💰 quantitative finance

Toward Black Scholes for Prediction Markets: A Unified Kernel and Market Maker's Handbook

This paper proposes a unified logit jump-diffusion framework with a risk-neutral drift that serves as a Black-Scholes analogue for prediction markets, enabling standardized quoting, hedging, and the creation of a coherent derivative layer for belief risk through a robust calibration pipeline.

Original authors: Shaw Dalen

Published 2026-04-07

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Shaw Dalen

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a world where you can bet on anything: "Will the unemployment rate hit 5%?" "Will a specific crypto upgrade happen by next Tuesday?" "Will Candidate X win the election?"

This is the world of Prediction Markets (like Polymarket). Right now, these markets work like a chaotic bazaar. Everyone is shouting prices, but there's no standard rulebook for how those prices should move when news breaks. If a surprise headline drops, prices jump wildly, and the people making the market (the "Market Makers") get burned because they don't have a map to navigate the chaos.

This paper proposes a new rulebook to bring order to the chaos. It's called the RN-JD Model, and here is how it works, explained simply.

1. The Problem: The "Black Box" of Prediction

Think of the stock market. If you want to bet on a stock, you have a famous tool called the Black-Scholes model. It's like a universal translator. It tells you exactly how much a stock option should cost based on how "jumpy" (volatile) the stock is. Because everyone uses this same translator, traders can buy and sell "fear" or "uncertainty" easily.

Prediction markets lack this translator.

The Issue: In prediction markets, the "price" is a probability (e.g., 60% chance of rain). But probabilities are weird. They can't go below 0% or above 100%.
The Consequence: When news hits, the price doesn't just wiggle; it teleports. Market makers are terrified of these jumps. They charge huge fees (spreads) to protect themselves, which makes the market expensive and slow for everyone else.

2. The Solution: The "Logit Jump-Diffusion"

The authors propose a new mathematical engine to fix this. Let's break down the fancy name into a simple metaphor.

The "Logit" Part: Unfolding the Box

Imagine a rubber band stretched between 0 and 1. If you try to stretch it further, it snaps. That's a probability.
The authors say: "Let's stop stretching the rubber band. Let's unfold it onto an infinite straight line."

The Metaphor: They take the probability (0 to 1) and stretch it out into "Log-Odds" (a number line from negative infinity to positive infinity).
Why? On this infinite line, standard math tools work perfectly. They can model smooth movements (diffusion) and sudden jumps without the math breaking.

The "Jump-Diffusion" Part: The Smooth Ride vs. The Earthquake

The model assumes two things happen to the market:

Diffusion (The Smooth Ride): Small, constant updates as people slowly change their minds. Like a car driving down a highway.
Jumps (The Earthquake): Sudden, massive shifts when big news breaks (e.g., a candidate drops out). Like an earthquake shaking the car.

The genius of this paper is that it separates these two. It says, "Okay, we know the market will shake when news hits. Let's build a model that expects the earthquake and prices it in before it happens."

The "Risk-Neutral Drift" Part: The Fair Coin

In finance, there's a concept called "No Free Lunch." If you know the future, you shouldn't be able to make infinite money.
The authors enforce a rule: The market price must be a "fair game."

The Metaphor: Imagine a coin flip. If the market says there is a 60% chance of heads, and you wait, the price shouldn't magically drift to 70% just because time passed. It should only move if new information arrives.
By forcing the math to respect this "fairness," they eliminate the bias that usually tricks market makers.

3. The New Toolkit: Trading "Fear" and "Surprise"

Once you have this new engine, you can build a whole new layer of products, just like the stock market has "Volatility Swaps."

Belief Volatility Swaps: Instead of betting on who wins, you can bet on how much the opinion will swing.
- Analogy: Imagine you are a weather forecaster. You can sell a contract that pays out if the temperature swings wildly between morning and noon, regardless of whether it ends up hot or cold. This lets market makers hedge their risk.
Correlation Swaps: If you are betting on two related events (e.g., "Will Team A win?" and "Will Team B win?"), you can bet on how much they move together.
Corridor Variance: You can bet on volatility only when the probability is in the "swing zone" (between 30% and 70%), where the action is hottest.

