This post is an excerpt from the white paper. It reproduces the abstract and introduction (section 1) word for word, with section headings and one pull quote added for reading on the web and the introduction’s footnote left out. The full paper, with data, calibration and results, is linked at the end.
Abstract
In this paper, we design and calibrate a shortfall pricing risk engine that enables solvent lending against collateralized prediction market positions. Prediction markets have experienced breakout success, yet locked capital and bounded returns leave substantial value on the table. Freeing this capital would strengthen informational quality by letting informed traders redeploy across markets, improve execution and retail profitability through deeper books, and open the door to institutional derivative products. However, because positions are bounded on [0, 1], reprice discontinuously, and exhibit especially heavy tails, solvency cannot be guaranteed from liquidation alone, rendering existing decentralized credit solutions unviable. We take a different approach by building an engine, calibrated on historical Polymarket data, that generates an at-origination-time loss distribution from market state features, enforces fixed-duration epochs, and prices lender shortfall ex ante through a Wang distortion. In a dual-pool Monte Carlo simulation on held-out market data, the risk engine pool earned a 92.5% annualized return versus −27.2% for a flat-rate baseline, limited max drawdown to 3.6% versus 85%, originated roughly twelve times as many loans, and halved the liquidation rate, with state-contingent premiums ranging from 63 bps to 4,000 bps per epoch under the most conservative lending parameters.
Why a credit layer
Prediction markets have emerged as powerful belief engines, turning monetary instruments into positions that innately carry predictive power (Hanson, 2003; Arrow et al., 2008). Their impact continues to grow as these platforms expand into increasingly niche subjects and broader cultural domains. For instance, their effect on political science and prevailing polling theory has already begun to reshape the discipline in the wake of the 2024 U.S. presidential election (Clinton and Huang, 2025).
As demand for trading these positions has skyrocketed, so has the cost of capital lockup. A trader who enters a position in a long-horizon market ties up cash in an asset with capped upside, no interim yield, and resolution risk. The longer the horizon, the larger the opportunity cost of holding that position rather than redeploying capital elsewhere. This pushes liquidity toward short-dated events, or simply off the platform, and leaves many longer-dated books thin. A credit layer that lets traders borrow against locked positions would address this directly, and the case for building one extends well beyond simple capital efficiency.
First, letting informed traders unlock and redeploy capital allows them to express and update views more frequently across more markets, strengthening the informational quality that makes prediction markets valuable (Wolfers and Zitzewitz, 2004). Second, thin books worsen execution for all participants; Della Vedova (2026) show that retail traders who correctly forecast outcomes still lose money because they arrive late to illiquid books and pay unfavorable prices, a problem that deeper liquidity directly mitigates. Third, the absence of a functioning credit layer blocks derivative products, hedges, and secondary instruments that traditional asset classes take for granted. Without the ability to borrow against positions, prediction markets remain a spot-only venue, and the full composability required for institutional participation cannot develop.
Beyond holding rewards
Existing platforms have attempted to solve the capital lockup problem by paying position holders and market makers “holding” and “liquidity” rewards from treasuries. These programs serve better as band-aids than sustainable long-term solutions. Locked capital still remains inefficient, reward rates are arbitrarily calculated, and their net contribution to market quality is still uncertain.
This paper proposes a direct fix: lending against prediction-market positions. A trader posts their position as collateral, borrows cash against it, and gets to keep the upside exposure while regaining liquidity. If such borrowing can be underwritten safely, many positions stop being dead capital, and liquidity can then recycle through the rest of the market.
Why liquidation alone fails
Standard lending solutions rely on the lending platform’s operational ability to liquidate underwater collateral in time. This makes them inherently optimistic. To this end, platforms enforce conservative loan-to-value (LTV) ratios, both at origination and for liquidation. These provide an additional safety margin and make lending over-collateralized. These solutions work because collateral price movements are generally smooth; losses typically occur over long enough time frames, and automated-market-makers (AMMs) and/or deep books help ensure that most price levels on the downward path likely present some opportunity to execute before reaching the solvency line. Thus, protocols shift their executional burden to after the adverse movement has already happened. The objective for these optimistic solutions is ex post: ensure solvency once a drawdown has occurred (Merton, 1976; Gould et al., 2013).
This does not work for prediction markets; the obstacle being that these positions are not ordinary collateral. Their prices are bounded on [0, 1], they often reprice in jumps rather than smooth paths due to how information arrives, and they exhibit excessive regime switching with esoteric, market-specific pricing dynamics. Unlike traditional lending, there is no optimistic solution; downward price movements are nearly instantaneous as informational shocks almost always precipitate a liquidity vacuum, followed by a brief period where the market attempts to find a new equilibrium. Once a drawdown has occurred, it is already too late. Thus, our objective must be ex ante: ensure solvency before the drawdown has occurred.
Once a drawdown has occurred, it is already too late.
The risk engine
Our solution is a calibrated risk engine that prices epoch shortfall risk directly. Instead of trying to model the full lifetime value of a loan with a single global process, we break the problem into set-length underwriting horizons, or epochs (Rasmussen, 2025). The risk engine is fed an at-origination-time state-conditional feature vector and estimates the conditional loss distribution of collateral over each epoch. It then prices that loss with a Wang distortion, and produces a premium that scales when tail risk or execution risk worsens.
Section 2 states the lending problem and the pricing setup. Section 3 summarizes the data and calibration pipeline. Section 4 presents the state-contingent premium schedule and the engine’s performance in a dual-pool Monte Carlo comparison against a baseline flat-rate model.
Read the full paper
Sections 2 to 4 cover the lending problem and pricing setup, the data and calibration pipeline, and the premium schedule and Monte Carlo results. Unlocking Liquidity on Prediction Markets (PDF) ↗
The protocol built around the engine is described in Introducing Lattica (PDF) ↗.