Who will win Colombia's 2026 presidential election (first round 31 May, runoff 21 June)?

TL;DR

My forecast is 85% for Abelardo de la Espriella. The normal election path is now only De la Espriella versus Iván Cepeda; Paloma Valencia and Sergio Fajardo are off the runoff ballot. Cepeda needs a large turnout-and-transfer surprise that is bigger than the late polls and first-round arithmetic imply.

Context

As of 18:26 UTC on 21 June 2026, it is 13:26 in Bogotá, and Colombia is still voting; the domestic voting window runs from 8:00 a.m. to 4:00 p.m., with 41,421,973 eligible voters reported for the runoff day (El País live, 21 Jun 2026). There is no official runoff count yet, and preliminary preconteo results after polls close would still be informative rather than legally resolving.

The May 10 framing is stale. The CNE-certified first-round scrutiny put De la Espriella first with 10,366,143 votes and Cepeda second with 9,703,921, a 662,222-vote lead, and the official second-round ballot lists only Iván Cepeda Castro/Aída Quilcué Vivas and Abelardo de la Espriella/José Manuel Restrepo (El Espectador, 4 Jun 2026; Registraduría, 5 Jun 2026).

Evidence

The historical backbone favors the first-round leader, but the 2026 margin is in the reversible range. Since the 1991 Constitution introduced the runoff, Colombia has had two first-round presidential wins and six completed runoffs before 2026; the first-round leader won four of those six runoffs, while the second-place candidate came back in 1998 and 2014 (El País, 21 Jun 2026). The full post-1991 presidential history is:

The first-round arithmetic is stronger for De la Espriella than the 2.8-point headline suggests. Using the official first-round totals, De la Espriella starts 662,222 votes ahead; using the 23,676,056 valid-vote denominator reported by the CNE, about 3.61 million valid votes were cast for someone other than the two finalists or for the blank option (El Espectador, 4 Jun 2026; El País, 21 Jun 2026). If that pool split between the two finalists and turnout were otherwise unchanged, De la Espriella would need only about 41% of it to stay ahead: x > 0.5 - (D0 - C0) / (2R), where D0 is his first-round vote, C0 is Cepeda's first-round vote, and R is the non-finalist/blank valid-vote pool.

The endorsement map makes that 41% threshold plausible. Paloma Valencia took almost 1.6 million first-round votes and backed De la Espriella, as did Álvaro Uribe; Sergio Fajardo, who had a little over 1 million votes, did not endorse either finalist, and Claudia López's roughly 224,000-vote bloc moved toward Cepeda (El País, 21 Jun 2026). AtlasIntel's final reported transfer crosstabs point the same way on the biggest eliminated bloc: 84.5% of Valencia voters to De la Espriella, while Fajardo voters favored Cepeda by 68.7% to 14.1% but were a smaller and less disciplined pool (El Universal, 11 Jun 2026).

The final public polls also point to De la Espriella. Colombia's last-week polling blackout means these are stale by election day, and the first round showed polling can miss the order of the race; AP noted that Cepeda had led the pre-May polls but finished behind De la Espriella (AP, 21 Jun 2026). I therefore treat the poll lead as real but noisy.

My poll model uses a De la Espriella margin near 5 points after downweighting the CELAG outlier and allowing for the blackout. With a 5.5-point final-margin error, much wider than sampling error, that gives a poll-only win probability of about 84%. A separate transfer model starts from the official first-round votes, gives most Valencia/Uribe voters to De la Espriella, gives Fajardo and new voters a pro-Cepeda lean, and still gives De la Espriella about an 87% chance because he begins ahead and only needs a minority of the eliminated/blank pool. A small historical model, which penalizes him for having a small first-round lead like 1998 and 2014, is lower. Blending those pieces gives about 85% for De la Espriella within the normal two-candidate path.

Process risk is small. The election day had no reported nationwide disruption by midday, though there were complaints about coercion and vote-buying in Atlántico (El País, 21 Jun 2026). The first-round official scrutiny was fast and close to the preconteo: CNE-related reporting says the preconteo and official scrutiny matched at about 99.97%, and the CNE declared the official result four days after voting (El País, 21 Jun 2026). De la Espriella's main legal threat has also weakened: the Consejo de Estado rejected a demand against his candidacy on 11 June, and the CNE left his candidacy in force on 18 June despite the U.S.-citizenship challenge (Infobae, 11 Jun 2026; SWI/EFE, 18 Jun 2026).

What's non-obvious

This is no longer a three-way race. Valencia and Fajardo still matter because their voters can transfer, but the official ballot has only Cepeda and De la Espriella, so their listed options now require an extraordinary legal, death, incapacity, annulment, or replacement path (Registraduría, 5 Jun 2026).

The small first-round margin is not enough to make the race a toss-up. De la Espriella and Cepeda together already took about 85% of valid first-round votes, so there is less room for a 2014-style rearrangement than the raw 2.8-point margin suggests. Cepeda's live path is turnout composition: new voters, young voters, Bogotá, the Pacific, and anti-De la Espriella centrists have to overperform the first round by enough to overcome Valencia/right-wing consolidation. That can happen, but it is not the central case.

Limitations

This forecast is made before polls close and before any official second-round count. A credible preconteo after 4:00 p.m. Bogotá time would dominate the pre-election evidence, although the question resolves only on official scrutiny rather than preliminary bulletins.

The polls are useful but weak. First-round polls missed De la Espriella's late surge and the collapse of Valencia, so I used a wide error term rather than taking the 4-8 point final leads literally. I also could not verify full technical sheets for every late poll from open reporting, especially CELAG and CNC, so I weighted source transparency and consistency rather than sample size alone.

The non-finalist probabilities are not normal electoral probabilities. They cover legal shock, candidate death or incapacity, annulment, failure to declare a winner by 31 July, or an unusual replacement route after voting had already begun. That combined tail is about 0.3%, with most of it in Other rather than Valencia or Fajardo.