Pith. sign in

REVIEW 3 major objections 2 minor 43 references

Predicting the 2026 FIFA World Cup with Sufficient Dimension Reduction of Elo Rating Histories

T0 review · 3 major / 2 minor · reviewed 2026-06-25 · grok-4.3

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Sufficient dimension reduction on recent Elo rating differences improves Poisson models for forecasting World Cup match outcomes.

desk verdict SDR on Elo histories gives small out-of-sample RPS gains versus baselines on 2018/2022 but the 2026 forecast is untested because of the new 48-team format. read the letter →

arxiv 2606.24171 v1 pith:B457U6XF submitted 2026-06-23 stat.AP

Mina Rezaei , S. Yaser Samadi This is my paper · ORCID
classification stat.AP
keywords sufficientdimensionreductionEloratingsPoissonregressionWorldCupprobabilisticforecastingsportsprediction
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tests whether a short history of Elo rating changes, compressed via categorical sufficient dimension reduction, supplies extra information for predicting goals in FIFA World Cup games. It fits Poisson regressions to the reduced scores and compares the resulting probabilities against eleven other forecasting methods. Out-of-sample tests on the 2018 and 2022 tournaments show lower ranked probability scores for the SDR variants. The comparison matters for the 2026 edition because its larger field and new stage structure may reward forecasts that capture momentum in team form. The central finding is that recent rating trajectories hold predictive value beyond the single latest rating difference.

What carries the argument

Categorical sufficient dimension reduction applied to short histories of Elo rating differences, which extracts informative directions for use as covariates in a Poisson model of match scores.

What would settle it

Direct comparison of ranked probability scores between the SDR-Poisson model and the best traditional model on the complete set of 2026 World Cup matches.

Watch

Extended reading notes

Core claim

Applying sliced inverse regression and sliced average variance estimation to sequences of Elo differences produces low-dimensional predictors that, when inserted into a Poisson double-regression model for home and away goals, yield better out-of-sample ranked probability scores than models relying solely on the current Elo difference or on time-series forecasts of the Elo series itself.

Load-bearing premise

That the predictive advantage observed on past World Cups will continue to hold when the tournament expands to 48 teams and adds a new round of 32.

Editorial extensions

If this is right

  • SDR-based Poisson models achieve lower ranked probability scores than standard Poisson or logistic regressions on 2018 and 2022 World Cup data.
  • Recent Elo histories contain useful information for goal prediction that the current difference alone does not capture.
  • The same reduced predictors can be used to generate full probability distributions for matches in the 2026 tournament.
  • Four variants of SDR (SIR and SAVE) all show gains over baselines including ARIMA, neural nets, and gradient boosting.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the improvement persists, forecasts that ignore rating trajectories will systematically undervalue teams on upward or downward trends.
  • The method could be tested on other rating-based sports by applying SDR to historical rating sequences.
  • With the expanded 48-team format, the extra predictive signal may matter more in the additional knockout rounds.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 2 minor

Summary. The paper proposes modeling team strength for the 2026 FIFA World Cup via categorical sufficient dimension reduction (SDR, using SIR and SAVE) applied to short histories of Elo rating differences, then feeding the reduced scores into a Poisson double-regression model for goal scoring. Eleven models (logistic regression, standard Poisson, ARIMA, neural-network Elo forecasts, gradient boosting, ensembles, and four SDR-Poisson variants) are compared out-of-sample on the 2018 and 2022 World Cups via ranked probability score (RPS); the central claim is that the SDR-Poisson models improve upon baselines, indicating that recent Elo history supplies predictive information beyond the current Elo difference alone.

