It was an excellent night for U.S. election forecasting models, as Nate Silver and others accurately forecasted the outcome of all 49 states that have declared a winner. Election prediction markets also did well, with the largest of which, InTrade, also predicting all 49 states.
In this election, a slight edge must be given to forecasting models, as the Democrats are likely to be declared winners in Florida, which would give all 50 states to Silver. InTrade, on the other hand, gave Romney a small advantage in the state.
But what does this tell us about the relative merits of prediction models over markets? Unfortunately, not much -- at least not yet.
There was a major difference between the InTrade and Silver forecasts – while both saw a victory by Barack Obama as the most likely outcome, InTrade gave a 30 per cent chance of a Mitt Romney upset compared to the 8 per cent chance given by Mr. Silver. The fact that Romney lost, however, does not necessarily mean, however, that forecasting models were more accurate.
Imagine if I asked two pundits: "If I draw a card from a standard 52 card deck, what are the chances I will not get a club?" The first pundit gives the correct answer of 75 per cent, while the second pundit gives the more "confident" answer of 95 per cent. If I draw a card and it is not a club, it does not make the first forecast less accurate because it was less confident. It would take many elections to be able to truly determine which model provides the higher level of accuracy.
One advantage of prediction markets over forecasting models is that they provide real-time analysis of changing conditions. The InTrade market was active through the night and traders were clearly analyzing the provisional results, identifying which polls were reporting, and using that to forecast the overall result. The media were analyzing the results as they came in, as well, but InTrade had the advantages of aggregating the opinions of thousands and being able to quantify the impact of specific events by the minute.
A problem for those who would argue in favour of the forecasting ability of markets is that the probability of Mr. Obama's victory on prediction market BetFair was consistently ten points higher than on InTrade. This presents a problem – if forecasting models are accurate, should the two markets not be forecasting the same thing? Furthermore, if both markets were truly efficient than this difference should be arbitraged away. In theory you should have been able to buy Obama futures on InTrade and Romney futures on BetFair and make a profit no matter who won. In practice, however, this is made difficult through transaction costs and barriers to entry. In order to comply with U.S. law, InTrade cannot accept payments via US issued debit or credit cards, an inconvenience to American users. But given that hundreds of thousands of shares were being traded on InTrade and transaction costs are minimal, it is quite an efficient market. BetFair, on the other hand, charges significant credit card fees and taxes on winning, which greatly diminishes returns. These transaction costs make BetFair a far-from-efficient market, so this helps explain the discrepancy.
Despite the good night for both systems, there is little we can conclude from a single election. Election forecasting models are adept at translating polls into to predictions. Election markets, however, do have one significant advantage – they incorporate election model predictions into their results, along with information not found in polls. So long as InTrade keeps transaction costs low, they will prove to be an accurate forecaster of election outcomes.