Video creator: John Collison
Compiled by: Peggy, BlockBeats
Editor's Note: Over the past few years, prediction markets have gradually moved from a relatively peripheral financial experiment to the center of discussions on technology, finance, and public policy.
Its widespread attention stems not only from the allure of "betting on the future" itself, but also from the fact that, against the backdrop of amplified noise on social media, inaccurate polls, and declining credibility of traditional information systems, a more fundamental question has emerged: can market prices become a signaling mechanism that is closer to reality than opinions, emotions, and narratives?
This conversation revolved around this very issue. Participants included Stripe co-founder John Collison, Paradigm co-founder Matt Huang, and Kalshi co-founders Tarek Mansour and Luana Lopes Lara.
Kalshi's two co-founders, Tarek Mansour (right) and Luana Lopes Lara (left).
As one of the most representative compliant prediction market platforms in the United States, Kalshi quickly gained widespread attention during the 2024 US presidential election. Prior to this surge in popularity, it had already gone through years of back-and-forth battles with the US Commodity Futures Trading Commission (CFTC), ultimately paving the way for the legalization of prediction markets in the United States through a key lawsuit.
The first part of the conversation focuses on the path to Kalshi's creation: why the two founders did not choose the Silicon Valley-common "do it first, talk later" approach, but instead insisted on "compliance first, growth later"; why they were willing to endure the pressure of lengthy approvals, layoffs, and external doubts in order to win the "election market"; and how this lawsuit against the CFTC became the turning point for the company's true takeoff.
The second part delves into the operational logic of prediction markets. Tarek and Luana explain the fundamental difference between Kalshi and traditional online entertainment platforms: it doesn't rely on a "market maker model" where users profit from losses, but rather is an exchange centered on transaction fees, encouraging liquidity and information to enter the market. They also point out a counterintuitive reality: Kalshi's liquidity doesn't primarily come from traditional large market makers, but rather from a large number of dispersed individual traders, "super-predictors," and small teams. In a sense, prediction markets are not just financial products, but also a mechanism that directly transforms dispersed knowledge into price signals.
In the latter half of the dialogue, the discussion extended further to the future boundaries of prediction markets: from elections and sports to AI, GPU computing power, macroeconomic variables and policy paths, can more and more real-world uncertainties be broken down into market problems that are tradable, feedback-based, and can assist in decision-making? At the same time, a series of unavoidable controversies also emerged—how to define insider trading, whether sports contracts amplify the risks of online entertainment, and how platforms and regulators should establish a new balance between innovation, transparency, and user protection.
Therefore, the significance of this dialogue extends beyond Kalshi himself. It truly attempts to answer the question: will prediction markets become the next generation of financial markets, or the next generation of information infrastructure?
The following is the original text (edited for easier reading and comprehension):
TL;DR
Kalshi chose an unconventional path of regulation first, then growth: spending three years obtaining a license and suing the CFTC to open up the election market. His core judgment was that the legality of the prediction market was more important than growth.
The essence of market prediction is to incentivize truthful information with money: compared to polls and social media, the market filters information through profit and loss mechanisms and is regarded as a signaling system that is closer to the truth.
Ordinary people, rather than institutions, constitute the core liquidity of the market: more than 95% of the matching comes from decentralized users and super forecasters, rather than traditional market makers.
Kalshi emphasizes that it is an exchange, not an online entertainment platform: revenue comes from transaction fees rather than user losses, and it encourages skilled players to participate, rather than restricting winners like the online entertainment industry.
Elections are the Holy Grail scenario, but the future market is far more than that: from sports and macroeconomics to AI, computing power and other variables, the team hopes to build a derivative system where everything can be priced.
Prediction markets are becoming a new information infrastructure: users are not only trading, but also consuming probabilities; 80% of users primarily use them to judge the world rather than place bets.
Behind its rise lies a distrust of traditional information systems: polarized social media and inaccurate polls are driving people to turn to price-based judgment mechanisms.
• Core long-term goal: To improve the efficiency of social decision-making, not just to be a trading platform. Through continuous pricing and feedback, to enable genuine consensus to be formed more quickly in political, economic, and other fields.
Interview transcript
John Collison (Stripe co-founder & interview host):
Tarek Mansour and Luana Lopes Lara are the co-founders of Kalshi, an emerging prediction market company that gained rapid popularity during the November 2024 US presidential election. They spent four years negotiating with regulators and seeking approval to establish the first homegrown, compliant prediction market in the US before its official launch. Today, Kalshi's monthly trading volume exceeds $10 billion.
So how do you two usually divide the work? But more than the division of labor, I'm curious about the differences in your ways of looking at problems.
Luana (Co-founder & COO of Kalshi):
Actually, our backgrounds are almost identical. We both studied math and computer science at MIT, and our internship experiences were similar, basically the same. But I'm a very optimistic person, I like taking risks, and I always feel that things will eventually work out; he, on the other hand, is very cautious, even a bit pessimistic. So I think this creates a very good balance. Looking back, besides our daily division of labor, this is actually what truly complements each other.
Tarek (Kalshi Co-founder & CEO):
Let me add a little background. I originally planned to become a trader; that was practically my chosen career path. If you've ever met people like that, you might understand that they always seem to have an expected return calculator in their heads.
Matt Huang (Co-founder of Paradigm):
A very typical trader.
John Collison:
Yes, but—
Tarek:
If you're the kind of trader who thinks about tail risks and worst-case scenarios all the time, you'll be thinking about them constantly. She doesn't usually think that way. I think it's precisely this difference that leads to good results.
Compliance First, Growth Later: Why Kalshi Chose the Most Difficult Path
John Collison:
That's exactly what I was going to ask. Your starting point is interesting. After Kalshi was founded, it couldn't really operate for several years until it got approval from the CFTC. Most companies don't start that way. On the other hand, there's a very common, though often criticized, but indeed prevalent model in Silicon Valley: the "do it first, then figure it out" approach used by PayPal and Uber in their early days—get things started first, then fill in the gaps in the structure and licenses. It's about doing it first and then seeking understanding, rather than requesting permission first.
