Evidence of Prediction Market Inefficiency From PredictIt’s 2020 U.S. Presidential Election Markets

By Aiden Singh 

Published August 28, 2022

I argue that persistent and exploitable pricing anomalies were present in PredictIt’s 2020 U.S. presidential election prediction markets. I claim that these pricing inefficiencies were the result of a subset of market participants acting on misinformation which erroneously led them to conclude that Donald Trump had won the election. I conclude by discussing the implications of traders struggling to correctly distinguish relevant information from misinformation for the efficient market hypothesis.

Table of Contents

Introduction

The Efficient Market Hypothesis 

PredictIt’s Political Prediction Markets

2020 U.S. Presidential Election

2020 Election Timeline

Expanded Mail-In Voting

Trump Claims Election Fraud

PredictIt’s 2020 U.S. Presidential Election Market

Part I: A Red Mirage

Pre-Election Day

Election Night

Conclusion

Part II: Alternative Facts 

Anomaly One – The Battleground States

Anomaly Two – Bets on Trump Winning the Electoral College By +280

Apparent Explanation

Implications for the Efficiency of Prediction Markets

Implications for the Efficiency of Financial Markets

Introduction

The Efficient Market Hypothesis

The efficient market hypothesis claims that asset prices in liquid securities markets reflect all publicly available information. As a result, markets only react to the emergence of new information which is relevant to asset prices, such as the outbreak of a pandemic, the start of a war, or the invention of new technologies. And, the theory continues, because these new events occur in a random fashion, market prices will follow a random walk. Therefore, because asset markets are priced correctly and because they move in a random manner as they continually absorb new relevant information, the theory concludes that it is impossible for an investor to consistently outperform the market. 

PredictIt’s Political Prediction Markets

PredictIt is an exchange which hosts event prediction markets that allow individuals to bet on whether an event will or will not occur (e.g. whether there be an economic recession within a particular timeframe) and on the outcome of events (e.g. who will win an election). In the run up to the 2020 U.S. presidential election, PredictIt hosted markets in which participants could bet on the outcome of the election. 

When an individual wants to place a bet on PredictIt, they do so by purchasing an event contract. For example, if an individual wanted to bet that Democratic Party candidate Joe Biden would win the 2020 presidential election, they could have done so by purchasing Yes contracts for Biden in PredictIt’s 2020 presidential election market. Alternatively, or additionally, they could have bought No Trump contracts. On the other hand, if an individual wanted to bet that Republican Party candidate Donald Trump would win, they could have purchased Yes Trump contracts and/or No Biden contracts. 

There are two potential ways to profit off of betting in PredictIt’s markets. Suppose that both Yes Biden contracts and No Trump contracts are trading at 50 cents a contract. The first potential way to profit would be to correctly wager on the outcome of the election and hold the contract to settlement. So, for example, assume an individual deposits funds with PredictIt and, wanting to bet that Joe Biden will win the election, buys 100 Yes Biden contracts at our hypothetical price of 50 cents each (a total wager of $50). Now assume that, some time after the election takes place, PredictIt determines that Biden has won. In this case, each Yes Biden contract and each No Trump contract will pay out $1. And all No Biden contracts and Yes Trump contracts will close at $0. Therefore, the individual holding 50 Yes Biden contracts will receive a payout of $100 for a before fees profit of $100 (payout) - $50 (amount wagered) = $50. And all holders of No Biden and Yes Trump contracts would lose the money they wagered. The second way to profit would be to sell a contract whose price has risen before settlement. So, for example, if a participant purchased 100 Yes Biden contracts at 50 cents each (a total wager of $50), and a week later the contracts are trading at 70 cents each, the participant could opt to sell the contracts at the market price for a before fees profit of $70-$50 = $20.

Note that PredictIt’s markets are capped in a way that the financial markets are not. No individual may wager more than $850 on a single contract type. So, for example, our hypothetical market participant wanting to bet on a Biden victory could buy, at most, $850 worth of Yes Biden contracts and $850 worth No Trump contracts for a total wager of $1,700. PredictIt does, however, sometimes offer multiple similar markets for large events. For example, for the 2020 election PredictIt hosted a market on who would win, a market on which party would win, a market on who would win the electoral college and by how much, and so on.

2020 US Presidential Election

2020 Election Timeline

Presidential elections in the U.S. occur every four years on the Tuesday after the first Monday in the month of November (Tuesday November 3, 2020 for the 2020 election). [1] Individual states and the District of Columbia must then certify their results. The electoral college then votes to determine the winner on the first Monday after the Second Wednesday in December (December 14, 2020 for the 2020 election). [2] Finally, the electoral college votes are officially counted by the U.S. Congress on the January 6th immediately following the election, and the winner is sworn into office on January 20th. [3, 4].

