On Thursday, June 30th, FiveThirtyEight released its 2022 Senate Elections Forecast. Nate Silver is the most famous election forecaster in the world, and it seemed likely that prediction markets and bookmakers would adjust to his model’s predictions.
We’re comparing the impact of FiveThirtyEight’s first public forecast by measuring the movement towards FiveThirtyEight’s state-level predictions on the following gambling platforms:
The Senate races we’re examining are the 12 seats where all three platforms are offering action, specifically looking at:
- How FiveThirtyEight’s forecast compares to the gamblers on the morning of June 30th.
- Whether gambling probabilities moved closer to FiveThirtyEight’s predictions following the 2022 Senate model being publicly released.
Prices and odds were captured from the three platforms at 10:47 AM EDT on June 30th and again at 11:03 AM EDT on July 1st. FiveThirtyEight’s prediction is being defined as “winner_Dparty” and “winner_Rparty” from its “Deluxe” state-level model output on June 30th, which represents the “chance that the corresponding party, regardless of candidate, wins the seat.”
FiveThirtyEight relative to gambling platforms
Below are differences in probabilities between FiveThirtyEight’s forecast on June 30th, and the other platforms that same morning:
| State | v. PredictIt | v. SBK | v. Smarkets |
| Alaska | +0.03 | +0.03 | +0.03 |
| Arizona | +0.08 | +0.08 | +0.07 |
| Florida | -0.01 | -0.02 | -0.02 |
| Georgia | -0.09 | -0.11 | -0.13 |
| Iowa | -0.06 | -0.07 | -0.07 |
| Missouri | -0.07 | -0.07 | -0.09 |
| Nevada | +0.06 | +0.05 | +0.05 |
| New Hampshire | +0.09 | +0.07 | +0.05 |
| North Carolina | +0.03 | +0.02 | +0.01 |
| Ohio | -0.05 | -0.05 | -0.06 |
| Pennsylvania | -0.05 | -0.06 | -0.07 |
| Wisconsin | -0.06 | -0.07 | -0.08 |
FiveThirtyEight was more optimistic about Democratic Senate chances in Alaska, Arizona, Nevada, New Hampshire, and North Carolina than gamblers, but more pessimistic in Florida, Georgia, Iowa, Missouri, Ohio, Pennsylvania, and Wisconsin. Please note that “more optimistic” doesn’t mean “optimistic;” in Alaska, we’re talking about estimating the Democratic Senate candidate’s chances of winning at 0.045 versus 0.02.
Gambling probabilities moving towards FiveThirtyEight predictions
The chart below shows how the markets moved closer to (or farther away from) FiveThirtyEight’s Senate forecasts:
Movement is defined as the changes in implied probabilities of the three gambling platforms on June 30th and July 1st becoming closer or farther apart from FiveThirtyEight’s June 30th predictions.
Of the 12 Senate races, six states moved closer to FiveThirtyEight’s forecast (Alaska, Arizona, Missouri, Nevada, New Hampshire, North Carolina), 3 states didn’t change (Florida, Iowa, Wisconsin), and three states moved farther away (Georgia, Ohio, Pennsylvania.) The three platforms moved in the same direction if they moved at all, albeit to differing amounts. Only in Missouri and Pennsylvania did all three not move in unison.
Georgia was the only state to move significantly farther away from FiveThirtyEight’s forecast, likely due to a confounding factor: the Quinnipiac GA Senate poll released on June 29th showed Democratic incumbent Raphael Warnock leading Republican nominee Herschel Walker by 10 points, 54% to 44%. The next day, FiveThirtyEight’s model went live with Walker having a 0.55 chance of winning the Senate race.
Again, movement is relative to where the gambling platforms started. The 0.01-0.02 change in the Alaska Democratic Senate candidate’s chances means little since all three platforms were at 0.02 when FiveThirtyEight went live.
Background
In Prediction markets, traders buy and sell binary option contracts: an event either happens or does not happen (Yes or No). The contract prices can indicate what traders believe is the probability of the event happening, and so it would seem likely that the prices would move to reflect FiveThirtyEight’s predictions.
PredictIt is a political prediction market run by Victoria University of Wellington, New Zealand, and Aristotle International of Washington, DC. PredictIt traders can buy “Yes” and “No” shares on whether each party (usually limited to “Democratic” and “Republican”) will win the U.S. Senate election in a particular state. A correct prediction receives $1/share, and a wrong prediction receives $0/share. PredictIt limits both investment amounts and the number of investors per market. Participation is limited to individuals 18 years of age or older who are U.S. Persons. There are no controls on insider information.
To estimate the implied probability of PredictIt shares, the following formula was used:
(bestBuyYesCost + (1-bestBuyNoCost) + BestSellYesCost + (1- BestSellNoCost)) / 4
Smarkets is a British online betting exchange, allowing peer-to-peer wagers on sports and politics. SBK is a British sports betting app created by Smarkets to disrupt the bookmaking industry. It is a mobile app sportsbook that draws on prices from the Smarkets exchange. SBK betting app draws on Smarkets exchange for pricing, and both share the same political betting guru, Matthew Shaddick.
The odds for both SBK and Smarkets are taken from oddschecker.com, converted into implied probabilities (accounting for the “vig” or bookmaker’s margin), and rounded to the nearest hundredth. To find the implied probability of fractional gambling odds, the following was used:
Stake / (Stake + Net Profit) = Implied Probability
Caveats
On June 30th, FiveThirtyEight went live with the 2022 Senate model at 10:07 AM EST while the gambling data was recorded at 10:47 AM EST. It’s certainly possible that the 40 minutes in between those two events was enough time to move share prices and odds.
Additionally, the Dobbs decision came out the previous Friday, June 24th, and that seems to have caused a spike in Democratic prices ahead of FiveThirtyEight’s launch.
Additional Analysis
Future analysis could look at both how FiveThirtyEight increased trading volume and how long the “bump” lasted in certain races. For example, in Arizona, the Democratic Senate bestBuyYesCost overtook the corresponding Republican price, but only for a few days:
