* Investors turn to AI tools to divine election outcome
* Fund managers also use private polling, sports betting
* Say they can't rely on public opinion polls after errors
* Dec. 12 vote is unpredictable and financially pivotal
By Saikat Chatterjee, Sujata Rao and Tommy Wilkes
LONDON, Dec 3 (Reuters) - Question: How do you predict the outcome of a snap election when so many polls have been so wrong, half of voters haven't made up their minds and the crucial factor may not be who wins, but how much they win by?
So financial investors, who could make or lose fortunes on the result of Britain's Dec. 12 vote, are taking matters into their own hands and arming themselves with every predictive tool at their disposal - from artificial intelligence analysis to private polling and sports betting techniques.
Faced with such an unpredictable and financially pivotal election, fund managers say they can't put their faith in pollsters who failed to accurately call the last two British elections and dramatically dropped the ball on Brexit and Trump.
"In the past opinion polls accounted for 85% of your input, now maybe it's 30%," said Stephen Jen, macro hedge fund manager at London-based Eurizon SLJ. "The world has become so complicated that polls, the standard metrics of the past, don't capture the picture anymore."
Reuters interviewed more than two dozen big investors, including Aviva, Legal and General, NN Investment Partners, State Street Global Advisors and M&G Investments.
Most said they had turned to AI-based tools that variously analysed news reports, social media, web traffic, opinion polls and betting odds to decipher how the vote might play out.
Brexit has scrambled traditional political allegiances, with Prime Minister Boris Johnson's ruling Conservatives pledging a swift split from the EU, the main opposition Labour promising another referendum and others vowing to stay in the bloc.
It will not be enough for Johnson simply to win more seats than any other party - he needs a parliamentary majority to ensure Brexit happens. Any other outcome is likely to lead to a second referendum on EU membership.
Given the stakes, recent studies show almost half the electorate are now considered undecided "floating voters".
"Frankly, this election is by far the most difficult to call in living memory," said Edward Shing, global head of equity derivatives strategy at BNP Paribas.
SCRAPING FOR SENTIMENT
Valentijn van Nieuwenhuijzen, who oversees almost 300 billion euros at NN Investment Partners, said he was using the social media analysis services of MarketPsych to monitor the mood among voters.
MarketPsych, which sells its products through data company Refinitiv, tracks almost 3,000 sites, searching for expressions related to certain subjects.
Ahead of elections, it can track public sentiment towards issues like a change of government and social tensions, said Eric Fischkin, director quant and feeds at Refinitiv, in which Thomson Reuters, the parent company of Reuters News, holds a 45% stake.
"On both those parameters, the index is at its highest since the Brexit referendum," he added. "If those indexes are marching upwards during an election it means that a higher share of the total media buzz is around those topics."
Using social media analysis to make political calculations is largely unproven. However Henrik Mueller, a professor at Dortmund University, argues it could be a key tool to gauge the mood of Britain's highly charged political scene.
In a paper for the Bruegel Institute think-tank, he detailed how Twitter-derived sentiment analysis he had conducted before the 2016 Brexit referendum reflected the swing towards Leave sooner and more accurately than opinion polls and bookmakers.
"The UK situation is unique, it's an ideal environment for this kind of analysis because the public is polarised around one question - Brexit - which is easier to track on social media," Mueller told Reuters.
But such techniques failed to yield clear signals for the 2018 Italian election, and before the 2017 French election at least one social media "scraping" company wrongly predicted victory for far-right Marine le Pen.
Demand from investors has spawned a raft of companies selling various brands of data analysis. U.S.-based Predata, for example, applies machine-learning algorithms to web traffic data to generate indicators that it sells to financial customers.
The company, which studies the sources and scale of traffic rather than content, said analysis in early November showed a spike in research into new voter registrations in Britain.
This was driven by people browsing on phones, rather than computers - with the mobile-based traffic four times higher than at a similar point before the 2017 election.
"One assumption we can make is, this is generally associated with a younger audience," said Eric Falcon, Predata's director of research.
Engagement by younger voters, who are likelier to be anti-Brexit, could be considered a negative for the Conservatives.
'MADE A TON OF MONEY'
British parliamentary seat projections are complicated by a first-past-the-post voting system - in each of the 650 electoral districts, the candidate with the most votes wins the seat. This can undermine gauges of the overall national vote.
While opinion polls show the Conservatives around 10 points ahead, BNP Paribas noted that at this point in the 2017 campaign, the party had an even bigger lead, only to see it evaporate by election day.
A model created by pollsters YouGov shows the Conservatives on track to win 359 seats but has a margin of error of 50 either way - the difference between a majority and a hung parliament, potentially Brexit and no Brexit.
"The uniqueness of the UK election makes polling less useful," said Peter Fitzgerald, chief investment officer for multi-asset at Aviva Investors. "What you actually need to do is effectively look at 650 mini-referendums."
Said Haidar, chief investment officer of macro hedge fund firm Haidar Capital Management, said he found bookmaker betting odds a more accurate predictor than opinion polls.
Using odds, as well as following news reports and polls, he said he correctly predicted the result of the Brexit referendum and "made a ton of money" in related bets.
Betting exchanges such as Betfair have indeed done a better job of predicting the outcomes of some votes; a Cambridge University study credited punters with cottoning on to the result of the Brexit referendum hours before financial markets.
'NO BAD THING FOR RISK-TAKERS'
Yet betting markets may not provide as good a gauge for elections as they do for binary referenda. The amount wagered is also tiny against the multi-billion-dollar currency and bond markets.
In an indication of investor uncertainty, average daily cash volumes for sterling traded against the dollar and the euro fell to $45 billion in November as players sat on the sidelines ahead of the election, according to CLS, a major settler of foreign exchange trades. That compares with the $87 billion traded before the Brexit referendum in June 2016.
Despite polls consistently showing a healthy Conservative lead, currently seen as bullish for sterling, it has remained below $1.30, pointing to markets' wariness about the projections.
Some investors, such as Aviva's Fitzgerald, have commissioned private polls, while others have sought to extract the maximum value from the public surveys using data analysis companies like Neudata.
Neudata provides hedge funds with metadata sets, assessed and ranked by their track record and can provide sets showing which pollsters have been most accurate during past elections, said founder Rado Lipus.
It remains to be seen which strategies, if any, will prove effective, but Jen of Eurizon SLJ said a lack of faith in opinion polls could work to investors' advantage.
"If polls are not reliable you can argue that's not a bad thing for risk-takers," he said. "If nothing is priced in, you don't need some extraordinary event to make money, you just need to beat the average."
(Additional reporting by Thyagaraju Adinarayan and Simon Jessop in London and Lawrence Delevingne in New York; Editing by Pravin Char)