Expanded Mail-In Voting

The 2020 US presidential election took place on November 3, 2020 in the midst of the SARS-CoV2 Pandemic.

Citing concerns about the virus spreading at in-person voting stations, thirty-one states and the District of Columbia sent mail-in ballots to all their eligible voters, without the usual application requirements. As a result, a record number of Americans - 64 million - made use of mail-in voting during this election.

The ease of voting offered by expanded mail-in balloting contributed to a record 154.8 million people - 2/3 of eligible voters - casting a ballot in the 2020 election, a roughly 7% increase in voter turnout from 2016. 

Rules for how the ballots were to be counted varied across states. For example, North Carolina began counting mail-in ballots weeks before election night. In contrast, the state of Pennsylvania would not begin counting mail-in ballots until election day.  

As a result of the expanded access to mail-in voting and the high voter turn out, counting the ballots and certifying the results took longer than usual. Major U.S. media outlets would therefore take longer than usual to declare a winner: CNN, Fox News, and the New York Times would not call the election for Joe Biden until November 7th, four days after election day. Typically, presidential elections in the U.S. are called the night of the election or in the early morning hours the next day. [5]

Trump Claims Election Fraud

The election was also atypical in that, even after large media outlets had projected a victory for challenger Joe Biden, the sitting president, Donald Trump, claimed that he had in fact won the election and carried out an audacious effort to overturn the results in key states needed to win the electoral college. 

In the weeks after the election, Trump would tweet that the election was ‘rigged’, tell Pennsylvania lawmakers on a phone call that ‘we won all of these swing states by a lot’, and state in a video posted on Facebook by the White House that America’s election system was ‘under coordinated assault and siege’. Trump would even place a phone call to the top election official in Georgia telling him to ‘find’ the votes needed to allow him to win the state.

Alongside his effort to pressure state-level election officials into overturning the election results, Trump’s legal team filed over 60 lawsuits in court arguing that election fraud had occurred. All but one failed. [6]

The events would culminate with Trump supporters breaking into the U.S. Capitol Building on January 6th, 2021 as Congress was set to tally the electoral college votes and officially declare Joe Biden the winner. The break-in delayed the tallying of the electoral college votes by several hours as legislators hid from the encroaching mob. The electoral vote counting resumed the evening of the 6th and made a Biden victory official.

The House of Representatives subsequently filed articles of impeachment against Trump on January 13, 2021 on the charge of ‘incitement of insurrection’.

Joe Biden was sworn in on January 20th, 2021. 

PredictIt’s 2020 U.S. Presidential Election Markets

PredictIt’s markets for betting on who would win the November 3, 2020 presidential election were not settled and closed until the meeting of the electoral college in December of 2020. This meant that its markets for predicting who would win the election remained open after the day of the election, as Trump sought to contest the results of the election and claimed, without evidence, that he had won.

Here I argue that several anomalous events occurred in PredictIt’s markets related to the 2020 US presidential election both on election day and in the chaotic period between election day and the meeting of the electoral college. And I claim that these anomalies are examples of the inefficiency of these markets. 

Part I: A Red Mirage

Due to partisan differences in concern over the risk from SARS-CoV2, it was widely anticipated that voters who were more likely to vote Democrat would also be more likely to opt for mail-in voting while voters who were more likely to vote for the Republican candidate would also be more likely to vote in person at a voting center. 

Additionally, the Supreme Court had ruled that mail-in votes in Pennsylvania could be counted as late as Friday November 6th (three days after the election), as long as they had been post-marked by November 3rd (election day). And the battleground states of Pennsylvania, Michigan, and Wisconsin, would not begin counting any votes - whether they were cast by mail-in ballot or in person - until polling stations had closed on election night. 

This created the possibility of a ‘red mirage’. Owing to their size, high voter turnout, and high use of mail-in ballots in the 2020 election, counting votes in large cities, which tend to lean Democrat, would take longer than counting votes in smaller more rural counties, which tend to lean Republican. And as smaller rural counties quickly reported their results, early election night voting tallies may misleadingly suggest that the Trump was posed to win these pivotal states by a wide margin, only to have the early margins reduced or surmounted by the subsequent completion of vote counting in larger counties. 

The possibility of such a ‘red mirage’ was widely discussed in the run-up to election night by American commercial media, including by pro-Republican outlet Fox News.

Pre-Election Day

In the lead up to the November 3, 2020 presidential election, PredictIt’s presidential election market had consistently expected a Biden victory. This can be seen in 90-day price chart below which spans August 19, 2020 to November 16, 2020. The blue line represents the price of buying a contract on Joe Biden winning the election and the red line represents the price for buying a contract on Donald Trump winning.

