AFOS-Analytics1/colombia-2026-electoral-divergence

AFOS · Colombia 2026 Electoral Divergence Dataset

🌐 English · Español · Português

Open dataset cross-referencing opinion polls × prediction markets for Colombia's 2026 presidential election (first round 31 May 2026; runoff 21 June 2026, Abelardo de la Espriella vs Iván Cepeda), built like the AFOS Brazil & Peru datasets: sources are reported side by side with explicit divergence, not blended into a single average.

Maintained by AFOS Analytics. Part of AFOS's expansion of its electoral-divergence method across Latin America. No personal data, only public electoral information.

License (dual): data → CC BY 4.0 (LICENSE-CC-BY-4.0); code/scripts → Apache 2.0 (LICENSE-APACHE-2.0), matching the repo root and the Hugging Face mirror. Please attribute AFOS Analytics and the original pollsters.

---

Press coverage layer

The qualitative third axis of the AFOS cross (market x polls x press) is now a structured file: news/colombia-2026-press-coverage.csv, 22 dated headlines from 15 outlets across the cycle (polls, first round, runoff, result, reactions), in ES/EN/PT. Headlines and links only (outlets retain copyright); dates are publication/coverage dates, best-effort. It complements the quantitative market-vs-poll divergence; it is not sentiment-scored.

---

Coverage map (illustrative, NOT a divergence signal)

Descriptive breakdown of the 22 curated headlines in news/colombia-2026-press-coverage.csv. This reflects AFOS curation, not total press volume, so it is deliberately NOT fed into the market-vs-poll divergence and must not be read as a quantitative press signal. A true coverage-volume cross would require systematic daily collection.

• By phase: pre-first-round 1, first round 4, runoff campaign 4, result + reactions 13
• By topic: result 9, reactions 5, first round 3, analysis 2, poll 1, debate 1, runoff poll 1
• By language: ES 16, EN 5, PT 1
• Outlets (15): CNN en Espanol (5), El Espectador (2), El Tiempo (2), El Colombiano (2), plus one each from Infobae, France 24, La Silla Vacia, CNN, NPR, Al Jazeera, CNBC, CBS News, Portafolio, La FM, Brazil Journal.

---

English

Path	Rows	Content
polls/colombia-first-round-polls.csv	170	First-round voting intention, long format (one row per candidate × poll), 10 candidates, 29 polls, 2025→May 2026.
polls/colombia-polls.json	n/a	Full structured polls with methodology.
data/colombia-market-odds-timeseries.csv	4,620	Daily Polymarket win-probability per candidate (19 candidates, Jul 2025→Jun 2026) from the "Colombia Presidential Election" market.
data/colombia-divergence-timeseries.csv	123	Market × poll divergence per candidate, first-round polls + the 15 Jun runoff snapshot, joined to the candidate's market odds.
polls/colombia-runoff-polls.csv	3	Runoff (2nd round) polls from named pollsters, AtlasIntel, Guarumo/Ecoanalítica, CNC.
data/colombia-results.csv	4	Official results, first round + runoff, votes and % per candidate, with the elected winner flagged.
data/colombia-poly-raw.json	n/a	Raw Polymarket payload (provenance).

Runoff RESULT (21 Jun 2026). De la Espriella won the presidency, 49.66% (12,959,542 votes) × Cepeda 48.70% (12,708,712), a about 0.96pp margin (about 250k votes) with nearly 100% counted (the official CNE escrutinio is the formality; inauguration 7 Aug). First round (31 May): de la Espriella 43.74% × Cepeda 40.90%.

The cross-reference, prediction market × opinion polls × press: at the close the market priced his win as near-certain (88.5% × 12.5%); runoff polls gave him a about 4-8pp lead (AtlasIntel 50.9×43.1; Guarumo/Ecoanalítica 52.6×45.0; CNC 48.6×44.7); the press reported a likely de la Espriella win with a tightening race. All three called the winner correctly, but market and polls overstated the margin: the finish was about 1pp near-tie, far below the poll lead and far from the market's near-certainty. International observers deemed the process transparent; the left (Petro, Cepeda) alleged irregularities.

The AFOS read: right direction, overstated confidence, the market more so than the polls. As in the first round, the spread was the signal: an average would have hidden how a market priced at 88.5% certainty resolved into a coin-flip. Official results in data/colombia-results.csv.