4. The Market Maker's Handbook

The paper also gives a "User Manual" for the people running these markets.

The Recipe: "When you see a sudden spike in order flow (toxicity), widen your prices. When you see a scheduled news event, prepare for a jump. When you are holding too many contracts, sell some 'volatility swaps' to offload the risk."
The Result: Market makers can now offer tighter prices (cheaper bets for you) because they have a way to sell their risk to someone else, rather than holding it until they get burned.

5. Why This Matters Now

Prediction markets are exploding. Big institutions (like the NYSE's parent company) are investing billions. But without a standard language, these markets are risky and inefficient.

This paper provides the standard language.

Before: "I think this will happen, but I'm scared of the news, so I'll charge you a huge fee."
After: "I know the news might cause a jump. I've priced the 'jump risk' into a separate contract. So, I can give you a fair price on the main bet."

The Bottom Line

Just as the Black-Scholes model turned options trading into a massive, liquid global industry by giving everyone a common language for "volatility," this paper aims to do the same for prediction markets.

It turns "belief risk" (the fear that opinions will change) into a tradable commodity. It allows the market to absorb shocks, keep prices stable, and let everyone—from the casual bettor to the giant hedge fund—participate in a fair, liquid, and efficient marketplace for the future.

1. Problem Statement

Prediction markets (e.g., Polymarket, Kalshi) aggregate dispersed information into tradable probabilities for binary events. However, unlike options markets which rely on the Black–Scholes (BS) framework, prediction markets lack a unifying stochastic kernel.

The Gap: Market makers face "belief risk" (movements in the implied probability $p_t$ ), volatility risk, jump risk (from news), and cross-event correlation risk without standardized tools to quote or hedge them.
Current Limitations: Existing mechanisms (LMSR, CP-AMMs, Order Books) determine prices but do not provide a shared model for how probabilities evolve over time or across related events. This leads to wide spreads, toxic flow exposure, and difficulty managing inventory near resolution boundaries (0 or 1).
Goal: To establish a "Black–Scholes for Prediction Markets"—a tractable, standardized language for quoting, hedging, and transferring belief risk, analogous to implied volatility surfaces in equity options.

2. Methodology: The RN-Jump-Diffusion Kernel

The authors propose a Logit Jump-Diffusion model with Risk-Neutral (RN) Drift.

A. State Variable Transformation

Instead of modeling the probability $p_t \in (0, 1)$ directly, the model maps it to the real line using the logit function:
$x_t = \text{logit}(p_t) = \ln\left(\frac{p_t}{1 - p_t}\right)$
This transformation allows the application of standard Itô–Lévy semimartingale tools while respecting the $(0, 1)$ boundaries of the probability space.

B. The Stochastic Process

The dynamics of the log-odds $x_t$ are modeled as a jump-diffusion process under the risk-neutral measure $Q$ :
$dx_t = \mu(t, x_t)dt + \sigma_b(t, x_t)dW_t + \int_{\mathbb{R}} z \tilde{N}(dt, dz)$

$\sigma_b$ (Belief Volatility): The diffusion coefficient representing the volatility of beliefs.
$\tilde{N}$ : A compensated Poisson random measure capturing news-driven jumps.
Risk-Neutral Drift Constraint: Since $p_t$ represents a discounted risk-neutral probability of a binary payoff, it must be a $Q$ -martingale. This constraint uniquely pins down the drift $\mu(t, x_t)$ :
$\mu(t, x) = -\frac{1}{S'(x)} \left[ \frac{1}{2}S''(x)\sigma_b^2 + \int_{\mathbb{R}} (S(x+z) - S(x) - S'(x)\chi(z)) \nu_t(dz) \right]$
where $S(x) = \frac{1}{1+e^{-x}}$ is the sigmoid function. This ensures the model is arbitrage-free.