So could you tell me how it all started? What was the entire approval process like? And I'd also like to discuss whether this approach is applicable to other companies.
Luana:
I think we were very clear from the beginning that if you're in financial services or healthcare, you can't just do it and worry about the consequences later. In the financial sector, once user funds are involved, the cost of problems is extremely high; FTX is a prime example. Healthcare is even more disastrous, with too many disastrous precedents. We wanted to do this the right way. More importantly, when we looked at the market, the core issue wasn't whether it would grow, but whether it was legal to do this in the US. So we decided to address this biggest problem head-on before moving forward. For a long time, many people thought this was the wrong strategy.
I think that before we won that lawsuit about the election contract, there was a lot of talk about how those who went to offshore markets were doing better and growing faster. But things really started to take off after we won that case, proving that our understanding of the law was correct and that the company could legally operate in the U.S. as we envisioned.
John Collison:
What is the timeline here? When did you start? And when did you win the lawsuit regarding the election contract?
Luana:
We founded the company in 2019 and joined Y Combinator that same year. It took us three years to finally get regulatory approval and launch, around 2022. Then, in late 2024, we won the election contract lawsuit, and from then on, the company really started to accelerate.
Tarek:
This issue actually has two aspects. First, there's a very practical consideration. We believe that if we want to achieve genuine mainstream and institutional adoption, the core issue we can't avoid is whether this can operate within a regulated, trustworthy, and secure framework. After all, this is a complex market involving the flow of user funds. We must solve this most difficult problem first; that's the path to success.
The second level is more principle-oriented. One thing that initially excited us was when we were writing that one-page document in Google Docs, we listed a series of questions: Why are we starting this company? Why is this so exciting for us? Our answer was that we wanted to build the next generation of the New York Stock Exchange. We wanted to build a credible, regulated financial market in the United States. We weren't really excited about building something similar offshore. The key question is, what kind of company do you want to build? Why are you doing this? There are many paths to success, but that other path wasn't what we truly wanted to take. We wanted this to happen here, in the United States.
John Collison:
You are the first prediction market to be approved by the CFTC and reach a certain scale.
Tarek:
Yes, that's right.
John Collison:
And even today, each of your contracts still needs to be approved separately, right?
Luana:
Yes. We submit every contract to the CFTC, and they have 24 hours to stop it.
John Collison:
In other words, they receive your contract information stream almost in real time?
Luana:
Yes, that's one way to understand it.
Tarek:
Yes. Reaching the state of today's contract processing network has been a very long process. Imagine when we first walked into the CFTC building, this was the concept in our minds, and the regulators had to work at breakneck speed. Because you were talking to them about a product without the backing of traditional financial underlying assets, and dealing with the possibility of dozens, even hundreds, of contracts every week. Of course, we've done much more now, but initially, this regulatory model wasn't designed for this kind of scenario at all.
So this process is actually quite similar to iterating on a product, except you're not building a product for customers, but rather exploring with regulatory agencies how to regulate this kind of thing, what their concerns are, and what we can do to address those concerns.
Luana:
In a sense, this is about finding a product-market fit at the regulatory level.
Matt Huang:
So you're pretty much used to this pace now. Send out the contracts first, unless they explicitly object. Have they vetoed anything else recently?
Luana:
Not recently. The biggest veto we faced was actually regarding the election contract, which ultimately forced us to sue them. They rejected us on that issue for two years. But by now, we've worked with them for so long that we both understand the boundaries, and they trust us, knowing that as a self-regulating entity, we understand what we can and cannot do. For example, we don't engage in markets like war or assassination. As long as we stay within the established boundaries, the entire process is much faster.
John Collison:
So let me confirm, the core of that election lawsuit was that they're generally willing to approve all sorts of contracts, but they're unwilling to approve contracts about who will win the election—and these are precisely the most popular types, especially during US presidential elections. So you sued the CFTC.
Tarek:
Yes. Actually, it's their own rules—
John Collison:
Generally speaking, suing one's own regulatory body is not considered best practice.
Tarek:
Indeed. Here's what happened: We started pushing into the election market at the end of 2021, communicating with policymakers, Congress, and regulatory agencies. Everyone said it sounded good. But then they consistently stalled, and we started to feel something was wrong. By the end of 2022, they had effectively delayed approval until after the election, essentially a veto. That period was incredibly difficult for the company; we had to lay off many people. Even more challenging was that the team, investors, and even most investors began to lose faith in this approach.
John Collison:
It's not that I don't believe in the idea itself, but rather that I don't believe in the strategy.
Tarek:
Yes, we stopped believing in this strategy and even started to doubt the idea itself. People felt things were getting a bit unhealthy; shouldn't we try something else? Clearly, this path didn't seem viable. But we just couldn't force ourselves to do anything else; we simply couldn't. So we said, okay, let's try one more time.
You can imagine that team morale had plummeted, and everyone was waiting for a new strategy. Many people left, many were laid off, because we had to downsize. Then at the next stand-up meeting, we told everyone that the strategy for 2023 was—we'll try again.
John Collison:
In other words, we will continue to do the same thing, only this time it will succeed.
Tarek:
Yes, that's exactly what I meant, this time it will work. Even though almost all the evidence points in the opposite direction. I have to say, a large part of this was really her doing. Of course, I really wanted this to succeed too, but my rational mind kept telling me, you should listen to these people, this path won't work. But she was more determined. So we tried again. By the end of 2023, they stopped it again. By then I almost thought—
John Collison:
Well, predicting the market is just not going to work.
Tarek:
Yes, that's exactly how I felt. Then she said that, of all the possible options, the only thing left to do was sue the government. My first reaction was: This is insane. We took this to the board to discuss it; I remember Alfred, Michael, and Seibel from my side were all there.
