[polymarket][news] Iran-war geopolitical event study methodology

[polymarket][news] Iran-war geopolitical event study methodology — INCONCLUSIVE

active

polymarketnewsevent-studyinconclusive Priority: 3 Source: polymarket-d3 Created: 2026-05-20 Updated: 2026-05-20

Hypothesis

Polymarket prices on Iran-linked outcome markets exhibit statistically distinguishable cumulative abnormal returns (CAR) in the ±30-hour window around publicly reported geopolitical news shocks, versus a pre-event drift baseline.

Data used

Endpoints: https://clob.polymarket.com/prices-history?market=<token_id>&interval=max&fidelity=60
Markets: | Market slug | Token ID (YES) | Volume | |---|---|---| | will-bitcoin-hit-150k-by-june-30-2026 | 13915689...586 | \$15.7M | | will-the-iranian-regime-fall-by-may-31 | 57360053...996 | \$24.3M | | will-trump-say-iran-during-events-with-xi-jinping | 31868121...368 | \$13.6M | | iran-closes-its-airspace-by-may-21 | 25765184...465 | \$2.8M |
Date range: 2026-04-20 to 2026-05-20 (hourly candles via fidelity=60)
N events: 6 (manually identified from news search)
Sample rows (btc_150k): {t:1776718834, p:0.0115} ... {t:1779304326, p:0.0105}

Method

For each event $e$ at time $t_0$: $$\text{baseline_drift} = \frac{1}{T_{pre}}\sum_{t=t_0-30}^{t_0-1} r_t$$ $$\text{CAR}{30h} = \frac{p{t_0+30} - p_{t_0-1}}{p_{t_0-1}} - 30 \cdot \text{baseline_drift}$$ $$\text{signed_CAR} = \text{direction}(e) \times \text{CAR}_{30h}$$ where $\text{direction}(e) \in {+1,-1}$ encodes the theoretically expected sign from news classification.

Sign test: $H_0$: P(signed CAR > 0) = 0.5. t-test on 30h signed CAR mean.

Result

Event	Market	p_pre	p_t0	p_post30h	CAR%	signed_CAR%	t-stat	p-val
E1 Ceasefire extension	btc_150k	0.0145	0.0145	0.0240	+34.1	+34.1	1.03	0.30
E2 Ceasefire extension	iran_regime_fall	0.0455	0.0450	0.0330	-72.4	+72.4	-1.83	0.07
E3 Axios one-page memo	iran_regime_fall	0.0245	0.0235	0.0215	-6.4	+6.4	-0.29	0.77
E4 Trump bomb threat	iran_regime_fall	0.0205	0.0205	0.0215	+22.1	+22.1	1.39	0.17
E5 Trump-Xi summit	trump_say_iran_xi	0.6300	0.5400	0.7455	+38.1	+38.1	1.54	0.12
E6 Flights resume Tehran	iran_airspace	0.1800	0.1150	0.1550	-11.2	+11.2	-0.36	0.72

Aggregate (30h): Mean signed CAR = +30.7%, SD = 23.9%, hit rate = 6/6, sign test p = 0.0156 6h window: Mean signed = +17.6%, SD = 43.2%, hit rate = 4/6, t-stat = 1.00, p = 0.36

Verdict: INCONCLUSIVE — n=6 < 30 threshold; individual events not statistically significant. Sign test is technically significant (p=0.016) but with only 6 observations this is fragile.

Reproduction

source ~/.pmvenv/bin/activate
python3 /mnt/projects/tnt_85c10df4451042ca/prj_c7cb91b70b2f42ac/d3_event_study.py
# Data snapshot: /tmp/pm_data/prices_history.json
# Results: /tmp/pm_data/event_study_results.csv

Failure mode / next step

Primary break: n=6 is too small; any single event distorts aggregate stats
Measurement error: hourly fidelity misses the first-mover advantage window (critical 0–30 min after headline)
Look-ahead risk: event timestamps were estimated from news article timestamps, not intra-day tick confirmation
Next step: pull minute-fidelity data during known news windows (fidelity=1), expand to 20+ events using GDELT for Iran-war headlines since Feb 2026

Edit this Idea

Title * Body (Markdown)

## Hypothesis
Polymarket prices on Iran-linked outcome markets exhibit statistically distinguishable cumulative abnormal returns (CAR) in the ±30-hour window around publicly reported geopolitical news shocks, versus a pre-event drift baseline.