Significance. If the reported RPS gains hold after proper handling of the new format, the work would demonstrate that low-dimensional summaries of rating trajectories can measurably improve probabilistic forecasts for knockout tournaments. The out-of-sample evaluation on held-out past World Cups and the explicit multi-model comparison (including both parametric SDR and nonparametric baselines) are positive features that strengthen the internal validity of the comparison.

major comments (3)
  1. [Evaluation and Results sections] The evaluation is performed exclusively on the 2018 and 2022 tournaments (both 32-team formats with fixed group stages). No simulation, covariate adjustment, or sensitivity analysis for the 2026 48-team format and new Round-of-32 stage is described; because the central claim concerns prediction for the 2026 edition, this omission is load-bearing.
  2. [Abstract and Results] The abstract and results summary state that SDR-Poisson models improve traditional approaches, yet supply neither the numerical RPS values, the chosen reduced dimension(s), nor any error bars or significance tests; without these quantities the magnitude and reliability of the claimed improvement cannot be assessed.
  3. [Methods (SDR application)] The methods description of the SDR step does not specify the criterion or procedure used to select the reduced dimension; because the reported gains depend on this choice, the absence of a reproducible selection rule undermines the claim that the improvement is attributable to the SDR component rather than to an ad-hoc tuning.
minor comments (2)
  1. [Methods] Notation for the reduced SDR scores and the Poisson rate parameters should be introduced with explicit equations rather than descriptive text alone.
  2. [Results] Table or figure captions for the RPS comparisons should include the exact number of matches used in each out-of-sample evaluation.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive comments. We address each major comment below and indicate the revisions we will make to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Evaluation and Results sections] The evaluation is performed exclusively on the 2018 and 2022 tournaments (both 32-team formats with fixed group stages). No simulation, covariate adjustment, or sensitivity analysis for the 2026 48-team format and new Round-of-32 stage is described; because the central claim concerns prediction for the 2026 edition, this omission is load-bearing.

    Authors: We agree that the absence of analysis tailored to the 2026 format represents a limitation, given the paper's focus on that tournament. While the 2018/2022 evaluations establish the method's performance under the prior structure, we will add a dedicated sensitivity analysis section. This will include Monte Carlo simulations of the 48-team group stage and Round of 32, propagating outcomes through the new knockout bracket using the fitted SDR-Poisson models to assess robustness. revision: yes

  2. Referee: [Abstract and Results] The abstract and results summary state that SDR-Poisson models improve traditional approaches, yet supply neither the numerical RPS values, the chosen reduced dimension(s), nor any error bars or significance tests; without these quantities the magnitude and reliability of the claimed improvement cannot be assessed.

    Authors: We will revise both the abstract and the Results section to report the exact RPS values for all eleven models, the selected reduced dimensions for each SDR variant, bootstrap-derived standard errors, and paired statistical tests (e.g., Diebold-Mariano) for the observed improvements. These additions will make the magnitude and reliability of the gains fully transparent. revision: yes

  3. Referee: [Methods (SDR application)] The methods description of the SDR step does not specify the criterion or procedure used to select the reduced dimension; because the reported gains depend on this choice, the absence of a reproducible selection rule undermines the claim that the improvement is attributable to the SDR component rather than to an ad-hoc tuning.

    Authors: The reduced dimension was selected by minimizing out-of-sample RPS on a validation set of pre-tournament international matches. We will expand the Methods section to document this cross-validation procedure in full, including the candidate dimensions examined (1–5) and the final choices, thereby making the selection rule explicit and reproducible. revision: yes

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; evaluation uses independent held-out tournaments

full rationale

The derivation applies categorical SDR (SIR/SAVE) to reduce Elo histories, feeds the reduced scores into Poisson double regression, and evaluates via RPS on the fully held-out 2018 and 2022 World Cups. No equation reduces the reported RPS gains to a fitted quantity defined from the evaluation data itself, no self-citation chain bears the central claim, and the dimension-reduction step is not defined in terms of the target outcome probabilities. The result is therefore self-contained against external benchmarks.

Assumptions & free parameters 0 free parameters · 0 assumptions · 0 invented entities

Abstract-only review supplies no explicit free parameters, axioms, or invented entities; the central claim implicitly assumes that Elo rating histories are stationary enough for SDR to extract stable predictive directions and that the Poisson assumption for goals remains valid across tournaments.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Predicting the 2026 FIFA World Cup with Sufficient Dimension Reduction of Elo Rating Histories." pith.science (2026). https://pith.science/paper/B457U6XF

@misc{pith2026260624171,
  author       = {Pith},
  title        = {Pith review of: Predicting the 2026 FIFA World Cup with Sufficient Dimension Reduction of Elo Rating Histories},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/B457U6XF}},
  note         = {Machine review of arXiv:2606.24171}
}
read the original abstract

