Spreads
The position is a difference. So is the forecast.
Cracks, crushes, feed margins and quality differentials, the things a desk actually carries, are differences between legs. Subtracting two published fan charts does not give you one, and this page shows exactly how that fails and what reading the legs together is worth.
01 · The quantity
A spread is its own market, with its own range.
WTI-Brent · settle difference, USD/bbl
- last difference
- -11.91
- middle outcome, day 30
- -11.93
- 3-in-5 band, day 30
- -14.53 – -10.11
- 9-in-10 band, day 30
- -16.88 – -8.15
A spread is what the desk actually holds
Almost nobody is long crude and nothing else. They are long one barrel and short another, long the crush, short the feed bill. The position is the difference, so the difference is the thing that needs a forecast.
The difference has its own middle
WTI minus Brent has spent this history negative, with WTI the cheaper of the two, and the middle path says where it settles next. That is one number about one quantity, not two forecasts left for you to reconcile.
And its own bands
The bands carry counted shares, the same way every other fan chart here does: three finished outcomes in five inside the inner band, nine in ten inside the outer. The published record counts whether they held.
02 · The trap
Subtract two fan charts and the interval turns inside out.
WTI-Brent · 4-in-5 band, two ways
- the spread's own 4-in-5 band
- -16.88 – -8.15
- subtracting rung from rung
- -14.69 – -9.71
- days where it reads backwards
- 0 of 31
The obvious move is to subtract two fan charts
Both legs are published. Take the high end of one, subtract the high end of the other, and you appear to have the spread. It costs nothing and it is the first thing anyone tries.
It comes out upside down
Subtracting rung from rung assumes the two legs move in lockstep. When the leg you are short swings wider than the leg you are long, the arithmetic hands back an interval whose top edge sits below its bottom edge. That is every day of the month here. Not a wide answer. Not a wrong answer. Not an answer.
And where it is not inverted, it is overconfident
On the spreads where the edges stay in order, the band it produces held far less often than the four-in-five it claims. A band that narrow looks impressive and costs money.
03 · The record
What reading the legs together is worth.
The honest comparison is a careful hand-combination
A quant who knows the trap flips the short leg before combining, which fixes the upside-down interval. That is the widest reading two published fan charts allow, and the opponent worth measuring against is that careful version, not the careless one.
Nearly twice as accurate, on data the model never saw
Across more than two thousand out-of-sample starts, forecasting the spread as one quantity cut the average error nearly in half against that careful hand-combination. Reading the legs together is what buys it.
Tighter and more accurate at the same time
A narrow band can flatter an error score by being wrong confidently, so the record reports both. Ours held about as often as it claims while the hand-combination had to stay wide enough to hold almost always. Tighter and truer is the only version of this worth selling.