Nicolas Liebaert Français

From a signal to a trade

A signal is not yet a trade. Below, the rule from the crack spread and butterfly pages is turned into a worked example for each spread: which legs the trade would have, in what size, and the levels the rule uses to leave it.

Personal research by a student. This is not investment advice, a recommendation or an offer, and I am not authorised to give any. The rule has never been traded. Prices come from free public sources and are delayed, at times by a week or more for oil, and the rule’s backtest assumes a fill at the close that produced the signal. These examples show how a trade is built. Do not use them to place one.

$44.27 3-2-1 crack spread a barrel, data to 3 November 2026
-0.16 Crack z-score standard deviations from its 250-day average
-8.0 bp 2s5s10s butterfly data to 3 November 2026
1.15 Butterfly z-score the rule enters beyond 1.5 either way

Oil refining margin

3-2-1 crack spread

Size

The crack on a typical day
--
Exact size for this risk
--
Size
--
Each $1 on the spread
--
A typical day at this size
--

The smallest size is 1 package, which is above this limit.

A typical day is the standard deviation of the last 60 daily changes. One package is 3 crude, 2 gasoline and 1 ULSD contracts, or 3,000 barrels. ULSD is ultra-low-sulphur diesel, and HO is the contract once called heating oil.

Legs

Contract Long the crack Short the crack Lots Barrels Spot price
WTI crude oil (CL) -- -- -- -- --
Gasoline (RB) -- -- -- -- --
ULSD (HO) -- -- -- -- --

At this size, the stop shown below is -- away. The worst closed trade in the backtest cost --.

Prices are spot prices from the EIA (WTI crude at Cushing, New York Harbor conventional gasoline and ULSD). Futures prices differ from spot.

Entry, exit and stop levels

Crack now
--
250-day average
--
Standard deviation, 250 days
--
z-score
--
Level Long the crack Short the crack
Would enter at -- --
The rule’s exit, if the average stays put -- --
The rule’s stop -- --

With no open trade, these levels use today’s average and standard deviation, so they move a little each day. The stop is fixed on the day the trade opens, 2 standard deviations against the entry level. A typical trade lasted 13 trading days (the median trade).

Backtest of this rule

Data to 3 November 2026.

Sharpe ratio
0.36
95% interval 0.03 to 0.70
Without the stop
0.48
Sharpe ratio, same rule
With a fill one day later
0.25
95% interval -0.09 to 0.58
Trades
71
2008 to 2026
Winning trades
77%
Profit in 14 of 19 traded years
Average trade
$1.68
per barrel
Worst closed trade
-$16.69
without stop: -$15.74
Worst open loss (before costs)
-$22.21
without stop: -$50.83
PeriodSharpe
2007 to 2011 0.42
2012 to 2016 0.69
2017 to 2021 0.79
2022 to 2026 0.11

What the rule did in similar markets

  • The crack has been moving -- on a typical day (last 60 days): --. In that kind of market the rule had a Sharpe ratio of --.
  • The latest data is in --. In that quarter of the year the rule had a Sharpe ratio of -- (about 20 years of data, so treat it with caution: the rule does not remove the seasons).

The calm and volatile groups are thirds of the whole history, so they use hindsight: nobody could have drawn those lines at the time. Nearly all of these slices have a 95% interval that includes zero.

What a desk would check first

  • Whether there is a lasting reason for the margin to be this high or this low, such as a war, a sanction or a plant closing. If there is, the gap may stay open.
  • The size. A desk would size for the worst trade, not the average one. The average trade is small and the worst closed trade is about 10 times bigger.
  • The contracts. A real trade needs real futures prices, roll dates and margin. None of them are here.
  • The costs. The backtest assumes 10 cents a barrel of the crack each time the rule trades, which is $300 on one package. A wide bid-offer spread (the gap between the buying and selling price) can cost more than a small gap earns.

US Treasury curve

2s5s10s butterfly

Size

DV01 is the change in dollars for a one basis point move in yield, and a basis point is 0.01 percentage point. The belly is the 5-year. The wings are the 2-year and the 10-year. Each wing carries half of that DV01, so a parallel move of the whole curve, up or down, has almost no effect.

