Mapping Prediction Markets to Securities: A Cross-Platform Event-Security Framework

We build a cross-platform framework that links prediction-market events to the securities they move. The system maps 300+ active events to 500+ affected securities, drawing probabilities from Polymarket, Kalshi, Limitless, and Metaculus. Probabilities are refreshed at 5-minute resolution, with history reaching back to June 2022.

Each event is mapped to between two and eight affected tickers, every link carrying an exposure type (positive or negative) and a sensitivity score from 0 to 1. Probability time series are stamped point-in-time at trade-execution timestamps and never restated, so the panel is free of look-ahead bias. On 500+ resolved events with probability moves greater than 10%, prediction-market probability changes tend to lead the mapped security price moves by hours to days.

Prediction markets price the probability of discrete future events directly, in real time, with money at stake. When a tariff becomes more likely, a deal more likely to close, or a rate cut more likely to land, that shift in odds shows up on a prediction market before it is fully reflected in the affected equities. The market context is no longer marginal: Polymarket and Kalshi combined for roughly $44.5B in volume in 2025.

The opportunity is also the hard part. Prediction markets and securities live in separate vocabularies. A market resolves on "Will the Fed cut rates in March?"; a portfolio holds banks, REITs, and rate-sensitive names. Nobody publishes the bridge between the two. The contribution of this work is that bridge: a structured, point-in-time mapping from event probabilities to the securities those events affect.

At the core is a taxonomy of event categories, each with a characteristic set of affected securities and exposure directions. An event is classified into a category, then linked to between two and eight tickers. Every link records two things: an exposure type (positive, where the security rises as the event probability rises, or negative, where it falls) and a sensitivity score from 0 to 1 that captures how strongly the security responds to that event.

The sensitivity score is what makes the mapping usable rather than merely descriptive. A tariff event might map to a domestic producer with positive exposure and a high sensitivity, and to an importer with negative exposure and a lower sensitivity. By attaching direction and magnitude to each link, a probability move on a single event resolves into a ranked, signed set of security exposures rather than an undifferentiated watchlist.

The four platforms identify their markets in entirely different ways. Polymarket uses an on-chain condition_id and token_id; Kalshi uses a CFTC-registered ticker; Limitless references a Base-chain contract; Metaculus uses a numeric question_id. Left alone, these are four disconnected silos with no common key.

We normalize all four into one schema. Each native identifier is resolved to a canonical event, so that the same underlying question tracked on more than one venue collapses to a single event with multiple probability sources. This unification is what allows the event-security mapping to sit above the platforms rather than being rebuilt for each one.

Every probability observation is stamped at the trade-execution timestamp at which it was true and is never restated afterward. This is the single most important methodological choice in the system. A probability series that is silently revised, or backfilled with values that were not knowable at the time, will manufacture predictive power that does not exist out of sample.

Because the series is point-in-time, any analysis run against it sees only what a trader could have seen at that exact moment. There is no look-ahead bias. The 5-minute resolution gives a fine-grained view of how odds move intraday, and the history reaching back to June 2022 means the panel spans multiple real event cycles rather than a single recent window.

To check whether the mapping carries information, we tested 500+ historical probability moves greater than 10% in events that have since resolved. The sample spans the 2024 US election cycle, 2025 tariff announcements, and Fed rate decisions. For each move we measured whether the mapped securities subsequently moved in the predicted direction implied by their exposure type.

The finding is qualitative and we state it as such: prediction-market probability changes tend to lead the mapped security price moves by hours to days. The lead is directional rather than instantaneous, which is consistent with prediction markets repricing discrete-event odds faster than the broad equity complex absorbs them. We deliberately do not report a single headline hit-rate, because the relationship varies by event category, by liquidity, and by how cleanly an event maps to its securities.

Three limits bound these results. The lead relationship is reported qualitatively, not as a fixed numeric edge, because it differs across categories and liquidity regimes. The mapping itself is a model of which securities an event affects, and a misclassified or incomplete link will degrade the signal regardless of how clean the probability data is. And thinly traded markets produce noisier probabilities, so not every event carries equal information. Each of these is a property of the domain to be managed, not a defect of the point-in-time method.

Prediction markets price discrete events directly, and those events move securities. By normalizing Polymarket, Kalshi, Limitless, and Metaculus into one schema, mapping 300+ events to 500+ securities with signed exposures and sensitivity scores, and holding the entire probability history point-in-time, we turn a scattered set of betting venues into a structured, look-ahead-free signal on the securities those events affect. The backtest on 2024-2025 events shows that probability changes tend to lead the mapped price moves by hours to days, and we hold that finding to an honest, qualitative standard.