Bookmakers not only offer lines on sporting events, but also on big political events such as the upcoming Brexit referendum. While the surveys taken by polling organization predict a close race (and the last polls taken before the assassination of Jo Cox saw the “Leave” supporters ahead), the odds offered by British bookmakers would suggest a more or less clear victory for the “Remain” vote.
I wonder what the track record is in general for bookmakers predicting the outcomes of elections or referendums in the past: Are bookies superior prognosticators?
Well, of course, the bookies aren’t making predictions. The odds are set by the weight of money so it’s the punters, collectively, who are making any predictions.
And they’re not making quite the same prediction as is made in a poll. Somebody responding to a poll is predicting how he or she will vote in a week or a month or whenever. But somebody placing a bet is trying to predict how other people will vote.
Here’sa link to a paper which looks at polls versus betting markets for US presidential elections. It suggests that, first, before polls were available, bettting markets were surprisingly good at predicting the outcome of elections. Secondly, since polls were introduced, they have been slightly better than the betting markets used to be. And thirdly, since polls were introduced the betting markets (presumably influenced by the polls) have tended to track the polls closely, but haven’t added anything by way of improved accuracy at forecasting the outcome.
Here, by contrast, is a Reuters article from last year quoting academics saying that, no, the betting markets are superior predictors. They’re looking at a wider range of events, in a wider range of countries, than just US presidential elections.
Believers in the Efficient Market Hypothesis should have no trouble believing in the efficay of prediction markets, at least if significant money is wagered. (And I’m reminded of the Monty Hall Puzzle — opinion is generally about evenly divided on that puzzle, but the offers to bet are one-sided. )
The pure weight of money odds must go through a few manipulations by the Bookies, though, to maximise their expected profit? ie the Bookies will try to present a picture that benefits themselves first and foremost, rather than the undistorted reality of where the money is going.
Not a betting man myself, so don’t really know how they operate - but making a race seem closer than it is would seem to be one way of doing this - shortening the odds on Brexit, for example, to entice people to back it when the money and informed opinion says otherwise.
Betfair.com which now oversees some £14,000,000 in money already wagered on the November result does nothing to set the odds. The odds are wholly determined by buyers meeting sellers, similar to a stock exchange.
Since OP’s question has been answered, here’s a good puzzle for the Board’s mathematicians. Suppose a bookie (or other market force) judges the probabilities of disjoint exhaustive events to be p, q, … r where p+q+…+r = 1 and offers fair odds on each event, i.e. 1-for-p payoff on the event he judges to have probability p.
You judge the probabilities to be, instead, p’, q’, … r’ where again p’+q’+…+r’ = 1. What bets should you place with the bookie? (Assume Kelly’s Criterion, i.e. that your goal is to maximize the geometric mean of your resultant bankroll, weighted over all possible outcomes.)
Bookies want to make a profit no matter what the outcome. Hence, whatever they do to maximise profits is unlikely to involve predicting the outcome themselves and taking action based on that prediction.
Obvious question: What were the betting odds on Donald Trump last summer? While he was leading the polls then, were people actually betting on him then? Were they betting, say, on JEB, or Rubio?
Bookmakers don’t (in general) do this. The bookmaker is trying not to take a position on the issue, and just make consistent profit not matter what the outcome. This is why if twice as much money is being bet on Hillary than Trump, the odds will be, (neglecting the bookmaker’s cut,) 2:1.
That way, if Trump wins, the bookmaker can take the money she made from Hillary bettors, distribute it to the Trump bettors, and come out even. And if Hillary wins, she can take the money she made from Trump bettors, distribute it to the Hillary bettors, and also come out even.
If the bookmaker sets the odds to anything else, either for honest reasons (she truly believes that the actual probability of Hillary winning is 75%), or dishonest reasons (she wants to make it seem like more of a horse race to encourage more betting), then she is exposed to the outcome – if the election goes one way she gets a windfall, but if it goes the other way she has to pay out bets with her own money.
Most commonly bookies keep all losses and take a 5% commission on wins called the vig. Ideally bookies want that 5% guaranteed money and don’t want the risk of a possible big pay off. The way they get that is to keep the line at a point where 50% of the bettors place their money on either side of the bet. Lines change when one side of the bet starts getting more money. It’s the same principle that casino games run off of. Each game has a slight house advantage built in. They don’t need more than that.
Thanks for the explanation (and others above). It sounds, though, like you’re describing how a one-man band bookie operates, wanting to run a balanced book and keep the vig.
Over here in the UK I’m pretty sure the big High St bookmakers are more active than that, and will risk exposure, take a view, and manipulate prices to make more money.
