It was 23:48 on Tuesday 7 April. My Week 4 forecast was locked at $123.53 with escalation at 47%. I'd called the Tuesday deadline the single most important variable.
Twelve minutes earlier, Sky News pushed a notification. Trump has agreed to suspend the bombing of Iran for two weeks, conditional on Iran reopening the Strait of Hormuz. The 1am UK deadline that was driving the day's panic is suspended.
I started writing this before I went to sleep, because the thing I've been trying to learn through this experiment is how forecasts decay in real time. And right now I'm watching one decay in front of me.
Twelve hours earlier Trump posted that "a whole civilisation will die tonight." Now he's posting that it's "an honor to have this longterm problem close to resolution." Same man. Same day. Two completely different rhetorical postures. The whiplash is the brand.
And the futures market, the thing my model is actually trying to predict, has been pricing in this exact pattern for weeks. That's where I let my logic override my gut. That's the bit Week 4 was already gesturing at in the "Where I Could Be Wrong" section, but I didn't lean into it hard enough when I locked the forecast.
What Changed
Pakistan brokered the pause. Specifically PM Shehbaz Sharif and Field Marshal Asim Munir, who reportedly asked Trump to hold off the strikes scheduled for tonight. Iran has tabled what Trump describes as "a 10 point proposal" and a "workable basis on which to negotiate." The deadline is suspended for a fortnight.
None of this is peace. All of it is provisional. But it's enough to move the markets when they open.
Brent dropped back to $92.04 at the time of writing. My $123.53 looks very high and the naive baseline wins for the fourth consecutive week. Four weeks, four losses, no wins for the Monte Carlo. That's the honest read.
But there's a more interesting read underneath the honest one. And I think it's the actual finding.
The Model Is Reading the Script. The Market Is Reading the Actor.
The model has been overcalling escalation because it's been treating Trump's rhetoric as price-moving information. The futures market, apparently, has not. The futures market has been treating Trump's rhetoric as performance. Discounted, anticipated, structurally priced out by traders who've seen this movie three times now and know how the third act goes.
Threat. Deadline. Climbdown brokered by an intermediary.
My model is reading the script. The market is reading the actor. They're not the same thing.
This is the wrong-frame possibility I flagged in Week 4 and didn't commit to. I should have committed harder.
The Spread Problem Gets Weirder
The futures vs physical spread is the other piece of evidence pointing the same way. Dated Brent physical was at $141 last week while futures sat at $109. If physical is the true signal and futures are compressed by financial positioning, then the model isn't wrong about escalation. It's right about something. But it's measuring the wrong number.
Yesterday's news doesn't resolve that question. It deepens it. Because if physical stays at $141 while futures fall on the ceasefire news, then we've got a permanently bifurcated oil market. That's a new structural state nobody's modelling cleanly.
Markers for Week 5
I'm not going to reweight today. But I want to put some markers down for whoever (probably weekend-me) sits down to lock the next forecast.
-
Escalation probability needs to drop sharply for the two-week pause window. Probably to the high 20s. The variable that mattered most just resolved provisionally in the de-escalatory direction. The model has to honour that.
-
Stalemate probability goes up. Two weeks of conditional pause with a "workable basis" but no signed agreement is the textbook definition of stalemate. That's the new modal scenario.
-
De-escalation probability goes up but not as much as a casual reader would expect. Pauses brokered by Trump have a specific failure pattern. They break when the next news cycle requires a new performance. I'd put de-escalation at maybe 15-18%, not 30%.
-
The 90% interval needs to get wider, not narrower. The situation is more uncertain than it was this morning, not less. The pause could hold and oil could drift to $90. The pause could break in 72 hours and oil could spike to $150. Both are now live possibilities. Compressing the interval would be a lie.
-
I need to track who the world credits with this pause. Pakistan brokered it. Trump will take credit. The Western press will probably split the difference and centre Trump. That distinction matters because it shapes how the next deadline gets read, which is itself a price-moving variable. If the world believes Trump de-escalated, his next threat gets discounted faster. If the world believes Pakistan saved the day, his next threat retains more credibility. I don't have a clean way to operationalise this in the model yet. But it's worth tracking as a qualitative input until I do.
That's the rough shape of where Week 5 is heading. I'll lock the actual forecast at the weekend once I've seen how the market responds, not how I predict it'll respond. That's the discipline.
The thing I keep coming back to, the thing that makes this experiment worth doing even when I'm losing four weeks running to a baseline that just says "use today's price," is this. I'm learning what kind of information actually moves oil. And the answer, increasingly, is not the news. Or at least not the news in the way I was trained to read it. The news is a story I can follow. The futures market is doing something else entirely. Three weeks ago I would have called that a model failure. Tonight I think it's a finding.
I wanted to know if systematic Bayesian updating could beat Goldman Sachs. I'm starting to think the more interesting question is whether anyone's model can beat the futures market on a story like this. And whether the futures market is even the right thing to be predicting in the first place.
This is not a forecast. Just a note from the middle of the decay 🙃. You can see the live dashboard here.
Previous: Week 4 Forecast