The Pricing Gap in Prediction Markets: Why Are There Four Different Probabilities for the Same Election Outcome?

Author: Coin Metrics Research Team

Compiled by: Deep Tide TechFlow

Deep Tide Overview: As the 2026 U.S. midterm elections approach, the cumulative trading volume in prediction markets has surpassed $1.4 billion, becoming the most sensitive barometer of winning probabilities globally, aside from traditional polls. However, most people focus solely on the single "winning probability" under the spotlight, overlooking the complex micro-market structure beneath the surface.

Crypto quant and on-chain data giant Coin Metrics, in this report, uses the Senate control event on the compliant prediction platform Kalshi as a case study, tracking 38 independent contracts linked to the same election outcome. Astonishingly, the politically binary question of whether "the Democrats can control the Senate" yielded four divergent implied probabilities across different contract matrices on the platform, with the maximum intraday price difference soaring to 9.7 percentage points.

By constructing a "Nine-Leg" combination package that buys control and shorts segmented seat positions, it is theoretically possible to lock in constant returns and capture arbitrage opportunities under any election outcome. However, cumbersome market-making frictions, extremely polarized order book liquidity, leg execution risks, and transaction fees as high as 5.2 cents vividly illustrate the brutal gap that prediction markets must cross from "theoretical arbitrage" to "realized returns."

Key Points

Kalshi prices the same core question: "Will the Democrats control the Senate?" through four different contract families.

Since August, the implied winning probabilities among them have averaged a difference of 3.3 percentage points, with the maximum spread even widening to 9.7 percentage points at one point. As of October 25, the probabilities given by different contracts still ranged widely from 62.5% to 65.3%.

Theoretically, the "Nine-Leg" arbitrage package (buying the Democratic Senate control contract while selling all independent seat contracts for the Democrats winning 51 seats or more) can yield a fixed $8 payout under any final election outcome if both contract families settle based on the same facts.

Based on end-of-day closing quotes, in 51 trading days, there were 19 days when the total buying cost of this package was below $8, creating theoretical arbitrage opportunities (on October 24, before accounting for fees, the peak arbitrage space for each $8 face value reached 8.1 cents).

Actual execution assumptions directly dominated final profits. After accounting for estimated trading fees, if the main leg contract was executed passively, simulated excess returns remained positive on 7 trading days (with a maximum of +4.7 cents); however, if all nine legs adopted aggressive active orders across the bid-ask spread, the number of days with positive returns plummeted to just 2.

The liquidity distribution within the platform shows an extremely severe extreme imbalance. Since August, the flagship contract for Democratic control has seen a trading volume of $3.3 million, while the cumulative trading volume for some segmented seat contracts required for pricing in the package was merely $2,000.

The Senate case is just a microcosm of the entire prediction market ecosystem. According to data captured by Talos CM, there are over 880,000 active contracts on the Kalshi platform, and this highly fragmented market structure generally gives rise to similar pricing gaps across different events and platforms. To truly capture these alphas, reliance on systematic event mapping identification and precise execution algorithms is essential.

Introduction

With only a week left until the midterm elections, the cumulative trading volume of contracts linked to the 2026 midterm elections on the Kalshi and Polymarket platforms has surpassed $1.4 billion.

Most media reports and market commentary typically focus only on a single headline figure: Kalshi currently shows that the probability of the Democrats winning the Senate is about 63%.

However, this widely known single data point is merely the tip of a massive iceberg. Kalshi has listed as many as 38 independent contracts under the topic of Senate control, each with its own independent order book and trading group.

In this article, we will delve into the real probability differences implied by these derivative contracts, calculate when these price differences can converge into a positive expected simulated arbitrage space, and deeply analyze what micro-market structures prevent these pricing gaps from being smoothed over in the long term.

The Same Question, Four Distinct Answers

On the Kalshi platform, investors can express their bets on the Senate control through four different paths:

First, there are binary contracts directly linked to the control outcome (whether the Democrats control the Senate); second, there are tiered contracts for the number of seats each party wins in the Senate (covering ranges from 51 seats to 57 seats and above); third, there are power balance combination contracts; and fourth, there are inverse contracts for Republican control.

Figure: Four types of contract families and event structures regarding Senate control on the Kalshi platform.

If every derivative contract in the prediction market had strict and consistent pricing efficiency, then the four curves reflecting the implied probabilities of the four types of contracts should completely overlap.

