TL;DR
Is Polymarket trustworthy?
Prediction markets have been at the forefront of this US election season. Polymarket, the biggest name in the space, is heavily cited by candidates, news outlets, and public figures in the same breath as polls from NYT/Siena, ABC/WaPo, and Marquette University.
We at Triton are not in the game of political prognostication nor are we trying to trade the election outcome given our long-term focus actively. Nonetheless, this is an interesting experiment to watch play out in real-time. Given most polling has the election near 50-50, highly dependent on essentially one state (Pennsylvania?) where the outcome is largely seen as a coin toss by most experts, the apparent assuredness of the market about a Trump victory is noteworthy. Prediction markets have been found to be more accurate than polls, so perhaps there is something there.
But to many, the question remains: is Polymarket trustworthy? Anecdotally, the answer to this question has seemingly fallen along political lines, depending on who is leading. Currently, those with left-leaning inclinations argue that the platform is awash in fraud and manipulation, while those with right-leaning inclinations argue that the platform is a far more accurate barometer of the likely outcome, especially when compared to polls that have seemingly lost accuracy in recent elections. With prediction markets, there is actual money on the line.
This post will simply explore some of the realities of the platform. We will hold off on taking a strong position either way and instead, let the reader ultimately decide how much weight to ascribe to Polymarket’s odds for his or herself.
The headline numbers are impressive
There has been over $3B collectively wagered on the main market: who will win the US Presidential Election in 2024? Across the broader platform? Closer to $5B. Bets on Trump have totaled $1.26B and Harris has totaled $790M. The daily volumes have grown substantially in the past 45 days as well, eclipsing $70M in bets on a single day. Impressive numbers, no doubt. But that does not mean there is currently $3B at risk on the platform.
Source: Dune (@rchen8)
Other metrics are more informative there, such as total value locked and open interest. TVL, referring to the actual amount of funds locked in contracts on the platform, is at an all-time high ($450M). Open Interest has followed a similar parabolic growth trend over the past few months, touching at closer to $440M. These numbers are reflective of the entire platform, and not just the presidential election market.
Source: Dune (@rchen8) and DeFiLlama
Other candidates (Biden, Haley, Newsom, Kennedy, DeSantis, etc.) have collectively received several hundred million in betting volume as well. For example, $250M in volume has been a bet on Kanye West, Hillary Clinton, and Michelle Obama's contracts alone.
Source: Polymarket
Volumes ($ billions) and value actually at risk at the time of writing ($ hundreds of millions) are both substantial. It is important to note that much of this is very new – open interest and TVL have nearly quadrupled since late September when each was just barely over $100M. This explosive rise has led many to question whether these numbers can be taken at face value. Obviously, this run-up has occurred as the election has drawn near and Polymarket is increasingly front and center of the election discourse. There is no doubt it is attracting new traders given its rise in prominence.
One chart here is particularly interesting to look at:
Source: Dune (@hashed_em)
The above graphs show the weekly volume of Polymarket over time. The four-year-old platform undoubtedly saw substantial increases throughout late 2023 and 2024 through July. At that point, volume growth seems to have somewhat tapered off at around $100M or so per week (the first step up in the graph). This first run-up occurred over a massively significant period in the election: the Biden-Trump debate on June 27, the assassination attempt of Trump on July 13, and the ultimate withdrawal of Biden on July 21. Each of those was a hugely important event, and it makes sense that volumes would increase in response. From the last week of July through the last week of September, though the campaigns continued and odds continued to change, Polymarket did not see any notable increase in activity in terms of betting volume.
Interestingly, something happened the week of September 23 to massively kickstart volumes on the platform that apparently attracted more activity than a Trump assassination attempt or Harris replacing Biden on the ticket.
Given the platform’s age and (relative) size at the time, all else equal one would expect that the next marginal user of the platform would largely act in the same manner as the several hundred thousand that came before them. Clearly, this is not even remotely the case here, given weekly volumes jumped to $434M from $123M in just a single week despite there being only a ~15% increase in new users. In the 5 weeks since, weekly volumes have continuously been above $500M, far above the trend seen for most of the summer.
Source: Dune (@hashed_em), September 23 is marked
Are the new users that the platform has recently attracted suddenly willing to bet 4x more on average than the 700K users that had come before them?
Perhaps, but if so, we’d expect to see TVL or open interest also instantly quadruple over that time as new users onboarded more funds to wager. But we don’t. Open Interest increased from $120M on September 23 to just $134M on October 1 and TVL from $119M to $132M. So, despite there being over 4x the betting volume, there was only a corresponding ~10% growth in funds at risk on the platform and a 15% increase in users. Also interestingly, daily active users increased by 27%.
To recap: the 15% of marginal new users onboarded a lower relative proportion of funds onto the platform (10%) but a disproportionately high share of those users are more active on a daily basis. Why such a massive change in pattern?
The general narrative one might get from the media is that the platform is ripe with manipulation. Many point to the single French trader, who through multiple accounts accumulated a nearly $50M position in Trump-Yes in October. While that explains a massive increase in TVL/OI on the platform, that individual is not swing trading that positions daily to drive volumes. Does an outsized position like that relative to general liquidity skew the odds on the platform?