John Collison:
That is, Alfred Lin and Michael Seibel.
Tarek:
Yes. I remember those board discussions, and it always started with, "We have to tell you clearly, this idea is terrible." There were many reasons: your opponent is the regulatory body; you only have about twenty people; if the government really wanted to take you down, there are countless ways—shutting you down, revoking your license, they can do it all. And this isn't just a theoretical risk. Even if you win, you might already be dragged down in the process.
I remember we had an internal meeting before we officially spoke with the board. It was the night before we'd contacted lawyers and were preparing to initiate litigation, and I suddenly backed out. I said, "Why don't we just become a liquidation firm, or focus more on financial products? Don't put everything on this, don't go all in." I don't remember the exact words from that phone call, but the gist was, "Are you kidding me?"
Luana:
That does sound like something I would say.
Tarek:
I realized then, okay, I can't win this argument. But another part of me knew we had to do it this way. Later we talked to the board, and their response was basically, this is obviously counterintuitive, a bad idea. But many great companies are built on some kind of counterintuitive pattern, and something unusual will always happen, maybe that's your unusual one.
John Collison:
That's a good point. Every company eventually emerges in some new, unusual way, and perhaps that's your way. So, what was the legal basis for your subsequent victory in the election lawsuit? Was there any particularly interesting policy angle?
Luana:
The core issue is actually quite simple: a government cannot arbitrarily ban a contract unless it determines that the contract violates the public interest, and this ban must fall into a specific category, such as war, terrorism, or assassination. The CFTC's position at the time was to try to force elections into these categories. They would argue that elections might be illegal under certain state laws, even citing state laws regarding bucket shops, trying to find any reason to block it.
But we are very clear about the law: elections have economic impact, and if they do, they should be tradable on futures or derivatives exchanges. That lawsuit ultimately told the CFTC that they can't do whatever they want.
John Collison:
In other words, a prohibited category must actually belong to one of those categories that are explicitly prohibited, and elections clearly do not fall into that category.
Luana:
That's right, exactly.
Tarek:
This is very important. We always say that the law constrains businesses, but the law also constrains governments.
John Collison:
Yes. Matt, were you referring to that point about suing the government over the past two or three years?
Matt Huang:
Yes. I initially thought suing governments seemed unusual in the crypto and prediction markets space, but I've found it's actually far more common than traditional Silicon Valley perceptions suggest. Coinbase has sued its main regulatory body; in the GovTech space, SpaceX, Anduril, and Palantir have all sued governments for various reasons. So I'm curious, given your extensive experience dealing with governments, what advice would you give to those looking to start a similar business? Under what circumstances do you feel it's the right thing to do to launch such a challenge?
Tarek:
I think it should only be done when there is no other choice. It is still very painful.
John Collison:
But do you really have no other choice? Can't you survive without the election market? Sure, elections are a very attractive and lucrative category, but I guess it's not the source of most of your contracts today.
Luana:
I think it's just too important. It might sound a bit obsessive, but it truly is the holy grail market. It best reveals the purpose of data in such markets and best demonstrates their value. Take the 2024 election, for example; the polls were wildly inaccurate, while the market clearly did a better job of integrating information. I think it's the brightest example of why prediction markets are beneficial and why the US needs them within a regulated framework. No other market offers such a strong demonstration.
The core logic of market prediction: using real money to produce information.
Matt Huang:
John just mentioned the "do it first, worry about it later" logic of PayPal and Uber. In fact, other prediction markets were already operating offshore at the time, demonstrating real demand. So I'm curious, did this help you in your lawsuit? For example, did it help demonstrate that election markets don't violate the public interest—after all, people were already doing them.
Tarek:
I'm not sure. But from a court perspective, the focus is actually more solid on the legal texts themselves. We're discussing the Commodity Exchange Act, one of the core laws governing financial regulation; the other corresponding law is the Securities Exchange Act. The key is to read and interpret these laws line by line, and then determine whether the regulatory agency overstepped its authority.
From our own perspective, the existence of offshore markets has indeed been helpful, as it allowed us to refer to external data even while following a "regulation first, product later" approach. At the time, we couldn't directly learn from our own products because we insisted on obtaining licenses before taking action. So, in a sense, external data and evidence did help us make decisions. It also helped more people understand what prediction markets are and how to use them. But in terms of policy, have offshore players been very helpful to us? I think not necessarily.
John Collison:
If Kalshi had appeared ten or fifteen years earlier, would it have never taken off? Is it because the CFTC is more open now, or would it require some kind of technological maturity, such as stablecoins?
Luana:
I do think there was a factor involved in crypto. Early prediction markets like Augur were already emerging. I think the existence of these things definitely made the CFTC realize that we needed a legitimate, regulated alternative. Previously, they could have simply said no. I think that did have some effect, but probably only 5% to 10%, no more.
Tarek:
On a broader scale, I think people's intellectual interest in prediction markets has always existed, dating back to the 1950s. Everyone has long known that it's a better source of information than many other mechanisms. But ten or fifteen years ago, the real-world pain points weren't as pronounced; in the last few years, those pain points have become very real. I think countries are more divided, and the world is more divided. Social media has fragmented the information flow into different camps, clickbait headlines are rampant, and much of what we read today—whether traditional news, social media, or elsewhere—is increasingly driven by incentives that prioritize grabbing attention. It is precisely because of this that more pressing problems have emerged, and these problems have spurred this wave of adoption of prediction markets. I don't think the situation would have been like this fifteen years ago, because the problems weren't this severe then.
Luana:
Moreover, most of our users aren't actually here to make transactions. About 80% of them are primarily consuming information. They'll come to see things like who will win yesterday's Texas primary; a quick glance tells them that polls say it's a close race, but that's not actually the case. This function as an information carrier is far more important now than before.
