pith. sign in

arxiv: 2605.12547 · v2 · pith:7FL6I2DKnew · submitted 2026-05-09 · 💰 econ.EM · cs.LG· q-fin.ST· stat.AP

The Payment Heterogeneity Index: An Integrated Unsupervised Framework for High-Volume Procurement Oversight and Decision Support

Kyriakos Christodoulides This is my paper

Pith reviewed 2026-05-20 23:02 UTC · model grok-4.3

classification 💰 econ.EM cs.LGq-fin.STstat.AP
keywords public procurementpayment monitoringunsupervised anomaly detectionheterogeneity indexGaussian mixture modelsprocurement oversightdistributional analysis
0
0 comments X

The pith

The Payment Heterogeneity Index identifies a small cohort of suppliers with structurally distinct payment patterns in high-volume public procurement.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces the Payment Heterogeneity Index as a composite, unsupervised statistic for characterizing payment distributions in procurement data. It combines parameters from Gaussian mixture models with non-parametric measures across four components: modality, asymmetry, tail behaviour that includes extreme concentration, and structural dispersion of latent regimes. When applied to UK municipal procurement records, the index isolates a financially notable group of 0.6 percent of all suppliers, or 10.1 percent of high-volume vendors, whose payment structures differ markedly from the broader set. Statistical comparisons and targeted human checks support that these differences are actionable for oversight and audit prioritization, while revealing separations that simpler metrics like the coefficient of variation miss.

Core claim

The Structural Heterogeneity Index, instantiated for payments as PHI, integrates Gaussian Mixture Model parameters with non-parametric statistics to form a decomposable measure of one-dimensional samples that captures modality, asymmetry, tail behaviour, and structural dispersion, thereby detecting latent payment regimes without labelled examples of error or fraud.

What carries the argument

The Payment Heterogeneity Index (PHI), a composite statistic that aggregates four interpretable components (modality, asymmetry, tail behaviour, and structural dispersion) derived from GMM parameters and non-parametric statistics to characterise payment structure and latent regimes.

If this is right

  • High-PHI suppliers exhibit payment patterns that differ statistically from the remaining population.
  • The index ranks cases for targeted human verification and audit in high-volume settings.
  • PHI detects regime separations that remain hidden when using the coefficient of variation alone.
  • The framework supplies a transparent, decomposable score for procurement integrity oversight.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same composite approach could be tested on other high-volume transaction streams such as invoice or expense data.
  • Adding time-series components might allow the index to flag emerging changes in payment structure.
  • Public bodies could incorporate PHI thresholds into routine monitoring dashboards to allocate limited audit resources.

Load-bearing premise

The four selected components are assumed to jointly capture meaningful latent payment regimes in an unsupervised manner, with detected differences more likely to reflect oversight-relevant anomalies than ordinary business variation.

What would settle it

An independent audit of the suppliers flagged by the highest PHI scores that finds no higher incidence of errors, irregularities, or fraud than in the low-PHI population would undermine the claim that the index supports effective prioritisation.

Figures

Figures reproduced from arXiv: 2605.12547 by Kyriakos Christodoulides.