…To do this, the bookmaker will increase the price on the opposition to make the price more appealing to punters while shortening the price of the runner which has been well backed in an attempt to discourage further investment on that selection.
Here we have a situation where a bookmaker is deliberately moving his odds away from the norm in order to make the amount of money bet on each team lopsided. So this particular bookmaker is not trying to balance their books but rather is having a bet themselves with the bookmakers’ margin still on their side.
Two sites have historical graphs, but they go back only to January of this year. Many sites are archived in the WayBack Machine but store their tables in ways that Wayback doesn’t grok. One site that does have a useful Wayback archive is Paddypower.com. (It’s an ordinary sportsbook, not a market like Betfair.)
On June 18, 2015 Trump was a 33-1 longshot to win the GOP nomination:
By August 31, 2015 Trump was 7-2 behind only Jeb! (13-8)
(Both of these pages show a date of “Sunday 24th April 2016, 22:00” but that must be some archive/render glitch.)
What this is describing IS the process by which a bookie “balances” his books.
My understanding is that Las Vegas sportsbooks give an early low-vigorish opportunity to certain high-rollers in order that they can then open to the public with a spread near its “correct” level.
Sure, the bookies balance their books by laying off bets to reduce their exposure. What I’m interesting in understanding is if the odds of a major bookmaker for a given event represent pure weight of money. The link I posted is clear that they do not - so a punter approaching the EU referendum tomorrow, to give the example in the OP, would be wrong to assume that the odds are a simple reflection of this.
You’ve told me that this is the case on the exchanges - thanks.
I’d also note that the link I put up is from an Australian, so carries weight when it comes to punting. If you’re Australian yourself, Septimus, then obv I won’t take this any further.
If the bookie does not balance their bets, then effectively they’re becoming gamblers themselves, just like if they balanced their bets and then independently placed a separate bet. I’m sure that some do this sometimes (since after all, gamblers come from all walks of life), but it’s not actually part of their business.
I glanced at this article and concluded that the “study” wasn’t done by anyone with statistical expertise.
For example, if a poll shows 53%-47% and your modelling implies an aggregate standard error of 3% than you predict that the leader will lose 16% of the time. Over an ensemble of such cases you expect to mispredict 16% of the elections — if you mispredict far fewer then your model is wrong! Nate Silver understands this. David Rothschild understands this. Whoever wrote the Bloomberg article couldn’t think beyond “How often does a poll leader win the election?”
Wow! I suppose if you stripped out the four biggest misses you’d get 95% accuracy. Double wow?
The Bloomberg article may have some value, e.g. comparing the efficacies of different polling organizations. But since it doesn’t even understand that in an ensemble of 80-20 prediction market events, the underdog “should” win 20% of the time, I think it’s safe to ignore its overall conclusions.
On the one hand yes you are right about what the models actually mean and that such is usually unappreciated in most portrayals of judging the accuracy of various models. Most polls though are not probablistic like 538’s is. You can’t use the RCP rolling average that way, for example.
OTOH the right/wrong ratio is likely a reasonable proxy for the sake of the question asked here. Better would be to see how often a polling aggregator like 538 was wrong when they gave 90% confidence vs how often the betting markets were … I did do that in one thread for 538 and found this cycle on the D side at least they were wrong one out of 11 times that they gave roughly 90% confidence. How did the betting markets do?
Of course when polling is available the market will mostly follow the polls as the betting markets use the polling data to inform their bets.
The op uses Brexit as its jumping off point. Latest there:
How do you score that based on what results today?
The stock market apparently is reacting to slightly better “stay” polling as the final and went up along with the poll going up … is the wisdom of the stock market a better measure or the wisdom of the betting market?
The question I guess is that assuming there is good polling data, is there any reason to believe that the betting markets, which presumptively adds the punditry of the masses, is more or less accurate?
I read long shot betting market odds of one in fifty of Christie winning the presidency … will those sort of events really happen one out of fifty times?
My guess is that the punditry of the masses does not add any value to reasonable volumes of polling data.
And returning with the results in hand (in case of under rock 52 to 48 for leave).
The betting markets had been consistently fairly confident of remain winning despite polls saying it was much less clear. Just one election but little anecdotal evidence here that the punditry of the polis added value to prognostication by the polls. (Mind you the betting markets’ relative confidence was better placed on the Scottish referendum.)
Personally if there is a strong difference in odds given by a strong poll aggregator (like 538 or PEC) and the betting market I’d go with the aggregators’ odds. At least for now.