- People are arguing over the accuracy of prediction markets versus election forecast models.
- Crypto markets have been volatile during the election, switching between Trump and Biden.
- The early Florida win for Trump shook bettors' confidence in a Biden victory.
Decrypt’s Art, Fashion, and Entertainment Hub.
There has been plenty of abuse hurled at Nate Silver and the FiveThirtyEight election forecast model he helped create. But at the moment, it gives similar odds for the 2020 presidential election as cryptocurrency-based prediction markets.
As of 1:15 a.m Eastern, FTX's presidential election market, which has over $8.6 million in trading volume, puts the price of shares in a Trump victory at $0.11. Catnip, built on Augur, is selling Trump tokens at between $0.14 and $0.15. However, Polymarket is relatively bullish at $0.18 for the president.
When adjusted for states that have already been called by NBC News, FiveThirtyEight's model currently sits somewhere in the middle: It gives Trump a 12% chance of winning, just two percentage points higher than its final pre-election projection of 10%.
Many cryptocurrency lovers are interested in prediction markets because they're an example of free markets at work. On the one hand, they rely on bettors with disposable income. On the other, bettors have an inbuilt incentive to get it right.
Thus, there's an ongoing tension about whether prediction markets—and cryptocurrency-based prediction markets, which are relatively young, untested, and low on liquidity—can outperform sophisticated (which isn't to say "correct") forecasting models, which use algorithms to factor in polling results and other variables.
I'm confused. I'm seeing contradictory claims in my timeline. Who won — pollsters or prediction markets?
— zooko (@zooko) November 4, 2020
Vitalik Buterin, co-creator of Ethereum, called the contest in favor of crypto prediction markets last night as Americans went to bed not knowing who would be their next president.
never doubt markets.
always doubt media figures with models.
— Evan Van Ness (@evan_van_ness) November 4, 2020
At that point, Democrats and Republicans had each taken turns freaking out (Libertarians were stuffing their faces with popcorn) as influential polls weren't accurate in some states.
But, in fact, only one thing had really changed. Florida, which FiveThirtyEight gave Biden a 69% chance of winning, went to Trump instead, a mild-to-moderate surprise. (FiveThirtyEight's other missed call so far, projected today for Trump, is Maine's 2nd district.)
That early Trump victory in Florida shook bettors' confidence in a Biden victory and established a new narrative for an election that didn't need any more narratives.
When adjusted for that Biden setback, the FiveThirtyEight model dropped Biden's chances to 66%. But bettors on Polymarket and FTX flipped the market, suddenly pricing Trump shares at 74 cents a pop before gradually coming back to their current levels.
In other words, inputs on cryptocurrency prediction markets were very sensitive to new inputs. For one thing, with relatively low liquidity, whales can move the market with a single bet. For another, while there's the presumption that bettors have accounted for all inputs, there's no guarantee that they're particularly knowledgeable about the intricacies of hundreds of local election processes and vote-counting procedures (although maybe they will be after this election). They could have over-reacted, until, of course, other bettors reacted to that over-reaction by buying low on Biden.
Indeed, some bettors might have been looking at preliminary vote counts in places like North Carolina and Georgia, which showed considerable early advantages for Trump but have narrowed as votes from blue-leaning areas are counted.
As it stands, there's still plenty in play. The election currently hinges on results from six states. With Biden stuck on 253 electoral votes and Trump at 214 (both unofficial counts based on projections from NBC), each still has a path to 270 through a combination of these states:
- Alaska: 3 electoral votes
- Arizona: 11 electoral votes
- Georgia: 16 electoral votes
- Nevada: 6 electoral votes
- North Carolina: 15 electoral votes
- Pennsylvania: 20 electoral votes
FiveThirtyEight's final forecast gave a higher probability to Biden in each of the outstanding states mentioned except Alaska. Therefore, there's still plenty of time for Silver and co. to look really, really wrong—and for savvy prediction market bettors to be rewarded for reacting early and decisively.
But Silver's team could equally end up going 49/51. Until each vote is counted, it's too early to run an autopsy on whether prediction markets outperformed models—and figure out whatever the hell is going on with polls.
One thing is for certain, though: As the forecast models and prediction markets update their priors, they'll eventually agree 100%.