The payment heterogeneity index: an integrated unsupervised framework for high-volume procurement oversight and decision support

Battle Count: 50

The paper addresses public procurement oversight and audit prioritization, not financial markets or trading. While GMMs and distributional analysis are used in quantitative finance, the specific application (supplier payment heterogeneity in government procurement) has minimal direct relevance to trading strategies, portfolio optimization, or market microstructure. The structural heterogeneity framework could theoretically be adapted for regime detection in financial time series, but this is not explored.

Moderate complexity. Core PHI computation involves: (1) data preprocessing pipeline (filtering, name harmonisation with TF-IDF/token matching, robust standardisation), (2) per-supplier GMM fitting with BIC model selection (constrained to k≤4), (3) computation of four component statistics (modality, Bowley skewness, tail ratio, structural dispersion), (4) multiplicative PHI score and log-decomposition. Uses standard scikit-learn GMM. Main complexity lies in the data harmonisation pipeline and ensuring sufficient transaction volume. No deep learning or complex optimization required. Computationally lightweight as stated by authors.

High reproducibility: publicly available dataset (City of York Council transparency portal), fixed RNG seed (seed=0), explicit scikit-learn implementation with all hyperparameters specified (init_params=k-means, n_init=1, tol=1e-3, max_iter=100, covariance_type=spherical, lambda=1e-6), detailed data preprocessing pipeline with exact thresholds for name harmonisation (TF-IDF≥0.76, token set ratio≥77, Jaccard≥0.36, ensemble>0.66), and transparent GMM constraints (k_max=min(4, floor(n/25)), weight exclusion π_i<0.05). Minor gap: no GitHub repository link provided for code.