What the Odds Do Not Show: Why Liquidity Matters in Prediction Markets

The odds are only the front door

A prediction-market screenshot can look convincing and still leave out the most important part.

A contract trades at 75 cents, so the crowd thinks the event has a 75% chance. Clean. Elegant. Screenshot-friendly.

But that number is only the front door. Behind it is the orderbook: the live stack of bids and asks showing who is actually willing to buy, who is willing to sell, at what price, and in what size.

If the price is the market's headline, liquidity is the supporting evidence. It tells you whether the market is relaxed, stressed, crowded, thin, directional, fragile, or simply easy to move.

That matters because prediction markets are increasingly used as public probability machines. People quote Polymarket odds in group chats, newsletters, trading desks, crypto Twitter, political analysis, and increasingly mainstream media. The temptation is to treat the displayed probability as a single clean forecast. But every market price comes with a hidden question: can you actually trade at that price?

If the answer is "barely," the headline probability deserves a discount. Not necessarily because it is wrong, but because it is less reliable than it looks.

This is why orderbook liquidity deserves more attention. A liquid prediction market can absorb disagreement and update cleanly when new information arrives. A thin one can jump on a small order, display stale consensus, or make a tiny amount of capital look like conviction.

Before markets became numbers

Before electronic markets, price discovery was loud. Buyers and sellers shouted across exchange floors. Market makers quoted prices by voice. Specialists managed inventory. The "market" was not just the last traded price. It was the entire social machinery around finding someone willing to take the other side.

Electronic orderbooks turned that shouting match into a ledger. A central limit order book, or CLOB, gathers limit orders from many participants, ranks them by price and time, and matches them when a buyer and seller cross. The best bid is the highest price someone is currently willing to pay. The best ask is the lowest price someone is currently willing to accept. The spread between them is the cost of immediacy.

This structure is not new. The Paris Bourse, one of the canonical electronic limit-order markets studied in finance, gave researchers a clean view into how orders, depth, spreads, and trades interact. In their classic Journal of Finance study, Biais, Hillion, and Spatt (1995)[1] found that order flow was concentrated near the quote, that thin books attracted new limit orders, that thick books tended to result in trades, and that large purchases or sales moved both bid and ask quotes in the expected direction.

That is the first important lesson: the book is not decoration around the price. It is the mechanism that produces the price.

Prediction markets inherited this machinery. A Polymarket market may feel like a probability widget, but under the hood it is a trading venue. Polymarket Trading Overview[2] describes its CLOB as a hybrid-decentralized system: orders are matched offchain, while matched trades settle onchain through the exchange contract. Polymarket Create Orders docs[3] distinguish resting limit orders from market-style orders that execute immediately against resting liquidity. Polymarket Orderbook docs[4] expose prices, spreads, midpoints, and book data, and note that the displayed midpoint is the average of the best bid and best ask, while wider-spread markets may display last traded price instead.

That small implementation detail says a lot. The number users see is not always a frictionless truth. It is a convention chosen under market conditions.

Liquidity makes a price believable

Liquidity is often explained as "how easy it is to trade." True, but too soft.

A better version: liquidity is the market's ability to accept a trade without forcing the price to lie.

In a deep market, a new buyer can enter without pushing the price far above where the market was already clearing. In a thin market, the same buyer may walk through several price levels, pay up, and leave behind a new "price" that mostly reflects their own footprint. The last trade happened, yes. But did it reveal new information, or did it reveal an empty book?

Financial economists have spent decades separating these ideas. Demsetz (1968)[5] framed the bid-ask spread as a transaction cost for immediacy. Glosten and Milgrom (1985)[6] showed how spreads can arise from asymmetric information: liquidity providers widen quotes because the next trader might know something they do not. Kyle (1985)[7] modeled price impact, where order flow moves prices because market makers infer information from trading pressure. Vayanos and Wang (2012)[8] emphasize that liquidity is multidimensional: spreads, depth, price impact, trading volume, and resilience all capture different pieces of the same problem.

• Spread: how much you pay to cross from buyer to seller.
• Depth: how much size is available near the current probability.
• Slippage: how far your realized fill moves from the price you thought you saw.
• Imbalance: whether buy-side or sell-side liquidity dominates near the top of book.
• Resilience: whether liquidity refills after a trade, cancellation, or news shock.

The spread tells you the immediate cost. Depth tells you how much the market can absorb. Slippage tells you what happens when you actually try to trade.

Prediction markets put liquidity under pressure

Prediction markets are not just miniature stock markets with stranger tickers. Their contracts have special properties that make liquidity analysis unusually important.