One basis point of the butterfly is worth half the 5-year DV01: at $10,000, the average trade of 2.8 bp is about -- and the worst, -33 bp, about --. The stop shown below is about -- away.

Legs

Leg Long the butterfly Short the butterfly Notional DV01 Futures
2-year note -- -- -- -- --
5-year note -- -- -- -- --
10-year note -- -- -- -- --

The butterfly is measured in yield and bond prices fall when yields rise, so being short the butterfly means being long the belly.

On these par-bond estimates the net DV01 is about -- per basis point. Real cheapest-to-deliver DV01s differ by several percent per leg, so the true net is larger. TU and FV are the 2-year and 5-year Treasury note futures. The 10-year leg uses the Ultra 10-year contract (TN), which is closer to a true 10-year than the standard contract (TY).

Entry, exit and stop levels

Butterfly now
--
250-day average
--
Standard deviation, 250 days
--
z-score
--
Level Long the butterfly Short the butterfly
Would enter at -- --
The rule’s exit, if the average stays put -- --
The rule’s stop -- --

With no open trade, these levels use today’s average and standard deviation, so they move a little each day. The stop is fixed on the day the trade opens, 2 standard deviations against the entry level. A typical trade lasted 20 trading days (the median trade).

Backtest of this rule

Data to 3 November 2026.

Without a stop, a short opened on 18 May 2026 was still open on 3 November 2026, at -8.5 bp. It is left out of the trade figures.

Sharpe ratio
0.26
95% interval -0.01 to 0.55
Without the stop
0.33
Sharpe ratio, same rule
With a fill one day later
0.14
95% interval -0.13 to 0.39
Trades
107
1991 to 2026
Winning trades
71%
Profit in 21 of 36 traded years
Average trade
2.8 bp
in basis points
Worst closed trade
-33.0 bp
without stop: -39.0 bp
Worst open loss (before costs)
-42.0 bp
without stop: -89.0 bp
PeriodSharpe
1991 to 1999 0.37
2000 to 2009 0.44
2010 to 2019 0.02
2020 to 2026 0.18

What the rule did in similar markets

  • The 10-year yield has been moving -- on a typical day (last 60 days): --. In that kind of market the rule had a Sharpe ratio of --.
  • The 2s10s slope (the 10-year yield minus the 2-year yield) is --: --. With that shape the rule had a Sharpe ratio of --.

The calm and volatile groups are thirds of the whole history, so they use hindsight: nobody could have drawn those lines at the time. Nearly all of these slices have a 95% interval that includes zero.

What a desk would check first

  • Whether the central bank is about to change course. If the whole curve is being repriced, a 5-year yield that is out of line can move further out of line.
  • The contracts. The futures sizes here are estimates. They use a par bond of each maturity, not the cheapest-to-deliver bond. A real trade needs the DV01 of the real contracts, real futures prices, roll dates and margin. None of them are here.
  • The costs. The backtest assumes half a basis point of the butterfly each time the rule trades. In quiet markets that can cost more than a small gap earns.

How to read an example

Signal. The rule counts how many standard deviations the spread is from its 250-day average (the z-score). It sells the spread above 1.5 and buys it below -1.5. It exits when the z-score is back between -0.5 and 0.5.

No open trade. The rule has no side yet, so the example shows both, each built at the rule’s real entry: the average minus or plus 1.5 standard deviations.

Open trade. The rule keeps a trade until it exits, so a spread that looks normal today can belong to a trade opened weeks ago. The example says which case applies.

Stopped out. If the trade goes 2 standard deviations against the entry, the stop closes it. The rule then waits until the z-score is back between -1.5 and 1.5. It does not re-enter the same move the next day.

Backtest. The Sharpe ratio is the average daily profit divided by its standard deviation, scaled to a year. Above 1 is good. Around 0.3 to 0.5 is weak.

The figures are computed in your browser from files refreshed every morning: Treasury yields from FRED and oil prices from the EIA. The calculation is the same as in the research repository, rewritten in JavaScript and checked against the Python by a test script. The backtest is fixed at the date shown above it.