PredictIt Currently Has Kirk Cox, Terry McAuliffe and Dems as Big Favorites in VA 2021 Governor’s Race. So How Accurate Has PredictIt Been the Past Few Years?

Since we’re apparently not going to be getting many public polls – let alone high-quality ones – here in Virginia for the Democratic and GOP statewide primaries/convention this cycle, let’s take a look at the prediction markets over at PredictIt (according to this WaPo article, “Prediction markets seek to harness the ‘wisdom of the crowds’ — the theory being that large groups of people are collectively smarter than experts and polls.”) to see where they’re putting their money.

• As you can see (see screen shot, below), the chances of Democrats winning the Virginia gubernatorial election this November have been hovering in the upper 80s to around 90 cents at least for the past 90 days.
• In the VA GOP gubernatorial contest, Kirk Cox has been in the lead consistently for at least the past 90 days (see screen shot, below), mostly in the 55-60 cents range, with Pete Snyder trailing at around 25-30 cents, and Glenn Youngkin pulling ahead of Amanda Chase into third place, at around 15 cents.
• Finally, on the Democratic side, Terry McAuliffe has slowly been increasing over the past 90 days (see screen shot, below), and is now at about 90 cents, with Jennifer Carroll Foy from a high of about 18 cents to around 9 cents now; Jennifer McClellan steady in the 2-4 cents range; and Justin Fairfax pretty stable at 1 cent. I’m not 100% sure why they aren’t including Lee Carter, but presumably – at least based on the comments – because PredictIt doesn’t think he’d be a factor…

So…the question is, should we care about these PredictIt prices at all? I was curious to see how well PredictIt has done in the past, so I did a bit of research. Here’s what I found.

• According to this, “Since at least August 7, approximately 90 days before Election Day, traders on PredictIt, the stock market for politics, have accurately been predicting the final outcome of the presidential election, including most of the most heavily contested and closely watched battleground states…In seven battleground states that have already been called—Florida, Iowa, Michigan, New Hampshire, Ohio, Texas, and Wisconsin—PredictIt accurately and consistently predicted the final outcome for at least the last two weeks leading up to Election Day. In one battleground state, North Carolina, the forecasted winner changed frequently over the last 90 days, showing an incredibly tight race for the final two weeks of the election. Depending on what happens in Arizona, Georgia, Pennsylvania, and Nevada, the traders could have been 100 percent accurate on the electoral count.”
• On the other hand, “Although prediction markets can work well, they don’t always. IEM, PredictIt, and the other online markets were wrong about Brexit, and they were wrong about Trump’s win in 2016. As the Harvard Law Review points out, they were also wrong about finding weapons of mass destruction in Iraq in 2003, and the nomination of John Roberts to the U.S. Supreme Court in 2005. There are also plenty of examples of small groups reinforcing each other’s moderate view.”
• Quoted in this article, “PredictIt’s spokesman Will Jennings says it’s difficult to make a ‘blanket accuracy assessment, given the variety of markets,’ he said the D.C.-based site’s handicappers are correct 70 to 80 percent of the time. It forecast, for example, that Brett M. Kavanaugh would be on the short­list to succeed Justice Anthony M. Kennedy on the Supreme Court. It also signaled that Kavanaugh would get the nomination hours before it was formally announced.” That same article also states, “Though research generally supports the idea that prediction markets are better forecasters than polls, Justin Wolfers, a University of Michigan professor who has studied prediction markets, said to take that with ‘a big asterisk’…Saying markets are better than the alternative is not saying markets are perfect. Most ways of predicting the future are bad. It’s possible markets are the least bad.”
• I also went back and checked some of PredictIt’s forecasts, including here in Virginia. See below for those…most of which were accurate, actually, although there certainly were some mistakes. Not bad overall, I’d argue, and as good or better in some ways than FiveThirtyEight’s forecasts, which were too optimistic on Cameron Webb in VA05, for instance (although The Economist nailed Bob Good’s victory in VA05, also Elaine Luria’s win in VA02 and Abigail Spanberger’s victory in VA07). What do you think of PredictIt specifically, and prediction markets in general?

From March 20, 2017, this prediction in the VA Dems’ gubernatorial primary between Tom Perriello and Ralph Northam was highly accurate, with Northam ending up winning in June by a 56%-44% margin.

From late March 2017, PredictIt nailed both of these…

From April 2, 2017…PredictIt was spot-on about Ed Gillespie winning the VA GOP gubernatorial nomination, as well as about Ralph Northam winning the VA Dem gubernatorial nomination and Democrats winning in November.

From late March 2018, this prediction of Democrats taking back the U.S. House of Representatives in November 2018 was accurate.

Also from late March 2018, this was accurate on the coming defeat of Rep. Barbara Comstock (R-VA10)

From September 17, 2018…PredictIt incorrectly predicts that Terry McAuliffe will run for president in 2020 (although he very much considered running, and it seems like it was a very close call).

This prediction, from 9/17/18, was definitely wrong, as Rep. Elaine Luria (D) defeated Republican Scott Taylor by nearly 6 points.

This prediction, on 9/17/18, was accurate that Rep. Barbara Comstock (R-VA10) would lose to Rep. Jennifer Wexton (D).

These predictions, on 8/14/19, were accurate…that Democrats would win majorities in the Virginia State Senate and House of Delegates in the November 2019 elections.

But this prediction – on 11/5/19 – was badly wrong, as Kirk Cox won by almost 5 points.

This 11/5/19 prediction, on the other hand, nailed the Clinton Jenkins (D) win (by almost 13 points) over Republican Chris Jones

This 11/5/19 prediction, for Democrat John Bell over Republican Geary Higgins in SD13, was correct as well.

This prediction, on 11/5/19, was a bit too optimistic, with Democrats actually ending up with 55 House of Delegates seats, not 56+ seats.

And this prediction, on 11/5/19, was too optimistic on how many VA Senate seats Democrats would win, with Dems underperforming and ending up with just 21 seats. Of course, almost *everybody* thought Dems would end up with 22, 23, 24 or even 25 Senate seats heading into that election, which was a big disappointment on the VA Senate side.

This 12/21/19 prediction was largly accurate, with Joe Biden leading for the Democratic presidential nomination, followed by Bernie Sanders.

This 1/26/20 prediction was also accurate, with Biden ending up winning with 53% of the vote and Bernie Sanders in second at 23% of the vote.

On the other hand, this prediction – from February 13, 2020 – basically fell for the Bloomberg boomlet, as well as for Biden’s losses in Iowa and New Hampshire, and got the Virginia Democratic presidential primary badly wrong.

PredictIt pretty quickly corrected itself, following Biden’s big win in South Carolina on 2/29, and vaulted Biden back solidly into the lead for the 3/3 Virginia Democratic presidential primary, with Sanders second.

This July 5, 2020 prediction correctly had Kamala Harris as the favorite to be Joe Biden’s VP pick.

This August 11, 2020 prediction was wrong, however, about Susan Rice…although probably not a bad guess, as she was one of the top candidates.