US Presidential Election 2020:
Guest article from J.P. Posted 3rd November 2020
Why are betting markets so much more bullish on Trump than opinion polls and models?
I make a living from figuring out the answers to these sort of questions as they relate to sports, and the global football betting market in particular. But the sort of difference between polling models and the betting market prices we have seen on the 2020 US election over the last few weeks basically never exists in football match betting these days.
That is because such a pricing difference is always closed by weight of money, towards the position of the models, by professional gamblers and particularly organised syndicates such as the one I work for. The pricing of high-profile professional football games is efficient. Not perfectly efficient mind - or else I wouldn't be able to make a living - but pretty damn efficient. The data-based models that we use to price a game are a far better guide to the likelihood of the result than any subjective opinions, including those of 'experts' such as ex-players and tipsters. These models are why we win, whereas virtually all 'amateur' football bettors lose.
The way we approach building our models, and the principles with which we use data within them are - very generally speaking - similar to the methods used by Nate Silver at 538.com. That is to say, if we were to model the US election with a view to betting on it, the output of the model we would build is likely to be similar to the 538 forecast. It would be a probabilistic expression of odds, rather than a prediction of who will win. It would use the results of opinion polls as its primary input, as they are best - albeit far from perfect - representation of the likely distribution of votes. And it would show, with a polling lead of around 10% pts, that the odds for Biden to win are around 90%, which we would usually express in their decimal form: 1.11.
The current odds (at 9am on November 3rd) available to back Biden to win on the biggest betting exchange Betfair are 1.62 (62%). In our world, that is a colossal difference. And it would mean we would be considering a 'maximum bet' on Biden, subject only to consideration of factors which could affect the result which are out-with the scope of factors we include in the model - such as the potential for corruption in the administration of the election, or the complicating impact of Covid.
The coffers in our 'politics betting' pool would be healthy, as we wouldn't yet have spent all the winnings we made on the 2016 US presidential election. On that occasion we would have backed Trump with a near-maximum stake, as our (read 538's) model gave Trump a much bigger chance of winning than the betting market, despite it projecting that Hillary Clinton was by far the likelier winner. This seeming logical contradiction is the key to a professional betting operation that most people don't grasp - we bet on the things that have a better chance of happening than the odds imply, not on the things that we think are most likely to happen.
So why is there this enormous difference between models and betting markets?
With the obvious caveat that politics isn't really 'my thing', and so it's probably smart to assign less weight to my opinion than if it was about a sporting market... for whatever they are worth, here are some thoughts;
Sometimes there is one big reason for things. This is not one of those times.
In sport a huge difference in model and market odds can occasionally occur because of a single factor which makes a large difference - such as Lionel Messi getting injured ahead of a Barcelona game. But that's not what seems to be going on here with the '20 election market. More likely it's a confluence of factors at play.
- Mistrust of polls. It has become fashionable to think that opinion polls are 'wrong' and can't be trusted - indeed even that the smart thing is to go with the opposite of what they say. This is mildly idiotic. The polls in the '16 US election and for Brexit weren't all that wrong - they only missed by a small amount. But the implication of the small polling errors was that the result changed from what the polls 'predicted'. And most people tend to think about the future in binary terms, not as a range of probabilities. Nobody bats an eye when the polls underestimate the favourite by a few points when that favourite still wins. So, the 'polls were wrong' has become an accepted truth. But that is a bad way to interpret what happened, and a bad way to think about what will happen in the future. Polls are not predictions. They are just data points, which when taken in the form of, say, a polling average, give you a reasonable guide to the event they are sampling.
- Fighting the last war. It is human nature that we give most weight to the thing that happened last, even if it is smarter to use a weighted average of a bigger sample size. So, 'Trump won last time despite trailing in the polls' is a powerful argument to many, including those who still bear the emotional scars from it. To them, refusing to count out Trump this time is a form of 'last disaster bias'. But this ignores the rather significant context though that Biden's lead is bigger than Clinton's was, and that he is less unpopular then she is. Dubious theories such as 'shy Trump & Tory' voters prosper because they fit the narrative of votes in the last few years, but rely on cherry-picking from a tiny recent sample.
- Playing with house money. People behave differently when they are able to bet with someone else's money, something known as 'moral hazard'. And there's an element of people feeling that way if they won money backing Trump in '16. Also, there is logical 'comfort' in repeating something that worked the last time. A small factor will also be that in a two way market, where people just want to have a bet in order to 'be involved' in a big event, they will simply back whichever side has the bigger odds. Not everyone who bets does so with a sophisticated view on the value offered by the odds.
- An emotional hedge. Few people inspire such passionate negative views as Donald Trump. Some folk will bet on Trump for the same reason that many Scottish football fans will bet on England to win the World Cup. Through betting on the thing that they passionately hope will not occur, they get some financial compensation if it does.
- Russia. Conspiracy theories thrive in the moist, putrid swamp of social media. And to be fair it's not such a ridiculous view to hold that that the way to win a modern election is to identify a base of 40% of the electorate, keep throwing them lots of red meat, and then do a deal with the Russians to get you over the line. Even if that's somewhere between an over-simplification and a falsehood regarding what happened in '16 and with Brexit, you can understand the logical appeal of believing it will happen again.
- The Apprentice's New Clothes. Donald Trump is charismatic. He inspires devotion and blind loyalty in around 40% of people. That the other 60% can blatantly see that he is a naked liar, cheat and a con man just does not matter to them.
- People think in stories. No amount of objective evidence can compete with the human failing that we will always find a way to believe what we want to believe. Confirmation bias is the 'daddy' of cognitive biases, and it ensures that we process new evidence only to confirm what we already know, rather than to challenge it. The idea of Trump winning again is just a good story - whether you read it as a romantic tale or as a horror story. It is what happened last time too, which engages recency bias. These biases trump(!) objective evidence like polls.
- Failure of the wisdom of crowds. The markets in the big betting sports (football, cricket, horse racing, tennis, NBA basketball, NFL etc) are shaped by dominant professionals and syndicates. Political betting does not have these forces being applied to it. Despite the colossal sums being bet on it (and so liquidity per se not being a problem) the US election market is inherently less 'smart' than say a Premier League football market with far less liquidity. In modelling terms we would say that political markets are 'less well calibrated' than the efficient sports markets.
- Talebism. People conflate uncertainty with a need for odds in a 2-way market to default to 50%. This is fuelled by 'experts' like Nasim Taleb who do pugnacious, influential and impressive-sounding repudiations of polling-based models, and promote the case for Trump having a 50% chance. This is pseudo-intellectual horseshit. Bookmakers and professional gamblers love guys like Taleb, just like they love tipsters and 'expert pundits', because the arguments they promote appeal to the gullible - and these are the people ultimately from whom they take money. Anybody who has priced up or bet on 2 year old maiden races (full of horses who have never run before) will know that there are plenty of ways to estimate their ability more accurately than to assume they will all have the same ability.
- Demographics Venn diagram. Inasmuch as there's a 'type' of person who likes Trump, a demographic dominated by white, working-class males is a decent match for the 'type' of person who bets on election markets.
- Anti-establishmentarianism. A vote for, or a bet on Trump is a way to say 'fuck you' to the establishment such as 'real' politicians and nerdy intellectuals like Nate Silver.
- A little bit of knowledge is a dangerous thing. Republicans currently enjoy an advantage in how the electoral college system of US general elections works, meaning they can win the presidency while losing the popular vote, as happened in '16. Some bettors will be giving this too much weight though, because the advantage only matters if the popular vote is close. It's a bit like giving a horse an extra chance of winning a race because it has a particularly long nose that will be useful in a photo finish. This is of no benefit at all if the horse is several lengths behind the winner, as Trump currently looks likely to be.
- Modern media. The influential mediums of our modern age make us less analytical, not more. Despite the ready access via the internet to quality objective evidence such as polling data and the output of election models, most people prefer to 'think' with their emotions. And social media thrives on emotion, as do partisan 'news' outlets such as Fox News.
The nearest equivalent to the 2020 US election in sports betting is probably where a market gets 'sentimental' about an ageing charismatic champion coming back for one last hurrah, and the public bets on the fairytale coming true. The weight of sentimental and hype-driven amateur public money outweighs the smarter pro money for once and the market is 'wrong'.
This has happened a few times in recent years with racehorses, and some sort of similar phenomenon occurred a couple of years ago when Conor McGregor fought Floyd Mayweather. McGregor was consistently way shorter in the betting for the fight than any strictly objective (model-type) assessment would have had him. This included those who figured that since we couldn't know how good McGregor would be as a boxer, therefore his chance must default towards 50%. The weight of 'mug' and Talebian money for the charismatic Irishman outweighed the 'pro' money which backed Mayweather and produced a market as illogical as what we've seen for the '20 US election.
Nate Silver explains the difference between his forecast and the betting market by saying prediction markets (by which he means betting exchanges) are dumb. This is one point on which we disagree. The landscape of sports betting is littered with the bones of the smug, brave and foolish who enter the fray with theories about how the market is stupid and wrong, confident that they know better. The sports betting market is, in fact, a fearsome foe. It is relentless and unforgiving, swallowing up and spitting out all who dare to underestimate it.
But markets on politics - and for bouts between boxers and MMA fighters - don't quite work by the same rules as those for the biggest sports. It is inconceivable that £300m could be matched on a football match and the market price be 'wrong'. But, through a combination of the factors listed above I think it's fair to conclude that on this occasion Nate is probably right, both about the '20 election market being 'dumb', and about Biden's true odds of victory - Biden is probably a true 1.11 shot.
My 'professional view' is that the bigger price at which it's possible to back him today represents the sort of great betting value which is available only very rarely in betting markets.