Timely information and behavioral tendencies
Successful predictions depend on informational efficiency. For instance, a study by Cambridge University economists found that during the Brexit referendum, betting markets priced in a Leave result at 3:00 a.m., a full hour before currency marketsdid, and well before the BBC called the election at 4:40 a.m. In theory, this resulted in a full hour of “unleveraged return.” As the lead author of the study put it, the betting markets “had a better sense that Leave could win, or that it could at least go either way.”
Behaviorally, this makes sense. When it’s your own money on the line, you’re more interested. In the case of Brexit, individuals trading their own money were faster to accept the idea that what had seemed impossible was unfolding beforetheir eyes. And they were quicker to act on that information. Meanwhile, London traders were afraid to break what had been a strong consensus and reacted more slowly to new information.
For political polling, the idea is much the same. Research has consistently shown that respondents may be prone to behavioral biases, such as giving the answer they think the pollster wants to hear. Polls can also fall victim to sampling error. And sometimes, respondents just change their mind.
Victory depends on your frame of reference
Is it more important that a poll be accurate in direction or magnitude? Usually, polls measure an outcome by a point margin—for example, Biden currently leads Warren by 11 points. If he ends up losing the nomination by 1 point, the poll will have had a 12-point error and been directionally wrong. And if he wins in a 23-point landslide, the poll’s margin of error would still be 12 points, but the outcome would be directionally correct.
Recent elections with surprise outcomes were simplyvery close races that were won on the margins. The UK voted to leave the EU in a 52% to 48% vote. President Trump was elected while receiving 46.4% of the popular vote to Clinton’s 48.5%. In a true toss-up election, a poll will be wrong half of the time. It’s just that in a 24-hour news cycle, being on the wrong side of that coin flip feels worse—the same as it does when investors feel the pain of a loss two times more than they feel the joy of a gain.
Polls’ statistical biases and antiquated methods will continue to pose forecasting challenges. However, recent research published in the Journal of Judgment and Decision Making shows that combining market-priced odds and polling can improve predictive ability.1
In any case, it’s way too early to adjust portfolios for the 2020 election. So, for now, sit back and enjoy the political theater. We’ll see plenty of shifts in the coming months. But if you want to see them coming, you may need to look beyond the polls.