As the chart makes clear, the consistent market expectation from August 19 to election day November 3 was for a Joe Biden victory. At 11:59pm EST on August 19, the price of a Joe Biden Yes contract was 59 cents while the price for a Trump Yes contract was 44 cents. The respective 11:59pm EST prices for these two contracts on November 2nd, the day before the election, were 61 cents and 44 cents. The market had therefore continually ‘predicted’ a Biden victory for months.

Election Night

But as voting results began to come in on election night, and early results misleadingly suggested a Trump victory, prices reversed sharply. The 11:59pm EST prices on election night now indicated a market expectation of a Trump victory: Biden Yes contracts were on offer for 42 cents while Trump Yes contracts were trading at 59 cents each. The 30-day price chart below depicts more clearly the abrupt temporary reversal in market expectations on election night. 

However, as battleground states began reporting the count of larger counties – which ultimately favored Biden – in the early morning hours of Wednesday November 4th, the market sharply reversed back to its original prediction of a Biden win. The November 4 11:59pm EST price for Biden Yes contracts was 87 cents while the price for Trump Yes contracts was 19 cents.

Conclusion

This night-of reversal suggests that these prediction markets may not constitute efficient processors of all available relevant information. Reversals in the markets due to them absorbing new information related to an election – such as performance in a presidential debate – would be consistent with them being efficient. However, these abrupt price changes were based, not on meaningful new information, but only on the sequence of voting counting: the market temporarily reversed its months-long prediction based on a ‘red mirage’ due to some ballots being counted before others. That the market completely reversed its assessment of the likely outcome of the election based only on a statistical illusion suggests that prediction markets do not always correctly distinguish between relevant information and noise.

Part II: Alternative Facts 

On November 7, 2020, four days after election day, major U.S. media outlets such as the New York Times, CNN, Fox News, and the Associated Press called the election for Biden/Harris ticket. Although vote counting was ongoing in many states, these outlets had projected - based on the districts/states that had finished counting and the districts/states outstanding - that the Democrats had won the election. However, Trump would continue to argue that he had won the election and would attempt to overturn election results in the key states of Georgia, Michigan, Pennsylvania, and Arizona. PredictIt hosted markets where speculators could bet on who would win each of these states.

Anomaly 1 – Battle Ground States

Commercial media outlets projected a Biden victory in Georgia on November 13th and an audit confirmed a Joe Biden victory in the state on November the 19th. Georgia election officials certified a Biden victory on November 20th. However, on November 22, 2020 PredictIt contracts on Joe Biden winning the state of Georgia were still trading at only 88 cents.

Markets for Michigan, Pennsylvania, and Arizona exhibited similar price behavior. Commercial media outlets had projected a Biden victory in Michigan as early as November 4th. And Michigan certified a Biden victory on the 23rd. However, contracts on Joe Biden winning Michigan were on offer for only 90 cents.

This constitutes an anomaly: if Joe Biden was deemed the consensus winner of Michigan weeks earlier, why were contracts which pay out $1 in the event of a Biden victory in the state on sale for just 88 cents on November 22? And why were contracts which pay out $1 in the event of a Biden victory in Georgia trading at just 90 cents on November 22 after he had been deemed the consensus winner of the state and Georgia election officials had already certified a Biden victory? Why had these contracts not been bid-up to PredictIt’s maximum price of 99 cents?

One possible explanation is that these markets were efficiently pricing in the probabilities that Biden doesn’t ultimately win those states. Under this interpretation, the market would be pricing in a 12% probability that Trump’s efforts to overturn the election result in Georgia would be successful. And it would be pricing in a 10% probability that Trump’s efforts to overturn the election result in Michigan would be successful. 

However, the price action in other prediction markets related to the 2020 presidential election suggests another explanation: the anomaly was a consequence of some speculators acting on ‘alternative facts’. That is, some market participants believed Trump’s claim that he had won the election.

Anomaly 2 – Bets on Trump Winning the Electoral College By +280

Evidence for this interpretation comes from looking at PredictIt’s market in which individuals could wager on what the spread in electoral college votes would be. In this market, participants could wager that Biden would win the electoral college by 100-149 votes, that Trump would win the electoral college by 100-149 votes, that Biden would win the electoral college by 150-209 votes, that Trump would win the electoral college by 150-209 votes, and so on.

Here we find another anomaly in PredictIt’s markets.

By November 22, the above-mentioned media outlets had called the election for the Democrats with 306 electoral college votes going to Biden and 232 electoral college votes going to Trump – a spread of Dems + 74. However, as shown in the figure below, contracts on Trump winning the electoral college with a whopping + 280 spread were still trading at 8 cents.

Why were contracts which settle at zero in the event of a Biden victory or a Trump victory by anything less than +280 not selling at PredictIt’s minimum price of 1 cent?

