Why do 90% of prediction markets not survive until the end of 2026?

Author: Azuma , Odaily Planet Daily

In the past two days, there has been a lot of discussion on X about the formula Yes + No = 1 in prediction markets. This can be traced back to an article written by the expert DFarm (@DFarm_club) that deconstructs the Polymarket shared order book mechanism, which sparked an emotional resonance among the group regarding the power of mathematics. The original article is titled "A Comprehensive Explanation of Polymarket: Why YES + NO Must Equal 1? ", and I highly recommend reading it.

In the derivative discussion, many experts, including Blue Fox (@lanhubiji), mentioned that Yes + No = 1 is another extremely simple but powerful formula innovation after x * y = k, which is expected to unlock a trillion-dollar information flow trading market. I completely agree with this point, but at the same time, I also feel that some of the discussions are a bit too optimistic.

The key issue is the construction of liquidity. Many people may think that Yes + No = 1 solves the barrier to entry for ordinary people to participate in market making, so the liquidity of the prediction market will rise like that of AMM with x * y = k. However, the reality is far from that.

Predicting market making is inherently more difficult.

In a practical setting, the ability to enter the market and build liquidity is not just a matter of entry barriers, but also an economic question of profitability. Compared to the AMM market, which is based on the formula x * y = k, market making based on market prediction is actually far more difficult.

For example, in a classic AMM market that strictly follows the formula x * y = k (such as Uniswap V2), if I want to create a market around the ETH/USDC trading pair, I need to simultaneously deposit ETH and USDC into the liquidity pool in a specific ratio based on the real-time price relationship between the two assets. Then, when that price relationship fluctuates, the amount of ETH and USDC I can retrieve will increase or decrease accordingly (i.e., the impermanent loss we are familiar with), while simultaneously earning transaction fees. Of course, the industry has subsequently made many innovations around the basic formula x * y = k. For example, Uniswap V3 allows market makers to pursue a higher risk-reward ratio by accumulating liquidity within a specified price range, but the fundamental model remains unchanged.

In this market-making model, if transaction fees can cover impermanent loss within a certain timeframe (and it often takes longer to accumulate fees), it's profitable—as long as the price range isn't too aggressive, I can be quite lazy with market making, checking back only occasionally. However, in the prediction market, if you try to market with a similar attitude, the likely outcome is that you'll lose everything.

To illustrate this further with Polymarket, let's consider a basic binary market. For instance, if I want to make a market in a market where "YES is currently priced at $0.58," I can place a buy order for YES at $0.56 and a sell order for YES at $0.60. As DFarm explained in its article, this is essentially equivalent to placing a buy order for NO at $0.40 and a sell order for NO at $0.44—that is, providing order support at specific price points with a slightly wider range above and below the market price.

Now that the order is up and running, can I just sit back and ignore it? When I check it next time, I might see one of the following four scenarios:

Neither of the two pending orders were executed.

Both pending orders have been filled.

An order in one direction has been filled, and the market price remains within the original order range;

An order in one direction has been filled, but the market price has deviated further away from the remaining orders—for example, a buy order for YES was placed at 0.56, and a sell order at 0.6 is still available, but the market price has already fallen to 0.5.

So, under what circumstances can you make money? I can tell you that in low-frequency trials, different situations may lead to different profit and loss results, but if you operate in such a passive manner in a real-world environment for a long time, the final outcome will almost certainly be a loss. Why is this?

The reason for this is that the prediction market is not based on the liquidity pool market-making logic of AMM, but is closer to the order book market-making model of CEX. The two have completely different operating mechanisms, operational requirements, and risk-return structures.

In terms of operation mechanism, AMM market making involves investing funds into a liquidity pool to jointly make the market. The liquidity pool will distribute liquidity in different price ranges based on x * y = k and its variant formulas. Order book market making, on the other hand, requires placing buy and sell orders at specific price points. Liquidity support is only available if there are pending orders, and transactions must be achieved through order matching.