Figure: Comparison of implied winning probability trends for different Senate control event contracts on Kalshi.

From a broad perspective, the four types of contracts do follow the same macro-evolution trend: the probability of a Democratic victory has risen from about 46% in early August to a low range of 60% to 65% in October. However, they rarely reach absolute consensus on specific points.

On as many as two-thirds of trading days, the implied winning probability derived from the specific seat tier for the Democrats was significantly higher than the headline control contract itself. On September 20, the winning probability estimates between the four answers even diverged sharply between 55.7% (derived from Republican seat numbers) and 65.4% (derived from Democratic seat numbers).

Figure: Trajectory of the maximum and minimum implied probability spread between different contract families.

Statistics show that the average spread between the highest and lowest implied probabilities is 3.3 percentage points. This gap was most exaggerated in September (averaging 4.5 percentage points, compared to 3.2 percentage points in August); in October, as trading volume continued to expand, the average spread remained at 2.5 percentage points.

This indicates that the pricing gap has not automatically closed with the injection of liquidity: on October 21, the maximum valuation difference between the four types of contracts was still as high as 4.0 percentage points.

Assembling a Nine-Leg Risk-Free Arbitrage Asset Package

When two contracts that should be equivalent disagree on the same election fact, theoretically, as long as one simultaneously buys the undervalued side, this price difference can be transformed into a certain risk-free profit. Here, the most intuitive arbitrage structure is intricately assembled from the "Nine Legs":

First Leg: Buy the YES contract for "the Democrats winning Senate control." If the Democrats win, this contract will pay out $1.

The remaining eight legs: Buy the NO contracts for each tier of the number of seats the Democrats win starting from 51 seats (51 seats, 52 seats, 53 seats up to 57 seats and above). As long as the final election result does not land precisely on that tier, the corresponding contract will pay out $1.

Figure: Simulated combination profit matrix for buying Democratic control and selling seat number contracts.

As long as these two contract families settle based on the same factual logic, this arbitrage asset package can reliably return $8 in payout under any election outcome.

In other words, as long as the total buying cost of these nine dimensions is below $8, the entire package mathematically locks in a positive arbitrage space.

Figure: Historical simulated performance of the Nine-Leg arbitrage package before and after accounting for trading fees.

Based on end-of-day closing price backtesting analysis since August (all calculations are theoretical deductions):

In an ideal state without accounting for trading fees, if each leg is bought at the ask price, there were 19 days in all 51 trading days when the total cost of this package was below $8.

However, after deducting estimated fees (averaging about 5.2 cents per arbitrage package), only 2 trading days maintained positive returns under aggressive order execution.

Taking the real order book depth snapshot from October 25 as an example, the total cost of building 10 sets of this arbitrage package was $79.65 (including estimated fees), while the fixed total payout at maturity was $80, resulting in a net profit of only a meager $0.35.

Most positive return opportunities concentrated in the last two weeks, primarily catalyzed by the approaching election and a sharp increase in market trading activity leading to pricing discrepancies.

Why Do Price Differences Persist Long-Term: Liquidity Disparities and Trading Frictions

Since the final cash flow of this strategy is locked, why hasn’t rational arbitrage capital quickly eliminated this price difference? The core of the answer lies in liquidity.

Figure: Severe imbalance distribution of trading volumes for various Senate contracts on Kalshi.

Since August, the nominal cumulative trading volume of all Senate contracts on Kalshi has been about $14.3 million, but over 90% of the funds have been completely absorbed into a single flagship Senate control contract.

The ultimate success or failure of the arbitrage package is entirely dependent on its most fragile leg's liquidity. The Democratic control contract has accumulated $3.3 million in trades, while the "Democrats winning 54 seats" contract has only seen a cumulative trading volume of $3,000, and the "Democrats winning 56 seats" contract has a cumulative trading volume of merely $2,000.

In addition to extreme liquidity scarcity, there are also many real friction costs that must be faced in actual operations:

First is legging risk. Traders must ensure that all nine orders are successfully matched before they can completely lock in theoretical profits; if they encounter slippage or unfilled orders on the weak leg, the remaining orders are exposed to high risks of severe price fluctuations.

Second is high trading fees. Kalshi's fee rates exhibit a parabolic distribution, peaking around a 50% win probability, and decreasing towards both 0% and 100%. The average fee expenditure of about 5.2 cents per package can instantly consume most of the meager price difference alphas.

