Measurement for central bank digital currency
Where readiness meets resolve.
CBDC Nexus brings together three open, reproducible indexes: one measures whether a country has the structural foundations to introduce a central bank digital currency, one measures the policy position its central bank communicates about it, and one maps how that conversation is framed, sentence by sentence. Explore them on a world map or a turning globe, then go as deep as the data allows.
The Nexus
Fifty economies appear in both indexes. Readiness runs left to right, communicated stance bottom to top, and the two dashed lines (median readiness, neutral stance) cut the world into four kinds of CBDC story. Pick a year to watch economies cross those lines.
The world, measured
The same globe, coloured three ways. Readiness colours the world by structural capacity, Policy by what each central bank has said, Discourse by how its speeches are framed. Hover any economy to read its score; click it to open the full profile.
Three indexes, one nexus
Readiness and rhetoric are different things. A well prepared economy may hold back; a less prepared one may announce early. Placing the two measures side by side is the point of this site: the gap between them is where the policy questions live.
Structural capacity
A five pillar, cross-country index of the digital, economic, financial, institutional and macroeconomic foundations a jurisdiction needs to introduce a CBDC safely and at scale. Weighted by CRITIC, aggregated geometrically, so the weakest pillar is always the binding constraint.
Communicated position
A disagreement-weighted Bayesian index of the policy position each central bank communicates about digital currency, estimated at sentence level from BIS speeches and carried through time with its own credible band.
Sentence-level discourse
Built on CentralBank-BERT, a language model adapted to central bank speech, with four classifiers that label every CBDC sentence by type (retail, wholesale), stance, sentiment and discourse (feature, risk-benefit, process). A map of who speaks, what they emphasise and how they frame it.
Vision, mission, and who we are
Every number on this site can be traced back to a public source, a documented procedure and, where estimation is involved, an interval that says how sure we are.
A world in which every step toward central bank digital currency is judged against measured foundations and clearly read intentions, rather than against announcements and headlines.
Our mission is to spread knowledge on central bank digital currency, a dynamic and vitally important area of monetary policy and financial innovation. We build open, rigorous measures of CBDC readiness and central bank communication, explain them in plain language, and put the data in the hands of policy makers, researchers, students and the public.
CBDC Nexus is an independent research initiative based in Doha, Qatar, run by a team of researchers working across monetary economics, financial technology, text analysis and data science. The team builds and maintains both indexes, keeps the data open, and publishes the methods in peer-reviewed journals: the Readiness Index appears in Quality & Quantity (2026), and the Policy Index is released as an open measurement dataset.
Every table on both consoles downloads as CSV, and the full source panel ships with the readiness index. Free to reuse with attribution.
The readiness console rebuilds scores, weights and ranks in your browser from the raw panel each time a control changes.
Bootstrap rank intervals for readiness; posterior credible bands for stance. A number without its interval is not a result.
Slow moving fundamentals and stated positions, read separately, so that identical announcements can be understood to have different effects.
Live traction
See how CBDC Nexus research, tools and evidence are being discovered and used around the world. Counts are aggregate and anonymous: no cookies, no personal data, and a visitor's country is inferred from their time zone.
Live counters appear once the site is hosted with its analytics endpoint. The tracker script and a ready-to-deploy worker are included with the site files.
Index one of three
Who has the structural foundations to introduce a central bank digital currency safely and at scale. Five pillars, CRITIC information weights, non-compensatory geometric aggregation, up to 183 economies, 2018 to 2024.
Readiness map
Hover an economy for its score and binding constraint. Click it for the full pillar profile. Switch years to watch readiness move; the average of all years is the paper's headline ranking.
Most economies cluster in a broad lower-to-middle band, while a small group that combines strong digital infrastructure, deep and inclusive financial systems, robust institutions and stable macro conditions sits well ahead. Small, highly interconnected financial jurisdictions and advanced economies occupy the top of the distribution; low-income economies with limited connectivity and weaker institutions fill the bottom decile.