In terms of operational requirements, AMM market making only requires depositing two-sided tokens into the pool within a specific price range, and it will remain effective as long as the price does not fall outside the range; order book market making, on the other hand, requires proactive and continuous order management, constantly adjusting prices to respond to market changes.

In terms of risk and return, AMM market making mainly faces the risk of impermanent loss and earns fees from the liquidity pool; order book market making, on the other hand, faces inventory risk under one-sided market conditions, and its returns come from bid-ask spreads and platform subsidies.

Continuing with the assumptions in the previous example, given that my main risk as a market maker on Polymarket is inventory risk, and my profits primarily come from the bid-ask spread and platform subsidies (Polymarket provides liquidity subsidies for some market-close orders; see the official website for details), the potential profit and loss scenarios for the four situations are as follows:

In the first scenario, you can't profit from the bid-ask spread, but you can receive liquidity subsidies.

The second scenario involves individuals who have already profited from the bid-ask spread but will no longer receive liquidity subsidies.

The third scenario involves having already taken a YES or NO position, which becomes a directional holding (i.e., inventory risk), but in some cases, one can still receive certain liquidity subsidies.

The fourth scenario is that the position has also become a directional position, and the position has already incurred unrealized losses, while the liquidity subsidy is no longer available.

Two points need to be noted here. First, the second scenario often evolves from the third or fourth scenario because usually only one side's orders are executed first, so it has also temporarily become a directional position. However, the risk ultimately fails to materialize, and the market price moves in the opposite direction, triggering the other side's orders. Second, compared to the relatively limited market-making profits (the spread profit and subsidy scale are often fixed), the risk of a directional position is often unlimited (the upper limit is that all the YES or NO positions in hand will become zero).

In summary, if I want to make money consistently as a market maker, I need to try to capture profit opportunities and avoid inventory risks. Therefore, I must actively optimize my strategy to maintain the first scenario as much as possible, or quickly adjust the order range after an order is triggered on one side to make it more like the second scenario, and avoid being in the third or fourth scenario for a long time.

Doing this well in the long run is not easy. Market makers need to first understand the structural differences between different markets, comparing subsidy levels, volatility, settlement time, judgment rules, and so on. Then, they need to track and even predict market price changes more accurately and quickly based on external events and internal capital flows. Subsequently, they need to proactively adjust their orders in a timely manner according to changes, while also making advance designs and implementing management for inventory risks. This is clearly beyond the capabilities of ordinary users.

A wilder, more dynamic, and less ethical market

If that were all, it would seem fine. After all, the order book mechanism is not a new invention. On CEX and Perp DEX, the order book is still the mainstream market-making mechanism. Market makers active in these markets can migrate their strategies to prediction markets to continue to profit, while injecting liquidity into the latter. However, the reality is not that simple.

Let's think about this question together: what is the situation that market makers fear most? The answer is simple—one-sided market conditions, because one-sided market conditions often amplify inventory risk, thereby breaking through the balance of allocation and causing huge losses.

However, compared to the traditional cryptocurrency trading market, the prediction market is a wilder, more volatile, and less ethical place, where one-sided markets tend to appear more exaggeratedly, abruptly, and frequently.

The wilder aspect is that in the conventional cryptocurrency trading market, if the timeline is extended, mainstream assets will still exhibit a certain degree of fluctuation, with upward and downward trends often rotating in cycles. However, the trading instruments in the prediction market are essentially event contracts, each with a clear settlement time. Furthermore, the formula Yes + No = 1 determines that only one contract will ultimately have a value of $1, while the other options will become zero. This means that bets in the prediction market will eventually end in a one-sided market trend starting from a certain point in time. Therefore, market makers need to design and implement inventory risk management more rigorously.

