Betting markets appear to have predicted the outcome of the U.S. presidential election correctly. But how good are they at predicting future events? And can they be used as a more accurate bellwether for the outcome of elections than opinion polls?
After the result of the presidential election was known, betting markets touted how well they predicted the result. The election was a binary decision: two candidates, only one winner. To call the toss of a coin correctly does not necessarily mean that betting markets can be relied upon.
Obviously, if you polled everyone who was going to vote and asked them how they were going to vote, if they gave you honest answers, you would know what the outcome was going to be. Undoubtedly, this would be time consuming, costly, and hugely inefficient.
Francis Galton is widely known as a Victorian-era statistician, but he was actually a polymath who was interested in and wrote papers on psychology, inheritance, and meteorology and devised a way of classifying fingerprints and many other things (including eugenics and you know where that ended up!). He studied the power of prayer and concluded it had no influence on the length of life of those prayed for.
As for statistics, Galton developed the ideas of correlation and standard deviation as a measure of how widely data sets vary, the latter a tool widely used in statistics today. It was he who first noticed that crowds could make guesses about something more accurately than individuals. Crowds could be smarter than the smartest people in them.
In 1906 at a country fair, there was a competition to guess the weight of a slaughtered ox. Eight hundred people entered the competition and Galton calculated that the average weight of all the guesses was 1,197 pounds. The actual weight of the ox: 1,198 pounds. This is known as the “wisdom of crowds,” something Galton did not develop further as a method of statistical sampling. Since that time, scientists have isolated the components that are critical to an intelligent crowd.
Today, statistical sampling has moved on quite a bit from the early days and the current techniques of election polling are sophisticated indeed. Pollsters attempt to create a smaller pool of people that will represent the intentions of the whole voting population. This is fraught with difficulty, because no one really knows until after the vote which demographic segments were more or less likely to vote, which way they would vote, and how much to believe their answers to the pollsters’ questions.
When a poll is published, it always specifies a margin of error and a confidence value. The more uncertain, the larger the margin of error. A margin of error is quoted as something like 3 percent at a 95 percent confidence level, which means that if the survey were conducted 100 times, it would be expected that the actual result would be within 3 percent of the actual result in 95 of those surveys. But everyone wants certainty, so the media fixes on the number and ignores the margin and confidence level.
It turns out that nearly all of the polls published on the eve of the presidential election got it “right”, i.e. the results were within the polls’ margins of errors.
And yes, most of the well-known betting markets got the result correct too, but to me this is a lucky guess. Why do I say that? A few elements need to be explored.
In an election, each vote carries the same weight; it is treated equally. In the case of the presidential election, discounting the electoral college, whoever gets more votes than the other candidate(s) wins the election.
With betting markets, each bet is not treated equally. What moves the needle is the weight of money. If more money is bet on one candidate than the other(s), that candidate is predicted to win. Each bet is not of the same value and so not equally weighted.
Polymarket, one of the more visible betting exchange markets, had a huge early swing to Donald Trump to win the election. It turns out the swing was the result of three very large bets placed by a French individual. Those three bets moved the price considerably and certainly more than three votes for one candidate would.
Despite the Federal Appeals Court overturning the CFTC’s ban on betting market operator Kalshi, U.S. residents are not allowed to place bets on the Polymarket site. Whether or not the big bettor was actually a resident of France is debatable. The site accepts cryptocurrency, so it may not be possible to ascertain the location of the individual making deposits. But can including bets from people who are not allowed to vote improve the reliability of the market?
The second thing is how representative are the people making the bets. Betting markets appeal to a tech-savvy and non-betting-averse section of society. Bettors on Polymarket, even if it allowed U.S. residents to bet, are a highly skewed demographic compared to the voting public. Cryptocurrencies attract primarily white males in the 20- to 40-year-old age bracket. It is unlikely you will find 70-year-old women represented amongst the customers on the site.
Since the Appeal Court’s decision, Robinhood Markets, the retail investor site responsible for the gravity-defying share prices of GameStop and AMC, has decided to get in on the action and offer bets on elections. Again, the userbase of Robinhood is not even close to being representative of the U.S. voting public.
The demographic makeup of the user base is important. Let’s assume the betting market allowed fiat currencies to be used, regular debit and credit cards. Even if the operator limited each person to a single $10 bet, the user base would still be badly skewed to a non-representative segment of society.
The reliability of the predicted result would be the same as a pollster going to an Ivy League university campus and canvassing a group of students in a mathematics class about how they intend to vote. As bookmakers know only too well, favourites do not always win.
