No, "Ordinary Users" Aren't Losing Half a Billion Dollars on Kalshi

Roosevelt Institute’s recent work on prediction markets has significant flaws that invalidate the findings.

Here is a deeper description of the study’s flaws: 

The problems with the report begin with the series’ title:

The Hidden House: Prediction Markets and How They’re Shaping Society

There is no “house” on Kalshi, hidden or otherwise. Kalshi works by matching orders together, like all financial exchanges do. The implication that there could be a “house” on a prediction market demonstrates a fundamental misunderstanding of the operation of financial exchanges, as well a fundamental misunderstanding of what “the house” is.

This second misunderstanding in particular is quite surprising, because the definition is quite simple: the house is the house because there is one of them. 

With casinos, there is only one entity setting prices — hence why the house is called the house, not “the homes”. 

This lack of pricing competition means casinos can set whatever prices they want. On Kalshi, immense competition for flow drives the prices down so that the most casual end consumer pays the lowest price possible. This is a key reason why Kalshi has seen such immense growth — consumers notice that their payouts are higher with no house — and is also the fundamental reason behind Kalshi’s competitive threat and the casinos’ subsequent legal, PR, and influence campaign against prediction markets.

The study conflates the presence of market makers — entities who are both buyers and sellers in a given market — with having a “house”. Here is this conflation, spelled out in the very first sentences of the report:

Prediction markets advertise that their platforms are better than traditional gambling because users are betting against each other rather than against a “house.” But the reality is that ordinary people participating in prediction markets are largely betting against professional traders using sophisticated methods.

The structuring of these sentences implies that these are somehow contradictory claims, which is false. 

Yes, there is a skill gap in prediction markets, as there is with every other activity in the history of the world. Yes, prediction markets do advertise that exchanges are an upgrade from gambling platforms because of the lack of a house (because they are). These are not contradictory claims; presenting them as such shows a misunderstanding of both financial markets and casinos. 

By conflating participant skill levels with market structure, Roosevelt obscures the key societal difference between exchanges and casinos. 

Casinos are an entity with monopolistic pricing power and an incentive to maximize user losses, while exchanges are neutral platforms that encourage competition to drive prices as low as possible. This is to say nothing of the more sophisticated traders, who are now empowered to actually have a chance to profit from their knowledge and skill, instead of being instantly banned by casinos for the crime of winning). 

Simply by being an exchange, Kalshi provides better prices for the most casual, least price-sensitive users. Also by virtue of being an exchange, Kalshi empowers more sophisticated traders to actually have a chance to profit from their knowledge and skill, rather than being instantly banned by casinos for the crime of winning. 

Exchanges are a better system for all user types involved. Kalshi is building a meritocracy as an alternative to a rigged system. The fact that Kalshi is a meritocracy, and not a charity, is not a knock on the system. It is the whole point. 

The headline of the Roosevelt study reads as follows:

Since Kalshi’s Launch, Ordinary Users Have Lost Half a Billion Dollars

The stated methodology does not establish this as true. Roosevelt has committed a severe methodological error that overstates this amount significantly. 

Their error is simple: they attempt to measure the economic success of “ordinary users” on Kalshi. This is not something that can be measured with Kalshi’s trade-level data. 

Their mistake involves these two issues:

The Dune snapshot that the authors use reports every trade with one “taker side/incoming order” and the opposite “maker side” is inferred. 

Using this taxonomy as a proxy for “casual traders” and "sophisticated users”, as the report does, is an incredible methodological flaw that obscures the truth in many ways. Here is why:

During events where all resting orders have been bought up, default/GTC/GTT limit orders can rest if not immediately filled. If later filled, they are categorized as the “maker” for that trade in the API. Additionally, if a GTC/GTT app order does not fully execute, any remainder can rest; if a later opposite incoming order hits it, that first order is then reported as the maker. 

Practically, what this means is that many app-based/”casual” trade orders end up categorized as maker trades — maker/taker is a description of order book classification, not necessarily a description of the types of users placing the trades. The report uses it interchangeably with demographics, invalidating the entire report’s results.