The more dramatic meaning is that the fluctuations in the conventional trading market are determined by the continuous game between emotions and funds. Even if the fluctuations are dramatic, the price changes are continuous, which gives market makers room to adjust inventory, control price spreads, and dynamically hedge. However, the fluctuations in the prediction market are often driven by discrete real-world events, and the price changes are often dramatic. One second the price may be at 0.5, and the next second a real-world event can directly push it to 0.1 or 0.9. Moreover, it is often difficult to predict when and what event will cause the market to change drastically, leaving market makers with very little reaction time.

What's even more unethical is that the prediction market is rife with insiders who are close to or even sources of information. They don't gamble with their predictions; they come with a clear outcome to reap the rewards. Market makers are naturally at an informational disadvantage against these players, and the liquidity they provide becomes their cash-out channel. You might ask, don't market makers have insider information? That's a classic paradox. If I knew the inside information, why would I bother with the market? I could just bet on the direction and make more money.

It is precisely because of these characteristics that I have long agreed with the statement that "the design of prediction markets is not very friendly to market makers in terms of structure" and I strongly advise against ordinary users easily entering the market to make markets.

So, is market making unprofitable in prediction markets? Not at all. Buzzing founder Luke (@DeFiGuyLuke) once revealed that, based on market experience, a relatively conservative estimate is that Polymarket market makers can earn about 0.2% of the trading volume.

So to put it bluntly, this is not a way to make money easily. Only professional players who can accurately track market changes, adjust their order status in a timely manner, and effectively implement risk management can operate continuously over a long period and make money through real skills.

Predicting market trends may make it difficult for a diverse range of sectors to flourish.

The challenges of market making in predicting the market place higher demands on market makers' capabilities, and also pose a challenge to platforms in building liquidity.

The difficulty of market making means limited liquidity creation, which directly impacts the user's trading experience. To address this issue, leading platforms like Polymarket and Kalshi have opted to subsidize liquidity with substantial financial incentives to attract more market makers.

In November 2025, Nick Ruzicka, an analyst specializing in prediction markets, cited a research report from Delphi Digital stating that Polymarket had invested approximately $10 million in liquidity subsidies, at one point paying over $50,000 per day to attract liquidity. As its leading position and brand effect have been consolidated, Polymarket has significantly reduced its subsidy efforts, but on average, it still needs to subsidize $0.025 for every $100 of trading volume.

Kalshi also has a similar liquidity subsidy program, and has already spent at least $9 million on it. In addition, Kalshi also leveraged its compliance advantages in 2024 (Odaily note: Kalshi was the first prediction market platform to obtain a CFTC regulatory license; Polymarket also obtained a license in November 2025) to sign a market-making agreement with Susquehanna International Group (SIG), a top market-making service provider on Wall Street, which greatly improved the platform's liquidity.

Whether for use as a reserve fund or for compliance requirements, these are solid moats for leading platforms like Polymarket and Kalshi. Just a few months ago, Polymarket received a $2 billion investment from ICE, the parent company of the NYSE, at a valuation of $8 billion, and there are reports that it is planning another round of financing at a valuation of over $10 billion. On the other hand, Kalshi has also completed a $300 million financing round at a valuation of $5 billion. The two leading platforms now have ample reserves of funds.

Prediction markets have become a hot startup trend, with numerous new projects emerging. However, I'm not optimistic about their future. The reason is that the leading companies in the prediction market have a stronger dominance than many people imagine. Faced with the continuous subsidies from giants like Polymarket and Kalshi, and their partners from the compliant world, what can new projects offer to compete head-on? How much capital do they have to sustain their efforts? While some new projects may have powerful backers and generate substantial profits, clearly not every one of them does.

A few days ago, Haseeb Qureshi, the bald guy from Dragonfly, posted his predictions for 2026. He wrote , "The prediction market is developing rapidly, but 90% of prediction market products will be completely ignored and gradually disappear by the end of the year." I don't know his logic, but I agree that this is not an exaggeration.

Many people are looking forward to a diverse range of sectors in the prediction market and dream of profiting from past experiences. However, this is unlikely to happen. Instead of spreading your bets, it is better to focus on leading companies.