pith. sign in

arxiv: 2405.20191 · v1 · submitted 2024-05-30 · 📊 stat.AP · econ.EM· stat.CO

Multidimensional spatiotemporal clustering -- An application to environmental sustainability scores in Europe

Caterina Morelli , Simone Boccaletti , Paolo Maranzano , Philipp Otto This is my paper

Pith reviewed 2026-05-24 01:25 UTC · model grok-4.3

classification 📊 stat.AP econ.EMstat.CO
keywords spatiotemporal clusteringESG ratingssustainability performanceEuropean firmshierarchical clusteringspatial analysistemporal dynamicsWestern Europe
0
0 comments X

The pith

European firms form distinct sustainability clusters that span countries and industries when location and time trends are jointly analyzed.

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

The paper uses a modified hierarchical clustering method on MSCI ESG ratings for Western European companies from 2013 to 2023, together with their geographic coordinates. It first performs spatial clustering that merges sustainability scores with location information, then extends the approach to spatiotemporal clustering that also incorporates how those scores evolve over time. The resulting groups cut across national borders and industrial sectors yet display clear differences in average sustainability levels. In the spatiotemporal version the clusters overlap substantially in geography, which the authors interpret as evidence that temporal changes add meaningful structure beyond static spatial patterns alone. The work aims to show how such groupings can help distinguish firms facing different sustainability risks in different places and periods.

Core claim

A modified version of the Chavent et al. (2018) hierarchical algorithm that fuses spatial dissimilarities with the time dynamics of multiple sustainability features produces clusters of Western European firms that exhibit homogeneous sustainability performance; these clusters are cross-national and cross-industry in composition, display marked differences in ESG scores, and show a high degree of geographical overlap when temporal information is included.

What carries the argument

Modified Chavent et al. (2018) hierarchical algorithm that combines spatial dissimilarities with temporal dynamics of multiple sustainability features.

If this is right

  • Cross-national and cross-industry clusters emerge with clear differences in sustainability scores.
  • Spatiotemporal clusters exhibit high geographical overlap, showing that temporal dynamics matter in a multidimensional setting.
  • The diversity of ESG ratings across Western Europe can be captured through these groupings.
  • Practitioners and policymakers gain a tool for assessing companies subject to different sustainability-linked risks in different areas.

Where Pith is reading between the lines

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

  • Policy interventions could be designed around cluster membership rather than national or sectoral boundaries.
  • Static annual ESG snapshots may understate risk if temporal trajectories differ inside the same geographic area.
  • The same method could be applied to other multi-year rating datasets to test whether temporal structure consistently dominates geography.

Load-bearing premise

The chosen distance measures and linkage criteria correctly combine spatial locations with temporal changes in sustainability scores without creating spurious groups.

What would settle it

Re-running the same algorithm on the identical ESG dataset but with a different distance or linkage choice yields clusters whose membership and score differences change substantially.

Figures

Figures reproduced from arXiv: 2405.20191 by Caterina Morelli, Paolo Maranzano, Philipp Otto, Simone Boccaletti.

Figure 1
Figure 1. Figure 1: Left panel: Number of observations per year between 2013 and 2023. We report [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Comparison of proportion of explained pseudo inertia from each dissimilarity matrices [PITH_FULL_IMAGE:figures/full_fig_p018_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Hyperparameters selection in spatial clustering. Left panel: increment in the weighted [PITH_FULL_IMAGE:figures/full_fig_p019_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Map of spatial clusters for 2023 using K∗ = 5 and α ∗ 5 = 0.40. Clusters are computed using two dissimilarity matrices: Geodetic spatial distance and Euclidean distance of ESG, Environmental and Carbon Emission scores. 20 [PITH_FULL_IMAGE:figures/full_fig_p020_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Cluster-specific centroids of the three economic variables used to define the spatial [PITH_FULL_IMAGE:figures/full_fig_p022_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Spatial clusters composition by country (specified using ISO code) and industry for [PITH_FULL_IMAGE:figures/full_fig_p023_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Top panel: proportion of explained pseudo inertia contained in each dissimilarity [PITH_FULL_IMAGE:figures/full_fig_p025_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Hyperparameters selection in spatiotemporal clustering. Left panel: increment in the [PITH_FULL_IMAGE:figures/full_fig_p026_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Map of spatiotemporal clusters between 2013 and 2023. Clusters are computed using [PITH_FULL_IMAGE:figures/full_fig_p028_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Spatiotemporal clusters (from 2013 to 2023) composition by country (specified using [PITH_FULL_IMAGE:figures/full_fig_p030_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Time series 2013-2023 of centroids (solid lines) and quantiles (dashed lines) of the [PITH_FULL_IMAGE:figures/full_fig_p031_11.png] view at source ↗
read the original abstract

