With thanks to Tarun Chitra, Xin Wan, Davide Rezzoli, Franklin Bi, Mason Nystrom, Gaussian Process

---

Prediction markets have entered a new growth phase in 2026.

Over the past year, notional trading volume has expanded by an order of magnitude, consistent with the rise in user activity and market count. Weekly aggregate volume across major prediction market platforms now regularly reaches billions of dollars. What was once a niche corner of crypto and fintech has increasingly become a mainstream venue for expressing views on real-world events — from politics and macro to, more recently, sports.

At the same time, market share has shifted meaningfully across venues. Different platforms now also attract activity in different segments—with some platforms concentrating volume in specific market types, such as sports, while others dominate political markets or support a broader diversity of markets. This divergence highlights that raw volume alone is no longer sufficient to understand how prediction markets function, or how well different platforms serve users when new information arrives and event dynamics shift.

The two leading platforms have made distinct market design choices. For example, Polymarket operates with an offchain order book but onchain settlement layer, while Kalshi runs a fully off-chain exchange. In addition, Polymarket applies a special rule for live sports markets as below:

With the Super Bowl approaching, we focus our analysis on NFL game markets to study how these design choices shape market behavior under real-time information flow—specifically comparing pricing speed, liquidity depth, and behavioral differences in market outcomes across platforms.

To ground the analysis, the table below summarizes the key market design differences between Polymarket and Kalshi that are most relevant for interpreting the empirical results that follow.

Dimension	Polymarket	Kalshi
Market orders	Yes (UI + API)	Yes (UI + API)
Market order delay	Yes — ~3s delay on marketable orders in live sports	No explicit delay
Limit orders creation	Yes (UI + API)	Yes (UI + API)
Limit order cancellation	Off-chain	Off-chain
Order book	Off-chain	Off-chain
Liquidity provision	limit orders mainly + (public Market Maker program with incentives for 15mins crypto market only)	partnered market makers + limit orders (via API)
Liquidity incentives	maker rebates, liquidity rewards ( + taker-fee for 15min crypto market only)	implicit agreements with professional Market Makers
Matching & execution	On-chain settlement after delay window	Immediate off-chain matching
Settlement layer	Polygon PoS (~2s block time)	Off-chain, instant settlement
Transparency of flow	Trades & order events publicly observable	Aggregate data public; no trader-level ID

We aim to answer the following questions:

• How do prices respond to live information across platforms?
• How does liquidity concentrate and distribute across platforms?
• Do market dynamics differ systematically across platforms for different market starting conditions and realized outcomes?

We would also like to examine trader-level behavioral profiles across platforms. However, due to limitations in Kalshi’s publicly available data—specifically the absence of trader identifiers—such analysis is not feasible in this study.

Datasets

To ensure an apples-to-apples comparison, we constructed a dataset restricted exclusively to NFL game markets, covering the following:

• Complete mapping of NFL events and their associated markets
• All NFL games in the 2025 season, including:
  • Regular season games
  • Playoff games through January 18

For each platform, we collected trade-level records including execution price, trade size, and timestamp, sourced from either on-chain transaction data or platform API endpoints.

To focus on periods of highest information intensity, we limited data collection to trades occurring from approximately 30 minutes before kickoff through the end of each game. The majority of trading activity in moneyline prediction markets occurs during live gameplay, making this window the most relevant for analyzing pricing dynamics.

The table below summarizes key descriptive statistics for the trades included in our sample.

	Polymarket	Kalshi
Time Range	9/4/2025 – 1/18/2026	9/4/2025 – 1/18/2026
Total # of Games	282	282
Total Notional Volume ($)	$359M	$1.3B
Avg # of trades / game	2,718	19,024
Avg bet size ($) across all games	$469.88	$224.83
Median bet size ($) across all	$18.00	$23.97

The 7s Lead in Pricing on Kalshi

How Do Markets React to Live Game Shocks? We begin with a single game-level example. Figure below shows a live win-probability time series for a Chicago Bears vs. Philadelphia Eagles NFL game, comparing prices on Kalshi and Polymarket—against timestamp of recorded events from ESPN.

