By Jemima Kelly
LONDON (Reuters) - With France's presidential election a potential banana skin-in-waiting for markets, forecasting sites that rely neither on voter surveys nor real-money betting are becoming a tool of choice for financiers and policymakers in search of an edge.
After years of anxiety about the accuracy of opinion polls, investors had increasingly come to rely on political betting odds as a more consistent barometer of election outcomes affecting billions of dollars of holdings.
But 2016 changed all that, as bookies and pollsters misfired over Britain's referendum vote to leave the European Union and Donald Trump's U.S. presidential election win - though they partly redeemed themselves in last month's Dutch election.
Some reckon France could deliver a similar shock next month: a presidency for the far right, anti-EU Marine Le Pen.
Though polling has consistently shown her losing the May run-off by around 20 percentage points, bookies and U.S. prediction markets put Le Pen's chances of winning at around 1 in 4 – comparable to the odds on a Trump victory on U.S. election eve.
So, with opinion surveys also consistently showing a third or more of French voters could still change their mind, how do markets navigate these murky waters?
"IT'S HOW YOU THINK"
Offering a different perspective are Hypermind, a company that allows punters to "bet" on the outcomes of geopolitical events using play-money, and other crowd-forecasters like U.S.-based Good Judgment Inc.
Hypermind, which selects amateurs for its panel via a test, says some eight in 10 of its almost 2,500 forecasters on the French election come from France.
Its system gives those who consistently predict outcomes correctly an increasingly strong voice, while eventually dropping those who do not - and it puts the chance of a Le Pen victory at around 1 in 8.
"What makes a good forecaster is not previous expertise in the domain; it's not how much you know about what you're trying to forecast," said Hypermind co-founder Emile Servan-Schreiber, who has a background in cognitive psychology. "It's how you think."
He declined to identify any of his clients but said they included hedge funds, banks and government agencies.
That may be down to Hypermind's track record.
Its research shows that, in the two months before the event, it assigned the highest chances of Brexit and Trump out of eight statistics-based and crowd-based forecasters, including U.S prediction market PredictIt, British betting site Betfair and website FiveThirtyEight.
In research after Trump's victory, The Washington Post put Hypermind and the Good Judgment Open - a project of Good Judgment Inc based on the same principle - as the most accurate forecasters.
Hypermind grew out of research funded by the Intelligence Advanced Research Projects Activity (IARPA), a U.S. government body set up in 2006 to develop new methods for the secret service work in the post-9/11 era.
Between 2011 and 2015, IARPA ran a contest to predict the outcomes of around 500 geopolitical events.
The winning group was the Good Judgment Project (GJP), which used similar methodology to Hypermind to produce collective “wisdom of the crowd” forecasts.
Hypermind was one of two commercial spin-offs by GJP participants. The other was Good Judgment Inc.
Once chosen, Hypermind's forecasters are given 100,000 units of "play money" to bet with. The more accurately they predict, the more they accumulate, but if they lose it they fall out of the panel, which happens to around 1 in 10.
Hypermind is UK-registered but was set up by two Frenchmen. Its website is in French and English and this has helped to attract many forecasters in France.
Hypermind does not ask the nationality of panel applicants, but given that most of them are based in France, Servan-Schreiber reckons Hypermind's forecast is more representative of the French vote.
"The further you are removed from French territory, the more you see Le Pen as a high probability," he said.
UK bookmaker Ladbrokes says some of the people who bet on Trump and Brexit - and reaped the rewards - are trying their luck again with Le Pen, despite the strong lead enjoyed by centrist candidate Emmanuel Macron.
That is skewing bookies' odds in her favour. There is also a lack of understanding of how France's two-round voting system works, and the bookies' reluctance to leave themselves out of pocket.
On Wednesday Ladbrokes online odds of a Le Pen victory were 3 to 1, reflecting around a 25 percent chance.
"Le Pen's odds are overestimating her chances of winning," said the firm's head of political betting, Matthew Shaddick.
"If the people who were actually voting and really understood the system were allowed to participate in these betting markets, I suspect her odds wouldn't be quite so favourable."
In France, betting is restricted to sports and horseracing.
While it might have done better than others over Brexit and or Trump, Hypermind gave a less than 50 percent chance of each outcome.
But for Servan-Schreiber, assigning 1 in 3 odds on an event that occurs is not "getting it wrong". Rather, it reflects that one in three times, it will materialise.
For a graphic on the French election with prediction market charts click http://fingfx.thomsonreuters.com/gfx/rngs/FRANCE-ELECTION/010031D933E/index.html
(Reporting by Jemima Kelly; editing by John Stonestreet)