Why will 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 in prediction markets Yes + No = 1. The origin can be traced back to a detailed analysis article on Polymarket’s shared order book mechanism written by the influential DFarm (@DFarm_club), which sparked a collective emotional resonance around the power of mathematics — the original article link: “A comprehensive explanation of Polymarket: Why must YES + NO equal 1?” Highly recommended reading the original.

In the subsequent discussions, several industry leaders, including Blue Fox (@lanhubiji), mentioned that Yes + No = 1 is another minimalist yet powerful formula innovation following x * y = k, with the potential to unlock a trillion-scale information flow trading market. I fully agree with this, but at the same time, I think some discussions are overly optimistic.

The key issue lies in liquidity building. Many might think that Yes + No = 1 solves the entry barrier for ordinary people to become market makers, so prediction market liquidity would rise like AMMs based on x * y = k. But the reality is far from that.

Market making in prediction markets is inherently more difficult

In practical environments, whether one can enter and provide liquidity is not just about participation thresholds but also an economic question of profitability. Comparing to AMM markets based on x * y = k, the difficulty of market making in prediction markets is actually much higher.

For example, in a classic AMM like Uniswap V2, if I want to market-make around the ETH/USDC trading pair, I need to deposit ETH and USDC into the pool at specific ratios based on the real-time price relationship within the liquidity pool. When the price fluctuates, the amount of ETH and USDC I can withdraw will change accordingly (the familiar impermanent loss), but I can also earn trading fees. The industry has also innovated around the x * y = k formula, such as Uniswap V3, which allows market makers to concentrate liquidity within specific price ranges for higher yield-risk ratios, but the fundamental model remains unchanged.

In this mode, if trading fees over a certain period can cover impermanent loss (which often requires longer periods to accumulate fees), then it’s profitable — as long as the price range isn’t too aggressive, I can be relatively lazy about market making, checking in occasionally. But in prediction markets, if you adopt a similar attitude, the likely outcome is losing everything.

Take Polymarket as an example. Suppose I want to market-make a binary market where the real-time YES price is $0.58. I could place a buy order for YES at $0.56 and a sell order at $0.6 — as explained by DFarm, this is essentially placing buy and sell orders for NO at $0.4 and $0.44 respectively — supporting the market at slightly wider points around the current price.

Once the orders are placed, can I just sit back and ignore them? When I check later, I might see four situations:

• Both sides’ orders remain unfilled;

• Both sides’ orders are filled;

• One side’s order is filled, but the market price remains within the original order range;

• One side’s order is filled, but the market price has moved away from the remaining order, e.g., bought YES at $0.56, but the price has dropped to $0.5.

So, which scenarios are profitable? I can tell you that in low-frequency attempts, different situations may lead to different profit or loss outcomes. But operating this lazily over the long term in a real environment will almost certainly result in losses. Why?

The reason is that prediction markets are not like AMM liquidity pools; they are closer to order book market making models used by CEXs. Their operation mechanisms, operational requirements, and risk-reward structures are completely different.

Mechanically, AMM market making involves depositing funds into a liquidity pool to jointly provide liquidity, spreading it across different price ranges based on x * y = k and its variants. Order book market making requires placing buy and sell orders at specific points, with liquidity supported only through active order placement and matching.

Operationally, AMMs only need to deposit tokens within a certain price range to keep the market active as long as the price stays within that range. Order book market making requires active, continuous order management, constantly adjusting quotes to respond to market changes.

In terms of risk and reward, AMMs mainly face impermanent loss risk, earning fees from the liquidity pool. Order book market makers face inventory risk in unidirectional markets, with profits coming from bid-ask spreads and platform subsidies.

Continuing the previous example, assuming that my main risk in Polymarket is inventory risk, and my income mainly comes from bid-ask spreads and platform subsidies (Polymarket provides liquidity subsidies for some near-market orders, see official site), then the profit and loss scenarios for the four situations are:

1. Miss out on bid-ask spread but earn liquidity subsidies;

2. Profit from bid-ask spread but no longer receive subsidies;

3. Already filled a YES or NO order, becoming a directional position (inventory risk), but in some cases still earning liquidity subsidies;

4. Also a directional position with unrealized losses, and no longer receiving liquidity subsidies.

Note two points here: First, scenario 2 is actually a progression from scenario 3 or 4, because usually only one side’s order is filled first, temporarily creating a directional position, but the risk doesn’t materialize if the market reverses and triggers the other side; second, compared to the relatively fixed scale of market-making profits (spread and subsidies), the risk of a directional position is often unlimited (the maximum being the total loss of the held YES or NO).

In summary, if I want to be a continuous profit-making market maker, I need to maximize profit opportunities while avoiding inventory risk — so I must actively optimize strategies to maintain scenario 1 as much as possible, or quickly adjust order ranges after one side’s order is triggered to turn it into scenario 2, avoiding long-term exposure to scenarios 3 or 4.

