A Better Way to Predict?

Investment Note #22 - 29th November 2024

A Better Way to Predict

.

Synopsis

• Prediction markets have become a popular tool for election forecasting, gaining significant attention during the 2024 U.S. presidential race. They are often praised for their perceived accuracy and stronger convictions about particular outcomes compared to traditional polling. But how reliable are they really?

• Interestingly with the Irish general election today, a similar pattern has emerged: prediction markets are assigning a much higher probability of victory to one party than most opinion polls. What explains this discrepancy?

• As media outlets increasingly reference these markets alongside traditional polls, some argue they represent a superior alternative. This note examines the mechanics of prediction markets, their limitations, and whether their growing reputation is deserved or inflated by their novelty.

Understanding Prediction Markets

• Prediction markets are essentially betting platforms where participants buy and sell shares tied to the outcome of future events. These range from elections to economic policies, with the price of shares reflecting the perceived likelihood of the event.

• The concept is simple: if a prediction is correct, the contract pays out $1; if wrong, it’s worth $0. For instance, Polymarket shows a 20% ($0.20) chance of snow in New York City on New Year’s Day - a $100 bet would buy 500 shares at $0.20 each. If the prediction is correct, each contract pays out $1, meaning you receive $500 in total - $400 in profit, plus your original $100 stake.

• Over $3.2 billion in transactions related to the 2024 US election were made on Polymarket alone (source: Polymarket). While these markets have expanded beyond politics, election-related activity dominates.

Skewed Odds

• Despite their growth in size, such platforms remain susceptible to manipulation by high-stakes gamblers. For instance, a French bettor wagered $45 million on a Trump victory, inflating his implied probability of success. As Rajiv Sethi, an Economics professor at Barnard College, noted, “There’s no reason why the price set by these few big players should necessarily reflect an accurate forecast.”

• Prediction markets are often lauded for their ability to react to new information. For example, after Biden’s poor debate performance, there was speculation about Kamala Harris’s chances despite her not being a declared candidate - and her probability of winning rose sharply. This was a shift that traditional polls, which were not explicitly focused on her at the time, largely missed. Yet prediction markets’ reliance on a small, affluent participant base often skews probabilities, reflecting individual biases rather than collective wisdom.

• Adding to these challenges is the issue of liquidity in smaller or international betting markets. Ireland's 2024 general election currently has only $430,000 in trading volume on Polymarket, limiting its capacity to reflect robust predictive insights.

Unearned Victory Lap?

• Despite the heightened attention prediction markets received during the 2024 U.S. election, their performance was far from exemplary. In the final hours before election day, Polymarket gave Trump a 59% chance of victory - hardly a definitive call and more akin to a coin flip than a confident forecast. Few would be comfortable betting much on anything where they would lose four times out of ten.

• Just days earlier, Trump’s Polymarket odds had dropped to 52% after a single outlier poll by famed pollster Ann Selzer showed Harris leading in Iowa. Known for her past accuracy, Selzer’s reputation amplified the impact of the flawed data point, even though it deviated sharply from other polling. Trump ultimately won Iowa by 15 points, underscoring how prediction markets can overreact to isolated, misleading data.

• Prediction markets also leaned heavily toward Harris as the likely winner of the popular vote (an aggregate of all voters from all States in America), despite national polls showing a near tie. While Trump ultimately won this too, prediction markets may have failed to account for the complexities of the Electoral College, which determines the presidency. These platforms assigned an overly high probability to the popular vote winner becoming president, neglecting the importance of specific states and leading to misaligned odds. Can such a misstep truly support the claim that bettors successfully forecasted the election outcome?

The Garzarelli Effect

• The term originates from Elaine Garzarelli, who rose to fame after predicting the 1987 stock market crash. While her call earned acclaim, her later predictions proved far less reliable, illustrating how one success can dominate perceptions.

• Like Garzarelli, Selzer’s earlier successes cemented her reputation as a “guru” of political polling, but reliance on her singular insight now appears misplaced.

• Prediction markets often mimic this phenomenon, where participants disproportionately trust individuals or platforms with a history of accurate predictions. This skews odds, amplifying biases rather than reflecting objective probabilities. As Niall Ferguson noted in the Wall Street Journal, the rise of prediction markets reflects a broader decline of traditional experts, with platforms encouraging users to believe they “know better than the pollsters.”

Betting Lag

• Prediction markets focus on long-term events, such as elections, allowing for the integration of evolving information over time. This extended time horizon introduces instability, as markets are prone to overreact to short-term developments and narratives - such as the Iowa poll. This tendency to amplify fleeting sentiment often detracts from their ability to reflect deeper voter trends.

• In contrast, sports betting revolves around events with immediate outcomes, where high volumes and quick resolutions stabilise prices. Platforms like Paddy Power cater to this by emphasising bets on games happening within hours or days, ensuring real-time price corrections.

• For example, while only £5,190 was wagered on the Wimbledon 2024 winner in late March, an in-play, low-profile tennis match between Francesco Maestrelli and Pierre-Hugues Herbert attracted £227,421 in bets. This preference for immediate outcomes ensures that live, in-game betting dominates sports markets, creating liquidity that sharpens price accuracy almost instantly.

Conclusion

• The French bettor credited his success in forecasting Trump’s 2024 victory to a method as unorthodox as asking people how they thought their neighbours would vote. This "neighbour polling" approach highlights the speculative strategies sometimes underpinning prediction market activity, offering a lens into how hidden voter sentiment might be uncovered, though not without its flaws.

• The Irish election provides a fresh opportunity to test whether prediction markets can overcome their limitations or if they remain as uncertain as predicting snow in New York. While they may capture sentiment shifts in real-time, their flaws continue to undermine their credibility. Suggesting they’ve rendered traditional polling obsolete is premature at best.