The Favorite-Longshot Bias in Prediction Markets: Evidence from Polymarket

URL SCAN: The Favorite-Longshot Bias in Prediction Markets: Evidence from Polymarket
FIRST LINE: Economics > General Economics

The Dissection

The paper measures how traders price unlikely and highly likely outcomes on Polymarket. Its central finding is real but unstable: longshots lose heavily in raw contract-level analysis, yet become profitable after related contracts are grouped by parent event. The supposed bias is therefore partly produced by the market’s architecture and the analyst’s unit of measurement. The account-level results further weaken the idea of a durable exploitable edge: habitual longshot buyers do not earn materially worse returns, while habitual favorite buyers perform worse.

The Core Fallacy

The paper risks treating a measurable return pattern as a clean behavioral bias before proving that it is not compensation for liquidity, settlement, selection, timing, contract structure, or trading frictions. The sign changes when contracts are grouped differently. That is not a minor statistical footnote; it means “the bias” is not a single stable object.

Relative to the Discontinuity Thesis, the deeper omission is more severe. The paper treats prediction markets as information-processing mechanisms while ignoring their systemic role as financialized uncertainty-management. Even if the market aggregates information well, it does not restore productive participation, wages, or ownership. It is a casino with better metadata, not a replacement for the employment-consumption circuit.

Hidden Assumptions

• Observed transaction returns are a valid measure of irrational pricing rather than a mixture of risk, liquidity provision, timing, and market mechanics.
• Contract-level and parent-event groupings each represent economically meaningful units, despite producing sharply different conclusions.
• Past trader tendencies are informative about future behavior and can be separated cleanly from changing market conditions.
• The sampled accounts and trades represent the market rather than a selected slice of participants and activity.
• A persistent aggregate pattern would remain exploitable after costs, competition, position limits, and execution constraints.
• Better prediction-market measurement has broader economic significance beyond reallocating money among participants.

Social Function

Classification: partial truth, prestige signaling, and ideological anesthetic.

The partial truth is valuable: Polymarket exhibits structured demand, and the favorite-longshot pattern is not automatically universal across contract types. The prestige signaling comes from converting mass speculation into a high-volume empirical object. The anesthetic is subtler: precision about who loses money on event contracts can create the illusion that uncertainty markets represent a durable economic frontier. They do not. They redistribute claims on uncertain outcomes while the underlying system’s productive base is being automated.

The Verdict

This is a competent autopsy of a market pathology, not evidence of a new economic foundation. Its strongest result is that the alleged bias depends materially on aggregation. Its broader implication is harsher: prediction markets are neither reliably efficient nor reliably exploitable, and their existence does nothing to solve the Discontinuity Thesis. They are transition-era financial machinery—useful for verification arbitrage and speculative extraction, but incapable of preserving mass economic necessity once cognitive automation severs labor from income.