Achieving this consistently over the long term is challenging. Market makers need to understand the structural differences of various markets, compare subsidy levels, volatility, settlement times, rules, etc.; then track and even predict market prices more precisely and quickly based on external events and internal fund flows; and timely adjust orders based on changes, while designing and executing inventory risk management in advance… This clearly exceeds the capabilities of ordinary users.

Wilder, more jumpy, and less ethical markets

If it’s just like this, it might still be manageable, since order book mechanisms are not new. They are still the main market-making mechanism on CEXs and Perp DEXs. Active market makers in these markets can transfer their strategies to prediction markets to continue profiting and inject liquidity, but reality is more complex.

Let’s think about the worst-case scenario for market makers: what is it? The answer is simple — a unidirectional trend, because such markets tend to amplify inventory risk, break the balance, and cause huge losses.

Compared to traditional crypto markets, prediction markets are inherently more wild, jumpy, and less ethical. Unidirectional markets tend to be more exaggerated, abrupt, and frequent.

More wild means that, over a longer timeline, mainstream assets still show oscillations, with trends cycling periodically. But in prediction markets, the underlying assets are event contracts, each with a clear settlement time, and the formula Yes + No = 1 ensures only one contract’s value will reach 1 dollar, while others will zero out — meaning bets in prediction markets tend to end in a unidirectional trend from a certain point, requiring market makers to design and execute more rigorous inventory risk management.

More jumpy means that, unlike traditional markets where volatility is driven by emotions and continuous capital battles, with prices changing continuously, prediction market prices are often jumpy — the price might be 0.5 one second, then a real-world event causes it to jump to 0.1 or 0.9 the next second. Many times, it’s very hard to predict when and how the market will suddenly shift due to an event, leaving market makers with very little reaction time.

Less ethical means that prediction markets host many insiders or players who are essentially insiders, not trading based on predictions but with a clear goal of harvesting the outcome — in front of these players, market makers are at an informational disadvantage, and the liquidity they provide becomes a channel for these insiders to cash out. You might ask, do market makers have insider info? That’s a classic paradox — if I already know the inside info, why would I bother making a market? I’d just bet directly to earn more.

Because of these features, I have long believed that “the design of prediction markets is structurally unfriendly to market makers,” and I strongly advise ordinary users not to jump into market making lightly.

Does this mean prediction markets are unprofitable? Not necessarily. Luke (@DeFiGuyLuke), founder of Buzzing, disclosed that based on market experience, a relatively stable expectation is that Polymarket’s market makers can earn about 0.2% of trading volume.

So, in essence, this is not an easy way to make a living. Only professional players who can precisely track market changes, adjust orders timely, and effectively manage risks can sustain operations over longer periods and profit through real skill.

Prediction market tracks, perhaps, are very hard to flourish in a hundred flowers

The difficulty of market making in prediction markets not only raises the bar for market makers’ abilities but also poses a challenge for platforms to build liquidity.

The difficulty in market making limits liquidity creation, which directly impacts user trading experience. To address this, leading platforms like Polymarket and Kalshi choose to subsidize liquidity with real money to attract more market makers.

Nick Ruzicka, an analyst focused on prediction markets, cited a report from Delphi Digital in November 2025, stating that Polymarket has invested about $10 million in liquidity subsidies, once paying over $50,000 daily to attract liquidity. As its leading position and brand effect solidify, Polymarket has significantly reduced subsidies, but on average, it still needs to subsidize $0.025 for every $100 traded.

Kalshi has a similar liquidity subsidy plan, having paid out at least $9 million. Additionally, in 2024, leveraging its regulatory advantage (Odaily note: Kalshi is the first prediction market platform licensed by CFTC; Polymarket also obtained licensing in November 2025), Kalshi signed a market-making agreement with top Wall Street market maker Susquehanna International Group (SIG), greatly improving its liquidity.

Whether for capital reserves or regulatory thresholds, these are real moats for platforms like Polymarket and Kalshi. A few months ago, Polymarket was valued at $8 billion and received a $2 billion investment from NYSE parent ICE, with rumors of planning a next round at over $10 billion. Meanwhile, Kalshi has completed a $300 million funding round at a $5 billion valuation. Both giants have substantial war chests.

Currently, prediction markets have become a hot startup track, with new projects emerging constantly. But I remain skeptical. The leading effect of prediction markets is actually stronger than many imagine. With continuous subsidies from giants like Polymarket and Kalshi, and the dimensionality reduction partnerships from the regulated world, what can new projects rely on to compete head-to-head? How much capital is willing to compete? Some new projects may have backing from big players and can explode tokens, but not all.

Haseeb Qureshi, the big bald head of Dragonfly, recently posted his forecast for 2026, stating: “Prediction markets are 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 reasoning, but I agree this is not an alarmist statement.

Many people look forward to a hundred flowers blooming in the prediction market track, hoping to profit from past experience. But this scenario may be very unlikely. Instead of spreading bets across many projects, it’s better to focus on the giants.