Economist Writing Every Day

Last night the major party candidates for Senate in Pennsylvania had their first and only debate. I didn’t watch it, since I don’t live in Pennsylvania. But judging by my Twitter feed, a lot of people did watch it, including (bizarrely to me) lots of people who don’t live in Pennsylvania. And overnight, tons of articles were written analyzing the debate, saying who “won” the debate, and so on (“5 Things You Need to Know About the Pennsylvania Senate Debate” etc.).

But this blog post is the only thing you need to read about that debate. And these charts are really all you need to look at.

These two charts come from the prediction market website PredictIt. The charts show the “odds” (more on that below) that each candidate will win the Pennsylvania Senate race, over a 90-day time horizon (first chart) and the last 24 hours (second chart). What do we see? The Democratic candidate has been leading for the entire race up until a week ago, though with his odds falling gradually over the past month or two.

Notice though the big jump last night during the debate. The Republican candidate moved up from odds of about 57% to odds of about 63%, close to where it stands as I write (67%). Based on this result, it’s safe to say that the Republican candidate “won” the debate, though not so decisively that the election is now a foregone conclusion. You don’t need to wait for the polls, which have consistently showed the Democratic candidate in the lead (though with the gap closing in recent weeks) — though of course, these betting odds could change as new polling data is released.

But where do these odds come from?

Take the 67% odds for the Republican candidate in Pennsylvania. What this number really means is that you can currently buy a contract for about 67 cents (sometimes with a small premium) that pays out 1 dollar if it ends up being correct. You could also buy the contract for the Democratic candidate, and this will only cost you 36 cents. It also pays 1 dollar if you are correct. We can reasonably interpret these odds as saying that there is a 2/3 chance that the Republican will win in Pennsylvania.

If you think the odds of the Republican winning are better than that… well, you can buy some contracts! And your purchase of the contracts will ever so slightly push up the price of that contract. Observers looking on can observe that price movement as some more information about the potential outcome of that race. Maybe your purchase is just a hunch, not based on any real information, but you are putting your money on the line. That tells us something, perhaps, or at least it tells us more than if you just freely offer up your opinion on Twitter.

That’s a nice theory of odds and information, but are these prediction markets any good? There has been a bit of academic research on the topic, but I will instead point you to the well-written essay and convincing data by Maxim Lott, who runs the site ElectionBettingOdds.com. His site aggregates data from PredictIt and other markets like it on a variety of political topics. For a race where the prediction market (such as PredictIt) says that a candidate has 60-70% odds of winning, they win… about 70% of the time. These results aren’t perfect. For a candidate with 20-30% odds of winning, they actually win about 40% of the time. But the key result is that prediction markets tend to perform better than pre-election polls, and better than “expert forecasts.” They also have the advantage of updating in real time (and they can change a lot, as new information comes in).

I joked earlier that it was strange for people outside of Pennsylvania to watch that debate last night. But let me walk that back a little bit. It’s not actually strange for people outside of Pennsylvania to care about that race. The US Senate is very close to a 50-50 party split, so any one race could tip the balance (there’s a betting market for the overall outcome too! Republicans currently at 70% odds, a big change from one month ago). Thus, it’s natural to wonder about the result of the debate and that election, even if watching the debate itself is probably not worth most people’s time.

So don’t watch the polls too much (unless you plan to bet in these prediction markets, then you should!). Don’t read too much analysis of the debates or the races. Just watch the prediction markets.

In closing, I should note that PredictIt is currently in danger of shutting down, as my co-blogger James pointed out back in August. This would be unfortunate, but I’m not too worried since other betting sites exist (remember Intrade?). But as you might expect, there is even a prediction market on this possibility, currently at a 23% chance of survival for PredictIt.