Why Bettors Take a Flier on Trump
Polls are one way of predicting which candidate is leading in a campaign. They’ve been pretty accurate in U.S. presidential elections, though in Europe they’ve often been misleading in recent years and in my country, Russia, they are all but useless.
Gambling is another indicator of public sentiment. You can’t place a legal bet on candidates in the U.S., but you can in parts of Europe, and there people have put their money on Donald Trump, ignoring his weakness in the polls. At William Hill, the U.K.’s largest bookmaker, 70 percent of the money is riding on Hillary Clinton -- but 65 percent of the individual bets are on Trump. In recent days, the money balance appears to have been shifting, too. On Monday, William Hill spokesman Graham Sharpe told me, 79.6 percent of the money bet on the U.S. election was for Trump, with 55.2 percent of individual bettors going in his favor.
Sharpe’s explanation is that Trump offers more attractive odds. “If you are going to bet, say, $20 on the outcome and you think Clinton will win, at the moment your $20 will make you a profit of $6.66 -- you probably won’t think that is an attractive return so won’t play,” Sharpe told me by e-mail. “But should you think Trump will win, your $20 can make you a profit of $50. Hence there are currently more bets for him.”
Serious amounts are being bet on the Republican nominee. The biggest individual bet on Monday was 7,000 pounds (about $8,500). People wouldn’t bank so heavily on Trump if they didn’t think he had a chance. Polls aren’t the only metrics favoring Clinton – most existing econometric models also forecast a Clinton win. Still, here are some serious and whimsical indicators to give Trump bettors hope.
- Social network engagement and Google searches. “Big data” analysis is an increasingly popular crutch for investors and analysts who don’t fully trust polls. More people are searching for Trump on Google, and "vote Trump” is a far more popular search request than “vote Clinton.” On the USA Today/Facebook candidate barometer, Trump demonstrates much better Facebook engagement, seen as the combined number of likes, comments, shares and posts. Most of those who “like” and share posts mentioning Trump could be haters, not backers. Yet a study of Twitter activity, which uses content analysis, shows that Trump has a consistently better ratio of positive to negative tweets than Clinton. That’s a warning to Clintonites: Brexit supporters dominated the social networks before the U.K. voted to leave the EU in June. An artificial-intelligence system that analyzes social network data, built by Shajiv Rai, founder of the Indian startup Genic.ai, in 2004, predicts a victory for Trump. It accurately called the last three U.S. presidential elections.
- Economic indicators. Several econometric models that predict electoral outcomes based on the incumbent party’s economic performance forecast a change of ruling party this year. Alan Abramovitz of the University of Virginia, whose model has accurately called the popular-vote winners since 1988, gives Clinton 48.6 percent of the two-party vote (with the caveat that Trump’s personality may well move the needle in the Democrat’s favor). A 2009 model by Yale’s Ray Fair reflects the relationship between economic indicators (output growth and inflation) and votes cast for the incumbent party’s candidate. This model has consistently predicted a loss for the Democrats this year. As of Oct. 28, Fair’s model gives Clinton 44 percent of the vote in a two-way race. In 2012, Fair’s equation gave Obama 49 percent; he ended up with 51 percent, within the model’s 2.5 percentage-point margin of error. If Clinton wins, it will be the first time the model will prove inaccurate. Unemployment dynamics and the ISM Manufacturing index, which have been good predictors of election outcomes, also give an advantage to Trump as the change candidate.
- Consumer confidence. The measure for October of the presidential election year has long been highly predictive of the outcome of the race, as Mitsubishi Bank vice president Michael Nemira argued in a 1992 paper. Nemira used a simple formula involving the University of Michigan’s Index of Consumer Sentiment that has only failed in 2000 (when the incumbent party’s candidate did win the popular vote but still lost the election) and in 2012 (when one could argue that consumer confidence hadn’t recovered after the 2008 financial crisis). This year, the formula predicts a loss for Democrats. There’s another, even simpler way to use consumer confidence to predict presidential race outcomes. Since 1968, the incumbent party has won nine of the 12 elections when the Consumer Confidence Index was above 100. The exceptions were 2012, 2000 and 1968, when President Lyndon Johnson decided to forego a re-election bid amid turmoil over the Vietnam War. It stands at 98.6 now, predicting a victory for the Republican nominee.
- The fiscal model. Based on the observation that growing federal spending relative to the gross domestic product reduces the incumbent party’s chances of winning, it failed to predict Barack Obama’s re-election in 2012. Its author, Alfred Cuzan of the University of West Florida, has since adjusted it. The model predicts a loss for the Democrats, with 48.2 percent of the two-party vote.
- Helmut Norpoth’s models. The SUNY at Stony Brook professor has two. One is based on the candidates’ primary performance: The one who does better in his party’s early primaries wins the election. This has been true in every election since 1912 except the 1960 one. The model predicts a strong likelihood of a Trump win. The other one, based on the view that “like sunspots, elections run in cycles” -- an approach to politics reminiscent of the technical analysis traders use -- projects a victory for Trump with 51.4 percent of the popular vote. Norpoth is on record as saying that polls are “bunk” because they are about opinions, not actions. He is convinced Trump is headed for victory and has bet on him on the Iowa prediction markets.
- Allan Lichtman’s “13 Keys to the White House.” The American University professor’s previously accurate approach is also largely based on the outgoing administration’s performance. Of the 13 indicators, including the state of the national economy, social unrest, scandals, the incumbent’s foreign policy success and both top candidates’ charisma, five should point in the challenger’s favor for him to win. Lichtman predicts a Trump victory.
- ISideWith.com. Since September, 2016, this site has offered a questionnaire allowing users to check their positions against those of the presidential candidates. More than 275,000 have done so thus far; according to the resulting data, Trump should carry 41 U.S. states. If the issues were at the forefront of this campaign and people actually voted for the candidate with whom they mostly agree, Trump would win, according to this large but unrepresentative sample.
There are good reasons to ignore each of these predictions. Big-data analysis is a young discipline, and the methods are still relatively untested. All the models based on economic performance and the previous administration’s performance could be unreliable in this year’s race because it is so intensely personalized and scandal-ridden. Besides, these models do not boast perfect historical accuracy.
On the other hand, the academic heft and logic behind the contrarian predictions gives a gambler enough to go on for a shot at bigger returns than Clinton offers.
I doubt most William Hill clients analyze all these data before they make a bet: Many of them are blue-collar Brits who couldn’t care less about U.S. professors and social-network content analysis. They have merely seen mainstream experts and pollsters fail in their own country, and they assume the same might happen in the U.S. They are betting on their disbelief in the establishment.
This column does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners.
To contact the author of this story:
Leonid Bershidsky at firstname.lastname@example.org
To contact the editor responsible for this story:
Jonathan Landman at email@example.com