The assessment of corporate sustainability performance is extremely relevant in facilitating the transition to a green and low-carbon intensity economy. However, companies located in different areas may be subject to different sustainability and environmental risks and policies. Henceforth, the main objective of this paper is to investigate the spatial and temporal pattern of the sustainability evaluations of European firms. We leverage on a large dataset containing information about companies' sustainability performances, measured by MSCI ESG ratings, and geographical coordinates of firms in Western Europe between 2013 and 2023. By means of a modified version of the Chavent et al. (2018) hierarchical algorithm, we conduct a spatial clustering analysis, combining sustainability and spatial information, and a spatiotemporal clustering analysis, which combines the time dynamics of multiple sustainability features and spatial dissimilarities, to detect groups of firms with homogeneous sustainability performance. We are able to build cross-national and cross-industry clusters with remarkable differences in terms of sustainability scores. Among other results, in the spatio-temporal analysis, we observe a high degree of geographical overlap among clusters, indicating that the temporal dynamics in sustainability assessment are relevant within a multidimensional approach. Our findings help to capture the diversity of ESG ratings across Western Europe and may assist practitioners and policymakers in evaluating companies facing different sustainability-linked risks in different areas.

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

2 major / 2 minor

Summary. The paper applies a modified version of the Chavent et al. (2018) hierarchical clustering algorithm to a dataset of MSCI ESG ratings and geographical coordinates for Western European firms (2013–2023). It performs spatial clustering that fuses sustainability scores with spatial information and spatiotemporal clustering that additionally incorporates temporal dynamics across multiple sustainability features, with the goal of identifying groups of firms exhibiting homogeneous sustainability performance. The authors report the discovery of cross-national and cross-industry clusters showing remarkable differences in sustainability scores; the spatiotemporal analysis additionally reveals a high degree of geographical overlap among clusters, which the authors interpret as evidence that temporal dynamics are relevant within a multidimensional approach.

Significance. If the reported clusters prove robust, the work would offer a practical demonstration of how spatial and temporal dimensions can be jointly incorporated into clustering of ESG data, potentially aiding practitioners and policymakers in identifying region-specific sustainability risks. The data-driven nature of the analysis and the focus on real-world corporate sustainability metrics are positive features; however, the absence of any reported validation metrics, cluster-stability diagnostics, or sensitivity checks substantially reduces the immediate utility of the findings.

major comments (2)
  1. [Abstract / Methods] Abstract and Methods: the central claim that the modified Chavent et al. (2018) algorithm produces clusters with 'remarkable differences' and 'high degree of geographical overlap' rests on the untested assumption that the chosen distance measures and linkage criteria correctly fuse spatial dissimilarities with multi-feature temporal dynamics without introducing artifacts. No sensitivity analysis to alternative distances or linkages is described, nor are any quantitative cluster-validity indices (e.g., silhouette scores, adjusted Rand index against external labels) reported.
  2. [Results] Results: the abstract states that cross-national and cross-industry clusters exhibit 'remarkable differences in terms of sustainability scores,' yet provides no numerical effect sizes, statistical tests, or comparison against a null model of random spatial assignment. Without these, it is impossible to judge whether the observed separation exceeds what would be expected from the spatial autocorrelation already present in the data.
minor comments (2)
  1. [Data] The abstract refers to 'Western Europe' without listing the exact countries or the total number of firms included; this information should be stated explicitly in the data-description section.
  2. [Methods] Notation for the spatiotemporal weighting parameter and the precise definition of the modified dissimilarity measure should be introduced with an equation rather than left implicit.