Figure 1
Figure 1. Figure 1: Distribution of robustly standardised payment amounts, shown as a histogram [PITH_FULL_IMAGE:figures/full_fig_p010_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Analytical sample characteristics: Distribution of payment counts per high [PITH_FULL_IMAGE:figures/full_fig_p017_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Distribution of supplier-level PHI scores. Vertical dashed lines indicate the 70th [PITH_FULL_IMAGE:figures/full_fig_p018_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: GMM profile for Supplier A. A single dominant payment regime is observed, [PITH_FULL_IMAGE:figures/full_fig_p021_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: GMM profile for Supplier B. Two distinct payment regimes are identified, with [PITH_FULL_IMAGE:figures/full_fig_p022_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Probabilistic activity zone decomposition for Supplier B, showing the service [PITH_FULL_IMAGE:figures/full_fig_p024_6.png] view at source ↗
Figure 6
Figure 6. Figure 6: Probabilistic activity zone decomposition for Supplier B, showing the service [PITH_FULL_IMAGE:figures/full_fig_p023_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Categorical boxplot for Supplier B, showing the financial scale, median tenden [PITH_FULL_IMAGE:figures/full_fig_p024_7.png] view at source ↗
Figure 7
Figure 7. Figure 7: Categorical boxplot for Supplier B, showing the financial scale, median tenden [PITH_FULL_IMAGE:figures/full_fig_p023_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: GMM profile for Supplier C. Three closely spaced payment regimes are observed [PITH_FULL_IMAGE:figures/full_fig_p025_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: GMM profile for Supplier D. Four distinct but closely aligned payment regimes [PITH_FULL_IMAGE:figures/full_fig_p027_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: GMM profile for the maximum PHI supplier. A dominant operational regime [PITH_FULL_IMAGE:figures/full_fig_p028_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Scatter plot of GMM centre payment amounts versus supplier PHI scores, [PITH_FULL_IMAGE:figures/full_fig_p031_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Proportion of each risk tier with GMM centres within [PITH_FULL_IMAGE:figures/full_fig_p032_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Permutation null distribution of the count of high-tier observations falling [PITH_FULL_IMAGE:figures/full_fig_p033_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Permutation null distribution of the count of high-tier observations falling [PITH_FULL_IMAGE:figures/full_fig_p034_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Box-and-whisker plot with overlaid strip plot showing the distribution of [PITH_FULL_IMAGE:figures/full_fig_p035_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Empirical cumulative distribution functions (ECDFs) of [PITH_FULL_IMAGE:figures/full_fig_p036_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Prevalence (proportion of each directorate) and absolute count of high-PHI [PITH_FULL_IMAGE:figures/full_fig_p040_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Mean and median log-contributions of PHI components across the full sup [PITH_FULL_IMAGE:figures/full_fig_p042_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Spearman correlation matrix of PHI components and overall score. Tail Be [PITH_FULL_IMAGE:figures/full_fig_p043_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: PHI structural driver decomposition for the top 12 suppliers, expressed as [PITH_FULL_IMAGE:figures/full_fig_p043_20.png] view at source ↗
read the original abstract

Public procurement is vulnerable to error, fraud, and corruption, particularly as high transaction volumes overwhelm oversight. While research often focuses on tender-stage anomalies, post-award payment monitoring remains underexplored. Since labelled datasets are rare and methods like Benford's Law face restrictive assumptions, there is a need for interpretable, unsupervised frameworks for high-volume procurement oversight and decision support. This paper introduces the Structural Heterogeneity Index (SHI), a composite statistic for one-dimensional samples, and its payment-specific instantiation, the Payment Heterogeneity Index (PHI), characterising payment structure and latent regimes. It incorporates Gaussian Mixture Model (GMM) parameters alongside non-parametric statistics, integrating four interpretable components: modality, asymmetry, tail behaviour, and structural dispersion. Uniquely, the tail-behaviour component captures both distributional heaviness and extreme-value concentration, while structural-dispersion combines the variability, prevalence, and separation of latent payment regimes. Applied to UK municipal procurement data, PHI identifies a financially significant cohort (0.6\% of suppliers; 10.1\% of high-volume vendors) with structurally distinct payment patterns. Statistical testing further supports these differences, and targeted human verification confirms the plausibility of prioritised cases. Comparative analysis shows PHI reveals regime separation obscured by the Coefficient of Variation ($\rho = 0.310$). PHI provides a transparent, decomposable, and computationally lightweight framework for procurement integrity oversight and targeted audit prioritisation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper introduces the Structural Heterogeneity Index (SHI) as a general composite statistic for one-dimensional samples and its payment-specific instantiation, the Payment Heterogeneity Index (PHI). PHI integrates Gaussian Mixture Model parameters with non-parametric statistics across four components—modality, asymmetry, tail behaviour (heaviness plus extreme-value concentration), and structural dispersion (variability, prevalence, and separation of latent regimes)—to characterize payment structures in an unsupervised manner. Applied to UK municipal procurement data, it identifies a financially significant cohort (0.6% of suppliers; 10.1% of high-volume vendors) exhibiting structurally distinct payment patterns, with supporting statistical tests for distributional differences, targeted human verification of plausibility, and a demonstration that PHI captures regime separation not reflected by the Coefficient of Variation (ρ = 0.310).

