Polymarket Up/Down Complement Discount
Paired-outcome complement discount
Inconclusive, strongest candidate — 1,716 first-seen paired prints implied +2.82% average after stated fees and held up chronologically, but historical prints do not prove simultaneous two-leg fills. Live atomic paper execution is required.
- Category
- Prediction markets
- Window
- 2026-05-27 → 2026-06-24
- Instruments
- Polymarket crypto Up/Down
- Timeframe
- 15-minute event markets
- Tested
- 2026-06-24
Measured equity history
Recorded directly from the chronological test ledger.
Shown in the backtest's native equity or cumulative P&L units. It is not scaled to an investment amount because the published artifact does not provide a defensible capital denominator.
How it works
BackfillingThe bet
What market behavior this strategy is wagering on.
Plain-English explainer is backfilling — check back tomorrow. We publish the explanation only once it's written and reviewed, never auto-filled.
How it decides
What makes it enter, size, and exit a position.
Plain-English explainer is backfilling — check back tomorrow. We publish the explanation only once it's written and reviewed, never auto-filled.
How it can break
The regimes and failure modes that turn the edge negative.
Plain-English explainer is backfilling — check back tomorrow. We publish the explanation only once it's written and reviewed, never auto-filled.
Gate scorecard — 3 / 11
auto-imported from results.json| # | Gate | Result | Pass |
|---|---|---|---|
| 01 | Minimum sample | 1,716 first qualifying paired-print events. | ✓ |
| 02 | Profit factor ≥ 1.20 | Lossless theoretical settlement makes PF undefined; mean gross return is the relevant screen metric. | ✗ |
| 03 | Sharpe ≥ 0.6 | Not reported for overlapping binary markets. | ✗ |
| 04 | Max drawdown ≤ 12% | No simultaneous-fill or capital-lockup model. | ✗ |
| 05 | Positive ≥ 60% of periods | Only four weeks of data. | ✗ |
| 06 | Bootstrap LB Sharpe > 0 | Holdout mean-return interval is +2.43% to +3.27% under the paired-print assumption. | ✓ |
| 07 | Placebo beats p95 | Not run. | ✗ |
| 08 | 2× cost stress PF > 1.0 | Two cents additional slippage per leg produces -1.25% mean return. | ✗ |
| 09 | Deflated Sharpe positive | Not evaluated. | ✗ |
| 10 | No component > 40% | Not evaluated by coin and week. | ✗ |
| 11 | Walk-forward OOS ≥ 0.9× IS | Holdout mean +2.83% closely matches training +2.82%. | ✓ |
VERDICT: INCONCLUSIVE — APPARENT EDGE NEEDS SIMULTANEOUS-FILL PROOF
Buying both Up and Down for a combined cost below $1 guarantees a $1 settlement payout. This screen paired the first actual Up and Down taker-buy prints within one second whose prices plus stated fees totaled less than $1. It deliberately avoids choosing the cheapest later pair with hindsight.
The screen found 1,716 events. The average implied return was +2.82%; the untouched 680-event holdout averaged +2.83%, with a 95% bootstrap interval of +2.43% to +3.27%. Median available paired size was only about five shares.
| Sample | Events | Mean implied return | Median implied return |
|---|---|---|---|
| Training | 1,036 | +2.82% | — |
| Holdout | 680 | +2.83% | — |
| Combined | 1,716 | +2.82% | +1.53% |
This is the strongest candidate, but it is not yet validated. Two historical prints within one second do not prove that one trader could fill both legs before either quote moved. An added 1¢ per leg leaves only +0.74% average and makes the median trade negative; 2¢ per leg reduces the mean to −1.25%.
The required next test is a live atomic paper executor that records both IOC outcomes, partial fills, latency, fees, and stranded-leg losses. Advance only if at least 200 paired attempts retain a positive lower confidence bound after every failed-leg cost.
Screened 1.12 million taker-buy prints across 7,267 15-minute markets from 2026-05-27 through 2026-06-24.
Frequently asked
Is Polymarket Up/Down Complement Discount profitable in 2026?
It is currently running through the 11-gate battery; the verdict will be published on this page, pass or fail.
Has Polymarket Up/Down Complement Discount been backtested honestly?
Yes — through The Validation Gauntlet, a pre-registered 11-gate framework (profit factor, deflated Sharpe, a random-permutation placebo, cost-stress and walk-forward) with the specification locked before any out-of-sample metric is computed.
Methodology: 11-gate validation — pre-registered spec, 11-gate battery, real market data. Full reproducible report: backtests/polymarket_complement_arbitrage/REPORT.md in the source repository.Author: Validated Research Team (Methodology v1.0 — 11-gate validation). Backtests are not investment advice.