The 2027 presidential forecast: line-ups, runoffs and simulation
Abstract. We follow two line-ups tested by French pollsters, average each pollster's latest poll, take round one as published and correct round-two head-to-heads for the RN's record of being overstated. Chances of qualifying and of winning come from 20,000 simulated elections in which each candidate's round-one error has the size of their family's past misses and the runoff is drawn from the corrected head-to-head with a spread of about 4.4 points. The paper sets out each step formally, with a worked example and the choices we tested and did not adopt.
Summary
This paper describes how the site forecasts the 2027 presidential election: which of the many line-ups tested by pollsters we follow, how the round-one polls are averaged, how the head-to-head polls for round two are averaged and corrected, and how 20,000 simulated elections turn those figures into each candidate's chance of reaching round two and of winning it.
Round-one polls are taken as published, because correcting them made past forecasts worse (Measuring French polling error, 1965–2024). Round-two polls are corrected, because the RN candidate has been overstated at every runoff the RN has reached: the correction is −1.8 points on the RN candidate's share, adjusted where the opponent's family has a round-two record of its own. On polls to 19 November 2026, Marine Le Pen has a 99% chance of winning if the election were held today. The same method is run on a second line-up, with Jordan Bardella standing for the RN instead of Marine Le Pen.
What has changed since this paper was written
Updated 9 November 2026. The rest of the paper is as first published, except where a sentence below says it was corrected.
- Only elections from 2000 shape the forecast (28 September). Earlier elections stay in the long view and on the history pages but no longer feed any test, correction or error size. In France the presidential tests now run on the five elections from 2002 to 2022: the round-one correction stays off and the round-two correction stays on.
What is forecast
Round one is on 18 April 2027 and round two on 2 May. Every candidate needs 500 sponsorships from elected officials, and the list closes in March 2027. Until then the field is a matter of speculation, and the pollsters deal with that by testing several line-ups (hypothèses) in each survey. A single poll may carry anything from one to eleven of them.
The forecast answers two questions for each candidate: the chance of finishing in the top two in round one, and the chance of winning round two. Both are for an election held today, on today's polls. The chances on polling day, which allow for how far polls move in the months before a vote, are a separate forecast described in From today to polling day: drift in French presidential polls.
This paper plays the part for France that Correcting the raw answers: the method behind our UK projection and Simulating UK polling error: where the chances come from play for Britain.
Choosing the line-up
We follow two reference line-ups, fixed in the database. The main one has ten candidates:
| Family | Candidate |
|---|---|
| National Rally | Marine Le Pen |
| Ensemble (Macron camp) | Édouard Philippe |
| France Unbowed | Jean-Luc Mélenchon |
| Socialists | Raphaël Glucksmann |
| Republicans | Bruno Retailleau |
| Reconquête | Éric Zemmour |
| Greens | Marine Tondelier |
| Communist Party | Fabien Roussel |
| Sovereigntists | Nicolas Dupont-Aignan |
| Far left | Nathalie Arthaud |
The variant is the same field with Jordan Bardella for the RN. The line-ups are chosen to match what most pollsters lead with, not what we think will happen. Candidacies are not settled, and when the pollsters' lead line-up changes, the reference line-up should change with it.
No poll matches a reference line-up exactly in every scenario, so for each poll we pick the scenario closest to it. Let \(L\) be the set of line-up candidates and \(C\) the set of candidates in a scenario. The scenario is scored
so that each line-up candidate present earns two points and each candidate from outside the line-up costs three. A scenario is eligible only if it includes the line-up's RN candidate, and the best-scoring scenario counts only if it scores at least \(2(|L| - 3) - 3\), which is 11 for a ten-candidate line-up. In words: the scenario must contain most of the line-up and few outsiders. A poll that tests Le Pen only against a different centre candidate and a different Socialist still counts, with those candidates simply missing from it; a poll that tests a very different field does not.
The rule matters because the choice of scenario moves the average. Dropping a candidate from a scenario sends that candidate's voters elsewhere, and the pollsters' scenarios differ most in exactly the candidates whose presence is uncertain.