John Collison:
So you mean that in the era of algorithmic information flow, markets like Kalshi are a perfect fit; if it were ten or fifteen years ago, people might not have been so interested.
Tarek:
Yes. I think a more accurate statement is that people's distrust of traditional sources of information is rising significantly and persistently. So you need a new source, and this mechanism does work. The incentive mechanism of prediction markets points to the truth; more trading volume, more liquidity, and ultimately, more accurate predictions. This process takes time; people need several rounds of verification before they start to believe it. But once it establishes a track record, people will no longer be willing to use a clearly inferior product.
John Collison:
Speaking of trading volume, could you give us a breakdown of Kalshi's growth? It seems to be growing exceptionally fast.
Tarek:
The transaction volume in February this year was $10.4 billion.
John Collison:
That is, the contract transaction amount is US$10.4 billion.
Tarek:
Yes. Compared to six months ago, it has increased by about 11 times, maybe even more.
John Collison:
The growth is so fast that you're too lazy to look back a year, because that feels like ancient times.
Luana:
A year ago, it was truly like ancient times. For example, a year ago we only had one sports market.
Tarek:
Yes, it was in February. In short, the growth was indeed very rapid.
Matt Huang:
Aside from AI, it is probably the fastest growing company.
Tarek:
I think so. It might even be comparable to some of the top AI companies. I don't know the latest data from Cursor or Anthropic, but—
John Collison:
Even within the realm of AI, an 11-fold increase over six months is already quite remarkable.
Tarek:
It's indeed very fast. I think the reason is that we are a true market, possessing the inherent attributes of a market, such as network effects. As the market categories increase and liquidity deepens, user retention improves, and user engagement and transaction volume continue to grow over time. This will certainly lead to their own usage growth, but it will also drive usage growth for others, because there is more liquidity in the system and the products are easier to use. Then, users will be more willing to share with others. So, these forces combined have driven the current growth.
Matt Huang:
A large portion of your early growth actually came from other brokerage platforms. That structure has changed today. What are your thoughts on the brokerage segment? What is its current proportion?
John Collison:
What does "broker" refer to here? Like Robinhood?
Tarek:
Yes. That's an interesting question.
Luana:
I can explain the broker part first, and he can provide the specific figures. Simply put, because we are essentially an exchange and clearinghouse, our role is more like the New York Stock Exchange, or more accurately, the Chicago Mercantile Exchange. That is, brokers can connect to us. You can trade stocks on Robinhood, and you can also trade Kalshi contracts on Robinhood; the same will be true on Coinbase and other platforms in the future.
From the very beginning, we were very clear that we are first and foremost an exchange and clearinghouse, and nothing more. Connecting with institutions like Goldman Sachs and Robinhood is also crucial for our understanding of the entire ecosystem.
At the beginning of last year, our first brokerage partners were Robinhood and Webull. During that initial period of rapid growth, brokerage channels accounted for a very high percentage of our business, which was actually a good thing because brokerages bring a lot of demand. When demand comes, market makers are willing to enter the market because they want to compete with retail investor traffic. In this way, we also bought ourselves time to gradually refine our direct-to-user product to what it is today.
Our current understanding is that the core will always be the exchange + clearinghouse. Users can access it directly through our app, website, and API, or through any brokerage firm. We are also increasing our investment in institutional and international brokerage firms. In the future, even if you are in Brazil, you can trade Kalshi directly; these things are coming soon. As for the numbers, you can provide them.
Tarek:
She declined to provide specific figures, but our so-called "direct" business—the part of our business that directly faces consumers through kalshi.com and the Kalshi App—has significantly outpaced the growth of intermediary channels, i.e., broker channels. I think this is mainly because the brand has expanded beyond its core audience. Now, when many people have a different opinion on something, their first reaction is to check the odds on Kalshi or place a bet on Kalshi. The brand has become synonymous with this behavior itself. There's already a lot of organic growth, and I think this trend will continue in the coming months.
A counterintuitive market: Ordinary people are more important than institutions
John Collison:
You just talked about the retail side, that is, how individual users grow. Some come through brokerages like Robinhood, and others come directly to the Kalshi website. But as an exchange, you have another issue that you must address: market making. The New York Stock Exchange doesn't have to worry too much about market making because the economic incentives are already strong enough; once the scale is large enough, it's no longer a major problem. But I'm curious about how you did it in the early days. Did you make the market yourself? Did you partner with external market makers? How did you incentivize them to participate? How did you build your market-making system from scratch?
Luana:
The markets on Kalshi can actually be divided into two categories, and their behavior is very different, so the market-making incentives are also completely different.
One type is the long-tail market, such as the market surrounding potential restructuring of One Direction. These are actually quite difficult to price, and demand is usually low, so we really need to attract market makers through various incentives, including recruitment incentives. One of our long-term biggest concerns is: how to establish stable and sustainable liquidity for these long-tail markets? We might have 10,000 markets now, but what if we have 50,000 or 100,000 markets in the future? How can we still guarantee liquidity?
Another type is the more classic market, such as crypto and sports betting. Market making is much easier in these markets because the demand is clear and the pricing logic is more mature. The incentive for market making here isn't direct cash, but rather a partial refund of transaction fees. However, we set very strict obligations, such as maintaining a certain spread or order book depth for a certain percentage of the time. This is because in these markets, we're more incentivizing the stability of the order book than simply incentivizing their presence.
John Collison:
What exactly do you mean by incentivizing order book stability?
Luana:
For example, live-streamed matches, or trading in crypto markets where transactions are settled hourly—
John Collison:
If no new information is coming in, you don't want prices to jump around wildly for no reason, right?