First, most contracts are bounded between zero and one. A YES share might pay $1 if an event happens and $0 if it does not. That makes the price intuitive as a probability, but it also compresses the entire disagreement of the market into a narrow range. A move from 70 cents to 75 cents looks small in dollar terms, yet it can be a meaningful change in implied probability.

Second, many prediction markets are event-driven. Liquidity can look stable for hours, then vanish around a debate, court ruling, sports injury, CPI release, token listing, weather update, or resolution ambiguity. In normal markets, uncertainty arrives constantly. In prediction markets, uncertainty often arrives in bursts.

Third, markets fragment across thousands of questions. A major presidential market may have meaningful depth. A niche cabinet appointment, regulatory deadline, celebrity outcome, or crypto price barrier may not. Even when a platform is large, most individual markets can still be thin.

Fourth, information is uneven. Some traders may have better models, faster news, domain expertise, or simply more patience. Liquidity providers know this. When they suspect informed flow, they widen or pull quotes. That is not noise around the market. It is part of how the market processes risk.

Finally, prediction-market traders often care about the displayed probability as information, not just execution. This creates a subtle trap. If users quote a price without asking how liquid it is, a thin market can influence beliefs more than it deserves.

A market with no liquidity is not a forecast. It is a rumor with a ticker.

A forecast needs a market behind it

Prediction markets have a strong empirical record, but the strongest papers do not say every displayed price is automatically reliable. They say markets can aggregate information well under the right conditions.

Wolfers and Zitzewitz (2004)[9]'s Journal of Economic Perspectives survey found that market-generated forecasts are often accurate and can outperform many benchmarks across domains. Arrow et al. (2008)[10] argued in Science that prediction markets can be powerful information-aggregation tools and should be more widely used. Berg, Nelson, and Rietz (2008)[11]'s Iowa Electronic Markets research showed impressive election-forecasting performance, including evidence that markets outperformed polls over longer horizons in U.S. presidential elections.

But there is an equally important second half to the literature: market design matters.

Healy, Linardi, Lowery, and Ledyard (2010)[12] tested prediction mechanisms in complex environments with few traders. Slamka, Skiera, and Spann (2013)[13] studied automated market makers for prediction markets and emphasized that low liquidity is a core design problem. Chakraborty, Das, and Peabody (2015)[14] found that adding a market maker lowered spreads and improved trader surplus, but did not always improve price discovery. Strijbis and Arnesen (2019)[15] found that prediction-market accuracy varied primarily with market setup rather than only the event or participant pool.

That is the mature view: prediction markets can be excellent, but their accuracy is not detached from liquidity, participation, incentives, and market structure.

So when someone says, "The market is at 75%," the better question is: what kind of 75%?

A 75% market with a one-cent spread, thick two-sided depth, repeated refills, and heavy participation is very different from a 75% market with a six-cent spread and $200 of visible size near the top.

Both may display 75%. Only one is backed by enough visible depth to deserve stronger confidence.

The book shows what the chart cannot

A price chart shows where trades happened. An orderbook shows where trades could happen next.

That distinction is central. The best empirical work on limit-order books repeatedly finds that information lives beyond the last trade. Cao, Hansch, and Wang (2009)[16] studied open limit-order books on the Australian Stock Exchange and found that book information beyond the best bid and offer contributed meaningfully to price discovery. They also found that imbalances between demand and supply schedules in the book were significantly related to future short-term returns, even after controlling for returns, spread, and trade imbalance.

For prediction markets specifically, Groeger (2016)[17] studied binary-option prediction markets and asked whether execution prices alone reveal consensus beliefs. His paper is cautious, which makes it useful: he finds little evidence for simple convergence in beliefs and explores how order submissions beyond executed prices can bound beliefs. In plain English, trades are not the whole market. The unfilled orders matter too.

This is where orderbook analytics become more interesting than simply asking which side has more size.

• Is the spread tight or wide relative to the contract's current probability?
• Is liquidity balanced on both sides, or is one side leaning hard?
• Is the depth close to the midpoint, or parked far away where it will not help execution?
• Did the price move because buyers lifted real asks, or because sellers canceled and the book went hollow?
• When a wall appears, does it stay, refill, and get hit, or vanish when price approaches?
• After a trade, does liquidity recover, or does the market remain fragile?

The answers do not guarantee the outcome. They tell you how much confidence to place in the current price and how expensive it may be to act on it.

On Polymarket, the book is observable

Polymarket is especially interesting because its CLOB data makes this analysis observable.

Polymarket Orderbook docs[4] expose public reads for orderbooks, spreads, midpoints, and price history, and describe fill-price estimation by walking the orderbook for a given order size, explicitly accounting for depth and slippage. Polymarket Create Orders docs[3] explain that market orders execute against resting liquidity using fill-or-kill or fill-and-kill order types, with a worst-price limit acting as slippage protection.