Simulated Author's Rebuttal

2 responses · 0 unresolved

Thank you for the opportunity to respond to the referee's constructive report. The points raised identify clear opportunities to strengthen the robustness and interpretability of the clustering results. We address each major comment below and will revise the manuscript to incorporate the suggested analyses.

read point-by-point responses

  1. Referee: [Abstract / Methods] Abstract and Methods: the central claim that the modified Chavent et al. (2018) algorithm produces clusters with 'remarkable differences' and 'high degree of geographical overlap' rests on the untested assumption that the chosen distance measures and linkage criteria correctly fuse spatial dissimilarities with multi-feature temporal dynamics without introducing artifacts. No sensitivity analysis to alternative distances or linkages is described, nor are any quantitative cluster-validity indices (e.g., silhouette scores, adjusted Rand index against external labels) reported.

    Authors: We agree that the absence of sensitivity checks and validity indices leaves the fusion of spatial and temporal components open to the concern of artifacts. The manuscript follows the mixed-data hierarchical procedure of Chavent et al. (2018) with an added spatial dissimilarity term, but does not test alternatives. In the revision we will add a dedicated sensitivity subsection that recomputes the clusters under alternative distance measures (including variants of Gower distance) and linkage criteria, and we will report silhouette scores together with a brief discussion of cluster stability. We will also expand the Methods section to justify the original parameter choices more explicitly. revision: yes

  2. Referee: [Results] Results: the abstract states that cross-national and cross-industry clusters exhibit 'remarkable differences in terms of sustainability scores,' yet provides no numerical effect sizes, statistical tests, or comparison against a null model of random spatial assignment. Without these, it is impossible to judge whether the observed separation exceeds what would be expected from the spatial autocorrelation already present in the data.

    Authors: We accept that descriptive presentation of the clusters is insufficient to substantiate the strength of the reported differences. The revised Results section will include quantitative effect sizes (cluster-wise mean ESG differences with confidence intervals) and formal statistical tests (multivariate ANOVA or non-parametric equivalents) for sustainability-score separation. We will also add a permutation test that randomizes firm locations while holding ESG features fixed, thereby providing a direct comparison against a spatially autocorrelated null; if computational constraints arise we will at least report the effect sizes and tests against a non-spatial baseline clustering. revision: yes

Circularity Check

0 steps flagged

No circularity: data-driven clustering with external method reference

full rationale

The paper applies a modified version of the Chavent et al. (2018) hierarchical clustering algorithm to MSCI ESG ratings and firm coordinates. The central outputs—cross-national clusters and spatiotemporal groupings—are produced by running the algorithm on the external dataset; no equations define a target quantity in terms of itself, no fitted parameters are relabeled as predictions, and the cited method is from unrelated authors. The analysis remains self-contained against the input data without self-referential reduction.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

Abstract-only review limits visibility into exact parameters; typical clustering requires choices for number of clusters, feature weights, and distance metrics that are not detailed here.

free parameters (2)
  • number of clusters
    Chosen to produce groups with homogeneous sustainability performance; value not specified in abstract.
  • spatial-temporal weighting parameter
    Balances spatial dissimilarities against temporal dynamics of ESG features; not quantified in abstract.
axioms (1)
  • domain assumption The hierarchical algorithm of Chavent et al. (2018) can be extended to spatiotemporal data while preserving cluster interpretability.
    Invoked as the base method for both spatial and spatiotemporal analyses.

pith-pipeline@v0.9.0 · 5766 in / 1252 out tokens · 22462 ms · 2026-05-24T01:25:05.231142+00:00 · methodology

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.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Mapping climate change awareness through spatial hierarchical clustering

    stat.AP 2024-09 unverdicted novelty 3.0

    Geographically-informed hierarchical clustering on climate change awareness data yields more stable and compact country groups than non-spatial clustering, with Western nations showing high awareness and Asian, Africa...

Reference graph

Works this paper leans on

30 extracted references · 30 canonical work pages · cited by 1 Pith paper

  1. [1]

    Alfalih, A. A. (2023). ESG disclosure practices and financial performance: a general and sector analysis of SP-500 non-financial companies and the moderating effect of economic conditions. Journal of Sustainable Finance and Investment , 13(4):1506 –

    work page 2023

  2. [2]

    Amores-Salvad´ o, J., Martin-de Castro, G., and Albertini, E. (2023). Walking the talk, but above all, talking the walk: Looking green for market stakeholder engagement. Corporate Social Responsibility and Environmental Management , 30(1):431 –

    work page 2023

  3. [3]

    Bucci, A., Ippoliti, L., and Valentini, P. (2023). Analysing spatiotemporal patterns of Covid-19 confirmed deaths at the NUTS-2 regional level. Regional Statistics, 13(2):214 –

    work page 2023

  4. [4]