We study probabilistic forecasting of the 2026 FIFA World Cup, the first edition with 48 teams and an added Round of 32. The main idea is to describe team strength not only by the current Elo rating, but by a short history of recent Elo differences. We then reduce this history to a few informative directions using categorical sufficient dimension reduction (SDR). The reduced scores are used in a Poisson double-regression model for home and away goals, which gives full outcome probabilities. We compare eleven models, including logistic regression, standard Poisson regression, ARIMA, and neural-network forecasts of the Elo series, gradient boosting, an ensemble model, and four categorical SDR variants based on sliced inverse regression (SIR) and sliced average variance estimation (SAVE). The models are evaluated out of sample on the 2018 and 2022 World Cups using the ranked probability score (RPS). The results show that SDR-based poisson models improve the traditional approaches, suggesting that recent Elo history contains useful predictive information that is not captured by the current Elo difference alone.

Figures

Figures reproduced from arXiv: 2606.24171 by the authors.

Figure 1
Figure 1. Projected 2026 group-stage standings under the SAVE model, ranked within d [PITH_FULL_IMAGE:figures/full_fig_p022_1.png] view at source ↗
Figure 2
Figure 2. Projected 2026 group-stage standings under the SIR model, ranked within d [PITH_FULL_IMAGE:figures/full_fig_p022_2.png] view at source ↗
Figure 3
Figure 3. Illustrative 2026 knockout bracket under the SAVE model, including the [PITH_FULL_IMAGE:figures/full_fig_p023_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Illustrative 2026 knockout bracket under the SIR model, including the Round [PITH_FULL_IMAGE:figures/full_fig_p024_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

43 extracted references

  1. [1]

    Maher, M. J. , title=. Statistica Neerlandica , volume=

  2. [2]

    and Coles, Stuart G

    Dixon, Mark J. and Coles, Stuart G. , title=. Journal of the Royal Statistical Society: Series C (Applied Statistics) , volume=

  3. [3]

    Journal of the Royal Statistical Society: Series D (The Statistician) , volume=

    Karlis, Dimitris and Ntzoufras, Ioannis , title=. Journal of the Royal Statistical Society: Series D (The Statistician) , volume=

  4. [4]

    Prediction and retrospective analysis of soccer matches in a league , journal=

    Rue, H. Prediction and retrospective analysis of soccer matches in a league , journal=

  5. [5]

    , title=

    Elo, Arpad E. , title=

  6. [6]

    International Journal of Forecasting , volume=

    Hvattum, Lars Magnus and Arntzen, Halvard , title=. International Journal of Forecasting , volume=

  7. [7]

    The predictive power of ranking systems in association football , journal=

    Lasek, Jan and Szl. The predictive power of ranking systems in association football , journal=

  8. [8]

    Journal of Quantitative Analysis in Sports , volume=

    Groll, Andreas and Ley, Christophe and Schauberger, Gunther and Van Eetvelde, Hans , title=. Journal of Quantitative Analysis in Sports , volume=

Show all 43 references
  1. [9]

    Learning to predict soccer results from relational data with gradient boosted trees , journal=

    Hub. Learning to predict soccer results from relational data with gradient boosted trees , journal=

  2. [10]

    and Thabtah, Fadi , title=

    Bunker, Rory P. and Thabtah, Fadi , title=. Applied Computing and Informatics , volume=

  3. [11]

    International Journal of Forecasting , volume=

    Leitner, Christoph and Zeileis, Achim and Hornik, Kurt , title=. International Journal of Forecasting , volume=

  4. [12]

    Bates, J. M. and Granger, C. W. J. , title=. Operational Research Quarterly , volume=

  5. [13]

    Handbook of Economic Forecasting , volume=

    Timmermann, Allan , title=. Handbook of Economic Forecasting , volume=

  6. [14]

    Box, George E. P. and Jenkins, Gwilym M. and Reinsel, Gregory C. and Ljung, Greta M. , title=

  7. [15]

    and Athanasopoulos, George , title=

    Hyndman, Rob J. and Athanasopoulos, George , title=

  8. [16]

    IEEE Transactions on Automatic Control , volume=

    Akaike, Hirotugu , title=. IEEE Transactions on Automatic Control , volume=

  9. [17]