⚖️ Notable divergences (why divergence beats the average)

The point of this dataset is the gap between what the market prices (probability of winning) and what polls measure (first-round vote share). Averaging the two would erase exactly the signal below. From the latest pre-election poll (Invamer, 20 May 2026):

• Abelardo de la Espriella, poll 31.6% × market 43.5% (+11.9pp). The market priced his win probability about 12 points above his first-round vote-share polling, and he won the first round with 43.7%. The market's signal matched the result; vote-share polls understated him.
• Iván Cepeda, poll 44.6% × market 43.5% (−1.1pp). Near-zero divergence: market and polls agreed he led on vote share, but priced him roughly even with Espriella to win, foreshadowing a tight runoff.
• Paloma Valencia, poll 14% × market 14.5% (+0.5pp): market and polls in lockstep.
• Earlier in the cycle the market ran below some candidates' vote share (e.g. Daniel Quintero, Nov 2025, poll 5.2% × market 0.6%, −4.6pp), it never believed they could win.

The reading: a single blended "market + polls" average would have shown Cepeda comfortably ahead and hidden that the market gave the eventual first-round winner (Espriella) a far higher chance than his vote-share suggested. The divergence was the signal.

Pollsters covered: Invamer, AtlasIntel, CNC, Guarumo, GAD3, CELAG, and others.

Provenance & method: poll figures compiled deterministically from the public Wikipedia aggregation ("2026 Colombian presidential election") and the AS/COA poll tracker, covering the first round (10 candidates) and the runoff (de la Espriella vs Cepeda, 21 June 2026); each figure traces to a named pollster. Market odds from the public Polymarket market. Nothing imputed or smoothed; missing values left blank. Underlying poll numbers are facts released by the named pollsters; the Wikipedia aggregation is CC BY-SA. Please attribute AFOS Analytics and the original pollsters.

---

Español

Dataset abierto que cruza encuestas × mercados de predicción para la elección presidencial de Colombia 2026 (primera vuelta 31 may; segunda vuelta 21 jun, de la Espriella vs Cepeda), con divergencia explícita en lugar de un promedio único.

Resultado (balotaje, 21 jun): de la Espriella ganó la presidencia, 49,66% (12.959.542) × 48,70% (12.708.712) de Cepeda, margen cerca de 0,96pp (cerca de 250 mil votos), casi 100% escrutado (escrutinio oficial del CNE es la formalidad; toma de posesión 7 ago). Primera vuelta: 43,74% × 40,90%. El cruce mercado × encuestas × prensa: mercado (88,5%) y encuestas (cerca de 4-8pp) acertaron al ganador pero sobreestimaron el margen (desenlace de cerca de 1pp). La prensa reportó la victoria; observadores internacionales lo consideraron transparente; la izquierda alegó irregularidades. Lectura AFOS: dirección correcta, confianza sobreestimada. Resultados oficiales en data/colombia-results.csv.

• polls/colombia-first-round-polls.csv, intención de voto en primera vuelta, formato largo, 10 candidatos, 29 encuestas (2025→may 2026).
• data/colombia-market-odds-timeseries.csv / colombia-divergence-timeseries.csv, probabilidad de Polymarket por candidato y divergencia mercado × encuesta.

⚖️ Divergencias destacadas (por qué la divergencia supera al promedio)

Lo importante es la brecha entre lo que valora el mercado (probabilidad de ganar) y lo que miden las encuestas (voto de primera vuelta). De la última encuesta preelectoral (Invamer, 20 may 2026):

• Abelardo de la Espriella, encuesta 31,6% × mercado 43,5% (+11,9pp). El mercado valoró su probabilidad de ganar cerca de 12 puntos por encima de su voto, y ganó la primera vuelta con 43,7%. La señal del mercado coincidió con el resultado; las encuestas lo subestimaron.
• Iván Cepeda, encuesta 44,6% × mercado 43,5% (−1,1pp). Divergencia casi nula: mercado y encuestas coincidían en que lideraba en voto, pero lo valoraban casi a la par con Espriella para ganar, anticipando un balotaje reñido.
• Paloma Valencia, encuesta 14% × mercado 14,5% (+0,5pp): mercado y encuestas al unísono.
• Daniel Quintero, (nov 2025) encuesta 5,2% × mercado 0,6% (−4,6pp): el mercado nunca creyó que pudiera ganar.