Luana:
That's right. Even if there's new information, like someone about to score a touchdown, you don't want the entire order book to instantly lose all liquidity. You can certainly allow spreads to widen appropriately, but you still want users to be able to trade. Especially after we started moving towards a brokerage model, brokers approached us with their expectations from the traditional market. They would say, "We want spreads and depth to remain at a certain level at all times." So we had to talk to market makers about how to design incentives in this situation. Because if we simply let the market decide, it might naturally widen a lot during periods of high volatility; but to serve all users, including brokerage channels, we had to do more design work on incentives.
Matt Huang:
In those periods when spreads would normally widen dramatically, would market makers lose money? Are they using profits from other, more stable periods to offset these periods?
Luana:
Currently, because overall demand is already very strong, even with narrower spreads, they can still profit from the spreads. But this is precisely the significance of our incentive program; you need to consider the benefits of the entire program together. You might lose a little at times, but as long as the overall returns are high enough, it's worth it.
Matt Huang:
Therefore, your goal is to consistently maintain relatively tight spreads in major markets.
John Collison:
In other words, major markets need to maintain tight spreads at all times. This actually requires careful planning.
Tarek:
Yes, it's difficult. But what's even more interesting is that the truly unique aspect of prediction markets is that a large amount of liquidity doesn't actually come from market makers as you usually understand it, but from ordinary people.
This brings us back to the initial logic. We've solved the regulatory issues, but then there's the liquidity problem. Traditional exchanges like the New York Stock Exchange and CME might say, "We're going to launch a grain futures contract," spend two years designing the product, gather fifty familiar market makers, prepare together months in advance, and spend several years promoting it. That's the traditional approach. Predictable markets are completely different because you have to constantly generate liquidity for new events, on a weekly, daily, or even hourly basis. How do you do that? Its rhythm is highly dynamic; there's always something new emerging.
John Collison:
Many people find this counterintuitive—that you need to incentivize market makers to provide liquidity? Because in the stock market, you don't need to incentivize high-frequency trading firms; they'll invest heavily in building low-latency links between New York and Chicago, vying to do it. So is this because prediction markets are still in their early stages, or is there some more fundamental difference?
Tarek:
This brings us back to the main point I made earlier. You're now dealing with a model that requires instantaneous liquidity creation, much faster and more dynamic than traditional markets. Traditional Wall Street market makers don't operate this way. You can't expect them to set up a desk within an hour to price political or cultural topics.
What's really interesting is that market prediction has a counterintuitive characteristic: in many markets, the people who are best at pricing it are not necessarily the so-called experts or authorities, but rather ordinary people.
Matt Huang:
An anonymous internet user.
Tarek:
Yes, that's right, those super forecasters. This ability is highly decentralized. It's difficult to say that a specific demographic group is the best at pricing. The reason we've gotten this far has taken so long is precisely because we've finally cultivated an entire community, a group of super forecasters on Kalshi who can quickly and efficiently price these things. Initially, it was difficult to get them to turn their hobby into a part-time job, and then from part-time to full-time. But as the market grew, it finally happened.
Luana:
Here's some data we can share: on the platform, the largest institutional market makers, traditionally considered large institutions, account for less than 5% of all traded market-making orders.
Tarek:
In other words, they only account for a very small portion of the matched liquidity.
John Collison:
Really?
Luana:
In other words, less than 5% of all executed orders came from well-known large institutional market makers. To put it another way, over 95% were peer-to-peer orders, or orders placed by small funds or teams with only two people.
Tarek:
This is very rare in exchanges.
Matt Huang:
How many of these small, full-time market-making teams are there?
Luana:
There are approximately 2,000 market makers on Kalshi.
John Collison:
Matt was actually asking, who exactly are the market makers on Kalshi? Are they institutions like Jane Street or Akuna, or are they just some guy drinking Red Bull and coding in his garage at 3 a.m.?
Tarek:
The kind of people in the garage are actually the most important.
Matt Huang:
And you just said that they account for 95%.
Tarek:
Yes. They are crucial to the entire system because they price quickly, constantly monitor the market, and continuously assess the situation. They are the true, original observers of the situation.
John Collison:
Therefore, Kalshi is built on a group of people who are constantly monitoring the situation.
Tarek:
That's right. Let me give you an example. Over the past few years, the most accurate inflation forecaster on Kalshi wasn't one of the institutions or well-known hedge funds, but a person living in Kansas. He had never traded in financial markets before; he just enjoyed reading the news, had a feel for inflation, and felt he could sense where it was headed. And you'll find that there are many people like this on the platform. There might be a few thousand people officially participating full-time, but broadly speaking, there are tens of thousands of such people. They know a lot about all sorts of topics, actively price these things, and are paid for this ability.
Luana:
Let me tell you about my favorite user.
Tarek:
By the way, I recently got a new favorite user.
John Collison:
Okay, then each of you should talk about your favorite user.
Tarek:
I was thinking about this this morning. Actually, the person in that Wall Street Journal article about taxes—
Luana:
Oh yes, he is indeed a very strong candidate. But my favorite user is a huge fan of Ariana Grande. He discovered Kalshi during election season, but he doesn't actually like elections and has absolutely no interest in them. Later, he discovered our Billboard chart market, you know, the kind of chart market.
John Collison:
This is a very important market in his view.
Luana:
This was very important to him. He had already earned over $150,000, which he used to pay off his student loans, complete his master's degree, and buy a car. He had never traded or done anything like this before, but he had an extremely strong, almost obsessive, interest in music charts. For the first time, he was finally able to monetize this hobby. And he was also very friendly to us on Twitter.
Tarek:
I actually like many things, but this one is currently ranked very high. Last week, the Wall Street Journal published an article about a tax accountant named Alan who is very active on Kalshi. When DOGE first came out, everyone was discussing how much it could actually cut costs. He went and studied a lot of tax laws and regulations in great depth, finally concluding that it was impossible to achieve the expected goals. He reached this judgment almost with certainty. Then he went back and told his wife, "I have extremely high confidence in this deal." In a way, he's a bit like Michael Burry in "The Big Short," except this time he's short DOGE. And he really did bet heavily, and ultimately won.
