Nicolas Liebaert Français

The 3-2-1 crack spread: a small edge, much weaker since 2022

A refinery buys crude oil and sells fuel. The gap between the two prices is its refining margin. I tested a simple rule on that gap to find out if it has an edge after costs.

0.48 Sharpe ratio after costs no stop, 95% interval 0.13 to 0.80
0.34 With a fill one day later 95% interval 0.02 to 0.63
87% Winning trades 60 closed trades in 18 years
-$15.7 Worst closed trade a barrel; worst open loss -$50.8

Data to 3 November 2026.

What I did
The rule sells the crack when it is far above its recent average and buys it when it is far below. I ran it on 20 years of daily oil and fuel prices, and the first trade came in July 2008. Each day it uses only data known that day, and every trade pays a cost.
Why it matters to a trader
Energy desks and commodity funds trade this margin. The trade is a bet on refining, with little exposure to the direction of oil. If the margin does return to normal, the trade can earn from the gap.
What I found
The edge is small. The Sharpe ratio (average daily profit divided by its standard deviation, scaled to a year) is 0.48 after costs without a stop, and 87% of trades win. But the worst closed trade lost 5 times the average trade. With a fill one day later, about half of the butterfly’s Sharpe ratio and a third of the crack’s disappear. From 2022 to 2026, the rule with a stop had a Sharpe ratio of 0.11.
What I could not show
That anyone could have traded this in practice. I use spot prices from the US government, not the prices of the futures you would trade. I did not model exchange margin (the collateral a broker requires), the cost of rolling from one contract to the next or the roll yield (the gain or loss the shape of the futures curve gives at each roll).

Try the rule on the crack

This demo reads the latest data: after 3 November 2026 its figures may differ from the text.

Stop
Fill

At these settings, a small edge: Sharpe ratio 0.36, and 0.11 in 2022 to 2026, its weakest period.

Cumulative profit of the rule on one barrel of crack, after costs. The thin line is the other stop setting.
Sharpe ratio
0.36
after costs
Winning trades
77%
of 71 trades
Average trade
$1.68
per barrel, after costs
Worst closed trade
-$16.7
9.9 times the average trade
Worst open loss
-$22.2
before costs
Total profit
$119
per barrel, all years
Maximum drawdown
-$54
fall from a peak
Time in a trade
33%
of all days

Profit by year

Profit in each calendar year. Bars above zero made money, bars below it lost money.

Profit in each year, as a table
YearProfit

Results by period

PeriodDaysSharpeProfit
2007 to 2011 1,261 0.42 $26
2012 to 2016 1,259 0.69 $42
2017 to 2021 1,255 0.79 $38
2022 to 2026 1,203 0.11 $13
Data to 3 November 2026. The rule is a JavaScript copy of the code in the research repository, so it responds at once on a phone. It is checked against the Python by a test script. The exit level moves with the entry level: it is always one third of it.

How the trade works

The crack spread is the price of the fuel minus the price of the crude. In the 3-2-1 version, 3 barrels of crude turn into 2 barrels of gasoline and 1 barrel of diesel. Fuel is quoted per gallon and a barrel is 42 gallons, so the spread is (2 × gasoline + 1 × diesel) × 42 / 3 minus crude, in dollars per barrel. The data are WTI at Cushing against New York Harbor conventional gasoline and New York Harbor ultra-low-sulphur diesel (ULSD). The crude is inland and the products are on the coast, so the spread also carries a location gap, and no single refinery earns exactly this margin.

The logic: when the margin is very high, refiners make more fuel and the margin falls. When it is very low, they cut back and the margin rises. This is a reason to test the trade, but it does not prove that the trade works.

When the crack is high, the rule sells the crack: long 3 crude oil contracts, short 2 gasoline contracts (RB) and short 1 ULSD contract (HO, the contract once called heating oil). When it is low, it does the opposite. I call this set of contracts a package. One package covers 3,000 barrels and earns or loses $3,000 for every $1 the spread moves.

The rule is the same as for the butterfly. The z-score counts how many standard deviations today’s spread is from its average over the last 250 trading days. The rule enters above 1.5 or below -1.5, and exits between -0.5 and 0.5. A stop at 2 standard deviations against the trade is optional. The relative value page explains the rule in full.

What the results say

A Sharpe ratio above 1 is good. Around 0.3 to 0.5 is small. The 95% confidence interval comes from a block bootstrap, explained on the relative value page. If it includes zero, the data cannot rule out a true Sharpe ratio of zero.

Data to 3 November 2026.

No stopWith stop
Sharpe ratio, after costs0.480.36
95% confidence interval0.13 to 0.800.03 to 0.70
Sharpe ratio, fill one day later0.340.25
95% confidence interval, fill one day later0.02 to 0.63-0.09 to 0.58
Closed trades6071
Winning trades87%77%
Average trade, per barrel$3.40$1.68
Worst closed trade, per barrel-$15.7-$16.7
Worst closed trade on one package-$47,226-$50,064
Worst open loss (before costs), per barrel-$50.8-$22.2
Traded years with a profit15 of 1914 of 19

The edge is small: the 95% confidence interval starts just above zero with the stop, and it includes zero if trades fill one day later. It may be nothing. A Sharpe ratio of 0.48 over 20 years is a t-statistic of about 2, below the hurdle of about 3 often asked of a rule found after many trials.

