Index construction and validation, on your data
Build a composite with intervals and rank ranges, see which choices move it, and test whether a signal warns in time. Nothing leaves your browser.
Composite builder and early-warning evaluator
Everything runs in this page. Your table never leaves your browser, and nothing is stored.
Paste a table, open a file or load the synthetic example, then build the composite.
Scores, intervals and rank stability
Statistical coherence
Distribution of each indicator
Which choice moves the ranks
Paste signals and outcomes, or load the synthetic example, then evaluate.
What each step does, and where it comes from
The same steps, with the same rules, build the index itself; the lab lets you see how each choice moves a result.
| Step | What the lab does | Reference |
|---|---|---|
| Outliers | An indicator is treated when its skewness exceeds 2 in absolute value and its kurtosis exceeds 3.5. Treatment winsorises one value at a time, up to five per indicator, then log-transforms if the rule still fails. | JRC Competence Centre on Composite Indicators, ten-step guide |
| Missing values | Left out with the unit scored on what is present, or imputed by the mean or median. Units below the coverage threshold are not scored, and coverage is shown for every unit. | OECD and JRC, Handbook on Constructing Composite Indicators (2008) |
| Normalisation | Min-max to 0 to 100, z-scores, ranks or percentiles. Each indicator is oriented so that a higher value means more risk. | OECD and JRC Handbook, step 5 |
| Weights | Equal, your own, or derived from principal components: components with an eigenvalue above 1 are kept, and each indicator is weighted by its squared loadings, scaled by the variance each component explains. | OECD and JRC Handbook, step 6; Nicoletti, Scarpetta and Boylaud (2000) |
| Aggregation | Arithmetic means let a high value offset a low one in full; geometric means penalise imbalance, so a unit cannot compensate a very low score in one indicator. | Munda and Nardo (2009) on compensability |
| Uncertainty | Every run perturbs the weights, and can draw another normalisation and aggregation. Each unit gets a median rank and the range holding 90% of its runs, so a rank is read with its spread. | Saisana, Saltelli and Tarantola (2005) |
| Sensitivity | First-order indices estimate how much of the variance in the average rank shift each choice explains, by conditioning on that choice alone. | Saltelli et al., Global Sensitivity Analysis: The Primer (2008) |
| Coherence | The correlation matrix, standardised Cronbach's alpha and the variance explained by each principal component show whether the indicators measure one thing or several. | Cronbach (1951); OECD and JRC Handbook, step 4 |
| Early warning | AUROC with its 95% interval, the ROC curve, and the threshold that maximises Youden's J and the one that minimises the noise-to-signal ratio. Usefulness weighs missed events against false alarms; the Brier score applies when signals are probabilities. | Hanley and McNeil (1982); Kaminsky and Reinhart (1999); Alessi and Detken (2011); Brier (1950) |
How to read a result
A rank is a range
Two units whose rank ranges overlap cannot be told apart with confidence. Read the median rank with its 90% range.
Choices are disclosed
When one choice, such as the aggregation, explains most of the movement, it deserves the most scrutiny and the clearest justification.
A signal is judged by its errors
An AUROC of 0.5 is no better than chance. The usefulness measure says whether acting on the signal beats ignoring it, given how costly misses are.
Not index values
Lab results are your own calculations. The synthetic example is random numbers, never data, and no result here is a value of the index.
Comment on the methodology
Every comment on the method receives a disposition in the consultation record.
Status
The methodology, structure and definitions of The Global Risks Index are a preparatory working draft (2026), circulated for review and comment and not for implementation. No value appears until an edition is released from a governed evidence cut; values then carry intervals, classes, sources and their correction pathway. Simulators run the published methods on numbers you choose and never show a NXS-GRX value.
What NXS-GRX is not
NXS-GRX is research and risk-analytics information. It is not a credit or ESG rating and uses no rating scale, not a benchmark for financial instruments, not official statistics, not an emergency warning or official classification, and not advice. It must not be used to evaluate natural persons. Risk classes are bands on a published scale, never rankings of countries or people.
Institutions and sources
The Nexus Consortium is stewarded by three independent not-for-profit institutions: GCRI Canada as Lead Institution, Consortium Secretariat and lead contracting organization, The Global Risks Forum (GRF) and The Global Risks Alliance (GRA). Technology, infrastructure and service providers work with the Consortium at arm's length under written agreements, like every other external party in the public-private-planet model, and hold no role in its stewardship.
Every value carries the attribution of its sources, which keep their own licences. External names are cited for interoperability only. Nexus-authored text is intended for CC BY 4.0 on release.
Avis en français
La méthodologie et les définitions du NXS-GRX sont un projet de travail préparatoire (2026). Le NXS-GRX est une information de recherche et d'analyse des risques : ni notation, ni indice de référence financier, ni statistique officielle, ni alerte, ni conseil. Les classes de risque sont des bandes sur une échelle publiée, non des classements.