The round-one average
For each line-up, let \(d^\ast\) be the fieldwork end date of the newest poll with a matching scenario. The average uses the latest poll from each pollster whose fieldwork ended within 45 days before \(d^\ast\), with equal weight per pollster, so that OpinionWay, which has at times published several polls a week, counts no more than a pollster that publishes monthly. For candidate \(c\),
where \(J_c\) is the set of those polls that include \(c\). Polls of a sub-population, such as a survey of a single community, are excluded.
The database holds 31 round-one polls for 2027 from eight pollsters, with fieldwork from 27 February to 16 September 2026, and 9 round-two polls from six pollsters. The full list, with each scenario, is on the polls page.
| Candidate | Family | Polls | Our bet | Reaches round two | Wins |
|---|---|---|---|---|---|
| Marine Le Pen | National Rally | 32.8 | 32.8 | over 99% | 99% |
| Jean-Luc Mélenchon | France Unbowed | 18.3 | 18.3 | 95% | under 1% |
| Édouard Philippe | Ensemble (Macron camp) | 12.2 | 12.2 | 3% | 1% |
| Raphaël Glucksmann | Socialists | 11.6 | 11.6 | 2% | under 1% |
| Bruno Retailleau | Republicans | 6.0 | 6.0 | under 1% | 0% |
| Éric Zemmour | Reconquête | 3.5 | 3.5 | 0% | 0% |
| Marine Tondelier | Greens | 2.6 | 2.6 | 0% | 0% |
| Fabien Roussel | Communist Party | 2.1 | 2.1 | 0% | 0% |
| Nicolas Dupont-Aignan | Sovereigntists | 2.1 | 2.1 | 0% | 0% |
| Nathalie Arthaud | Far left | 0.8 | 0.8 | 0% | 0% |
The "Polls" and "Our bet" columns are the same, because round one is not corrected. Correcting would have been further out at 4 of the 5 past elections when each was predicted from the others. The page also shows today's round one with each past election's miss replayed, as a sense of scale: the 2002 miss added to the RN, the 2022 miss added to Mélenchon, and so on. We do not use those replays for the projection, because the misses change direction.
Round two: the pairs
For each pair among the six leading candidates, we collect every head-to-head poll of that pair and average it the same way: each pollster's latest poll within 90 days of the newest poll of that pair, equal weight per pollster. The window is longer than in round one because head-to-head polls are rarer. The first-named candidate's share of the two-way vote is
Most possible pairs have never been polled. Pollsters test the pairings they think likely, which at present means the RN candidate against each of the others.
| Match-up | Polls | Our bet | Chance it happens | First-named wins |
|---|---|---|---|---|
| Le Pen – Mélenchon | 68.1 | 66.2 | 95% | over 99% |
| Le Pen – Philippe | 55.0 | 53.1 | 3% | 76% |
| Le Pen – Glucksmann | 57.0 | 56.1 | 2% | 92% |
| Le Pen – Retailleau | 57.0 | 55.8 | under 1% | 91% |
| Mélenchon – Philippe | 34.0 | 32.1 | 0% | under 1% |
The round-two correction
At the three runoffs with an RN candidate, the final polls overstated that candidate by 3.2, 4.5 and 3.3 points. The general rule takes off half the mean past miss of each family with at least two elections of round-two evidence:
| Family | Round-two elections | Mean error | Adjustment to its share |
|---|---|---|---|
| National Rally | 3 (2002, 2017, 2022) | +3.67 | −1.84 |
| Ensemble (Macron camp) | 2 (2017, 2022) | −3.90 | +1.95 |
| Republicans | 3 (2002, 2007, 2012) | −1.20 | +0.60 |
| Socialists | 2 (2007, 2012) | +0.19 | −0.10 |
In a two-way race one side's error is the other's opposite, so each family with a record gives its own estimate of the correction to candidate A: \(a_A\) from A's family, and \(-a_B\) from B's. Where both exist we average them:
using whichever of the two estimates exist. For Le Pen against Philippe both exist, and they agree: \(-1.84\) from the RN record and \(-1.95\) from the Macron camp's, giving \(-1.9\). For Le Pen against Mélenchon only the RN record applies, since France Unbowed has never reached round two, giving \(-1.84\). For Le Pen against Glucksmann the Socialists' small record pulls the other way, giving \(\tfrac{1}{2}(-1.84 + 0.10) = -0.87\). For Le Pen against Retailleau the Republicans' record, in which their candidate was understated, adds to the RN's: \(\tfrac{1}{2}(-1.84 - 0.60) = -1.22\).