That's the power of market prediction. If you truly possess that kind of knowledge—and that knowledge is often quite niche—like, I bet none of us here have read through all those tax laws—then you can actually do research, gain a deeper understanding of the world, and reap the rewards. That's fantastic.
John Collison:
One important early application of AI was poker bots. Have you seen any truly powerful AI market makers now? After all, the idea that nobody reads through all the tax laws is probably no longer valid, except perhaps for Claude.
Luana:
That's true. Perhaps we should ask it.
Tarek:
We are indeed seeing more and more people using agents for trading, especially on the API side, which is quite evident.
John Collison:
Do you have any users who are already successfully operating highly agency-driven market-making businesses? That is, where most of the process is driven by agents?
Tarek:
Users typically don't tell us the details of their strategies.
John Collison:
But do you chat with them?
Tarek:
Yes, John, the answer is yes. My understanding is that Renaissance Technologies, in its early days, wasn't already using some kind of proxy-based model for trading? Of course, that was a very early version. I think today it's just continuously evolving, and it will become increasingly powerful. Many traders on our platform already have AI-based summarization and comprehensive judgment modules in their systems.
John Collison:
What I'm really curious about is a completely autonomous system, without human intervention, that takes in information and sets prices accordingly. It feels like that's coming soon, if not already. For example, your Claude is already a market maker.
Luana:
I'm not sure if completely unmanned systems exist now. For example, in scenarios like international elections, I know many systems already automatically translate documents, conduct polls, and perform various other processing tasks. But I don't know if they're fully automated yet.
Tarek:
We don't actually know yet if our model has reached that stage. We recently launched Kalshi Research, and one of our goals is to collaborate with research labs to create a brand new benchmark to see which models are better at predicting the future. This benchmark is unique because it doesn't test memory patterns, but rather the model's understanding of the world itself. I'm really looking forward to the results.
John Collison:
So how do you plan to conduct the assessment?
Tarek:
It's not completely finalized yet. But the general idea is to run the models on the same market for one or two consecutive months, and then see which performs better, for example, in terms of prediction accuracy, long-term PnL, etc.
The fundamental difference between online and online entertainment: trading vs. betting.
John Collison:
Another issue regarding market making. There's a well-known phenomenon in online sports betting: online betting companies crack down on so-called "sharps"—those who are too smart and adept at betting. Many people don't realize that, for online betting companies, the ideal bettor is actually someone who isn't particularly professional and simply supports their own team; while the worst are those who are particularly adept at finding mispricing in niche markets. Because online betting companies might offer odds for ten thousand markets, if they're wrong once or twice, professional players will specifically target those errors. Therefore, they will identify you through behavioral signals. If you've just registered and only bet on your own team, that's fine; if you act too professionally, they will ban you.
This is actually quite interesting. You think you're just betting according to the odds they offer, but if you're too clever, they don't want you. It's a bit like a Las Vegas casino asking you to leave. Could Kalshi have the same problem with Sharp? I initially thought not, because you should welcome them. But wouldn't market makers worry about having too clever opponents?
Tarek:
The problem is that Sharp itself is part of the market. Let me clarify one point first—
Luana:
We won't limit the winners. We won't do that. We want the smartest people to come.
Tarek:
We need these sharps. Without them, how can the market be more accurate? This is the biggest difference between us and online entertainment companies.
John Collison:
That's true, but it's not entirely the same. What you need is liquidity—maintaining a narrow spread throughout the entire match and election process. Sharp, on the other hand, might just jump in when the odds are wrong, make a huge profit, and then disappear. So, providing liquidity and making accurate predictions are ultimately two different things.
Tarek:
But many Sharps could actually earn more by providing liquidity. They could absolutely become part of the liquidity ecosystem. This is crucial. Many people will say, "I'm not gambling, I'm trading." While this may sometimes sound like self-comfort, I think it does highlight a fundamental difference. The business model of gambling is that the house profits from the players' losses; your revenue comes from the users' losses. So the behaviors you just described are perfectly reasonable for online entertainment companies. If someone wins, I have to stop them because that's a direct loss on my profit and loss statement; if someone loses, I have to find a way to bring them back.
This is completely different from traditional financial markets. In financial markets, the core of institutional design is fairness and transparency. You need to create a set of rules that are fair to all participants. Maybe Matt is better than Luana, maybe Luana is better than Matt; they can compete and determine the winner themselves.
John Collison:
Therefore, your incentive systems are completely different. Unlike casinos that profit from one side losing money, you earn transaction fees.
Luana:
That's right. For us, the best outcome is that users feel the market is fair, the prices are good, and the liquidity is stable, so they are willing to trade here. Of course, to achieve this, we also need to design different incentives for different roles. That's why we have various liquidity programs. If you are providing liquidity and you take on a higher risk of being targeted, then your fees should be lower; if you are actively taking orders or targeting, then your fees should be higher because you have to pay for that behavior.
John Collison:
In other words, you use transaction fees to incentivize prosocial behavior.
Luana:
Yes, I think that's how financial markets are supposed to operate.
Tarek:
The traditional financial market operates on the same logic.
Luana:
It's just that it wasn't said so bluntly.
Tarek:
Essentially, it's about giving a slight advantage to those who truly create value for the market, while reducing the advantage for those who merely extract value.
John Collison:
What behaviors are considered prosocial, and what behaviors are considered antisocial?
Tarek:
Insider trading is, of course, an antisocial act.
Luana:
This is the most typical example.
Tarek:
And it's illegal. As for sniping, it's actually part of the market. Someone suddenly gets new information and uses it to trade; this happens every day in traditional financial markets. It's just that if you want liquidity to persist and for these market makers to be willing to invest resources, you must give them certain incentives.
