Algorithmic Forensics of the Democratic Purse: A Causal Heterogeneous Temporal Graph Neural Network with Explainable Multi-Modal Intelligence Fusion for Detecting Illicit Political Finance, Beneficial Ownership Opacity and Corporate–Electoral Capture in Hybrid Democratic Regimes
Journal of International Law & Human Rights · 17.10.2026 · 5 (8) · 174-198
DOI: 10.59075/ilhr.v5i8.326 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of International Law & Human Rights, online erschienen am 17.10.2026. Verfasst von Hamza Sajjad und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).
Abstract (Originalsprache)
Purpose
Hybrid democratic regimes hold competitive elections while leaving the money that finances politics weakly observable. This study builds and tests an algorithmic-forensic framework that detects jurisdictions where illicit political finance, beneficial ownership (BO) opacity and corporate–electoral capture jointly raises illicit-finance risk, with Pakistan as the focal case. Design/methodology/approach: Using a real-data cross-section of all 36 Economist Intelligence Unit (EIU) 2024 hybrid regimes plus Pakistan (EIU Democracy Index, Transparency International CPI, Basel AML Index 2025, FATF grey-list status and fourth-round BO ratings, and International IDEA political-finance rules), we construct a heterogeneous temporal graph (country, rule, region and assessment-body nodes; 455 typed country–country edges) and propose a Causal Heterogeneous Temporal Graph Neural Network (C-HTGNN) that couples NOTEARS causal discovery with relation-specific message passing and a deep neural head. Explanations use exact modality-level Shapley values, relation ablation and counterfactual simulation. Results are triangulated with multiply-imputed OLS (Rubin's rules), wild and wild-cluster bootstrap, spatial autoregressive and spatial Durbin models, Lewbel heteroskedasticity-based instrumental variables, double/debiased machine learning, augmented inverse-probability weighting, quantile regression, Oster coefficient-stability bounds, Bayesian model averaging, Firth bias-reduced logit and sup-Wald structural-break tests.
Findings
BO opacity is the only variable with a fully stable causal edge to illicit-finance risk (bootstrap frequency 1.00) and its effect strengthens in the upper tail (quantile coefficient 0.58 at τ = 0.9 versus 0.22 at τ = 0.1). Corporate–electoral capture carries a posterior inclusion probability of 0.94 and a DML effect of 0.195 per standard deviation (p = 0.053), but it is not robust to clustering by region. De jure political-finance rules show no detectable association with risk. Heterogeneous message passing outperforms a flat deep network for both AML risk and BO opacity (best BO-opacity model, out-of-sample R² = 0.38), yet gradient boosting remains the most accurate AML-risk predictor in a sample this small, and FATF grey-listing is not predictable from structural indicators. Pakistan shows a significant 2023 democratic break (bootstrap p = 0.014) and ranks 16th of 37 on the fused Illicit Political Finance Risk Score. Originality/value: The study offers the first causal graph-learning framework for political-finance forensics in hybrid regimes, and reports where deep learning helps, where it does not, and why. Practical implications: Supervisors and electoral management bodies should prioritize effective BO verification over adding formal donation bans.
Zitieren
APA
Sajjad, H., Malik, D. M. U., & Khan, D. Z. (2026). Algorithmic Forensics of the Democratic Purse: A Causal Heterogeneous Temporal Graph Neural Network with Explainable Multi-Modal Intelligence Fusion for Detecting Illicit Political Finance, Beneficial Ownership Opacity and Corporate–Electoral Capture in Hybrid Democratic Regimes. Journal of International Law & Human Rights, 5(8), 174-198. https://doi.org/10.59075/ilhr.v5i8.326
Vancouver
Sajjad H, Malik DMU, Khan DZ. Algorithmic Forensics of the Democratic Purse: A Causal Heterogeneous Temporal Graph Neural Network with Explainable Multi-Modal Intelligence Fusion for Detecting Illicit Political Finance, Beneficial Ownership Opacity and Corporate–Electoral Capture in Hybrid Democratic Regimes. Journal of International Law & Human Rights. 2026;5(8):174-198. doi:10.59075/ilhr.v5i8.326
BibTeX
@article{Sajjad2026,
author = {Sajjad, Hamza and Malik, Dr. Muhammad Usman and Khan, Dr. Zulfqar},
title = {Algorithmic Forensics of the Democratic Purse: A Causal Heterogeneous Temporal Graph Neural Network with Explainable Multi-Modal Intelligence Fusion for Detecting Illicit Political Finance, Beneficial Ownership Opacity and Corporate–Electoral Capture in Hybrid Democratic Regimes},
journal = {Journal of International Law & Human Rights},
year = {2026},
volume = {5},
number = {8},
pages = {174--198},
doi = {10.59075/ilhr.v5i8.326}
}
EUVIMED