Smartphone on top of list of figures

If you look at the polls, the election is close, and getting closer. But Australia’s bookies think the Coalition has it in the bag.

Last week Australia’s major bookies were offering odds of $1.28 to $1.30 on a Coalition win this election. The shortest odds for Labor were offered by Unibet, at $3.40.

Bookies and pollsters generally ask fundamentally different questions: two-party polling asks respondents who they intend to vote for if an election were held today, while bookies ask them who they think will win — sometimes years in advance.

But Newspoll also occasionally also asks voters who they think will win the next election. Last Monday’s Newspoll had 44% of respondents picking the Coalition, with 33% believing Labor would be victorious. That’s up from 25% backing Labor the last time the question was asked in March (when 55% thought the Coalition would win). We can express the odds offered by betting markets in percentage terms to offer a comparison. Odds of $1.30 translate to a 76.92% chance of a Coalition victory. Even with a similar question, the bookies are far more certain Turnbull will win than poll respondents.

Last election, the odds of a Labor victory drifted as long as $12.50 (or percentage odds of 8%). This election is relatively closer. But still, though Coalition’s odds have lengthened slightly this year, the bookies have consistently favoured the it, beyond the tightening polls.

Why is this? And with the bookies predicting a different result to the pollsters, how seriously should they be taken?

Election betting in Australia has a long history, but it isn’t until recently it was done on a commercial scale. It was illegal until the Commonwealth Electoral Act was amended in 1983 (it used to contain a clause banning “wagering on the result of any election”). Today, it isn’t explicitly dealt with in Commonwealth or state regulation, with the exception of New South Wales, where election betting is illegal but the ban is unenforceable for practical reasons.

There were two main reasons for why governments were, and sometimes still are, reluctant to allow political betting, says Graeme Orr, a professor of law at at the University of Queensland, who has studied political betting in Australia.

The first was moral. “It was seen as unseemly,” Orr said. “Politics isn’t a game.”

The second reason was more technical. When electorates were smaller (as they were in Australia’s early history, when only men could vote), political betting could be a way to disguise bribery. A candidate could enter a wager with voters that he would lose an election. The voters would then vote for that candidate and receive a payoff from the candidate (due to the losing bet) once he won — a bribe disguised as a bet. Or it could be used by those in power to profit those close to them. Many political decisions are known only to a small group before they are made public — novelty betting pools on these decisions could allow them to profit from this information, creating a major moral hazard for decision-making.

Those days are gone now. Online sports betting has increased the visibility and volume of election betting. Every election, the figures get bigger. In 2013, more than $5 million was placed with Sportsbet on the federal election — a figure the bookie says it’s going to easily surpass this time round. Online bookies in Australia generally make at least half of their revenues from bets on horse racing, but outside of that, the election brings in a lot of punters, and a not insignificant amount of media attention to boot. In election years, it’s eclipsed only by the NRL and AFL grand finals in terms of the amount of money placed, according to Sportsbet’s Ben Bulmer.

How does political betting work?

Asked how exactly Sportsbets election odds are decided, Bulmer explains that they’re picked and calibrated by a full-time politics trader who looks at a range of factors, including how the punters are betting, polling, and media coverage, to decide on the odds. It’s a subjective process. And when it comes to the market on who’ll win the election, it’s built on individual seat predictions, rather than a national two-party preferred vote.

Sportsbet offers odds on the 150 individual seats — and then adds the expected result on these up to reach its predicted election results. Based on this analysis, Sportsbet believes the Coalition will win 79 seats — a clear election victory.

Offering odds on all 150 seats is a tricky business, and unlikely to be very accurate whether done through polling or bookmaking because of the difficulty of getting enough people, or enough information, about the dynamics in every seat. But while the two-party preferred vote (what’s measured by pollsters) is more rigorous, seat-by-seat outcomes are important because only a handful of seats change hands every election, and the outcome of these changes determines who is in government.