I think this is one of the reasons why market prediction is becoming increasingly accepted. People like this mechanism—your strengths are directly proportional to the depth of your research, the extent of your information, and the time and effort you invest. Traditional financial markets are similar, but for many, those markets aren't as interesting. Compared to analyzing IBM's quarterly financial reports, I find studying the DOGE, elections, how people view an election, and why they vote in a particular way much more interesting.
John Collison:
Kalshi is building a new type of market where real-world outcomes can be traded. For example, whether the US will confirm the existence of extraterrestrial life by 2027. Thousands of participants can open positions, transfer funds, and settle accounts in real time, while the underlying system is actually a very complex multi-party fund flow system. This fund orchestration behind Kalshi is driven by Stripe Connect, which handles participant account registration, payment processing, fund routing, and withdrawal management. When fund flows themselves become programmable, new products and even new market structures can emerge. Therefore, if you are building a completely new product with complex fund flows, Stripe Connect is for you.
Tarek:
Sorry, everyone.
Can everything be priced? Predicting the future boundaries of the market.
Matt Huang:
Let's talk about different market verticals. Everyone understands markets like elections, sports, and economic indicators. But I think predictive markets are like a search function, operating on all the interesting markets that people want to trade. In the past, traditional exchanges like the CME decided what commodities to list, usually things like wheat, oil, and corn. But now you can list a thousand markets a day. So, with this happening, what do you think we'll ultimately discover?
Luana:
There's one direction we're particularly excited about, and we've already started moving in that direction: watches, bags, and other collectible items. In fact, creating derivative products around these items is entirely feasible.
Another area worth mentioning is GPU computing power. This is also a very interesting direction. We are increasingly thinking that many assets are better suited to traditional futures structures rather than binary yes/no markets. That is, it's not a yes/no question about whether the price will reach a certain point, but more like true futures, where you can have margin and more institutional liquidity. This direction is important because once we move out of pure binary markets and into more traditional derivatives structures, we are essentially expanding the trading instruments from grains to computing power.
Matt Huang:
Historically, the most successful futures markets have always been for commodities, because those are inherently huge spending categories. And we are currently in an era where humanity is spending the most money on a new commodity.
John Collison:
A new category of goods.
Matt Huang:
Yes, a new product. The traditional market hasn't really been aggressively targeting the computing power category.
Luana:
Our perspective on this is that we want to become the world's largest derivatives exchange. To achieve this, there are four key things in our product roadmap.
First, there's the breadth of market themes. That means we need to cover computing power, sports, elections, securities, and so on.
Second, there's the market structure. Currently, we only have binary yes/no products, but in the future, we need futures, swaps, options, and so on.
Third is the margin system. The current system is actually quite poor; you have to deposit all your funds first.
John Collison:
In other words, too much capital is tied up.
Luana:
Yes. This will make many markets very difficult to operate in, such as those facing hurricanes this year. If you want to make market-making decisions or sell these contracts, the capital efficiency will be almost unacceptably low. Fourthly, there's the issue of liquidity.
Our internal assessment is that if we can win in these four areas—the broadest range of market themes, the richest market structure, an excellent and efficient margin system, and good liquidity—we'll win at everything we do. Therefore, everything in the company ultimately comes down to one of these four areas.
The so-called thematic expansion is essentially about matching the right themes to the right market structures and margin models, and then ensuring adequate liquidity. You're absolutely right. If we can rebuild these margin systems from scratch, we can launch new margin models and list more new markets faster in the future.
Matt Huang:
Since you have mobile products directly targeting retail investors, I'm curious whether you naturally tend to favor markets that retail investors find interesting, rather than more professional or institutional markets? For example, computing power, in my view, is a market more geared towards institutional investors. How do you consider liquidity and cultivating interest in these types of markets?
Luana:
To be honest, internally we almost divide the company into two parts. One part is about improving existing markets, such as sports, crypto, and other established ventures; the other part is about constantly pushing for the next new thing. Because I think what truly brought Kalshi to where it is today isn't regulation, or anything else, but our constant pushing for the next new thing. First elections, then sports. For us, continuously defining the next thing and doing it well is essential. If we stop doing this, the company won't win.
The company's current organizational structure is roughly the same: the market operations and engineering teams focus more on maintaining and optimizing existing systems; while newer teams, such as the institutional team, margin team, and international team, are more focused on pushing things forward. Above that is a platform layer responsible for core exchange, compliance, and other infrastructure. We've been constantly trying to find a balance between these different themes. But ultimately, we only have 120 people, and there's a lot to do, so it's indeed very challenging.
Matt Huang:
So now, institutions have started to drive your interest in certain themes?
Luana:
Yes, of course. For example, we just launched a feature called Block Trades a week ago. You know block trades, right? It's a very institutional trading method. I can call you directly, agree on a trade, and then send it to the exchange for matching and confirmation, instead of everything having to go through the open order book. We're currently working on many of these features, hoping to attract more institutions.
Matt Huang:
Are they trading the same thing as Kalshi retail clients? Or will you be offering new products?
Luana:
Yes and no. For example, if they are interested in a certain topic—like the previous interest in tariffs, whether or not tariffs will be increased; or the recent focus on the future of oil reserves—as long as we hear we want to trade on this market, we usually launch it directly and make it available to everyone. However, I do think there will be a significant difference between what institutions will ultimately trade and what ordinary retail investors are interested in. But at least for us, launching the market is now very low-cost.
John Collison:
When Uber first appeared, it wasn't actually bad for the taxi industry, as it initially mainly exploited idle supply and met incremental demand. However, as it scaled up, it began to truly impact the taxi industry. What about Kalshi and other prediction markets? Are there any existing industries that will increasingly feel your influence? Because once the market becomes larger and more liquid, it may replace some of their functions. For example, traditional futures exchanges might find it more convenient to hedge soybean prices through Kalshi in the future. Or online sports entertainment companies and political polling agencies—the latter might find that you can provide similar information more cheaply. So, who do you think will be the first to feel surpassed by prediction markets?
