PredictWise Blog

Iowa Electronic Market

Bookmark and Share

The Iowa Electronic Market (IEM) is a prediction market run by the University of Iowa and sanctioned by the Commodity Futures Trading Commission (CFTC). PredictIt is the other exchange sanctioned by the CFTC; it just opened in the fall of 2014. IEM is capped at $500 per person. Being the oldest continuously running prediction market on politics in the USA, people often wonder why PredictWise does not show it more prominently in the aggregation of political prediction market data. There are several reasons.

1) IEM only has three markets: two-party vote share of the national popular vote for president, winner-takes-all for the popular vote for president, and formal balance of power of Congress. PredictWise builds all of its presidential and Congressional predictions on a state-by-state level (i.e., it predicts the Electoral College by the aggregation of the 51 states and DC), so IEM does not have data on this level.

2) The contracts that IEM do sell are not the ideal contract for the outcomes we care about predicting.

The two presidential contracts are exclusively constructed around the national popular vote. We only care about the national popular vote insofar as it relates to the Electoral College. Most recently in 2000 the winner of the national popular vote did not win the Electoral College and become president (this has happened 4 times total: 1824, 1876, 1888, and 2000). The national popular vote is very correlated with winning the presidency, but it is not the same thing.

Similarly the Congressional contract has two peculiar aspects. First, the contracts mix Senatorial and House control. They trade on five mutually exclusive outcomes: Democrats for both houses, Republicans for both houses, both directions of Democrats and Republicans each controlling a house, and other. Second, the “other” outcome is a bit confusing because either the Democratic or Republican party is going to control the houses of Congress! There are numerous candidates and sitting members of Congress whose formal party is not the Democratic Party, but caucus with them exclusively. They do not count in these contracts. For example, current candidate for the Democratic nomination for president, Bernie Sanders is a Socialist, so he does not count for this contract. If the Democrats actually controlled the senate, with votes from Sanders and King (ME-I), the contract would state “other”, rather than Democratic. Sometimes this is easy to adjust for, but other times it is not.

Differences between “bookie” and “market” data

Bookmark and Share

Betfair has two locations on the same website (1) exchange or market (2) sportsbook or bookie. The first, the exchange or market, is a place where people bet against each other. The second, the sportsbook or bookie, is a place where people bet against the house. Both sell the same product, which are contracts that are, canonically, worth $1 if something occurs and $0 if it does not occur.

While both sets of prices are subjected to market forces, they tend to diverge; the key distinction is where Betfair makes its money. In an exchange Betfair makes a guaranteed profit of 2-5% of the winnings, but in a sportsbooks it makes an expected profit of 5%+ of the total wager.

An example helps explain what that means. In the exchange, two people come together and agree on a price for any given contract and Betfair takes a percentage of the winner’s winnings. For example, imagine a contract for Candidate A winning an election; user X believes that there is a K% likelihood of Candidate A winning and user Y believes that there is a K-5% (where K is between 0 and 100) likelihood of Candidate A winning. User Y will sell a contract for Candidate A to win to user X for K-2.5% (or somewhere between K and K-5%). Both of them think they have a better deal in expectation. In the sportsbook, Betfair overcharges for the contract and charges everyone the same price. For example, Betfair may charge K%, higher than most people’s subjective probability, and hope that there are some people with subjective probabilities that high to make a deal.

This difference is most obvious in full sets of mutually exclusive outcomes. Consider a two candidate race between Candidate A and B. Due to arbitrage traders, if people are trading Candidate A at K-2.5%, Candidate B is going to trade very close to 102.5–K%. So that Candidate A + Candidate B = 100%. But, the sportsbook is likely to sell at something closer to K% for Candidate A and 105-K%. So in the sportsbook Candidate A + Candidate B = 105%. This means that Betfair makes 5% in expectation if it sells equal amounts  of both Candidates.

One further note is that in the exchange we see the price that buyers and sellers come together, but in the sportsbook we just see the price they are willing to sell for. So, while we can approximate the “price” on an exchange as the mid-point of the bid/ask on the order book, we can only see the sell price on a sportsbook. Thus, another reason it is a little higher than the true underlying probability.