    International Journal of Forecasting , volume=

    Goddard, John , title=. International Journal of Forecasting , volume=

  10. [18]

    Journal of the Royal Statistical Society: Series D (The Statistician) , volume=

    Crowder, Martin and Dixon, Mark and Ledford, Anthony and Robinson, Mike , title=. Journal of the Royal Statistical Society: Series D (The Statistician) , volume=

  11. [19]

    Eddy and Hu, Michael Y

    Zhang, Guoqiang and Patuwo, B. Eddy and Hu, Michael Y. , title=. International Journal of Forecasting , volume=

  12. [20]

    , title=

    Loeffelholz, Bernard and Bednar, Earl and Bauer, Kenneth W. , title=. Journal of Quantitative Analysis in Sports , volume=

  13. [21]

    Transactions on Knowledge and Data Engineering , volume=

    Tax, Niek and Joustra, Yme , title=. Transactions on Knowledge and Data Engineering , volume=

  14. [22]

    Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , pages=

    Chen, Tianqi and Guestrin, Carlos , title=. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , pages=

  15. [23]

    , title=

    Friedman, Jerome H. , title=. Annals of Statistics , volume=

  16. [24]

    , title=

    Fisher, Ronald A. , title=. Annals of Eugenics , volume=

  17. [25]

    Radhakrishna , title=

    Rao, C. Radhakrishna , title=. Journal of the Royal Statistical Society: Series B (Methodological) , volume=

  18. [26]

    Anderson, T. W. , title=

  19. [27]

    Journal of the American Statistical Association , volume=

    Li, Ker-Chau , title=. Journal of the American Statistical Association , volume=

  20. [28]

    Dennis and Weisberg, Sanford , title=

    Cook, R. Dennis and Weisberg, Sanford , title=. Journal of the American Statistical Association , volume=

  21. [29]

    Game-related statistics that discriminated winning, drawing and losing teams from the

    Lago-Pe. Game-related statistics that discriminated winning, drawing and losing teams from the. Journal of Sports Science and Medicine , volume=

  22. [30]

    Journal of Human Kinetics , volume=

    Castellano, Julen and Casamichana, David and Lago, Carlos , title=. Journal of Human Kinetics , volume=

  23. [31]

    Match statistics related to winning in the group stage of 2014

    Liu, Hongyou and G. Match statistics related to winning in the group stage of 2014. Journal of Sports Sciences , volume=

    2014

  24. [32]

    Agresti, Alan , title=

  25. [33]

    Annals of Applied Statistics , volume=

    Gelman, Andrew and Jakulin, Aleks and Pittau, Maria Grazia and Su, Yu-Sung , title=. Annals of Applied Statistics , volume=

  26. [34]

    Journal of the Royal Statistical Society: Series B (Methodological) , volume=

    Tibshirani, Robert , title=. Journal of the Royal Statistical Society: Series B (Methodological) , volume=

  27. [35]

    Journal of Statistical Software , volume=

    Friedman, Jerome and Hastie, Trevor and Tibshirani, Robert , title=. Journal of Statistical Software , volume=

  28. [36]

    , title=

    Epstein, Edward S. , title=. Journal of Applied Meteorology , volume=

  29. [37]

    , title=

    Gneiting, Tilmann and Raftery, Adrian E. , title=. Journal of the American Statistical Association , volume=

  30. [38]

    , title=

    Brier, Glenn W. , title=. Monthly Weather Review , volume=

  31. [39]

    Chikuse, Yasuko , title=

  32. [40]

    International football results from 1872 to 2026 , year=

    J. International football results from 1872 to 2026 , year=

    2026

  33. [41]

    Dennis and Yin, Xiangrong , title =

    Cook, R. Dennis and Yin, Xiangrong , title =. Australian & New Zealand Journal of Statistics , year =

  34. [42]

    Dennis , title =

    Cook, R. Dennis , title =. Communications in Statistics -- Theory and Methods , year =

  35. [43]

    Technometrics , year =

    Zhang, Xin and Mai, Qing , title =. Technometrics , year =

Pith tools

Reviewed June 25, 2026 · model on record in the stance chip above.