That is exactly the layer most casual users skip.

Imagine a YES market shown around 75 cents.

If the best bid is 74 and the best ask is 76, the midpoint is 75. A buyer pays roughly 76 to get filled immediately. A seller receives roughly 74. The spread is two cents.

Now imagine the best bid is 70 and the best ask is 80. The midpoint is still 75, but the market is not really saying "75" with the same confidence. A buyer pays 80. A seller receives 70. The displayed midpoint may be mathematically correct, but economically it sits inside a wide uncertainty band.

Now add size.

Maybe there is $5,000 available within two cents of mid. Or maybe there is $80. In the first case, a modest trader can interact with the market without changing it much. In the second, even a small order can become the market event.

• Probability asks: what is the market implying?
• Tradability asks: can I enter or exit near that implication?
• Reliability asks: does the book support the implication, or is it thin, jumpy, and easy to move?

Prediction-market analysis gets much better once those three are treated separately.

Liquidity can move before the price

Liquidity does more than measure execution quality. It can reveal how participants are preparing for information.

When uncertainty rises, market makers often protect themselves. They may widen spreads, reduce size, or move quotes away from the midpoint. That does not necessarily predict direction. Sometimes it predicts volatility. The book may be saying: something is coming, and no one wants to be the generous counterparty right now.

Directional pressure is different. If bids keep refilling while asks get lifted, the market may be absorbing sellers and leaning upward. If ask-side liquidity thickens while bids retreat, downside pressure may be building. If both sides disappear, the signal is not bullish or bearish; it is fragility.

This is where the best orderbook work becomes a time-series problem, not a screenshot problem. A single snapshot can fool you. A wall can appear for a moment. A market maker can cancel. A thin market can look imbalanced because one trader posted a lazy order. But persistent behavior across many updates is harder to fake and more informative.

• Repeated bid replenishment after sells.
• Ask depth evaporating near the top of book.
• Spread widening before a known catalyst.
• Depth collapsing after a price move.
• Walls moving closer to or farther from mid.
• Slippage worsening before the headline price moves.

The point is not that liquidity predicts the future by itself. It is that liquidity shows the current negotiation between information, risk, and urgency.

Posted liquidity can disappear

Visible liquidity is useful, but it is not the whole truth.

Orders can be canceled. Market makers can quote only while conditions are calm. Some displayed size is opportunistic: happy to earn spread or rewards, not necessarily willing to stand through news.

Polymarket Liquidity Rewards docs[18] describe incentive programs that reward makers for posting qualifying limit orders near the midpoint. Incentives can improve market quality by encouraging tighter, deeper books, but they also mean some liquidity may be present because the reward design pays for it.

This does not make the liquidity fake. It means analysts should ask why it is there.

A tight book in a high-volume market with organic two-sided flow says one thing. A tight book in a quiet sponsored market where liquidity disappears around volatility says another. The same spread can have different meaning depending on participation, event timing, and refill behavior.

Emerging Polymarket-specific research is beginning to map this terrain, though much of it is still working-paper evidence rather than peer-reviewed literature. Yang (2026)[19] studies skilled liquidity provision using Polymarket transaction data, while Dubach (2026)[20] examines maker/taker behavior, wash-trading constraints, and decentralized prediction-market microstructure. These are useful early field notes, but they should be cited as emerging evidence, not settled consensus.

The safer conclusion is already supported by older market microstructure research: the orderbook contains information, but reading it requires context.

That is the story so far. The price is the visible answer, but liquidity tells you how that answer was formed, how hard it is to trade, and how quickly the market might change when pressure arrives. From there, the practical task is straightforward: read the probability, then read the market quality behind it.

A practical liquidity checklist

For anyone using Polymarket or similar CLOB-based prediction markets, here is a practical checklist.

Better decisions start with better context

Understanding liquidity does not mean treating the orderbook as an oracle. It means refusing to flatten a market into a single number.

This is the next layer of prediction-market literacy. The first wave was teaching people that prices can be probabilities. The second wave is teaching them that probabilities have market quality.

The future is odds with market quality attached

Prediction markets are becoming part of the information layer of the internet. That is exciting. It also raises the bar.

If prediction markets are going to be quoted as forecasts, the quote needs context. A clean probability is useful, but a probability with spread, depth, slippage, imbalance, and resilience is far more useful. It tells us not just what the market thinks, but how strongly the market can support that thought.

This is where orderbook monitoring becomes fundamental. It is not a niche trading toy. It is the infrastructure for interpreting these markets responsibly.

The more prediction markets matter, the less acceptable it becomes to look only at the last price.

The real question is not "What are the odds?" The real question is: what are the odds, and how liquid is that belief?