C. Multi-Event Dependence

The framework extends to multiple events by introducing:

Diffusive Correlation: Correlated Brownian motions ( $\rho_{ij}$ ) between events.
Co-Jumps: A joint Lévy measure $\nu_{ij}$ capturing simultaneous jumps across related events (e.g., election results for different candidates).

3. Key Contributions

A. The Derivative Layer

The paper defines a coherent menu of derivatives to hedge specific risk factors, analogous to variance and correlation swaps in equity markets:

Belief Variance Swaps: Exchange realized quadratic variation of $x$ (or $p$ ) for a fixed strike.
Correlation Swaps: Hedge cross-event exposure (diffusive and co-jump).
Corridor Variance: Accrues variance only when $p_t$ is in a "swing zone" (e.g., $[0.35, 0.65]$ ), targeting the most toxic liquidity areas.
First-Passage Notes: Payoffs triggered if $p_t$ crosses a threshold before resolution, hedging gap risk.

B. Calibration Pipeline

A data-driven pipeline is proposed to map raw market data (mid/bid-ask/trades) to a stable Belief-Volatility Surface $\sigma_b(\tau, m)$ :

Filtering: Uses a heteroskedastic state-space filter (Kalman Filter) to separate the latent efficient logit signal from microstructure noise.
Jump Separation: Employs an Expectation-Maximization (EM) algorithm to distinguish between diffusion increments and jump increments, rather than using ad-hoc thresholds.
Drift Enforcement: Re-smooths the path to strictly enforce the Risk-Neutral martingale condition.
Surface Construction: Fits a smooth surface over time-to-resolution ( $\tau$ ) and moneyness ( $m$ ) using penalized splines with shape constraints (non-negativity, edge stability).

C. Market-Maker's Handbook

The paper provides practical operational guidelines for market makers:

Greeks in Logit Space: Defines Delta, Gamma, and "Belief-Vega" ( $\partial V / \partial \sigma_b$ ) in the $x$ -domain.
Inventory-Aware Quoting: Adapts the Avellaneda-Stoikov model to logit units, adjusting spreads based on inventory, belief volatility, and toxicity.
Hedging Strategies:
- Calendar Hedges: Using variance strips to manage exposure around scheduled news.
- Cross-Event $\beta$ -Hedges: Neutralizing correlation and co-jump risks using correlation swaps.

4. Experimental Results

The authors conducted end-to-end experiments using synthetic RN-consistent paths and real Polymarket data.

Task: Forecasting short-horizon (60s) realized logit variance.
Baselines: Random Walk (constant $\sigma$ ), Logit Diffusion, Wright-Fisher/Jacobi diffusion, and AR-GARCH models.
Performance: The proposed RN-JD model achieved the lowest Mean Squared Error (MSE) and QLIKE loss compared to all baselines.
- Key Finding: Models lacking the RN drift constraint or jump separation (e.g., standard diffusion) failed to capture volatility bursts near information shocks.
- Stability: Boundary-respecting models (Wright-Fisher) mapped to logit space resulted in numerical instability (massive MSE inflation), whereas the RN-JD model remained stable.
Interpretation: The gains stem from enforcing the martingale property (eliminating systematic bias) and the EM-based separation of diffusion and jumps, allowing for adaptive volatility forecasting.

5. Significance and Impact

Standardization: Just as implied volatility standardized the options market, this Belief-Volatility Surface provides a common language for prediction markets. It allows liquidity to concentrate around a few tradable risk factors (volatility, jump intensity, correlation) rather than individual event prices.
Risk Transfer: It enables market makers to offload "belief risk" (volatility and jump exposure) rather than holding it, reducing adverse selection costs and allowing for tighter spreads.
Institutionalization: With major institutional interest (e.g., ICE's investment in Polymarket) and rising volumes, this framework provides the necessary infrastructure for scaling prediction markets into a mature asset class.
Practical Utility: The paper moves beyond theory to provide a "handbook" for implementation, including specific algorithms for calibration, quoting, and hedging, making it actionable for quantitative desks.

In summary, the paper bridges the gap between prediction markets and traditional derivatives by providing a mathematically rigorous, empirically validated, and practically implementable framework for pricing and hedging event risk.