Try forecasting this election
When election season rolls around, Americans want the news before it happens. The journalistic innovation of poll aggregation tries to meet that demand with political forecasts, and it has become a booming business. But there is a problem.
The credit for this business, and perhaps the blame, belongs to FiveThirtyEight, which earned its reputation by using others' polls to forecast the outcomes of the 2008 and 2012 presidential elections. Presidential and Senate candidates send out fundraising messages using FiveThirtyEight probabilities as carrots or sticks. Journalists use the latest FiveThirtyEight forecast to show the state of the race. In the final days of the 2012 race, up to 20 per cent of The New York Times' web traffic was attributable to FiveThirtyEight, which had been acquired by the Times at that point and has since been spun off to ESPN.
FiveThirtyEight's reputation has taken a hit in this election cycle. Like most pundits and analysts, the site played down Donald Trump's chances in the Republican primary contest. In the general election, FiveThirtyEight's forecast has been more volatile than its competitors', and has given Trump a better chance of winning the election than most other poll aggregators have.
It is worth thinking about the forecasting site as a business. As it has grown, FiveThirtyEight has more incentive to protect its reputation by hedging. Any large investment firm that had its reputation tied up in an infrequent binary event would be insane not to hedge. As FiveThirtyEight grew, the payoff for being right increased, and the cost of being wrong increased as well.
To its credit, FiveThirtyEight has been transparent and straightforward about its model. The higher volatility is because it gives more weight to new polls than its competitors do. Because there are more undecided voters than in past elections, their accounting for that uncertainty means that the model both gives better odds for a Trump win than most models do, but also gives more of a chance for Hillary Clinton winning in a landslide. In options parlance, this is like being “long the tails, short the belly” — betting on either a Trump win or a Clinton landslide, rather than a modest Clinton win.
So what if FiveThirtyEight is an outlier from other poll aggregators? The problem is that investors appear to be using it to assess the electoral outcome probabilities. Nate Silver, the site's founder, has defended the model by saying that betting markets are in line with the FiveThirtyEight forecast. Yet on Friday, a Silver tweet coincided with a rally in equity markets and a drop in implied volatility in options markets. The causality may go the other way: markets may be responding directly to changes in the FiveThirtyEight forecast.
This election may end up being a strange confluence of events: the race itself is unprecedented, with two candidates who could not be less alike. The financial markets are worried about a repeat of the reaction to Brexit. And through it all, one forecasting firm dominates the public's perception of what is likely to happen today. The stock market's recent nine-day losing streak, its first in 36 years, could be seen as markets' response to FiveThirtyEight's relatively high odds for a Trump victory. Those odds themselves could be seen as FiveThirtyEight's hedge so that it can point to some degree of prescience regardless of who wins — a lesson that pollsters learnt the hard way when British voters unexpectedly chose to leave the European Union.
We should think carefully about what it means if news organisations are going to be at the centre of the election forecasting business. What are their incentives, and what havoc do they wreak? Players such as FiveThirtyEight act as a sort of hybrid between a news organisation seeking an audience and a big bank research organisation seeking to move markets.
• Conor Sen is a Bloomberg View columnist