Polymarket ETH Up/Down: How Often Does It Match Binance Spot Direction?
Resolved ETH Up/Down markets are an unbiased estimator of direction over their horizon — sometimes. Here's the actual match rate by event_type, and where Polymarket ETH books lead vs lag the spot move.
Posted by
BenRelated reading
Polymarket vs Binance: What 30 Days of Live Arbitrage Audit Show
Logical arbs (YES+NO < $1) win 100% by construction. Endgame model signals win ~95%. Here's the 30-day public audit of every call we made — fills, fees, and realised PnL.
Polymarket BTC Up/Down 5-Min Markets: Walk-the-Book Reality Check
Most BTC up/down backtests assume mid-price fills. The real book is 11¢/89¢ for half the market's life. Here's what walk-the-book actually pays — and why your strategy probably doesn't survive contact with it.
The Real Cost of Trading Polymarket: Fees, Slippage, and Walk-the-Book Math
Polymarket charges 2% taker. That's the easy part. The hard part is walking the book, paying the spread, and discovering your strategy needed a 65% win rate not 51% to break even.
If Polymarket's ETH Up/Down markets perfectly tracked Binance spot, the market YES probability at τ = 0 would equal the realised outcome (0 or 1). It doesn't, quite. The gap between "what the market believed 60 seconds before resolution" and "what Binance actually printed at resolution" is the entire trading opportunity in short-dated prediction markets.
We've persisted every ETH Up/Down resolution outcome since early 2026, alongside the synchronised Binance spot 5-minute klines. This post is what those joined tables say about direction match rates — broken out by event_type so you can see where Polymarket leads, lags, or just disagrees with the underlying.
The four event types
Polymarket runs Up/Down markets on ETH at four horizons:
- 5-minute — high noise, low edge unless you have a sub-second feed
- 15-minute — the sweet spot for endgame signals; vol is high enough to disagree with the market, low enough to be modellable
- 1-hour — deepest book, slowest updates, where logical arbs tend to surface
- Daily — longest dated, dominated by sentiment and option-implied moves
What the data says
15-minute markets are the most informative
On 15-min ETH markets, the YES probability at τ = 60s is correct ~93% of the time when we filter to markets where the underlying had moved > 0.15% in the previous 5 minutes. Below that movement threshold, the book mostly stays at 50/50 and the resolution is a coin-flip from a model standpoint — the market knows it doesn't know.
1-hour markets reward patience
ETH 1-hour markets have ~5× the average book depth of 15-min markets. The spread is tighter (typically 30¢ throughout the market's life rather than 78¢). The trade-off: at τ > 30 min, even a 2% Binance move doesn't move the market price by a full pricing point because there's plenty of vol budget left to retrace.
5-minute markets are mostly noise
Direction match on 5-min ETH markets at τ > 90s is barely better than 55%. Inside the last 60 seconds, when realised vol shows up in actual print, the match rate jumps to ~91%. The edge is real but only available to clients with sub-second execution.
What this means for a trader
The takeaway, in one sentence: trade ETH 15m and 1h, stay away from ETH 5m unless you have low-latency infrastructure. The vast majority of retail-accessible edge in ETH prediction markets lives in those two horizons. 5-min is dominated by liquidity providers and a handful of latency-sensitive bots; daily is dominated by macro views you could express more cheaply on options.