Likely yes – there is a substantial, growing premium reflected in Polymarket’s odds relative to the next more liquid market on Kalshi since the beginning of October. Kalshi’s market itself sees volumes in the hundreds of millions, so it is difficult to argue that it too is not reflective of the market’s general prediction. But outsized bets can definitely move the market: one trader cleared Polymarket’s entire order book with a $3M bet, filling a substantial amount of his bid at 0.99 despite the contract price being 0.63. Some of this premium may be due to a lack of ability to arbitrage this away across markets, but it is widely known that Polymarket’s restrictions are somewhat…avoidable for a determined US trader.
Source: Polymarket and Kalshi
The general takeaway is that there is definitely noise and inefficiencies in these markets that can lead to skewed results. And given the impact that polls and increasingly prediction markets have on general discourse and their ability to influence voters, it is fair to raise these concerns. Plus, with the frequency with which state actors attempt to influence other countries’ elections, this is a very obvious method through which to do so. Your author has not seen evidence of this but is rather just raising a possibility.
Another interesting perspective is to look at prediction market spending as a form of advertising buys. Kalshi strictly prohibits any participation from anyone who holds elected office in the US or who is associated with any campaigns or elections (news desks, pollsters, PACs, etc.), but Polymarket’s permissionless nature makes it far harder to police who uses the platform. Is there a world where interested parties want to influence public perception about a candidate’s positioning? It is tough to argue no given the propensity for campaigns to selectively blast out favorable, non-contextualized polling results. Plus, the math seems to work out far, far better than just straight ad spend.
Imagine a world where an entity with $1M to spend can choose between A) sponsoring new polls or ad buys the traditional way and B) spending $1M to prop up or harm a candidate in an election via prediction markets. For our purposes, let’s assume that the benefit achieved is the same either way (e.g. the candidate appears marginally stronger/weaker in the eyes of the electorate). It is just the means that are different.
In scenario A, the entity spends $1M and achieves its outcome. Importantly, they have a 100% chance of losing $1M in the process – money spent on ads or polls is gone. The net outcome here is the target benefit achieved, but $1M is gone.
In scenario B, the entity spends $1M to purchase contracts on a betting platform and achieves the same outcome. Pollsters generally find that the election is a coin toss (50/50) but prediction markets point to one candidate being favored (60/40). $1M is spent on contracts trading at $0.60. That means, there is a 60% chance of the market resolving to ‘Yes’ and the entity receiving $1 for every $0.60 spent on the platform – a potential 67% payoff. Naturally, that also means that there is a 40% chance that the market resolves to ‘No’ and the entity loses all $1M. The expected value there is $0. Thus, the net outcome here is the target benefit achieved, with only a 40% chance of losing that $1M but a 60% chance of ending up with $1.67M (your initial capital + $670K in winnings). If the odds are actually closer to 50/50, the net payoff is naturally lower.
Plus, that $1M initial spend is not locked up. Polymarket allows you to sell your position at any time. So, the entity could actually close out that position shortly before an election resolves, recover all 100% of those funds, and take almost zero risk at all, aside from the contract price moving in between buy and sell. Not a bad ad spend strategy.
There have also been some recent concerns about “wash trading” on the platform, with some researchers estimating that perhaps up to one-third of the volume on the platform is fake. The underlying research has yet to be released so it is difficult to validate their findings. But some level of manufactured activity would not be surprising. Is it necessarily malicious? There doesn’t seem to be proof of that.
Rather, there is a far easier, far less sinister reason that explains the change in user behavior and overall activity, including the above-mentioned ‘wash trading’. As is often the case with crypto: news broke that Polymarket was going to release a token. As we’ve highlighted in previous posts, this is typically one of the single biggest drivers of activity for a crypto project in its early days of adoption, and see no reason why this time would be any different. Following the election and the airdrop, it would not be surprising to see activity and volumes on Polymarket evaporate, returning to volumes/user levels closer to what we saw prior to the September announcement. One can be fairly confident in this just by looking at retention numbers. The below shows how many addresses interacted with the contracts in subsequent months after first using Polymarket. While still relatively strong, there is some cause for concern here.
Source: TokenTerminal, below the red line aligns with October
First, notice the spike from 42K to 159K between August and September – a massive departure from ordinary trends. The October cohort returns to a similar magnitude as before, dropping to a level not seen since June, before all of the runup. Secondly, notice that far fewer users are returning the next month. For much of 2024, Polymarket saw phenomenal retention rates with 50-60% or more in the first month. Since the airdrop announcement, those rates have fallen by 2/3. This is a common pattern we see with many airdrops and if and when a snapshot date is announced, one can be fairly certain that these retention rates will drop again even further. How much noise are these airdrop farmers adding? Difficult to say, and likely much of their impact is netted out, so it could be very minimal. More of the impact will be on user counts, trade volumes, and transaction counts rather than skewing odds.
Everything in this post indirectly highlights something important: the ability to evaluate the platform like this is why on-chain prediction markets make so much sense. The transparency and data granularity that is available are unmatched versus centralized competitors. Given the stakes and frequency of which Polymarket is cited in mainstream election coverage, the ability for anyone to look into user behavior and key drivers is a strong value proposition of the platform and undoubtedly a positive development if, as it appears will be the case, prediction markets are going to play a larger role in elections going forward.
Conclusion?
That’s for the reader to decide.