pmxt-counterfactual-replay-v1/v1.0 · CC-BY-4.0

PMXT Counterfactual Replay v1 — Engine and Resolution-Zone Protocol Comparison

Per-(engine/mechanic, leverage, class) summary statistics from two counterfactual simulation experiments on 13,000+ resolved Polymarket binary-event markets. E2 compares three margin-engine variants on survivability and drawdown; E3 compares four resolution-zone mechanics (R0–R3) on final-hour liquidation rates, bad-debt frequency, and average PnL. Bundle 2 of the PMXT family; companion to Paper 1 of the Event-Linked Perpetuals research programme.

GitHub repository →

Per-(engine/mechanic, leverage, event class) summary statistics from two counterfactual simulation experiments run on the resolved-market subset of the pmxt-stylized-facts-v1 corpus. Released as the empirical companion to Paper 1 of the Event-Linked Perpetuals research programme. DOI: 10.5281/zenodo.20108387

Experiments

E2 — Margin formula recalibration (CC-007b) Three engine variants compared on survivability, drawdown, and trader PnL across leverages 1×–10×.

Engine	Description
E0	Baseline — constant volatility estimate, no dynamic margin
E1	EMA-vol — exponential moving average vol, dynamic margin
E2	Recalibrated — vol term without index-price multiplier (CC-007b)

E3 — Resolution-zone protocol comparison (CC-008) Four resolution-zone mechanics compared on final-hour liquidation rates, bad-debt frequency, and average PnL across leverages 2×–10×.

Mechanic	Description
R0	Naive forced expiry — no resolution-zone rules
R1	Leverage compression only — max leverage ramps to 1× over final hour
R2	R1 + boundary funding correction when \|mid − 0.5\| > 0.10
R3	Full multi-stage halt — zone compression, notional taper, position freeze, circuit-breaker

Key results

Experiment	Metric	Result
E2 (L=5×)	Liquidation rate reduction E2 vs E0	−5.96%
E2 (L=5×)	Pre-registered floor (≥10% reduction)	FAIL
E3 R3 (L=5×, all classes)	Final-hour liquidation vs R0	−80.4%
E3 R3	Pre-registered floor 1 (liq reduction ≥50%)	PASS
E3 R3	Pre-registered floor 2 (bad-debt reduction ≥75%)	FAIL (−2.4%)

Corpus

Attribute	Value
E2 markets	13,306
E3 markets	13,115
Snapshot	2026-04-21 to 2026-04-27 UTC
E2 simulation run	2026-05-08
E3 simulation run	2026-05-09
Build run	2026-05-10

Files

File	Rows	Description
`data/e2a-survivability-v1.parquet`	75	E2a: survivability and liquidation rate by engine × leverage × class
`data/e2b-drawdown-v1.parquet`	15	E2b: total drawdown by engine × class
`data/e2c-pnl-distribution-v1.parquet`	3	E2c: synthetic trader PnL percentiles by engine
`data/e3-liquidation-rate-v1.parquet`	80	E3: final-hour liquidation rate by mechanic × leverage × class
`data/e3-bad-debt-v1.parquet`	80	E3: bad-debt event frequency
`data/e3-pnl-v1.parquet`	80	E3: average terminal PnL (USDC)
`data/aggregates.json`	—	Key summary statistics
`data/engine-parameters-v1.json`	—	Locked engine and mechanic parameters

Related datasets (PMXT family)

Bundle	DOI	Contents
Bundle 1: pmxt-stylized-facts-v1	10.5281/zenodo.20107449	13,314 markets, SF1–SF9
Bundle 2: pmxt-counterfactual-replay-v1	10.5281/zenodo.20108387	This dataset
Bundle 3: pmxt-behavioral-clusters-v1	10.5281/zenodo.XXXXXXXX	43,116 markets, behavioral clusters

Citation

@dataset{nechepurenko_pmxt_counterfactual_replay_v1_2026,
  author    = {Nechepurenko, Maksym},
  title     = {{pmxt-counterfactual-replay-v1: Engine and Resolution-Zone Protocol
                Comparison Outputs on Polymarket Binary-Event Markets}},
  year      = {2026},
  publisher = {Zenodo},
  version   = {1.0.0},
  doi       = {10.5281/zenodo.20108387},
  url       = {https://doi.org/10.5281/zenodo.20108387}
}