Week 4 Outcomes: The Ceasefire
- Forecast: $123.53
- Actual (April 10 close): $95.20
- Error: $28.33 (29.76%)
- Naive baseline error: $13.85
- Skill score: -1.05
The biggest miss of the experiment. Not close.
Trump announced a two-week ceasefire at 11:30pm on April 7, roughly twelve hours after posting that "a whole civilisation will die tonight." Brent crashed 12% overnight. By Friday it was sitting at $95.20, down from $109.05 when the forecast was locked.
The model had escalation at 47% and a point estimate of $123.53. The market got a ceasefire instead.
What Actually Happened
The forecast window ran from Saturday 5 April to Friday 10 April. It was, in a word, whiplash.
Saturday and Sunday were pure escalation theatre. Trump's Easter Sunday Truth Social post announced "Power Plant Day and Bridge Day" for Tuesday. The 48-hour Hormuz ultimatum was ticking. Iran said the strait stays shut until war damages are paid. An F-15E crew had just been rescued from inside Iran after a firefight.
Then Monday night, hours before the deadline, Pakistan brokered a pause. A two-week ceasefire. Brent cratered from $110 to $96 on the open. Trump posted about the "Golden Age of the Middle East" at 5am.
But the ceasefire started fracturing immediately. Israel launched its largest strike on Lebanon since the war began, killing 254 people. Netanyahu explicitly excluded Lebanon from the deal. Iran re-closed Hormuz. The strait saw just 2 ship transits on Saturday, the lowest since the ceasefire was announced. Vance flew to Islamabad for talks. After 21 hours, he walked out. "We have not reached an agreement."
So the ceasefire exists. The progress doesn't.
What the Model Got Right
-
Confidence intervals held. $95.20 sits within both the 68% CI ($81.53 to $140.39) and the 90% CI ($69.23 to $162.23). That's 4/4 weeks for both intervals. Calibration remains perfect.
-
The physical/futures divergence prediction was directionally correct. Dated Brent was at $131.97 on April 10 while futures sat at $95.20. The spread widened to $37, up from $32 the week before. The model has been tracking closer to physical reality than futures give it credit for.
-
The wide 90% interval ($93 span) was justified. A ceasefire announcement was exactly the kind of regime-change event that made the tails meaningful. The model's uncertainty was appropriate even if its central estimate wasn't.
What the Model Got Wrong
-
Direction. Again. Second consecutive directional miss. Predicted up from $109.05 to $123.53. Actual went down to $95.20. Two weeks of pointing the wrong way.
-
Couldn't predict a ceasefire. This is the fundamental problem. The model assigns probabilities to scenarios but it can't anticipate a discrete political event that flips the entire regime in hours. At 47% escalation and 7% de-escalation, the model was betting against peace. Peace (of a sort) happened anyway.
-
Escalation bias persisted despite the Week 3 correction. Dropped escalation 8 points from 55% to 47%. Should have dropped it further. The market was already pricing in ceasefire rumours that the NOVA briefings (which tagged every day as ESCALATORY) weren't capturing.
The Physical Price Problem, Week 4
The gap keeps growing.
| Week | Futures Close | Dated Brent Physical | Spread |
|---|---|---|---|
| 1 | $106.41 | ~$110 | ~$4 |
| 2 | $106.84 | ~$115 | ~$8 |
| 3 | $109.05 | $141.37 | $32 |
| 4 | $95.20 | $131.97 | $37 |
The ceasefire crashed futures 13%. Physical barely moved. The spread nearly doubled from Week 2 to Week 4. The model keeps losing to the naive baseline on futures, but it's been closer to physical prices than either the naive baseline or the futures market for three straight weeks.
This doesn't fix the score. But it's worth tracking.
The Running Score
| Week | Forecast | Actual | Model Error | Naive Error | Skill Score | Direction |
|---|---|---|---|---|---|---|
| 1 | $120.16 | $106.41 | $13.75 | $7.50 | -0.83 | Correct |
| 2 | $108.16 | $106.84 | $1.32 | $0.40 | -2.30 | Correct |
| 3 | $118.88 | $109.05 | $9.83 | $5.83 | -0.69 | Wrong |
| 4 | $123.53 | $95.20 | $28.33 | $13.85 | -1.05 | Wrong |
Four weeks. Zero wins against baseline. Two consecutive directional misses. Perfect CI calibration. The model is well-calibrated for uncertainty but structurally biased toward escalation on the point estimate.
The cumulative average model error is $13.31. The cumulative average naive error is $6.90. The naive baseline is outperforming by nearly 2x. At some point, "the model captures uncertainty well" stops being a consolation prize.
Methodology Notes for Week 5
This is the biggest regime change in the experiment. Everything shifts.
-
The ceasefire changes the scenario structure. De-escalation was 7% for four straight weeks. That's over. First direct US-Iran talks since 1979 happened this weekend, even if they failed. Goldman cut Q2 Brent to $90 on the ceasefire. The market is pricing in a world where peace is at least plausible.
-
Stalemate becomes the dominant scenario. The ceasefire exists but produced nothing. Hormuz is at its lowest traffic since the pause started. Negotiations failed. This is grinding stalemate with a ceasefire label on it.
-
Escalation must drop but can't drop too far. The ceasefire is bilateral. Israel excluded Lebanon. Iran's nuclear programme is the sticking point. Two ships through Hormuz today. The ceasefire could collapse next week.
-
Demand destruction keeps climbing. Seven weeks of prices above $95. Goldman base case $80 Q4. JPMorgan base case $60. The institutional consensus says this war premium is temporary. History says war premiums always feel temporary until the war changes the underlying economics.
Scored April 12, 2026. Live dashboard here.
Previous: Week 4 Forecast | Next: Week 5 Forecast