Betting Markets v Pollsters in Forecasting the 2020 Election

Leading up to the 2020 election, countless streams of polls reported candidates’ respective leads. When it came to election day, many of the highly trusted polls systematically underpredicted the presence of Republican voters. For example, Biden was polling +8 percentage points over Trump in Michigan, but when it came to election day, he won the state by a much smaller margin. Alternatively, betting markets proved to be fairly accurate election forecasting metrics. Trump carried Texas by a fairly large margin, a state where betting markets were bullish for Trump, and many polls considered it a tossup. Additionally, Trump handily won Florida, a state pollsters expected to go Blue but bettors gave to Trump. How could polling ‘experts’ consistently get it wrong, but the betting market get it right?

 

Given no systemic bias in the sample, a large enough sample size, and some knowledgeable voters, polls and betting markets should on average predict election outcomes correctly. The repeated underprediction of Republican representation points to significant systematic bias present in the polling system. Whether poll respondents were lying or disproportionally democratic, polling errors were not purely random and undermined the election forecasting process. On the other hand, gamblers put their real money behind a political candidate. With real losses/gains on the line, bettors have incentive to be rationally knowledgeable and back the candidate that they truly believe will win, regardless of their preferences. In fact the marginal cost for a bettor is much higher than it is for a poll respondent, who might let their ideological ideas influence their response. While there were outliers in the betting market (guy that bet 5 Million on Trump), for the most part, random errors cancel out and the proportion of respondents that are well informed dominate the political betting market. With the betting markets better eliminating systematic bias and creating disincentives to be ignorant, they produce a result suggestive of the Miracle of Aggregation and prove to be a better forecasting method than polls.