The Digital, Economic and Financial pillars move together, but Institutions and Regulation correlates only weakly with them and Macro Resilience adds a partly independent layer. This is the empirical warrant for keeping five pillars rather than one score. Because aggregation is geometric and non-compensatory, strength elsewhere cannot offset a weak pillar: the weakest pillar is the binding constraint, and two economies at the same rank can need entirely different reforms.
Across alternative weighting schemes, aggregation rules, coverage thresholds, leave-one-variable and leave-one-pillar exercises and a 500-draw bootstrap, the top and bottom groups are essentially invariant. Positions in the middle band move by a few places under reasonable specification changes, so a difference of a few ranks there should be read as sampling noise. The index is a diagnostic dashboard, not an ordinal league table.
Against the independent project, search-interest and speech dataset of Auer, Cornelli and Frost (2020), covering 134 jurisdictions, the index correlates positively with observed CBDC project activity (Spearman 0.327 overall; 0.517 for the Digital pillar). Economies with an active project average 0.283 against 0.176 without one. Holding income constant the association survives (partial rank correlation 0.206), and the index adds explanatory power beyond GDP per capita: 12.5 percent of variance each on their own, 17.9 percent together.
The association with central bank speech stance is small and statistically insignificant. Structural readiness and rhetorical posture are distinct, which is exactly why this site pairs the Readiness Index with a separate Policy Index.
Clustering economies on the shape of their pillar profile yields four archetypes. Each implies a different binding constraint, so the typology translates more directly into policy sequencing than an ordinal rank.
Strong across all five pillars; room for ambitious designs.
Malaysia, Denmark, Italy, Netherlands, China, Japan
Digitally and financially advanced; the depressed score reflects binary regulatory flags under non-compensatory aggregation, and points to regulatory codification.
United States, Singapore, Canada, France, UAE
Priority is digital and financial deepening before retail issuance.
Indonesia, Kenya, Bangladesh
Low scores across most pillars; foundational infrastructure and macro stabilisation come first.
How a score is built
183 economies, 2018 to 2024, 44 indicators mapped to five pillars from public sources including the World Bank, IMF, ITU and BIS. Indicators are admitted only where there is a clear channel to CBDC design, adoption or transmission; design-specific variables such as remuneration or holding limits are excluded by construction.
Each indicator is trimmed at the 1st and 99th percentiles within its year, so a single extreme observation cannot set the scale for everyone else.
Centre on the median and scale by the median absolute deviation (MAD x 1.4826), falling back to the mean absolute deviation when the MAD is degenerate.
Cost indicators (income inequality, inflation, cash-based government payments) are inverted. Scores are clipped to [-3, 3] and rescaled to [0, 1]. Binary policy flags pass through as 0 or 1.
Weights are estimated from each year's own data: indicators that vary widely across countries and carry information the others do not receive more weight. Weights are an output, re-estimated whenever new data lands.
Indicators combine into pillar scores by weighted geometric mean, with epsilon = 1e-6 guarding the logarithm. A missing indicator is dropped, not penalised.
Pillar scores are weighted by CRITIC once more and combined geometrically. An index is returned only where at least three pillar scores are finite. The geometric mean is deliberate: it is non-compensatory, so the weakest pillar binds.
Quality & Quantity, 2026
Open access under a Creative Commons Attribution 4.0 licence. Full author and funding details are on the publisher page.
CBDC readiness, policy design, multi-pillar index, central banks, financial stability. JEL E42, E58, G20, O33.
Index two of three
Where central banks stand on digital currency, measured. A disagreement-weighted Bayesian index of CBDC policy position in central bank speeches: the position each institution communicates, estimated from 5,718 CBDC sentences across 60 institutions, 2015 to 2025.
Policy map
Blue is committed, brick red is sceptical. Hatched economies said nothing that year, so their estimate is carried by the model rather than by speeches. Press play to watch the global turn from 2015 to 2025.
