Why the gap between bookmakers and pollsters could bode well for Trump

If the betting markets accurately reflect the true state of play, then the US president stands a reasonable chance of victory.

Sign Up

Get the New Statesman's Morning Call email.

Throughout this 2020 US election cycle, bookmakers have consistently offered higher odds on President Donald Trump’s chances of winning a second term than those afforded him by prominent forecasters. 
 
A quick look at a betting aggregator site will tell you that the odds being offered on a Trump victory are hovering around 9/5 - or an implied probability of 36 per cent.  In contrast, the New Statesman’s own model currently gives Trump a 13 per cent chance (a figure in line with other forecasts also using polling data and historic voting patterns to calculate likely outcomes).  
 
Pollsters and forecasters have experienced some big perceived misses in recent elections, leading to doubts about their accuracy. But are betting markets to be better trusted as a marker of voters’ intent? And do the improved odds bookmakers are giving Trump suggest he has reason for quiet optimism after all?

 

Before the advent of modern political polling, betting markets were used in the US as a way of gauging the likely election winners. Despite often being illegal, such betting was widespread in the late-19th and early-20th centuries, with newspapers known to print near-daily price information in the final weeks of races. 
 
As an article in The New York Times explained on 7 October, 1924, “The Wall Street odds represent the consensus of a large body of extremely impartial opinion that talks with money and approaches Coolidge and Davis as dispassionately as it pronounces judgment on Anaconda and Bethlehem Steel”.  
 
The perception of the betting markets at the time was that they could aggregate all kinds of information and boil it down to a bottom line: the amount of their own cash that punters were willing to risk. If polling was flawed because it relied on respondents following through with what they say they intend to do, betting markets represented a more concrete form of action. And according to analysis by Paul W. Rhode and Koleman S. Strumpf, election betting patterns in this period were good at predicting winners. 
 
In 2020, if the betting markets accurately reflect the “true” state of play, then Trump stands a reasonable chance of winning a second term. Although the odds of his overall victory have diverged over the past month or so, Trump leads Biden in a number of key swing states, including Iowa and Ohio, and he has a better chance of winning in Florida, North Carolina and Pennsylvania than he did at the same stage in 2016, according to historical betting patterns from the Betfair Exchange. 

 

Yet betting markets can also be poor barometers for election outcomes. They can be manipulated, intentionally or otherwise, by the size and volume of bets being placed and the demographics of those betting. Big bets backing Mike Bloomberg during the Democratic primaries, for instance, could help explain the brief period when he was considered the favourite. 
 
The fact that it is still illegal for US bookmakers to offer political betting could also play a part here. Though it is still technically possible for US citizens to wager their money online with a foreign bookmaker, most bettors on US politics are not American and may not, therefore, predict the outcomes of American political races as accurately.
 
Betting markets' powers of prediction also appear to be waning. You can still find successes – like the better chance of victory given to the UK Conservative Party in the 2015 election that that given by the polls – but there are also some big misses. In 2016, the markets went the wrong way on Brexit and on the presidential race: one bookmaker, Paddy Power, famously paid out more than $1m on a Clinton victory weeks before polling day. 

[See also: Downing Street isn’t ready for a Joe Biden victory – but not for the reason you think]
 
The proliferation of modern polls could be partly to blame. Research by Robert S. Erikson and Christopher Wlezien in the journal Political Studies found that betting on US elections became less accurate as a predictor of their outcomes from 1936 onwards, as polling became more widespread. “Even when the polls were inaccurate (e.g., 1948) the market followed the polls. If there was information beyond the polls in 1948 to suggest that Truman might win, bettors in the election markets did not see it coming,” the authors write. The same could be said of Trump’s win in 2016, an outcome that the betting markets – possibly driven by the polls and forecasting models  didn’t see coming. 
 
This year’s odds are slightly different, however, because they are more upbeat on Trump’s chances than the polls and forecasting models. Research by the Betfair Exchange has found that events such as debates tend to move their markets more than outlying poll results do, for instance, suggesting that bettors are considering other information as well. 

 

“Bettors on the Exchange take into account reams of data, such as current affairs, historical trends, debates, and, yes, polling data,” explains Betfair spokesperson Darren Hughes.  “There are swings in betting markets based on polls, but also based on many other events. For example, the first US presidential debate saw Joe Biden’s chances improve by over five per cent on the Betfair Exchange market.” Other set-piece campaign events also tend to result in big swings in the odds, with Betfair finding that the biggest movements in this year’s cycle occurred during each party’s convention.

Given the size of the gap between the poll-based forecasts and the betting markets this year, a Trump victory could be excellent PR for the betting markets’ supposed abilities to predict the future. But even if he does lose, the power of the crowd to predict events is something that has become fashionable of late and is unlikely to go away. Philip E. Tetlock’s book Superforecasting: the art and science of prediction has caused a stir on both sides of the Atlantic, with Boris Johnson’s chief aide Dominic Cummings being a fan. (Tetlock’s Good Judgement Project website – which invites users to make predictions on all manner of events and weights each user’s prediction based on how successful their previous predictions have been currently gives Trump a 30 per cent chance of victory.) 
 
The idea that the collective power of the masses could result in a purer form of knowledge than that possessed by the pollsters, who have been, until now, the elite gatekeepers of political prediction, is a very Trumpian one after all.

Patrick Scott is the data projects editor for the New Statesman Media Group

Free trial CSS