Third is settlement benchmark and timing mismatches. This strategy assumes that "controlling the Senate" and "specific seat numbers" settle based on the same facts. However, the control contract settles immediately after the election results are clear, while the seat number contracts settle based on the official congressional composition on February 1, 2027. If a resignation, party switch, or disputed recount occurs during this period, the settlement benchmarks for the two may diverge, leading to maturity returns deviating from $8.

Fourth is capital occupation and opportunity costs. The majority of the capital used for building positions will be forcibly locked in the seat number contracts until the final settlement on February 1, 2027, leading to significant interest costs over the months.

Profound Implications for Prediction Market Participants

The structural gaps exposed in the Senate event are not isolated cases but rather common structural features present during the rapid expansion of the entire prediction market. The monthly trading volume of the entire prediction market has explosively grown from less than $5 billion in September 2025 to about $24 billion in April 2026, accompanied by unprecedented market fragmentation:

Single topics give rise to a massive number of contracts. According to Talos CM prediction market reference data, there are currently over 880,000 active event contracts on the Kalshi platform. From election politics, Federal Reserve decisions, macroeconomic data to sports events, each macro theme has been fragmented into win/lose, ranges, thresholds, and various combinations of contracts.

Multiple trading venues coexist in a fragmented manner. Platforms like Kalshi and Polymarket often launch their competing contracts for the same objective event, but there are subtle differences in contract text definitions, trading fees, and settlement rules.

Extreme disparities in capital depth. As demonstrated by the Senate data, mainstream headline contracts absorb the vast majority of trading funds, while the liquidity of derivative related contracts is often only a fraction of that. This depth scarcity of weak order books is a breeding ground for significant price deviations, but it is also a killer that quickly erodes arbitrage profits due to execution slippage.

For quantitative market makers and hedge funds, systematically capturing such pricing gaps depends on building capabilities in two dimensions:

First is precise event identification capability. Rapidly uncovering contracts linked to the same objective outcome across different platforms and contract matrices, converting them to a unified benchmark, and accurately calculating the real executable prices including fees and depth. This requires establishing a standardized event map across platforms, thoroughly reviewing settlement terms to prevent factual conflicts, and real-time access to high-frequency order books and fee models.

Second is precise trading capture capability. The theoretical opportunities on the screen do not equate to real money in the account. In practice, this requires algorithms to patiently place passive orders on the most liquid main legs, and then quickly penetrate the eight shallow order books to complete hedging before prices drift, while precisely controlling slippage and order execution fee losses.

This rigorous analytical paradigm is also applicable to any significant prediction event with multiple contracts or cross-platform coexistence.

Conclusion

In public perception, prediction markets are often simply viewed as a black box that provides a unique probability for each question. However, Kalshi's Senate contracts profoundly reveal that the essence of prediction markets is actually a collection of interlinked but independently operating order books.

These order books, while generally following the same macro narrative, consistently exhibit cognitive friction of two to three percentage points in localized areas. It is thought-provoking that even against the backdrop of explosive trading volume growth, these gaps have not been completely closed. In several trading days this month, the aggregation of these small price differences successfully evolved into mathematically arbitrage opportunities with positive expectations.

The 2026 midterm elections are just a standard sample in this wave. As the scale of prediction markets continues to expand and the ecosystem becomes increasingly fragmented, similar opportunity windows will continue to emerge across various fields. To successfully venture into this track, institutional participants need to possess two core foundations: the ability to gain insights from underlying data assets to discover opportunities, and a calm and rational cognitive framework regarding real execution costs to make prudent bets.

Research Methodology: This study is based on daily K-line data of 38 Senate-related contracts obtained from Kalshi's public API from August 1, 2026, to October 25, 2026, with trading days defined according to Eastern Time. All contracts included in the package were officially listed on Kalshi before August 2026. Implied probabilities are calculated based on the mid-price between the end-of-day bid and ask prices, and seat numbers and power balance probabilities have been normalized through tiered summation to eliminate excess premiums caused by bid-ask spreads.

The cost of the package for YES legs uses the ask price, while for NO legs, it uses 1 minus the bid price. The nominal trading volume is the number of contracts traded multiplied by the price. The contract universe is sourced from the Talos CM prediction market reference database. All deduced results are theoretical simulations and do not include additional slippage beyond the quoted prices, nor do they constitute any investment advice.