Even more curiously, speculators had skipped over contracts on Biden winning by less than + 74 or Trump winning by a thin margin; those contracts were trading at lower prices. As seen in the figure below, contracts on Trump + 10-29 were trading at 3 cents, Trump + 30-59 contracts were trading at 4 cents, and Trump + 60-99 contracts were trading at 5 cents.

If the market was efficiently pricing in a possibility that Trump’s efforts to overturn the election result may be successful, we would expect to see the opposite price action: it would be more likely that Trump has success in overturning enough states to eke out a small electoral college win than to overturn results in numerous states and decisively win by + 280 electoral college votes. And so, contracts on Trump + 10-29 should have been priced higher than those on Trump by + 30-59, and so on. 

Additionally, as shown in the image below, PredictIt’s rules for this market explicitly stated that the actions of ‘faithless electors’ would not have a bearing on this market. That is, even if Biden won the election but electors in the electoral college who were obliged to vote for him opted to vote for Trump, the contracts in this market would resolve to a Biden victory. This eliminated faithless electors as a potential route by which Trump could overturn the results of the election.

So what could explain speculators wagering on Trump winning the electoral college when it had already been widely called for Biden with a + 74 spread? And why the particular interest in contracts on a Trump + 280 victory?

The answer seems to lie in some traders acting on erroneous information.

Apparent Explanation

As noted above, Republican leaning media outlet Fox News had called the election results in line with other large media outlets. 

In response, Trump encouraged his supporters to abandon Fox News in favor of media outlets which bolstered his claims of election fraud. One of the alternative outlets Trump endorsed was One American News Network (OANN), which saw an increase in viewership during this period.

Reports indicate that OANN was suggesting Trump may have actually won 410 electoral college votes to Biden’s 128 (a margin of Trump + 282 electoral college votes). And this included Trump winning both Georgia and Michigan, the two states that are the subject of our first anomaly.

Some speculators acting on these ‘alternative facts’ would explain all the above-mentioned anomalies: that contracts on Biden winning Georgia and Michigan were trading at only 88 cents and 90 cents respectively after both states were called for the Democrat, that contracts on Trump winning the electoral college by + 280 were trading at 8 cents after major media outlets had called the election for Biden, and that contracts for Trump winning the electoral college by a lesser margin were trading at lower prices.

This theory is further bolstered by polling data gathered in November 2020 indicating that an overwhelming majority of Trump voters believed he had won the election and by the ‘Stop the Steal’ break-in to the Capital Building on January the 6th, 2021: just a small subset of these individuals acting on their beliefs in the prediction markets would explain all the anomalies I’ve identified. 

Implications for the Efficiency of Prediction Markets

If this explanation is correct, it has some interesting implications for the efficiency of political prediction markets. Efficiency would require that these markets incorporate new publicly available data and, in doing so, continually move contracts to the ‘correct’ price. But what if market participants are consuming entirely separate streams of information, essentially occupying alternative realities: it can’t be the case that markets are efficiently incorporating publicly available information when market participants don’t even agree on what the facts are and these differences express themselves in market prices as noticeable, persistent, and exploitable anomalies. 

The efficient market theory treats information as a homogenous given which is correctly processed by traders. But, as both part I & II of this essay detail, in reality humans often disagree on how to distinguish fact from fiction and in their determination of which of the facts are truly relevant. This heterogeneity in how information is processed poses a challenge to the idea of efficient markets.

Implications for the Efficiency of Financial Markets

It might be argued that such concerns do not apply to the financial markets. These prediction markets were, after-all, capped in size. This prevented the ‘smart money’ large institutions and investors from partaking, allowing such anomalies to manifest and persist.

History provides us with many reasons to believe that large investors are not immune to challenges in information processing. But even if we were to assume for a moment that they are, my finding in this prediction market still has interesting implications for the financial markets. Recent events have shown that small investors can, under certain conditions, cause significant movements in the prices of liquid securities in the financial markets. (I will detail this in my series of forthcoming articles on what I dub the ‘Meme Asset Frenzy’.) So even if we assume that such inefficiencies are generated only when ‘dumb money’ investors have significant sway over asset prices, we can still identify substantial and persistent inefficiencies in the financial markets. In my forthcoming series of articles on this subject, I shall argue that such inefficiencies can indeed be identified and exploited to generate alpha. 

Footnotes

[1]  2 U.S.C. §7

[2] 3 U.S.C. §7

[3] 3 U.S.C. §15

[4] U.S. Const. amend. XX. 

[5] The disputed election of 2000 between Republican candidate George Bush and Democratic candidate Al Gore is a notable exception.

[6] The Trump campaign won a case in Pennsylvania which resulted in a small number of ballots not being counted.