Averaged across institutions with CBDC discourse, communicated stance rose from a clearly sceptical -1.14 in 2016 to +0.50 in 2025. Attention, measured as CBDC sentences per thousand sentences spoken, peaked in 2023 and has fallen since, while stance kept climbing. The two are reported separately and never combined: a central bank that says nothing is silent, not neutral, and is absent from these means rather than counted as zero.
Plotting attention against stance isolates a quadrant no single tone score can find: institutions that discuss CBDC at length and remain sceptical. In the 2025 wave the most committed communicators include the Bank of Greece, the Banque de France, the Deutsche Bundesbank, the Bank of Latvia and the Bank of Thailand; the most sceptical include the Federal Reserve System, the National Bank of Cambodia, the Swiss National Bank, the Central Bank of Chile and the Central Bank of Brazil.
The index measures the position a central bank communicates toward CBDC, not the affective tone of its language. The two are related but not the same construct, and the corpus shows by how much: sentences dominated by risk discussion read markedly colder than the position they actually express, so a tone reading can invert a favourable position. The two released encoders correlate 0.76 with each other on favourability, but only 0.51 and 0.22 with an independent rubric reading of the same sentences: they share a base model, so their agreement is method variance rather than evidence that either captured the construct.
Each institution's series is a posterior path with a 90 percent credible band. Years with CBDC discourse are estimated from the sentences; years of silence are propagated by the random walk and the band widens to say so (average interval width 1.65 for observed cells, 2.45 for silent ones). Risk salience loads negatively on stance (-0.27, with the 94 percent interval excluding zero), a prediction of the design rather than a sign imposed by hand.
Dropping the two dominant institutions, the ECB and the Bundesbank, leaves the ranking essentially intact (r = 0.967). Switching off disagreement weighting barely moves the point estimates (r = 0.998) but inflates residual standard deviation by roughly 40 percent: the mechanism improves uncertainty quantification rather than the rankings.
Sources and construction
All BIS central bankers' speeches from 2015 to 2025: 9,172 speeches, filtered to 5,718 hard-keyword CBDC sentences across 680 speeches and 60 institutions. Every sentence carries a source URL, so any score can be traced back.
A rubric annotation, a purpose-built CBDC stance dictionary, and the released CBDC-Stance and CBDC-Sentiment encoders from the CentralBank-BERT family (Zafar, 2026). CBDC-Sentiment measures affective tone rather than position, so it is scored and released but held out of the favourability consensus.
Each sentence is scored on favourability, urgency and risk salience, and classified by design orientation (retail versus wholesale, which is not valenced and not part of the index). Scores aggregate to institution-year cells.
Cell means are treated as noisy indicators of an unobserved stance. The variance of each indicator grows with how far the engines disagree about it, and the latent stance follows a random walk through time.
Hamiltonian Monte Carlo, four chains, 2,000 draws after 2,500 tuning iterations. Loadings and disagreement weights are estimated, not assigned, and reported with 94 percent highest-density intervals.
Annotation stability from two independent passes of the rubric over a 322-unit stratified sample (exact agreement 0.89 to 0.93). Robustness to dropping the ECB and Bundesbank, to switching off disagreement weighting, and to adding or removing engines.
CBDC Policy Index, version 2.0
The underlying speeches are the property of the Bank for International Settlements and the issuing central banks and are not redistributed; the scored corpus carries a source URL for every sentence.
Index three of three
How central banks talk about CBDC, sentence by sentence. Built on CentralBank-BERT, a language model adapted to central bank speech, with four classifiers applied to 5,376 CBDC sentences from 588 BIS speeches by 157 speakers across 57 jurisdictions, 2016 to 2024.
Discourse map
Choose what to colour the map by: the balance of stance (pro minus anti), the balance of sentiment, the retail focus, or sheer activity. Click a jurisdiction for its full composition and to read its sentences.