Significance. If the central claim holds, PHI would supply a transparent, decomposable, and computationally lightweight unsupervised tool for post-award payment monitoring in high-volume procurement settings where labelled fraud or error data are scarce. The integration of parametric and non-parametric elements, explicit attention to tail behaviour and structural dispersion, and the reported orthogonality to the Coefficient of Variation constitute genuine strengths. The practical significance for oversight and audit prioritisation, however, depends on establishing that the flagged differences correspond to integrity risks rather than normal business heterogeneity.

major comments (3)
  1. [Methods / PHI definition] The manuscript provides no explicit equations or algorithmic description for constructing the PHI composite from the four components (modality, asymmetry, tail behaviour, structural dispersion). In particular, the normalization, weighting, and scaling of GMM-derived parameters and non-parametric statistics are not specified, which directly affects reproducibility and the identification of the 0.6% cohort.
  2. [Results / Application to UK data] The application section reports statistical tests confirming distributional differences and human verification of plausibility for the prioritised cases, yet supplies no labelled data on errors or fraud and no baseline comparisons against alternative unsupervised methods. Consequently, it is not shown that the flagged suppliers are enriched for integrity issues rather than reflecting legitimate variation by contract type, industry, or scale.
  3. [PHI components] The tail-behaviour component is described as capturing both distributional heaviness and extreme-value concentration, and structural dispersion as combining variability, prevalence, and separation; however, the precise non-parametric statistics chosen for each sub-element and their integration into the composite are not detailed, leaving the load-bearing claim of 'structurally distinct payment patterns' difficult to evaluate.
minor comments (2)
  1. [Comparative analysis] The correlation coefficient ρ = 0.310 with the Coefficient of Variation should specify the exact measure (Pearson, Spearman, etc.) and the sample on which it is computed.
  2. [Abstract / Data description] The abstract and results would benefit from a brief statement of the number of suppliers, transactions, and time period covered by the UK municipal dataset to allow readers to gauge the scale of the 0.6% cohort.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments, which identify key areas for improving transparency and contextualizing the unsupervised results. We will revise the manuscript to add explicit equations, algorithmic details, and expanded discussion of limitations and alternative methods. Point-by-point responses are provided below.

read point-by-point responses

  1. Referee: The manuscript provides no explicit equations or algorithmic description for constructing the PHI composite from the four components (modality, asymmetry, tail behaviour, structural dispersion). In particular, the normalization, weighting, and scaling of GMM-derived parameters and non-parametric statistics are not specified, which directly affects reproducibility and the identification of the 0.6% cohort.

    Authors: We agree this detail is essential for reproducibility. The revised manuscript will include a new subsection with full algorithmic pseudocode and equations: each component will be defined mathematically (e.g., modality via GMM BIC differences and Hartigan's dip test; asymmetry via skewness and GMM mean separation), followed by explicit min-max normalization to [0,1], equal weighting across the four components, and linear aggregation to PHI. This will enable exact replication of the cohort. revision: yes

  2. Referee: The application section reports statistical tests confirming distributional differences and human verification of plausibility for the prioritised cases, yet supplies no labelled data on errors or fraud and no baseline comparisons against alternative unsupervised methods. Consequently, it is not shown that the flagged suppliers are enriched for integrity issues rather than reflecting legitimate variation by contract type, industry, or scale.

    Authors: We acknowledge the limitation: labelled integrity data is scarce, which is why the paper develops an unsupervised method. The reported KS tests, human plausibility checks, and low correlation with CV (ρ=0.310) demonstrate that PHI isolates regime structure beyond simple dispersion. In revision we will add baseline comparisons (e.g., against isolation forest and k-means on payment vectors) and a limitations subsection discussing potential confounders such as contract type. Direct enrichment for fraud cannot be shown without labels, but the framework is positioned for prioritization rather than definitive detection. revision: partial

  3. Referee: The tail-behaviour component is described as capturing both distributional heaviness and extreme-value concentration, and structural dispersion as combining variability, prevalence, and separation; however, the precise non-parametric statistics chosen for each sub-element and their integration into the composite are not detailed, leaving the load-bearing claim of 'structurally distinct payment patterns' difficult to evaluate.

