dataset actually drive the target signal.
  • Advanced Series Decomposition: Goes beyond basic filtering by offering a suite of methods including Empirical Mode Decomposition (EMD) for non-linear trends, Wavelet Analysis (MODWT) for multi-resolution variance, and the HP-GC Bayesian Filter (Grant & Chan, 2017) for robust cycle extraction.
  • Persistence Characterization: Features an automated "Unit Root Battery" (ADF, Phillips-Perron, KPSS, ERS) to rigorously classify the stationarity and memory of your data.

Why is it useful?
SignalY solves the "Inverse Problem" of applied econometrics: identifying the structure, drivers, and persistence of a system when the theoretical model is unknown or contested. By leveraging a high-performance Stan backend for its Bayesian modules, it allows researchers to:

  1. Robustly separate noise from signal using Total Variation regularization and shrinkage priors.
  2. Decompose economic or physical indicators into their constitutive trends and cycles without imposing rigid functional forms.
  3. Automate the "Which variables matter?" question with statistical rigor.

Getting Started:

The package vignettes provide detailed workflows, from recovering logarithmic trends in noisy data to isolating business cycles using the HP-GC filter. I hope SignalY becomes a valuable asset in your exploratory data analysis and structural modeling.

I welcome any feedback, bug reports, or contributions on GitHub. What are your challenges with extracting signals from noisy panel data?

Best,

José Mauricio Gómez Julián