We categorize in-game events into different levels of informational shocks (as indicated in the legend below), including scoring events, turnovers, and penalties. These events differ in both magnitude and immediacy of their expected impact on win probability.

We observed consistent patterns in this chart — in how live markets respond to different types of shocks:

• Possession effects: During most possession periods, the team with the ball experiences a increase in win probability—ranging from 10% at game start, to a bigger swing up to 30% near game end. If the drive stalls without scoring, this effect gradually decays, reflecting the market’s reassessment of scoring likelihood as opponent’s defense holds.
• Scoring events: Scoring events—particularly touchdowns—represent the largest potential point gains, but are often priced in during the preceding possession, with the official scoring primarily to stabilize prices rather than trigger discrete jumps.
• Turnovers: Interceptions and fumbles often cause abrupt probability reversals, as they simultaneously remove a scoring opportunity from one team and transfer possession to the opponent. These events typically result in sharper price movements than penalties or routine plays.
• Penalties: Generally lead to smaller, more incremental adjustments, unless they materially alter field position or occur late in the game when remaining time amplifies their impact.

Across all event types, the chart highlights how live prediction markets continuously translate on-field events into probabilistic beliefs.

One clear distinction, however, is timing. Across the majority of large probability moves, Kalshi’s prices begin repricing earlier, while Polymarket’s price series often appears visibly lagged. At the league level, Kalshi leads the initial repricing in roughly 80% of large moves, with a median lead time of approximately 7 seconds.

Importantly, this lead is measured from the onset of repricing, not from the completion of the move. In many cases, Polymarket follows with a smoother adjustment path rather than an immediate jump. This pattern is consistent across shock sizes, from moderate (5–10pp) to large (20pp+) swings.

Swing Size	Events	Kalshi Leads
5–10pp	1,236	78.6%
10–20pp	459	86.7%
20pp+	126	85.7%

Intuitively, we ask whether faster repricing on Kalshi is associated with higher in-game volatility. To examine this, we compute the difference in realized in-game volatility between Kalshi and Polymarket for each game and find that the gap is consistently positive across the sample—indicating higher price variability on Kalshi.

We further classify games by pregame expectations and eventual outcomes, grouping them into:

• Underdogs: teams with a pregame win probability below 40% that ultimately won
• Evenly matched: games with pregame win probabilities in the range [0.4, 0.6]
• Heavy favorites: games where the eventual winner was strongly favored at kickoff

Across all game profiles, the volatility gap remains positive, suggesting that Kalshi exhibits higher in-game volatility regardless of starting expectations. We also observe a weak trend whereby the volatility gap is larger in more surprising outcomes—particularly when heavily unfavored teams win.

However, this relationship is modest in magnitude. The linear fit explains a limited share of the variance (R^2 \approx 0.118), indicating that while outcome surprise contributes to volatility differences, it is not the primary driver.

It takes 3.5x more liquidity to move a NFL Market on Polymarket

We then examine the liquidity profile of each platform. Because our dataset include executed trades rather than full order book snapshots, we cannot directly observe how much liquidity is posted at each price level. Instead, we infer liquidity indirectly using a **Kyle-style price impact framework**, which measures how prices respond to trading flow after execution, while controlling for individual games.

Under this approach, liquidity is defined in terms of trade-induced price impact, not visible depth in the order book. Kyle-style liquidity depth (LD) is computed as follows:

• Trading pressure: For each trade, compute signed dollar flow (price × size), with buys treated as positive pressure and sells as negative pressure.
• Price response: Measure how market prices change over a fixed horizon (60 seconds) following each trade.
• Price impact (Kyle λ): For each game, estimate how much prices move per dollar traded by regressing price changes on trading pressure.
• Liquidity depth (LD): Take the inverse of price impact. Higher LD implies that more trading flow is required to move prices—indicating a more resilient market.
• Distribution: We compute one LD value per game and analyze the histogram / distribution across platforms to compare liquidity conditions at the market level.