    Authors: We will expand the methods section with precise specifications: tail behaviour uses the Hill estimator for heaviness and a top-5% concentration ratio; structural dispersion combines within-component variance, GMM mixing weights for prevalence, and pairwise Wasserstein distance for separation. Integration equations will show how these sub-elements are scaled and combined with GMM parameters into the four components and final PHI score. revision: yes

Circularity Check

0 steps flagged

No significant circularity; PHI is a definitional composite with independent application

full rationale

The paper introduces PHI as a new composite statistic explicitly defined from GMM parameters and chosen non-parametric measures across four components (modality, asymmetry, tail behaviour, structural dispersion). This is a transparent construction of an index rather than a derivation that reduces to its inputs by construction or renames a fitted result as a prediction. Application to UK procurement data, identification of the 0.6% cohort, statistical tests for differences, and human verification are downstream uses of the defined index; they do not feed back into its definition. No self-citation chains, uniqueness theorems, or ansatzes smuggled via prior work are referenced in the abstract or description. The framework remains self-contained against external benchmarks as an unsupervised oversight tool, with the reported orthogonality to CV (rho=0.310) providing an independent comparative check.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

Abstract-only review limits visibility into exact construction; the index rests on standard statistical assumptions plus choices about component integration that are not detailed here.

free parameters (1)
  • Component integration weights or scaling factors
    Abstract does not specify how modality, asymmetry, tail behaviour, and structural dispersion are numerically combined into the final PHI value.
axioms (1)
  • domain assumption Gaussian Mixture Models can reliably identify latent payment regimes in procurement transaction data
    The framework uses GMM parameters as a core input to the index.

pith-pipeline@v0.9.0 · 5801 in / 1433 out tokens · 56639 ms · 2026-05-20T23:02:28.889157+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

32 extracted references · 32 canonical work pages

  1. [1]

    , year =

    Fazekas, Mihály and Tóth, István János and King, Lawrence P. , year =. Anatomy of Grand Corruption: A Composite Corruption Risk Index Based on Objective Data , journal =

    work page

  2. [2]

    Uncovering High-Level Corruption: Cross-National Objective Corruption Risk Indicators Using Public Procurement Data , journal =

    Fazekas, Mihály and Kocsis, Gábor , year =. Uncovering High-Level Corruption: Cross-National Objective Corruption Risk Indicators Using Public Procurement Data , journal =

    work page

  3. [3]

    Anomaly Detection: A Survey , journal =

    Chandola, Varun and Banerjee, Arindam and Kumar, Vipin , year =. Anomaly Detection: A Survey , journal =

    work page

  4. [4]

    Detecting Fraud in Public Procurement: A

    Tyska Carvalho, Jônata and Castro, Márcio and Machado dos Santos, Matheus and Ferrão, Lívia and Schmitz, Fernando Augusto , year =. Detecting Fraud in Public Procurement: A. Anais do XXXIX Simpósio Brasileiro de Banco de Dados (SBBD 2024) , pages =

    work page 2024

  5. [5]

    Public Procurement Fraud Detection and Artificial Intelligence Techniques: A Literature Review , booktitle =

    Nai, Roberto and Sulis, Emilio and Meo, Rosa , year =. Public Procurement Fraud Detection and Artificial Intelligence Techniques: A Literature Review , booktitle =

    work page

  6. [6]

    Prediction of Public Procurement Corruption Indices Using Machine Learning Methods , booktitle =

    Rabuzin, Kornelije and Modrušan, Nikola , year =. Prediction of Public Procurement Corruption Indices Using Machine Learning Methods , booktitle =

    work page

  7. [7]

    2025 , url =

    Payments over. 2025 , url =

    work page 2025

  8. [8]

    , title =

    Nigrini, Mark J. , title =. 2012 , publisher =

    work page 2012

  9. [9]

    Dempster, A. P. and Laird, N. M. and Rubin, D. B. , title =. Journal of the Royal Statistical Society: Series B (Methodological) , volume =. 1977 , doi =

    work page 1977

  10. [10]

    and Peel, David , title =

    McLachlan, Geoffrey J. and Peel, David , title =. 2000 , doi =

    work page 2000

  11. [11]

    The Annals of Statistics , volume =

    Schwarz, Gideon , title =. The Annals of Statistics , volume =. 1978 , doi =

    work page 1978

  12. [12]

    American Economic Journal: Economic Policy , volume =

    Palguta, Ján and Pertold, Filip , title =. American Economic Journal: Economic Policy , volume =. 2017 , doi =

    work page 2017

  13. [13]

    American Economic Review , volume =

    Bandiera, Oriana and Prat, Andrea and Valletti, Tommaso , title =. American Economic Review , volume =. 2009 , doi =

    work page 2009

  14. [14]

    Information Systems Management , volume =

    Janssen, Marijn and Charalabidis, Yannis and Zuiderwijk, Anneke , title =. Information Systems Management , volume =. 2012 , doi =

    work page 2012

  15. [15]

