IGNIS: A Robust Neural Network Framework for Constrained Parameter Estimation in Archimedean Copulas

Classical estimators, the cornerstones of statistical inference, face insurmountable challenges when applied to important emerging classes of Archimedean copulas. These models exhibit pathological properties, including numerically unstable densities, a restrictive lower bound on Kendall’s tau, and vanishingly small likelihood gradients, making MLE brittle and limiting MoM’s applicability to datasets with sufficiently strong dependence (i.e., only when the empirical Kendall’s $τ$ exceeds the family’s lower bound $\approx 0.545$). We introduce \textbf{IGNIS}, a unified neural estimation framework that sidesteps these barriers by learning a direct, robust mapping from data-driven dependency measures to the underlying copula parameter $θ$. IGNIS utilizes a multi-input architecture and a theory-guided output layer ($\mathrm{softplus}(z) + 1$) to automatically enforce the domain constraint $\hatθ \geq 1$. Trained and validated on four families (Gumbel, Joe, and the numerically challenging A1/A2), IGNIS delivers accurate and stable estimates for real-world financial and health datasets, demonstrating its necessity for reliable inference in modern, complex dependence models where traditional methods fail. To our knowledge, IGNIS is the first \emph{standalone, general-purpose} neural estimator for Archimedean copulas (not a generative model or likelihood optimizer), delivering direct, constraint-aware $\hatθ$ and readily extensible to additional families via retraining or minor output-layer adaptations.

💡 Research Summary

The paper introduces IGNIS, a neural‑network‑based estimator designed to overcome the severe limitations of classical methods when estimating the dependence parameter θ of Archimedean copulas, especially the recently proposed A1 and A2 families. Classical approaches such as Maximum Likelihood Estimation (MLE) suffer from numerically unstable densities and flat likelihood surfaces, while the Method of Moments (MoM) is inapplicable for A1/A2 because their Kendall’s τ lower bound (~0.545) excludes many real‑world data sets.

IGNIS sidesteps these issues by learning a direct, constrained mapping from a set of robust, data‑driven summary statistics to θ. The input consists of a five‑dimensional vector of empirical dependence measures (Kendall’s τ, Spearman’s ρ, Pearson r, upper‑tail and lower‑tail dependence coefficients) concatenated with a four‑dimensional one‑hot encoding that identifies the copula family (Gumbel, Joe, A1, A2). These features are fed through a D‑layer fully connected network. The output layer applies a softplus activation plus one (softplus(z)+1), guaranteeing the domain constraint θ ≥ 1 automatically during training.

Training data are generated synthetically: for each family, θ is sampled uniformly from the practical range