La lectura: un promedio "mercado + encuestas" habría mostrado a Cepeda cómodamente al frente y ocultado que el mercado le dio al ganador efectivo de primera vuelta (Espriella) una probabilidad mucho mayor que su voto. La divergencia era la señal.

Encuestadoras: Invamer, AtlasIntel, CNC, Guarumo, GAD3, CELAG. Fuente: Wikipedia + AS/COA. Licencia: CC BY 4.0 (atribuir a AFOS Analytics y a las encuestadoras). Investigación observacional; no es asesoría de inversión ni orientación de voto.

---

Português

Dataset aberto cruzando pesquisas × mercados de previsão para a eleição presidencial da Colômbia 2026 (1º turno 31/mai; 2º turno 21/jun, de la Espriella × Cepeda), com divergência explícita entre fontes. Pesquisas (10 candidatos, 29 do 1º turno) compiladas da Wikipedia + AS/COA; odds do Polymarket. Licença CC BY 4.0 (atribuir AFOS Analytics + institutos). Pesquisa observacional; não é recomendação de investimento nem orientação de voto.

Resultado (2º turno, 21/jun): de la Espriella venceu a presidência, 49,66% (12.959.542) × 48,70% (12.708.712) de Cepeda, margem cerca de 0,96pp (cerca de 250 mil votos), quase 100% apurado (escrutínio oficial do CNE é a formalidade; posse 7/ago). 1º turno: 43,74% × 40,90%. O cruzamento mercado × pesquisas × imprensa: mercado (88,5%) e pesquisas (cerca de 4-8pp) acertaram o vencedor mas superestimaram a folga (desfecho de cerca de 1pp). A imprensa noticiou a vitória; observadores internacionais consideraram o processo transparente; a esquerda alegou irregularidades. Leitura AFOS: direção correta, confiança superestimada. Resultados oficiais em data/colombia-results.csv.

⚖️ Divergências em destaque (por que a divergência supera a média)

O ponto é a diferença entre o que o mercado precifica (probabilidade de vencer) e o que as pesquisas medem (voto de 1º turno). Da última pesquisa pré-eleição (Invamer, 20/mai/2026):

• Abelardo de la Espriella, pesquisa 31,6% × mercado 43,5% (+11,9pp). O mercado precificou a chance de vencer dele cerca de 12 pontos acima do voto, e ele venceu o 1º turno com 43,7%. O sinal do mercado bateu com o resultado; a pesquisa o subestimou.
• Iván Cepeda, pesquisa 44,6% × mercado 43,5% (−1,1pp). Divergência quase nula: mercado e pesquisa concordavam que ele liderava em voto, mas o precificavam quase empatado com Espriella para vencer, antecipando um 2º turno apertado.
• Paloma Valencia, pesquisa 14% × mercado 14,5% (+0,5pp): mercado e pesquisa em uníssono.
• Daniel Quintero, (nov 2025) pesquisa 5,2% × mercado 0,6% (−4,6pp): o mercado nunca acreditou que ele pudesse vencer.

A leitura: uma média "mercado + pesquisas" mostraria Cepeda confortavelmente à frente e esconderia que o mercado deu ao vencedor efetivo do 1º turno (Espriella) uma chance bem maior que o voto dele. A divergência era o sinal.

---

Sources / Fuentes: Pollsters (Invamer, AtlasIntel, CNC, Guarumo, GAD3, CELAG, …) · Wikipedia, 2026 Colombian presidential election · AS/COA poll tracker · Polymarket. Column definitions in DATA_DICTIONARY.md.

Structural context (World Bank)

Beyond the divergence data, this dataset ships data/colombia-structural-context.csv: official, open World Bank indicators that frame the country, governance (Worldwide Governance Indicators, 0-100 scale) plus economy & education (World Development Indicators: population, GDP, GDP per capita, inflation, public education spending, expected years of schooling). These are annual structural indicators that contextualize the country; they do not predict the electoral outcome. Columns are documented in DATA_DICTIONARY.md.