5 Polymarket Strategies Backtested on Real BTC, ETH, and SOL Books
Threshold entry, mean reversion, time-before-resolution, condition_based, and logical arb only — backtested on real walk-the-book fills with real fees. Which survive the slippage, which don't.
Posted by
BenRelated reading
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.
Paper-Trade Polymarket Without Risking Capital (and Without Lying to Yourself)
The honest way to paper-trade Polymarket: live snapshots, walk-the-book fills, real fees, and a paper-vs-backtest baseline that catches when reality stops matching the simulation.
We have five strategy primitives in the PolyQuantLab backtest engine. They're not the "best strategies" — they're the orthogonal patterns that show up when you stare at enough Polymarket order books and ask "what kind of signal could plausibly cross the spread plus fees?"
Each one below is a single sentence describing the mechanism, followed by an honest read on whether it actually pays once walk-the-book fills and Polymarket's 2% taker fee are subtracted from the gross PnL. Reproducible: every spec below has a corresponding template in the Strategy Builder.
1. Threshold entry
Mechanism: buy YES whenever YES probability crosses a threshold (e.g. > 0.6), exit on opposite cross or at resolution.
Reality: dies on fees. The implied break-even probability on a $0.65 YES buy with 2% taker is ~0.67. You're buying right at the model's break-even. Without a leading signal, threshold entry on the mid is a 50/50 bet wearing a costume.
Where it works: as a confirmation filter stacked on top of a leading signal (e.g. Binance spot move > 0.4% AND YES > 0.6). Then the threshold isn't the edge — the spot move is.
2. Mean reversion
Mechanism: buy YES whenever it dips X points below its rolling mean; sell when it reverts.
Reality: doesn't fire on the data we have. Polymarket order books don't mean-revert at the microstructure level — aggressive orders punch through stale quotes and the new equilibrium holds. The signal needs history to compute a rolling mean; we don't store enough tick-level history per market to fit a stable estimator. Paper trader logs zero entries for this strategy on most days.
Verdict: kept in the registry for completeness; not in any template you should actually trade.
3. Time before resolution
Mechanism: at τ = T minutes, buy whichever side the underlying is currently above/below the strike on.
Reality: the simplest naive "directional" strategy and surprisingly profitable at τ = 60s with strict event_type filters. On BTC 15m markets with τ = 90s entry, this strategy logs ~62% win rate net of fees in walk-book mode — small per-trade edge, decent frequency.
Risk: all the win rate comes from low-vol windows. When Binance prints a 0.5% candle in the last 60 seconds, you're on the wrong side and pay the spread twice (no time to exit).
4. Condition-based (the composable one)
Mechanism: an arbitrary AND/OR tree of primitive conditions (token price thresholds, spot moves, spread depths, time-to-resolution). The most expressive of the five.
Reality: this is what most users actually end up running once they spend an hour in the Builder. The generated Python code is documented and can be lifted into a live bot.
Worked example. Entry: (Binance ETH 5-min return > 0.4% within the last 3 min) AND (YES probability < 0.55) AND (τ < 4 min). Exit: YES > 0.85 OR τ = 0. On the last 200 ETH 15m markets, walk-book + fee-adjusted: ~58% win rate, ~$0.014 net PnL/share. Modest but positive, vs textbook threshold entry which prints negative PnL on the same universe.
5. Logical arb only
Mechanism: buy YES + NO simultaneously when their walk-book asks sum below $1; hold to resolution.
Reality: the only strategy in this list with a 100% win rate by mathematical construction. Limited frequency (5–15 fires/day across all coins and horizons), small per-trade edge (typically 1–5¢ after fees on a $1 outlay = 1–5% return per pair).
Capital efficiency: each pair locks $1 of capital for up to the full market duration (5 min to 24h). Annualised on a 5-minute market at 3¢ edge: ~$31,500 theoretical, but the capital constraint is real because Polymarket isn't margined.
What we're NOT publishing
We have a sixth template that's consistently profitable across all coin/event_type combinations, with low drawdown and reasonable frequency. We don't publish that one here because (a) doing so kills its alpha and (b) it's the strongest argument for upgrading from Free to a paid tier. Pro tier and above can see the spec; Premium can paper and live-trade it directly.
We mention it because this category of post tends to claim to publish "the best strategy" and never does. We're being explicit: there's a better strategy we intentionally aren't including, and the Builder documents exactly where to find it.