This phenomenon is especially prevalent in highly active events. When taker orders arrive very close together, the engine still serializes them: for any emitted trade, one order is the incoming taker and one was the resting maker by the time the match happened. Given how much of Kalshi’s volume comes during these types of events — election days, mid-sporting events, etc. — this represents a pretty serious flaw in the methodology of the authors.

Here is a table explaining the errors:

Any classification structure that categorizes some standard app orders from casual traders as activity from “sophisticated users” and some high-frequency trades from institutions as activity from “ordinary users” is clearly flawed beyond repair.

Overall, what the study is trying to measure is the economic differences between casual users and sophisticated users. This is why the findings are invalid. Any sort of narrative attempting to measure the split between "everyday", "ordinary", or "casual" users, and "professional" or "sophisticated" users can not be determined from Kalshi's trade data. 

Lumping every maker order into a bucket of “sophisticated traders” paints a false picture of who these traders actually are. “Sophisticated traders” is not a synonym for Wall Street; often these traders are subject-matter experts with ordinary day jobs. 

In the API, “maker orders” can be resting orders placed on the orderbook. You do not need to be very “sophisticated” to place a resting order. There is some degree of sophistication to be sure; it takes ~4 seconds instead of ~1 second, and involves pressing ~6 buttons instead of ~2. But the truth is that anyone — from casual app user to casual web user to advanced API high-frequency trader — can go on Kalshi and very easily place a resting order. 

Among Kalshi’s top earning traders are a public schoolteacher from Pennsylvania, an IT professional in Minnesota, and a retiree in Kansas, to say nothing of the ordinary people who have been able to leave other jobs to become full-time Kalshi traders — which is a thing that is only even possible because you can actually win on Kalshi, unlike a casino. 

The notion that only Jane Street and SIG know how to press buttons on the Kalshi app is not only ridiculous, but insulting and patronizing to the millions of people who use Kalshi’s features. 

Roosevelt’s focus on “total dollar amount lost by orders classified as takers in Kalshi’s API” is in and of itself a misleading statistic. 

It is true that there are categories of users on prediction markets who are not price-sensitive, and thus participate in the market in ways that do not maximize expected value. These participants are not a monolithic group, and likely have different motivations for not caring to maximize their statistical expected value; perhaps they are hedging an external risk, perhaps they are confident enough in the outcome they are okay with a few cents of price differential, or perhaps they simply enjoy having exposure to events and are not concerned with the exact price of the contract. 

They key thread for all of these groups: there is demand for event exposure regardless of how profitable that exposure may be. 

Given this demand exists, there are two options of where it can be filled:

For people who do not care about the pricing, especially the most casual of casual users, which Roosevelt’s study purports to care about, fulfilling this demand on a prediction market is clearly the superior option, both for them and society as a whole. It is a difference that likely represents  millions of dollars in savings for the most casual users. 

Roosevelt’s study chooses to ignore this reality, and instead focus on a misleading (and incorrect) metric of “total dollar amount lost by orders classified as takers in Kalshi’s API”. The proper metric, and a proper headline for the report, would be measuring the amount of money saved by the public’s exodus from casinos to prediction markets. 

Another key metric that the report rests on is “ratio of users that lose money compared to those that win money”. They use both Polymarket and Kalshi data for this. Here is their claim about Polymarket data:

Existing research, largely on Polymarket, suggests that the vast majority of people participating in prediction markets (retail traders) lose money, with the profits being captured by an extremely small number—the top 0.1 to 1 percent—of sophisticated users.

Logical alarm bells should be going off for anyone who sees this data. 0.1% of traders capturing 67% of gains is an extreme statistic. If true, it would make Polymarket one of the most concentrated and unequal markets of any kind, ever. Anyone who sees it should immediately quadruple-check why it would be that way. 

The authors of the study did not include what is widely known as the plausible explanation for these market dynamics: wash trading.

