Yves here. Rajiv Sethi stresses a key point: that presidential-election-impinging events have been so far out of band as to render many of the correlations in statistical models useless. Sethi has more faith in prediction markets. I would not be so positive. Less bad does not necessarily mean all that good. One can point to many prediction market failures, some highly visible like the forecast that the Brexit referendum would fail by four points when it won by two. Barry Ritholtz has been a long-standing skeptic. You can take a gander through his many posts and articles on this topic here.
By Rajiv Sethi, professor of economics at Barnard College. Originally published at his website
This has already been the most astonishing election cycle in living memory, and we are still three months away from the finish line. A history of the season could fill volumes. But many of the major events that have shaken up the presidential race can be detected in the movement of prediction market prices, which compress and reveal a lot of information in a very compact space.
The following two charts, viewed together, tell much of the story so far and allow us to contemplate counterfactual scenarios in an interesting way. The figure on top shows the prices of three contracts in the presidential winner market on PredictIt, while the one below shows the prices of contracts that reference the winning political party:
Many significant events—the debate, the pressure on Biden to step aside, his resistance to this pressure, the attempt on Donald Trump’s life, and the coalescing of party support for Harris, can all be seen in the oscillations above.
In early May, markets gave Biden a slight edge in a contest that was expected to be close. By mid-May Trump had moved ahead of Biden in the probability of victory, but the Democrats remained ahead in the party contracts—traders were assigning some probability to Harris (or someone else) replacing Biden at the top of the ticket.
In early June the party contract prices crossed, but the gap remained relatively narrow until the debate on June 27.
During the debate the price of the Biden contract fell sharply, as did the price of the Democratic contract, while the price of the Harris contract rose. These trends continued as pressure mounted on Biden to step aside, with Harris eventually overtaking Biden in the presidential winner market in early July.
The prospects for the party kept tumbling during this period of uncertainty, with increasingly public calls for a new nominee even as Biden and his supporters insisted that the matter was closed.
The assassination attempt came on July 13, and led to a sharp increase in the likelihood of a Trump victory. This probability peaked a couple of days later, and then started to decline in tandem with a rise in the price of the Harris contract. By the time Biden formally stepped aside on July 21, the gap between the parties had already narrowed substantially.
As soon as the decision to step aside was announced, the price of the Biden contract collapsed completely and the price of the Harris contract surged. The gap between the two party contracts narrowed, and continued to narrow over the next few days.
Yesterday, for the first time since early June, the likelihood of a Democratic victory edged past that of a Republican victory in this market.
One can use these charts to consider a counterfactual scenario. By the time of the debate, Biden had already trailed Trump for six weeks. Had there been no debate, this gap would likely have persisted, but remained small enough to prevent a serious challenge. What the debate did is to expose vulnerabilities early enough for action in the face of panic, rather than resignation and paralysis. When people look back on this period, the debate debacle might be seen as a blessing in disguise for the party. It opened up the only possible avenue for a competitive election in the November.
This election season has exposed the limitations of statistical models when we are faced with significant departures from historical norms. Once it became clear that Biden would not be contesting this election, models based primarily on opinion polls were suspended and had to be revised and recalibrated to accommodate the change in presumtive nominee. One of these is now back online, but the others remain silent.
Markets, meanwhile, adjust instantaneously to changing circumstances. They tell an uninterrupted story, accommoding all manner of twists and turns. And they establish a precise historical record that can supplement other sources of information once these events start to recede into the distant past.