Luana:
You know that meme, right? The bodyguard at the door asks, "Who's next?" We don't really want to name names, but those directions you mentioned are definitely there. The traditional online entertainment industry is one, of course. We just discussed many of its structural problems, and we're completely different. The traditional futures market is another; we're increasingly entering their territory. Political polling organizations will also be affected. I think since the last election, many campaign teams have been using our data. And then there's parametric insurance. Once we have a more mature margin system, we can enter real-world risk pricing scenarios like hurricane and disaster insurance.
Matt Huang:
Could polls lead to a tragedy of the commons? After all, part of the accuracy of prediction markets lies in their interpretation of polls. In other words, polls are sensors, and prediction markets are the mathematical interpretations of those sensor outputs. So, if everyone stopped conducting polls, wouldn't that weaken the prediction markets themselves?
Luana:
My view is the opposite; polls will only get better. Because people will realize that if my polls are more accurate, I can make money. So they'll be willing to commission better polls and better methods. You can actually put many different polling models into the same market to compete.
John Collison:
Like the meta-pollometer interpreter that 538 made.
Luana:
Yes, even further. You eventually get a single number that brings all these things together. Actually, someone did this in the last election; he specifically commissioned a new polling method, similar to the Nearest Neighbors approach, though I don't remember the specifics, but the result was that he made more money in the market. That's the power of real money being there—it aligns incentives with the truth. Polls are no longer about telling me what I want to hear, but about giving real numbers.
So I think they are complementary, and the same goes for news. Many people say that predicting the market will destroy journalism, but I think it's more about complementarity. For example, when you're discussing an election, news commentators will still offer their opinions, but the market doesn't offer opinions; it offers a number. You still need commentators, but now they can say in their commentary, "This is the probability given by the market, and this is my personal judgment." I don't think opinions will disappear.
John Collison:
You mentioned insider trading earlier, and there's a very complex policy issue here: how should the boundaries of insider trading in prediction markets be drawn? Even in the stock market, this is complex enough. For example, everyone knows the SEC frequently enforces laws against illegal insider trading, but there are also legal cases. For instance, a hedge fund might use its proprietary satellite data of Walmart parking lots to predict financial reports—this is clearly information others don't have, yet it's legal. Similarly, the boundaries in prediction markets are very complex. For example, we generally agree that government officials shouldn't trade before military operations. But what if someone knew in advance how long the Bad Bunny halftime show would last before the Super Bowl? Some people do have that kind of information beforehand. So, how do you think the boundaries of insider trading should be drawn?
Luana:
Your statement is actually quite accurate. This is a very complex issue. The stock market is even more complex and larger in scale. Our current principle is to comply with federal law.
John Collison:
This includes both CFTC rules and other financial regulatory rules.
Luana:
Yes, both the CFTC and SEC have relevant frameworks. Simply put, if you've signed a non-disclosure agreement promising not to disclose certain types of data, then you can't use that information for trading. For example, if I work for the Bureau of Labor Statistics, my confidentiality obligations require me not to disclose inflation figures before the data is released, so of course I can't trade. But if you know the Super Bowl rehearsal is on Thursday, and you're just standing outside and hear Lady Gaga singing, then that's fine. Traditional hedge funds use the same logic for alternative data like Starbucks store traffic. The market's purpose is to encourage information to enter the market. We welcome information into the market, but only if it's not unfair. If you obtain it unfairly, then you shouldn't trade.
John Collison:
In other words, if you have an obligation to keep this type of information confidential, you cannot use it for trading.
Luana:
Yes. Actually, we've gone a bit further. For example, if you're a government official, you can't trade whether a bill will pass, even if I'm not sure if they've signed some kind of formal confidentiality agreement.
John Collison:
It's interesting because, as is widely known, members of Congress are now allowed to trade stocks.
Luana:
Yes. So we've actually gone a step further than the current system in this regard. We've been communicating with regulatory agencies because this is a new issue for both of us. We're a regulated market, so we have a complete monitoring department that watches every anomaly almost 24/7, trying to investigate everything. About two weeks ago, we just announced two insider trading cases. Because we're within a regulated framework, we can impose hefty fines on these individuals—more than five times their illegal gains—and permanently ban them, among other things.
John Collison:
What I find interesting is that you yourselves are already doing this. In the public stock market, the SEC has always been very proactive in enforcing insider trading laws. So what is the CFTC's stance on this issue?
Luana:
That's a good question. You can think of the whole mechanism as three layers. The first layer is our own monitoring and enforcement. The second layer is the CFTC's own monitoring and enforcement. The third layer, if it's more serious, goes to the Department of Justice. The reason people think SEC cases are more prominent is that in many cases, exchanges themselves will investigate, handle, fine, and freeze accounts first before proceeding to the next stage. Our situation is similar. Every transaction on Kalshi is synchronized to the CFTC; they can see everything, every case. We also submit these cases to the CFTC for review. As for whether they will take further action, we don't know, but at least the ball is now in their court.
John Collison:
So you will transfer the case to them, but first take the first layer of protection.
Luana:
Yes, that's it.
Matt Huang:
I'm also thinking about another scenario. In some markets, the future outcome is unknown, so by definition, insider trading is almost impossible. But in other markets, a single person can change the outcome—for example, whether a politician will mention a certain word in a speech, or whether a player will make a few shots. So what's your take on this spectrum?
John Collison:
For example, is the so-called mention market inherently a bad idea because it is too easily manipulated?
Matt Huang:
Or is it that such markets are inherently incapable of being established on a large scale?
John Collison:
Just like whether Brian Armstrong mentions a certain word in Coinbase's earnings call, the same applies to this type of market.
Luana:
I actually think the mention market is very good. Just think of the Federal Reserve. So many hedge funds sit there, watching which way the chairman tilts his head