The punters disagree with Sportbet’s odds in several key battlegrounds. In David Feeney’s seat of Batman, for example, Sportsbet him likely to retain it with odds of $1.50. But Bulmer reveals 90% of the money placed so far is on Greens candidate Alex Bhathal — Sportsbet has shortened her odds from $9 to $3 in response. In Brisbane, the odds on Coalition candidate Trevor Evans winning are $1.62. But again, 75% of the money placed has been on Labor’s Pat O’Neill. Sportsbet can adjust its odds to take this into account, but how the money is flowing isn’t the only factor it looks at, Bulmer says.

How reliable is it to watch the betting markets, as opposed to the polls?

In America, betting markets have had success in easily and confidently predicting election winners when pollsters have assessed a close race. In 2012, for example, most national surveys had Mitt Romney and Barack Obama neck and neck, but betting markets considered Obama a “heavy favourite”, as The Economist stated in its January edition, which explored how this has happened several times in recent memory (though in Australia, bookies didn’t predict Campbell Newman’s demise at the last Queensland election).

Economists expect bookies to be reliable. Polling forces people who might not have thought much about the election to say who they’re personally backing, while betting forces people who’ve thought about the election a lot to put their money where their mouth is. And if a pollsters odds are off, it stands to lose a lot of money. For these and other reasons, studies have shown betting markets make excellent predictors. Economists have been known to advocate betting markets on where the next terrorist attack or assassination will be — while it’s morally repugnant to give those with knowledge of crime the capacity to benefit from that knowledge, it could help law enforcement prioritise their efforts on what insiders think are the greatest risks.

But Orr is a critic of such thinking.

People like having a punt, and Orr says there’s nothing wrong with it. But when betting turns from a wager between friends to global commoditised markets, the opportunity for corruption blooms. Betting markets are theoretically more predictable than polling because those with greater, or inside, information have every incentive to bet. But Orr says this wouldn’t be a good thing — people shouldn’t be benefiting from their knowledge of democracy’s inner workings. The extent to which they can provides ample opportunity for corruption.

And anyway, Orr adds, betting markets and most political polling do very different things. “The point of opinion polls isn’t to predict the outcome of elections — it’s to measure the mood of the electorate at a particular point in time. That’s good — it can give feedback and put pressure on governments about policy.” Betting on who’ll win an election doesn’t do this, and feeds into the horse-race narrative of political discussion (as betting on horse racing was the only form of betting allowed in Australia for much of its history, the metaphor is particularly apt).

There are other criticisms of political betting markets. Last election, former Fairfax scribe Michael West called up Tom Waterhouse (now William Hill), Centrebet and Sportsbet to ask them whether they’d take a large bet of $10,000. Over the phone, they wouldn’t, which is why West was sceptical Australia’s election betting markets really existed. “If the market exists at all [it’s] to lure journalists into writing it up for free advertising in the press,” he wrote (Sportsbet sent Crikey a list of some of its largest on the election, which included $30,000 at $1.34 — last month — on the Coalition to win the election, and $7,500 at $1.90 — in late 2014 — on Labor to win the election).

It’s undeniable that as an industry, political betting markets in Australia are not very transparent. Markets reveal information well when demand and supply are allowed to alter prices unhindered, as happens, for example, in the share market. That’s not what happens with betting through a bookie — the bookies set the odds. While how the punters are voting will influence the odds, it doesn’t happen instantaneously, and isn’t the sole factor. Those watching the prices externally can’t gauge demand and supply transparently. While polling methods are also proprietary and take a bit of experience to make sense of, they are the result of decades of institutional experience in assessing the mood of the electorate. And pundits have been dealing with polls for decades and are familiar with how to interpret them.

One 2014 study found Australia’s bookies themselves had a heavy reliance on polling in pricing their markets. Simon Jackman, of the United States Studies Centre at Sydney University, has tracked the polling-to-betting odd relationship in the 2013 election. In a 2014 paper on the subject, he wrote of how the data “clearly indicate the markets are responding to the polls”, incorporating their movement into the odds within 48 hours.

He concludes his paper by pointing out that given pollsters ask a hypothetical “who would you vote for today” question, it’s perhaps no wonder betting markets will diverge from polling results. And as an election draws nearer and the question gets less hypothetical, it’s no wonder the two draw nearer.