Arbitrage Scanner Wiki

01 — What Is Arbitrage?

Arbitrage is the practice of exploiting price differences for the same asset across different markets. In prediction markets like Polymarket, arbitrage occurs when the prices of mutually exclusive outcomes are mispriced — their probabilities don’t add up correctly.

Think of it like concert tickets: if a venue has 3 sections and you can buy a ticket to every section for less than the guaranteed payout, you profit no matter which section ends up being “the winner.” In prediction markets, the “sections” are outcomes (candidates, teams, results) and the “ticket price” is the YES share price.

A sports betting analogy: if a sportsbook offers odds on a tennis match where the implied probabilities of both players winning sum to less than 100%, you can bet on both and guarantee a profit. This is exactly what our scanner detects in prediction markets.

02 — The Three Types

03 — Formal Definitions

From “Unravelling the Probabilistic Forest” (Saguillo et al., 2025)

|1 - Σᵢ p(oᵢ)| > τ   where τ = minimum deviation threshold

If Σ < 1: LONG signal → buy all YES shares (cost < $1, payout = $1)
If Σ > 1: SHORT signal → buy all NO shares (cost < $1, payout = $1)

D(E₁, E₂) = 0.4 × J(tags₁, tags₂) + 0.6 × J(kw₁, kw₂)

where J(A, B) = |A ∩ B| / |A ∪ B|  (Jaccard similarity)
      kw = extracted title keywords (stopwords removed)

04 — Algorithm Pseudocode

Algorithm 1: Market Rebalancing

function detectRebalancing(events, config):
  for each event E in events:
    if not E.negRisk or |E.markets| < 2: skip

    midpoints ← [getYesMidpoint(m) for m in E.markets]
    if any midpoint > config.maxPosition: skip

    sum ← Σ midpoints
    deviation ← |1 - sum|
    if deviation < config.minDeviation: skip
    if deviation < config.minProfit: skip

    direction ← sum < 1 ? LONG : SHORT
    orders ← generate buy orders for all markets
    emit Signal(direction, deviation, orders)

Algorithm 2: Combinatorial Arbitrage

function detectCombinatorial(events, config):
  for each pair (E₁, E₂) in events:
    if not overlapping dates: skip

    score ← D(E₁, E₂)  // dependency score
    if score < config.minJaccardSimilarity: skip

    for each pair (m₁ ∈ E₁, m₂ ∈ E₂):
      qSim ← J(keywords(m₁.question), keywords(m₂.question))
      if qSim < 0.4: skip

      diff ← |price(m₁) - price(m₂)|
      if diff < config.minProfit: skip

      direction ← price(m₁) < price(m₂) ? LONG : SHORT
      emit Signal(direction, diff, cross-market orders)

05 — System Architecture

The scanner is a Next.js application with server-side API routes that proxy requests to Polymarket’s two APIs (Gamma for market discovery, CLOB for order books). The arbitrage engine runs server-side to keep API keys secure and reduce client-side compute.

06 — Key Findings

Statistics from the paper’s analysis of Polymarket data (2020–2024):

07 — Assumptions & Limitations

08 — Layman’s Guide