Greek Polling

Bookmark and Share

Pollfish, a Greek-based mobile polling platform, has been releasing a series of fascinating polls over the last few weeks. Pollfish operates as a pop-up poll inside of third-party mobile applications. Prior the Sunday, July 5 referendum they polled 7,275 Greeks. And, in the last week, they have polled 2,400 Europeans. This poll is random, but not fully representative.

While 61% of Greek voters voted NO on the referendum, 75% of Pollfish respondents (who indicated a choice for either NO or YES) indicated they would vote NO in the week leading up the voting. This result stands in sharp contrast to the majority of polls that underestimate the support for NO. The discrepancy in favor of NO, is likely driven by an undercount of older respondents in this poll.

Understanding the poll is over-counting NO voters, it is still incredibly meaningful to learn more about the divide between NO and YES voters. 75% of NO voters and 88% of YES voters indicated they thought Greece would stay in the Euro. The NO voters, confident in victory, clearly did not think the vote meant the end of the Euro in Greece. The big divide between the two camps was on the government, with 81% of NO voters supporting the government and 84% of YES voters not supporting the government.

In the current poll, in the field from July 7 to July 10, 55% of Europeans want Greece to be part of the Euro, with 26% saying no, and 20% providing no answer. Again, ignoring the topline numbers for a second, it is extremely constructive to jump into the differences between the two camps. Those respondents that support Greece staying in the Euro have both positive view of Greece and of the Greek government. Those respondents that support Greece leaving the Euro also have a positive view of Greece, but an overwhelmingly negative view of its government.

There is a strong hint of wishful thinking (or possibly bandwagon) in the expectations of the respondents. Those that want Greece staying with the Euro think it would be good for the Eurozone for them to stay, and they are likely to stay; while those that want Greece out of the Euro have mixed feelings on the benefit of Greece leaving the Eurozone, but they think that it is likely they will. Those respondents that wish Greece to stay on the Euro think it will stay with 58%. And, they feel that the Eurozone will be impacted negatively with a departure by 62%, with just 11% saying it would be positive and 27% saying either no affect or no answer. Those respondents that wish Greece to leave the Euro think it leave with 77%. And, they feel that the Eurozone will be impacted negatively with a departure by 37%, with 28% saying it would be positive and 35% saying either no affect or no answer.

The latest market numbers are 77% that Greece will stay on the Euro through 2015.

Survey on Greeks: https://www.pollfish.com/dashboard/results/2784/-1362401898

Survey on Europeans: https://www.pollfish.com/dashboard/results/2848/281738220

Prediction Markets and the UK Election

Bookmark and Share

In the final stages of the UK election on May 7, 2015 the polls showed Labour in the lead for seats. It is reasonable to say it was a statistical dead heat. Despite the polling, prediction markets had the Conservative party over 65% to get more seats than Labour every day from March 10, 2015 onward.

To be clear, Labour never had a majority of the seats in the polling (neither party did) and, with several parties and the prospect of a coalition, determining who would take power was non-trivial. So, FiveThirtyEight, YouGov, and others created complicated models and determined possible coalitions. I cannot determine from their sites a direct probability of government for each party, but prediction markets certainly provided that for me. Prediction markets gave the Conservatives 55% to hold the government (coming from both Betfair and HyperMind markets) on the eve of the election, despite polling showing them not even getting a plurality of the vote.

I do not have the time to build models for every: political, entertainment, sports, and economic outcome, and prediction markets consistently do as well or better than the best modelers (even mine). To be clear, I am a huge fan of the modeling that FiveThirtyEight, YouGov, and others do to translate polling into predictions. I do it myself for USA elections, Oscars, and few other events. But, prediction markets crowd source many models, from many people, who are willing to bet real money on their information.

Prediction market-based forecasts are scalable because in exchange for expected winnings users do a lot of work aggregating information. Prediction market-based forecasts are timely because in exchange for expected winnings users do a lot of work in getting information into the market quickly. Prediction market-based forecasts are flexible because in exchange for expected winnings users do a lot of work in answering all types of questions and on all types of outcomes. Prediction market-based forecasts are accurate because in exchange for expected winnings users do a lot of work in ensuring that they incorporate all of the most accurate information available, and then some more …

NCAA Men's Basketball - March 30

Bookmark and Share

Kentucky is 55% to win the tournament, same as they were before the Elite Eight. They did not look dominate so advancing a round was not as helpful as it should!