Two points about this rule deserve to be stated plainly. First, the Ensemble record and the RN record for 2017 and 2022 are the same two runoffs seen from either side, so averaging them does not add evidence; it only keeps the rule symmetric. Second, the correction applies to any RN candidate against any opponent, although the evidence comes from runoffs against Jacques Chirac and Emmanuel Macron. Whether an RN candidate would be overstated against a candidate of the left, or against Édouard Philippe, is not something three runoffs can show.
The leave-one-out test on the five past runoffs, each predicted with the correction worked out from the other four, gives this:
| Election | Polls' average miss | Corrected, leave one out | Verdict |
|---|---|---|---|
| 2002 round two | 3.21 | 1.62 | closer |
| 2007 round two | 0.73 | 1.22 | further out |
| 2012 round two | 1.11 | 0.78 | closer |
| 2017 round two | 4.48 | 3.45 | closer |
| 2022 round two | 3.32 | 2.24 | closer |
| Average | 2.57 | 1.86 |
Correcting was closer at 4 of 5. The one exception, 2007, was the smallest miss and had no RN candidate, so the correction there came only from the Republican and Socialist records.
The simulation
The chances come from 20,000 simulated elections. The random number generator is seeded, so the same polls always give the same chances; a change on the page reflects a change in the polls, not simulation noise.
Round one
In run \(k\), each candidate \(c\)'s share is drawn around our bet:
where \(\text{RMSE}_{f(c)}\) is the root mean square of the final-poll error of the candidate's family at past elections, and a family with no record is given 2.5. The floor of 1.5 points stops a family whose past polls happened to be close from being treated as certain. For the main candidates, the spreads are about 2.5 points for Le Pen (the RN's misses of 3.9, 3.3 and 2.0 points), 3.1 for Mélenchon (misses of 2.5 and 4.8), 2.7 for Zemmour (one miss, in 2022), 1.9 for Retailleau, and the floor of 1.5 for Philippe and Glucksmann.
The errors are drawn independently for each candidate. That is a simplification: in reality, when one candidate is understated, others must be overstated. Simulating UK polling error: where the chances come from explains how the British model keeps that correlation by replaying whole past error patterns. With five French elections, different fields each time and misses that change sign, we did not find a pattern worth replaying, and independent draws of the right size are the simpler choice.
The two candidates with the largest \(x_{k,c}\) go through.
Round two
If the pair \((A, B)\) has been polled, A wins run \(k\) when
The first term under the root is the largest round-two RMSE of any family, which is the Macron camp's 3.94 points (misses of 4.5 and 3.3). The second adds 2 points for movement in the fortnight between the rounds and for the time still to run before the campaign. Together they give \(\sigma_2 \approx 4.4\) points. Taking the largest family RMSE rather than the relevant one is deliberately cautious.
If the pair has never been polled, the winner is a coin toss. That is not a forecast; it is an admission that we have nothing to go on. When more than 1% of simulated runoffs fall into that case, the presidential page says so. At present almost every simulated round two includes the RN candidate, whose pairs have all been polled, so the coin toss matters little. It would matter a great deal if, say, a centre candidate and a left candidate were to overtake the RN in round-one polling.
Chances
A candidate's chance of qualifying is the share of runs in which they finish in the top two. Their chance of winning is the share of runs they win. Because round two is drawn independently of the round-one draw, the winning chance splits neatly by opponent:
The "Chance it happens" and "First-named wins" columns of the round-two table are the two factors in that sum. Reading the table this way shows where a candidate's chances come from. A candidate who is sure to reach round two can still have only a modest chance of winning if their likeliest opponent is close to them head to head; a candidate who is behind in round one can have a real chance of winning if they would beat everyone in round two, as long as they get there.
A worked example with round numbers shows the arithmetic. Suppose Le Pen and Philippe meet in round two and the head-to-head polls average 53–47 for Le Pen. The correction takes 1.9 points off, giving 51.1. With \(\sigma_2 = 4.42\), Le Pen's chance of winning that pair is \(\Phi(1.1 / 4.42) = \Phi(0.25) \approx 0.60\). A lead that looks comfortable in the polls becomes little better than a coin toss once past runoff misses are taken into account.