The stop lowers the profit and does not improve the worst closed trade (-$16.7 with it, -$15.7 without). What it cuts is the worst open loss: -$50.8 a barrel without the stop, in 2022, and -$22.2 with it, in 2026. A win rate means little without this line: the rule without a stop wins often because it sits through open losses two to three times the size of its worst closed trade. Many small gains and a few large losses are the usual shape of a mean-reversion trade (a bet that a gap will close). This is why position size matters more than the signal.

When could the rule work?

Each point is marked Tested if I measured it in this data. It is marked Idea if it is how the market works in theory and I did not test it. Each subset of the data is small, so treat the numbers with caution. The Sharpe ratios and the losing years below are for the rule with the stop.

The calm and busy groups are thirds of the whole history, so they use hindsight: nobody could have drawn those lines at the time. With lines drawn only from past days the crack’s Sharpe ratio is 0.97, 0.35 and 0.22 (calm, middle, busy). Nearly all of these slices have a 95% confidence interval that includes zero.

When it could work

  • Idea After a short shock, such as a refinery outage, a hurricane on the US Gulf Coast or a cold snap. Refining margins jump for a few weeks, then the plants restart and the margin falls back.
  • Idea When refining margins are very high. High margins lead refiners to run their plants harder and restart idle ones. This adds fuel and pushes margins down.
  • Tested In the fourth quarter. The rule had a Sharpe ratio of 0.99 from October to December and -0.05 from July to September. This may be only the seasonal fall of the crack, as explained below.
  • Tested When the crack is calm. The Sharpe ratio was 0.71 when the spread had been moving little, against 0.22 when it had been moving a lot.

When it can lose money

  • Tested In some years. The rule lost money in 2011, 2013, 2020, 2023, 2026. The worst year was 2011 (-$17.2 a barrel), and from 2022 to 2026 the Sharpe ratio was 0.11. I did not test the cause. My guess is that a lasting shock keeps the margin away from normal: without a stop, a short opened in January 2022 lasted 130 days and closed at -$15.7.
  • Tested Extreme losses. The worst closed trade lost $16.7 a barrel with the stop and $15.7 without it, against an average trade of $1.68 with the stop. On one package that is $50,064 and $47,226. Without the stop, a trade was down $50.8 a barrel while open, about $152,000 on one package.
  • Idea Around the September change from summer to winter gasoline. The front-month gasoline contract (the nearest expiry) switches grade and the spread can jump by several dollars overnight. This is a futures effect and does not show in the spot prices used here.

The latest position. The worked example applies the rule to the latest data and shows these situations next to it.

How much should you trust it?

Other settings. This table shows the Sharpe ratio of the rule without a stop for other windows and entry levels. Across the 12 settings it goes from 0.32 to 0.63.

Data to 3 November 2026.

Window \ entry level1.01.52.0
60 days0.620.630.54
120 days0.630.610.55
250 days0.480.480.34
500 days0.400.410.32

The outlined cell is the setting used on this page. The stronger the tint, the higher the Sharpe ratio.

Costs. With a cost of $0, $0.05, $0.10 and $0.25 a barrel each time the rule trades, the Sharpe ratio (with stop) is 0.41, 0.39, 0.36, 0.30. It falls as the cost rises but stays above zero at these four levels. You can try other costs in the demo above.

The seasons. The rule does not remove the seasons. The crack averages about $24 in May and $17 in December, and the fourth-quarter result above may be that fall and nothing else. With a seasonal adjustment built only from past data, the Sharpe ratio is 0.44 without the stop and 0.22 with it, against 0.48 and 0.36.

I removed two days. On 20 April 2020 the price of WTI crude went below zero, to -$37. The crack on that day is about $66. On about four days out of five it is between $9 and $36. That one value distorts every statistic. I drop any day when crude was below $5 on that day or the day before. This removes 20 and 21 April 2020 and nothing else. Keeping them would raise the Sharpe ratio to 0.59 without the stop and 0.51 with it, so this choice works against the rule.

What I did not model. The prices come from the US Energy Information Administration (EIA). They are spot prices, and the gasoline futures contract is RBOB, a different grade. A real trade uses futures, with a bid-offer spread, exchange margin, a contract roll every month and the roll yield of the curve. I did not model any of that. A rough check on front-month futures since 2015, with no roll adjustment, gives a Sharpe ratio of 0.34 (0.50 with the stop) against 0.28 (0.24) on spot. The check is rough, but it does not point to spot prices as the source of the result. Daily changes of the two series correlate only 0.65.

Research only

This is research. The rule has never been traded, and nothing here is investment advice. The rule earned a small amount in the past, and it may not do so in the future.

Source and tests at github.com/Nicolas8330/relative-value-strategies. Every figure outside the demo comes from one run of examples/make_results.py, or from the rule itself, on oil prices to 3 November 2026. The one-off checks use data to 3 November 2026: the intervals with a later fill, the seasonal adjustment, the groups drawn from past days and the test that keeps the two days of April 2020. The futures check stops on 21 September 2026. See also the butterfly page.