Predicting the Brexit Vote


Polls, Prediction Markets and More on June 22

Will Britain decide to leave the European Union?

No one honestly knows. It’s just too close to call.

I mean really, REALLY close, well within the margin of error of current polling.

Bloomberg puts the poll numbers at 44.7% “remain” and 45% “leave .”

brexit bars

So, what other input do we have at our disposal? Some forecasters are using predictive models. At the moment, NCPolitiks.UK has put the probability of remaining at 63%.

Another type of input comes from prediction markets, in which people bet on the outcomes of events. One such market, PredictIt, puts the chances of remaining at 66% (that is, 66 cents).

Data from PredictIt, June 22


And another data point comes from PredictWise, which reflects economist David Rothschild’s academic research into prediction markets as well as polling and online/social media data. This source predicts that there’s a 70% chance that the voters will elect to stay in the European Union.

PredictWise, in turn, relies on a variety of other data. Rothschild points to the following sources:

Prediction Markets (Betfair, PredictIt, Hypermind), Polling (HuffPost Pollster), Bookie (OddsChecker). Click here for details of the data collection and aggregation process. Want some additional insight on the values? Research papers are at ResearchDMR or email David directly.

Summary of Predictions on June 22

Despite the neck-and-neck nature of the race according to polling, it appears that other sources are giving an edge to the “stay in the EU” possibility. This does not mean, of course, that anyone should bet on that result. In such a dynamic and crazy-close race, anything could happen.  Still, it will be interesting to look back on these numbers after the fact, once final results are in on Friday, June 24.

Update: Looking Back at the Brexit Vote on June 24

In the end, the “leaves” had it. The final vote was 48,1% for “Remain” and 51.0% for “Leave,” according the Economist. Today’s Reuters headline: “Britain Votes to Leave EU, Cameron Quits as Markets Dive.”

There will be terabytes of digital ink spilled on the questions of “how it happened” and “what it means for the future of the EU and Britain.” But what does the final vote say about the prediction tools outlined above?

First, polls turned out to be more accurate than what we can call the polls-plus sources. On June 22, the polls were giving a razor-thin edge to the “leave the EU” vote, while the polls-plus sources were giving a somewhat thicker edge to the “stay in EU” prediction.

So, why did the polls beat out the polls-plus? Here’s one interpretation: the polls succeeded because they were not influenced by a statistical “memory” of what came before. In the case of polls-plus tools, the many older polls showing that Brexit would fail were “baked into” the numbers. It was only in the final days of the vote that the polls showed a closing of the gap. The prediction markets and other devices, however, did not have time to react. The polls themselves, however, were not constrained by previous forecasts. A good poll has no memory and contains as little bias or wishful-thinking as possible.

This suggests that in such a very tight race where there’s a sudden swing in momentum, polls-plus predictive tools are not yet ready for prime time.

Maybe that’s wrong. We can’t speculate too much based on a single vote. And, even if that theory is correct, it may not stay true. If prediction markets, for example, ever become truly mainstream and their numbers of users grow, then their crystallized  “wisdom of the crowds” may become more accurate.

In the case of the Brexit, however, straightforward polling won the day, and I’m sure there are a ton of exhausted but vindicated pollsters breathing a sight of relief.


Leave a Reply