    Chavent, M., Kuentz-Simonet, V., Labenne, A., and Saracco, J. (2018). ClustGeo: an R 34 package for hierarchical clustering with spatial constraints. Computational Statistics, 33:1 –

    work page 2018

  5. [5]

    Chen, Z., Hu, L., He, X., Liu, Z., Chen, D., and Wang, W. (2022b). Green financial reform and corporate ESG performance in China: Empirical evidence from the green financial reform and innovation pilot zone. International Journal of Environmental Research and Public Health , 19(22). Chipalkatti, N., Le, Q. V., and Rishi, M. (2021). Sustainability and soci...

    work page 2021

  6. [6]

    2022 UK greenhouse gas emissions, provisional figures

    DESNZ (2023). 2022 UK greenhouse gas emissions, provisional figures. National Statistics. Fu, T. and Li, J. (2023). An empirical analysis of the impact of ESG on financial performance: the moderating role of digital transformation. Frontiers in Environmental Science,

    work page 2023

  7. [7]

    R., Klotzle, M

    Gonzaga, B. R., Klotzle, M. C., Brugni, T. V., Rakos, I.-S., Cioca, I. C., Barbu, C.-M., and Cucerzan, T. (2024). The ESG patterns of emerging-market companies: Are there differences in their sustainable behavior after covid-19? Sustainability (Switzerland), 16(2). He, X., Jing, Q., and Chen, H. (2023a). The impact of environmental tax laws on heavy- poll...

    work page 2024

  8. [8]

    M., and Cernat-Gruici, B

    Iamandi, I.-E., Constantin, L.-G., Munteanu, S. M., and Cernat-Gruici, B. (2019). Mapping the ESG behavior of european companies. a holistic kohonen approach. Sustainability (Switzer- land), 11(12). Ishizaka, A., Lokman, B., and Tasiou, M. (2021). A stochastic multi-criteria divisive hierarchical clustering algorithm. Omega (United Kingdom),

    work page 2019

  9. [9]

    I conti delle emissioni atmosferiche (aea)

    Istat (2022). I conti delle emissioni atmosferiche (aea). Istituto Nazionale di Statistica . Jaya, I., Ruchjana, B., Andriyana, Y., and Agata, R. (2019). Clustering with spatial constraints: The case of diarrhea in bandung city, indonesia. volume

    work page 2022

  10. [10]

    and Rousseeuw, P

    Kaufman, L. and Rousseeuw, P. J. (1990). Finding Groups in Data: An Introduction to Cluster Analysis. Wiley, United States. Lai, X. and Zhang, F. (2022). Can ESG certification help company get out of over-indebtedness? evidence from China. Pacific Basin Finance Journal ,

    work page 1990

  11. [11]

    and Li, S

    Li, J. and Li, S. (2022). Environmental protection tax, corporate ESG performance, and green technological innovation. Frontiers in Environmental Science,

    work page 2022

  12. [12]

    Lian, Y., Li, Y., and Cao, H. (2023). How does corporate ESG performance affect sustainable development: A green innovation perspective. Frontiers in Environmental Science,

    work page 2023

  13. [13]

    A., Santos-Rodrigues, H., Qui˜ no´ a-Pi˜ neiro, L., and Pi˜ neiro-Chousa, J

    L´ opez-Cabarcos, M. A., Santos-Rodrigues, H., Qui˜ no´ a-Pi˜ neiro, L., and Pi˜ neiro-Chousa, J. (2023). How to explain stock returns of utility companies from an environmental, social and corporate governance perspective. Corporate Social Responsibility and Environmental Management, 30(5):2278 –

    work page 2023

  14. [14]

    Ma, J., Gao, D., and Sun, J. (2022). Does ESG performance promote total factor productivity? evidence from China. Frontiers in Ecology and Evolution ,

    work page 2022

  15. [15]

    and Franses, P

    Mattera, R. and Franses, P. H. (2023). Are african business cycles synchronized? evidence from spatio-temporal modeling. Economic Modelling,

    work page 2023

  16. [16]

    36 Mo, Y., Che, Y., and Ning, W. (2023). Digital finance promotes corporate ESG performance: Evidence from China. Sustainability (Switzerland), 15(14). Montero, P. and Vilar, J. A. (2014). Tsclust: An r package for time series clustering. Journal of Statistical Software, 62(1):1 –

    work page 2023

  17. [17]