Sentence explorer
Just over half of all CBDC sentences concern retail CBDC (51.3 percent), against 7.6 percent for wholesale and 41.1 percent that stay general. On discourse, Process is the modal frame (47.3 percent), ahead of Feature (27.0) and Risk-benefit (25.7). Central banks talk about rulebooks, intermediated distribution, holding limits, privacy models and offline use: the concrete tasks of a general-purpose CBDC rather than the case for one.
Sentiment is predominantly neutral (59.8 percent), with 32.4 percent positive and 7.8 percent negative. Stance divides between Pro-CBDC (47.5 percent) and Wait-and-see (42.9 percent), with outright Anti-CBDC at 9.7 percent. The two move together: the strongest association among the four classifiers is stance with sentiment (Cramer's V 0.637), consistent with technocratic communication that separates factual updates from evaluative claims.
CBDC sentences rise from 48 in 2016 to 614 in 2020, peak at 1,309 in 2022 and remain elevated (964 in 2023, 838 in 2024). The expansion is not a scale-up of generic statements: later years carry more retail design and program management, a shift in official narratives from whether to how.
The Deutsche Bundesbank (1,018 sentences) and the European Central Bank (1,011) together account for more than a third of the corpus, followed by the Federal Reserve Board, the Banque de France, the Bank of England and the Banca d'Italia. Asia-Pacific contributes a diverse group led by the Bank of Japan, the Reserve Bank of India, the Hong Kong Monetary Authority and the Reserve Bank of Australia.
From speeches to structured indicators
19,609 speeches posted by the Bank for International Settlements, 2016 to 2024. A controlled lexicon screens candidate CBDC sentences; the screen is validated against the relevance classifier below.
A masked-language model initialised from bert-base-uncased and pretrained on the full BIS corpus (about 2.09 million sentences, 66 million tokens). Perplexity on held-out central bank text falls from 13.06 to 4.66, and this adapted encoder underpins every classifier.
Detects whether a sentence concerns CBDC. Trained on 11,000 manually labelled sentences with in-domain negatives; accuracy and F1 about 0.995, ROC-AUC 1.000, Brier 0.0024.
Retail, wholesale or general. 1,417 labelled sentences, inverse-frequency class weights; accuracy about 0.887, macro-F1 about 0.898.
Stance: Pro-CBDC, Wait-and-see, Anti-CBDC, from 1,647 labelled sentences with focal loss; accuracy about 0.849, macro-F1 about 0.852. Sentiment: positive, neutral, negative, from 2,405 labelled sentences; accuracy about 0.822, macro-F1 about 0.812.
Feature, Risk-benefit or Process framing. Label shares, dispersion (entropy and HHI) and cross-task dependence (Cramer's V, mutual information, Spearman on scores) turn 5,376 labelled sentences into comparable indicators by year and jurisdiction.
Journal of Economics and Business, 2026
The prediction dataset behind this page (5,376 sentences with labels and scores) is released with the paper's replication package.
Central bank digital currency, central bank communication, BIS speeches, NLP, BERT, stance and sentiment. JEL E42, E58, E52, D83, C55.
Publications
Briefs and reports from CBDC Nexus that turn the two indexes into readable analysis for policy makers and researchers.
This section will list CBDC Nexus reports as they are released, each with a summary, a PDF and the data behind it. Until then, both indexes are fully open on their own pages and every table can be downloaded as CSV.
Get in touch
Enquiries, data requests, collaboration proposals and corrections all start here.
Share the institutional challenge, project stage and support you require. Your information is used only to assess and respond to this enquiry.
Sponsor or support our mission
CBDC Nexus publishes its indexes, data and methods free of charge. Support from institutions, foundations and partners funds the annual data updates, new jurisdictions, reports and the open consoles that keep every number checkable.
Compare
Pillar profiles, readiness over time and what each central bank has said. Pick any two economies; the link updates so you can share the comparison.