Wash trading is a form of market manipulation where an entity simultaneously buys and sells the same financial asset — essentially sending money back and forth to yourself via the order book. According to economists at Columbia University, Polymarket is rife with wash trading – at times representing 90% of all activity on the platform. 

Wash trading, by nature, incurs a small loss — you’re only able to wash trade at scale by buying and selling orders with yourself at similar prices, incurring fees with each trade. So, while the majority of wallets on Polymarket lose money, it is not because of sophisticated trader advantages, it is because the majority of activity is likely wash trading, which incurs small losses when done. This checks out with other public analyses of Polymarket’s data that show the median loss is about $2. 

Most wallets are likely wash trading, which incurs tiny losses and would specifically skew the “ratio of winners” statistic. The issue is that this is not representative at all of what the authors try to make it represent, which is something akin to “platform equality dynamics for average users”.

Also, the authors make the mistake of conflating “wallets”, which are anonymous pieces of code that can be controlled by anyone, with “retail traders”. Those are not the same thing! 

In addition to Polymarket data, Roosevelt uses Kalshi “winner ratio” data in its report, and hilariously uses the phrase “admitted” when referencing Kalshi’s own self-published data about this, falsely insinuating that our own data that we voluntarily released publicly should be something we should be ashamed of.

Kalshi itself recently admitted that there are nearly three times as many people losing money as making it on its platform.

A one to three ratio of net profitable users makes Kalshi one of the most equitable, egalitarian marketplaces in existence. At casinos, ~0% of users are profitable, because anyone that proves they can win is banned from the platform — despite the casinos being allowed to advertise their products as if you can win. This is to say nothing of binary options or equity or forex markets, where large, institutional market makers have monopolized nearly all order flow and have developed proprietary algorithms that systemically outperform non-institutional traders to the tune of billions of dollars per year. 

Also, these numbers they cite are only from trade data, which does not include the interest that you earn from being a Kalshi user! In reality, these numbers underestimate everyone’s profit on Kalshi by around ~3-4%. 

Roosevelt chooses to ignore this context, misleading readers about the true reality of Kalshi’s market dynamics.

Irrespective of the report’s lack of context and proper framing, it is indeed true that there are, by raw metrics, more losers than winners on Kalshi, and likely on all prediction market exchanges. The true error of the report is that, by ignoring essential context, the authors try to make this fact seem like something that is bad or dangerous about prediction markets. This is wrong. The current dynamics are markets working as they should. The most critical context missing of all is that the inverse is much, much worse.

Prediction markets are binary options. Every dollar lost is also a dollar won. There are two options for how this distribution can work:

It’s tempting to look at these options and say, as the report implies, ‘The best option for society here is that more people win than lose’. But remember, there is a finite pool of money at stake. If there are more winners than losers, the tradeoff is that those losses are significantly larger.

The dynamic that prediction markets operate under now, which is “smaller losses subsidizing a smaller group of winners who get rewarded for being right”, is better for society than the inverse of “a broader group of winners get smaller benefits at the expense of a smaller group of losers who have significant, life-altering losses”. 

This second option is largely what the casino dynamic is, except the casino dynamic is even worse. There is a small group of extreme losers; typically addicts who the casinos prey on via targeted “VIP” bonuses and other predatory tactics that make up the majority of casino revenue. The benefits, however, don’t go to a broader group of winners, but to a single entity: the house.

The system of “small group with life-changingly large losses” is the root cause of gambling’s negative effects on society. This is the system of the casinos, which exchange models like Kalshi do not operate under. The report’s own cited data backs this claim up, but Roosevelt instead opts to frame this information as negative for prediction markets, misleading their readers. 

Individually, each of these issues could be dismissed as a simple mistake or oversight. However, when you take…

…all together, it paints a very clear picture. Whether directly or indirectly, the casino industry has influenced the work of the Roosevelt Institute, calling into question the integrity of this report and all future reports in the four-part series they plan on releasing.