The Bardella variant
The same calculation runs on the line-up with Jordan Bardella in place of Marine Le Pen. The RN's family record applies to him unchanged: the round-one spread and the round-two correction are the same, because they belong to the family, not the candidate. Whether polls would miss Bardella as they missed the Le Pens is an assumption, not something the record can show.
The variant rests on fewer pollsters, and its newest poll is older than the main line-up's: at the time of writing, no poll in the database has tested Bardella since late June 2026, while the main line-up was polled several times in September. Each line-up's 45-day window is counted back from its own newest poll, so the variant can describe the state of opinion several months earlier than the main line-up does. Its polling-day forecast also starts from that older date, and so carries more drift (From today to polling day: drift in French presidential polls). The two line-ups are shown side by side on the presidential page, not combined. We do not know the probability that Bardella stands instead of Le Pen, and a weighted average would hide that we don't.
What we tested and did not use
Correcting round one. The same rule that corrects round two would, applied to round one, move no family by more than half a point: about 0.4 points onto France Unbowed and the Macron camp, 0.5 off the Communists, 0.3 off the far left, 0.2 onto the RN. Reconquête, with one election, would not be corrected at all. Tested on each past election without knowing its result, it was further out at four of the five. The reason is visible in the record: the families with the largest misses (the RN, the mainstream right and France Unbowed) are the ones whose misses have changed sign.
Raw figures. In Britain, our projection starts from the weighted answers of everyone asked, before the pollster's turnout and don't-know adjustments (Why we start from British pollsters' raw answers rather than the published headline). Of the French pollsters, only OpinionWay and Cluster17 print raw vote shares for 2027 in their notices, and they point in opposite directions. Across the scenarios on file, OpinionWay's raw figures put the RN candidate on average 2.6 points above its published figure and Mélenchon 2.9 below; Cluster17's raw figures put the RN candidate 3.9 points below and Mélenchon 1.5 above. A single "raw answers lean this way" correction cannot fit both panels, and the historical raw record (a handful of final polls from OpinionWay, Ifop and Kantar) is too small to test one. The raw figures are shown on the polls page.
Weighting pollsters by their record. The pollster ratings (Rating French pollsters: accuracy, the lead and openness) are shown beside the polls but do not weight the average. Pollsters' records of final round-one polls span between one and five presidential elections, and at that length the differences between them are small beside the differences between elections.
What the forecast does not capture
Who stands. The forecast follows two line-ups. A candidate outside them, or a line-up that pollsters have not tested, is outside the model.
Correlated round-one errors. Candidates' round-one errors are drawn independently, so the simulation understates the chance that two candidates of the same camp are both missed in the same direction.
Transfers between the rounds. Round two is taken from head-to-head polls, not modelled from where each eliminated candidate's voters go. For pairs that are polled, that is the better evidence. For pairs that are not, we have nothing.
The thinness of the correction. The round-two correction rests on three runoffs. The pattern is consistent, but a fourth runoff that went the other way would halve it.
Time. Everything here is for an election held today. 148 days before round one, polls move far more than their final-week error, and the polling-day chances in From today to polling day: drift in French presidential polls are the ones to read for the real election.
References
Efron, B. and Morris, C. (1975). Data analysis using Stein's estimator and its generalizations. Journal of the American Statistical Association, 70(350), 311–319.
Jennings, W. and Wlezien, C. (2018). Election polling errors across time and space. Nature Human Behaviour, 2(4), 276–283.
Stone, M. (1974). Cross-validatory choice and assessment of statistical predictions. Journal of the Royal Statistical Society: Series B, 36(2), 111–147.
Loi n° 77-808 du 19 juillet 1977 relative à la publication et à la diffusion de certains sondages d'opinion.
How to cite
Polling Forecast (2026). The 2027 presidential forecast: line-ups, runoffs and simulation. Retrieved 21 November 2026 from https://www.pollingforecast.com/research/france-presidential. Live figures change as new polls arrive; the date above is when the text was last revised.



