I scanned 3,000 Polymarket markets and couldn't find a single arbitrage opportunity.
I just finished a live test run, using data entirely from real-time order books:
1) YES + NO in a single market
Across about 3,000 active binary markets, buy both YES and NO, and you’re guaranteed to receive $1 at resolution. The best cost was $1.001, or -0.10%. Every result was negative. Zero opportunities.
2) Full baskets in multi-outcome events
For 1,145 negRisk events, I calculated the VWAP using the ask ladders for all YES legs:
Mississauga mayoral election (7 legs) +5.46% → +0.08% three minutes later
Israel foreign minister pick (15 legs) +4.02% → +0.09%
Who will list a stock first (5 legs) +3.48% → +0.52%
January FOMC (5 legs) +0.95% → +0.06%
The price gaps I found disappeared within three minutes, which suggests bots are watching. The remaining “edge” is just quote noise caused by thin order-book depth; once you account for gas, it’s gone.
3) Cross-platform: Polymarket vs. Kalshi
October FOMC: PM unchanged at 0.94/0.95, Kalshi at 0.93/0.94. A one-tick spread isn’t enough to cover Kalshi’s taker fee of 0.07×p×(1-p).
One apparent “opportunity” of around 23% was actually a question of interpreting the rules: “one more rate hike in 2026” versus the “number of rate hikes in 2026” basket for 1 or more hikes. The former means “one additional hike,” while the latter means the total number for the year. The numbers do line up; it’s not arbitrage. Misunderstanding this could cost you dearly.
Bottom line: By 2026, risk-free arbitrage is basically the domain of market makers and latency bots. Retail traders have only three types of opportunities left, and all require infrastructure:
· Latency capture: monitor via WebSocket and place an order when you see a basket drop to 0.98; no need to predict direction
· negRisk conversion: wait for a news shock to briefly throw a basket out of balance
· Cross-platform inventory hedging: you need capital on both platforms, and the profit is just a few dozen basis points
#预测市场 #Polymarket #Arbitrage
I just finished a live test run, using data entirely from real-time order books:
1) YES + NO in a single market
Across about 3,000 active binary markets, buy both YES and NO, and you’re guaranteed to receive $1 at resolution. The best cost was $1.001, or -0.10%. Every result was negative. Zero opportunities.
2) Full baskets in multi-outcome events
For 1,145 negRisk events, I calculated the VWAP using the ask ladders for all YES legs:
Mississauga mayoral election (7 legs) +5.46% → +0.08% three minutes later
Israel foreign minister pick (15 legs) +4.02% → +0.09%
Who will list a stock first (5 legs) +3.48% → +0.52%
January FOMC (5 legs) +0.95% → +0.06%
The price gaps I found disappeared within three minutes, which suggests bots are watching. The remaining “edge” is just quote noise caused by thin order-book depth; once you account for gas, it’s gone.
3) Cross-platform: Polymarket vs. Kalshi
October FOMC: PM unchanged at 0.94/0.95, Kalshi at 0.93/0.94. A one-tick spread isn’t enough to cover Kalshi’s taker fee of 0.07×p×(1-p).
One apparent “opportunity” of around 23% was actually a question of interpreting the rules: “one more rate hike in 2026” versus the “number of rate hikes in 2026” basket for 1 or more hikes. The former means “one additional hike,” while the latter means the total number for the year. The numbers do line up; it’s not arbitrage. Misunderstanding this could cost you dearly.
Bottom line: By 2026, risk-free arbitrage is basically the domain of market makers and latency bots. Retail traders have only three types of opportunities left, and all require infrastructure:
· Latency capture: monitor via WebSocket and place an order when you see a basket drop to 0.98; no need to predict direction
· negRisk conversion: wait for a news shock to briefly throw a basket out of balance
· Cross-platform inventory hedging: you need capital on both platforms, and the profit is just a few dozen basis points
#预测市场 #Polymarket #Arbitrage