    Computers & Security , volume =

    West, Jarrod and Bhattacharya, Maumita , title =. Computers & Security , volume =. 2016 , doi =

    work page 2016

  16. [16]

    Ngai, E. W. T. and Hu, Yong and Wong, Y. H. and Chen, Yijun and Sun, Xin , title =. Decision Support Systems , volume =. 2011 , doi =

    work page 2011

  17. [17]

    and Jaeger, Paul T

    Bertot, John C. and Jaeger, Paul T. and Grimes, Justin M. , title =. Government Information Quarterly , volume =. 2010 , doi =

    work page 2010

  18. [18]

    The Law and Economics of Public Procurement Reform , editor =

    Spagnolo, Giancarlo , title =. The Law and Economics of Public Procurement Reform , editor =

    work page

  19. [19]

    and Austin, Jim , title =

    Hodge, Victoria J. and Austin, Jim , title =. Artificial Intelligence Review , volume =. 2004 , doi =

    work page 2004

  20. [20]

    Management Science , volume =

    Coviello, Decio and Guglielmo, Andrea and Spagnolo, Giancarlo , title =. Management Science , volume =. 2018 , doi =

    work page 2018

  21. [21]

    and Varoquaux, G

    Pedregosa, F. and Varoquaux, G. and Gramfort, A. and Michel, V. and Thirion, B. and Grisel, O. and Blondel, M. and Prettenhofer, P. and Weiss, R. and Dubourg, V. and Vanderplas, J. and Passos, A. and Cournapeau, D. and Brucher, M. and Perrot, M. and Duchesnay, E. , title =. Journal of Machine Learning Research , volume =. 2011 , url =

    work page 2011

  22. [22]

    and Iossa, Elisabetta and Mollisi, Vincenzo and Spagnolo, Giancarlo , title =

    Decarolis, Francesco and Giuffrida, Leonardo M. and Iossa, Elisabetta and Mollisi, Vincenzo and Spagnolo, Giancarlo , title =. The Journal of Law, Economics, and Organization , volume =. 2020 , doi =

    work page 2020

  23. [23]

    Journal of the European Economic Association , volume =

    Szucs, Ferenc , title =. Journal of the European Economic Association , volume =. 2024 , doi =

    work page 2024

  24. [24]

    Journal of Political Economy , volume =

    Carril, Rodrigo , title =. Journal of Political Economy , volume =. 2021 , doi =

    work page 2021

  25. [25]

    2021 , number =

    Bandiera, Oriana and Bosio, Erika and Spagnolo, Giancarlo , title =. 2021 , number =

    work page 2021

  26. [26]

    Data Science and Management , year =

    Chen, Yisong and Zhao, Chuqing and Xu, Yixin and Nie, Chuanhao and Zhang, Yixin , title =. Data Science and Management , year =. doi:10.1016/j.dsm.2025.08.002 , note =

    work page doi:10.1016/j.dsm.2025.08.002 2025

  27. [27]

    Journal of Economic Behavior & Organization , year =

    Caglayan, Mustafa and Talavera, Oleksandr and Zhang, Wei , title =. Journal of Economic Behavior & Organization , year =

    work page

  28. [28]

    Harvard Journal of Law & Technology , volume=

    Counterfactual explanations without opening the black box: Automated decisions and the GDPR , author=. Harvard Journal of Law & Technology , volume=

    work page

  29. [29]

    Journal of Business Research , volume=

    Factors influencing big data decision-making quality , author=. Journal of Business Research , volume=. 2017 , publisher=

    work page 2017

  30. [30]

    International Journal of Project Management , volume=

    Corruption in public projects and megaprojects: There is an elephant in the room! , author=. International Journal of Project Management , volume=. 2017 , publisher=

    work page 2017

  31. [31]

    International Conference on Information Processing and Management of Uncertainty in Knowledge-Based Systems , pages=

    Human-in-the-Loop Anomaly Detection and Explanation , author=. International Conference on Information Processing and Management of Uncertainty in Knowledge-Based Systems , pages=. 2022 , publisher=

    work page 2022

  32. [32]

    2017 IEEE International Conference on Big Data (Big Data) , pages=

    Human-in-the-loop learning-oriented anomaly detection , author=. 2017 IEEE International Conference on Big Data (Big Data) , pages=. 2017 , publisher=

    work page 2017