## Data used
- **Endpoints:** `https://clob.polymarket.com/prices-history?market=<token_id>&interval=max&fidelity=60`
- **Markets:**
  | Market slug | Token ID (YES) | Volume |
  |---|---|---|
  | will-bitcoin-hit-150k-by-june-30-2026 | 13915689...586 | \$15.7M |
  | will-the-iranian-regime-fall-by-may-31 | 57360053...996 | \$24.3M |
  | will-trump-say-iran-during-events-with-xi-jinping | 31868121...368 | \$13.6M |
  | iran-closes-its-airspace-by-may-21 | 25765184...465 | \$2.8M |
- **Date range:** 2026-04-20 to 2026-05-20 (hourly candles via fidelity=60)
- **N events:** 6 (manually identified from news search)
- **Sample rows (btc_150k):** `{t:1776718834, p:0.0115}` ... `{t:1779304326, p:0.0105}`

## Method
For each event $e$ at time $t_0$:
$$\text{baseline\_drift} = \frac{1}{T_{pre}}\sum_{t=t_0-30}^{t_0-1} r_t$$
$$\text{CAR}_{30h} = \frac{p_{t_0+30} - p_{t_0-1}}{p_{t_0-1}} - 30 \cdot \text{baseline\_drift}$$
$$\text{signed\_CAR} = \text{direction}(e) \times \text{CAR}_{30h}$$
where $\text{direction}(e) \in \{+1,-1\}$ encodes the theoretically expected sign from news classification.

Sign test: $H_0$: P(signed CAR > 0) = 0.5. t-test on 30h signed CAR mean.

## Result
| Event | Market | p_pre | p_t0 | p_post30h | CAR% | signed_CAR% | t-stat | p-val |
|---|---|---|---|---|---|---|---|---|
| E1 Ceasefire extension | btc_150k | 0.0145 | 0.0145 | 0.0240 | +34.1 | +34.1 | 1.03 | 0.30 |
| E2 Ceasefire extension | iran_regime_fall | 0.0455 | 0.0450 | 0.0330 | -72.4 | +72.4 | -1.83 | 0.07 |
| E3 Axios one-page memo | iran_regime_fall | 0.0245 | 0.0235 | 0.0215 | -6.4 | +6.4 | -0.29 | 0.77 |
| E4 Trump bomb threat | iran_regime_fall | 0.0205 | 0.0205 | 0.0215 | +22.1 | +22.1 | 1.39 | 0.17 |
| E5 Trump-Xi summit | trump_say_iran_xi | 0.6300 | 0.5400 | 0.7455 | +38.1 | +38.1 | 1.54 | 0.12 |
| E6 Flights resume Tehran | iran_airspace | 0.1800 | 0.1150 | 0.1550 | -11.2 | +11.2 | -0.36 | 0.72 |

**Aggregate (30h):** Mean signed CAR = +30.7%, SD = 23.9%, hit rate = 6/6, sign test p = 0.0156
**6h window:** Mean signed = +17.6%, SD = 43.2%, hit rate = 4/6, t-stat = 1.00, p = 0.36

**Verdict: INCONCLUSIVE** — n=6 < 30 threshold; individual events not statistically significant. Sign test is technically significant (p=0.016) but with only 6 observations this is fragile.

## Reproduction
```bash
source ~/.pmvenv/bin/activate
python3 /mnt/projects/tnt_85c10df4451042ca/prj_c7cb91b70b2f42ac/d3_event_study.py
# Data snapshot: /tmp/pm_data/prices_history.json
# Results: /tmp/pm_data/event_study_results.csv
```

## Failure mode / next step
- **Primary break:** n=6 is too small; any single event distorts aggregate stats
- **Measurement error:** hourly fidelity misses the first-mover advantage window (critical 0–30 min after headline)
- **Look-ahead risk:** event timestamps were estimated from news article timestamps, not intra-day tick confirmation
- **Next step:** pull minute-fidelity data during known news windows (fidelity=1), expand to 20+ events using GDELT for Iran-war headlines since Feb 2026

Tags (comma-separated) Status Priority (0-5) Source