    K., and Woon, L

    Mukhtar, B., Shad, M. K., and Woon, L. F. (2023). Predicting the effect of environment, social and governance practices on green innovation: An artificial neural network approach. Lecture Notes in Networks and Systems , 550 LNNS:527 –

    work page 2023

  18. [18]

    Ortas, E., ´Alvarez, I., Jaussaud, J., and Garayar, A. (2015). The impact of institutional and social context on corporate environmental, social and governance performance of compa- nies committed to voluntary corporate social responsibility initiatives. Journal of Cleaner Production, 108:673 –

    work page 2015

  19. [19]

    and Ray, K

    Panda, A. and Ray, K. K. (2023). Equity market performance: The role of environmental protection and corporate social responsibility efforts. Business Strategy and Development . Phillips, P. and Sul, D. (2007). Transition modeling and econometric convergence tests. Econo- metrica, 75(6):1771–1855. Phillips, P. and Sul, D. (2009). Economic transition and g...

    work page 2023

  20. [20]

    M., Bernardo, M., and Roman´ ı, J

    Ronalter, L. M., Bernardo, M., and Roman´ ı, J. M. (2023). Quality and environmental man- agement systems as business tools to enhance ESG performance: a cross-regional empirical study. Environment, Development and Sustainability , 25(9):9067 –

    work page 2023

  21. [21]

    Saraswati, E., Ghofar, A., Atmini, S., and Dewi, A. A. (2024). Clustering of companies based on sustainability performance using ESG materiality approach: Evidence from indonesia. International Journal of Energy Economics and Policy , 14(2):112 –

    work page 2024

  22. [22]

    and Ta¸ skın, D

    Sariyer, G. and Ta¸ skın, D. (2022). Clustering of firms based on environmental, social, and governance ratings: Evidence from bist sustainability index. Borsa Istanbul Review, 22:S180 – S188. 37 UBA (2024). Annual carbon dioxide emissions in Germany from 1990 to

    work page 2022

  23. [23]

    the environment agency of the German government . Wang, Q. (2023). Herding behavior and the dynamics of ESG performance in the european banking industry. Finance Research Letters,

    work page 2023

  24. [24]

    Wang, X., Elahi, E., and Khalid, Z. (2022). Do green finance policies foster environmental, social, and governance performance of corporate? International Journal of Environmental Research and Public Health, 19(22). Ward, J. H. (1963). Hierarchical grouping to optimize an objective function. Journal of the American Statistical Association, 58(301):236 –

    work page 2022

  25. [25]

    and Tham, J

    Wu, Y. and Tham, J. (2023). The impact of executive green incentives and top management team characteristics on corporate value in China: The mediating role of environment, social and government performance. Sustainability (Switzerland), 15(16). Xue, Q., Wang, H., and Bai, C. (2023). Local green finance policies and corporate ESG performance. Internationa...

    work page 2023

  26. [26]

    and Xiao, K

    Yu, X. and Xiao, K. (2022). Does ESG performance affect firm value? evidence from a new ESG-scoring approach for chinese enterprises. Sustainability (Switzerland), 14(24). Zhang, D. (2022). Environmental regulation and firm product quality improvement: How does the greenwashing response? International Review of Financial Analysis ,

    work page 2022

  27. [27]

    Zhang, D., Meng, L., and Zhang, J. (2023). Environmental subsidy disruption, skill premiums and ESG performance. International Review of Financial Analysis ,

    work page 2023

  28. [28]

    B., Zhang, Y., and Lin, M.-S

    Zhang, H., Ho, T. B., Zhang, Y., and Lin, M.-S. (2006). Unsupervised feature extraction for time series clustering using orthogonal wavelet transform. Informatica (Ljubljana), 30(3):305 –

    work page 2006

  29. [29]

    and Cai, L

    Zhao, X. and Cai, L. (2023). Digital transformation and corporate ESG: Evidence from China. Finance Research Letters,

    work page 2023

  30. [30]

    Zheng, M., Feng, G.-F., Jiang, R.-A., and Chang, C.-P. (2023). Does environmental, social, and governance performance move together with corporate green innovation in China? Business Strategy and the Environment , 32(4):1670 –

    work page 2023