Importantly, higher Kyle-LD does not imply more posted liquidity. Rather, it indicates lower price sensitivity to trades—i.e. more concentrated liquidity —that is, prices move less in response to comparable trading pressure.

The distribution of implied liquidity depth shows a clear rightward shift for Polymarket relative to Kalshi. Across NFL games, Polymarket markets tend to exhibit higher Kyle-style liquidity depth, indicating that prices on Polymarket generally require more trading flow to move by the same amount.

At the median, Polymarket’s Kyle-implied liquidity depth is approximately 10⁶·⁹⁶, compared to 10⁶·⁴² on Kalshi—this implies that—on a typical game—Polymarket requires roughly 3–4× more notional volume to generate a comparable price move over a 60-second horizon. This gap reflects materially lower price sensitivity to trades on Polymarket.

While the two distributions overlap substantially, Polymarket exhibits:

• A heavier right tail, with more games displaying very high depth
• Fewer low-depth outliers relative to Kalshi
• More consistent price resilience across games

Importantly, this does not imply that Polymarket has uniformly higher activity or faster price discovery. Rather, it suggests that liquidity on Polymarket is more concentrated near the prevailing price, allowing the market to absorb trading pressure with smaller price dislocations.

To examine whether liquidity systematically improved over the course of the season, we tracked the distribution of implied liquidity depth over time. We do not observe a meaningful upward trend as the season progresses. Across weeks, the overall shape of the distribution remains broadly stable, with similar central tendencies and only modest changes in dispersion. While later-season games exhibit slightly fewer extreme outliers, the core distribution is largely unchanged.

Conclusion

Across all NFL games in our sample, we find that:

• Kalshi leads price adjustments by a median of ~7 seconds, exhibiting faster and more volatile reactions to in-game information shocks.
• Polymarket, despite lower volume and open interest, consistently exhibits higher implied liquidity depth, allowing prices to absorb comparable trading pressure with smaller dislocations.

Zooming back out to check against the basic metrics, publicly available data (e.g., Dune dashboard) shows that:

• Kalshi consistently processes ~2× higher notional volume than Polymarket in sports markets.
• Kalshi also maintains ~2× higher open interest across sports categories.

Focusing specifically on NFL game markets, our dataset shows a similar pattern: Kalshi settles roughly 2–3× more notional volume per game, which is also consistent with the 3x open interest ratios observed for football markets.

At first glance, these aggregate statistics may appear “contradictory”. However, volume and open interest do not directly measure liquidity depth—as high volume reflects trading activity, while open interest reflects outstanding positions.

Taken together, these results point to a structural difference between the two platforms. The underlying drivers of this divergence remain an open question. Potential contributing factors include market design choices:

• Polymarket’s ~3-second delay on marketable orders in live sports may reduce adverse selection for liquidity providers, allowing them to quote tighter prices with lower risk. This design can support deeper concentrated liquidity near the prevailing price, as liquidity providers face less exposure to instantaneous, information-driven order flow. In contrast, Kalshi’s immediate execution model exposes resting liquidity to faster information shocks, potentially leading market makers to quote more conservatively.
• These design differences may also shape participant composition and trading behavior—as a secondary effect. Differences in execution risk and latency can influence: 1) who comes to the platform to provide liquidity, how aggressively liquidity is supplied, 2) the ratio between informed vs uninformed trader, and how users trade in response to live information. —leading to divergent liquidity provision strategies and user profiles across platforms.

Overall, these findings suggest that pricing speed and liquidity depth need not move together. Rather, they reflect deliberate design choices that shape how markets respond to information, volatility, and surprise—highlighting that faster markets are not always deeper ones, and deeper markets are not always faster.