The Diesel Crack Spread Is at a Record. What Happens to Stocks Next?

Research note: New York Harbor diesel and WTI crude daily since 1986, and S&P 500 daily closes, through September 29, 2026. Frozen October 4, 2026. How we checked it · How we use AI.

• The first year after a spike was weak. Across nine spikes since 1989, the S&P 500’s median twelve-month return was +3.4%, against +12.0% in all months.
• From six months in, it was closer to normal. The next year’s median was +9.5%, and drawdowns were about ordinary. The next-quarter result depends on how spikes are defined.
• Drawdowns ran deeper in the first months. The median peak-to-trough decline within six months of a spike was 10.2%, against 7.6% for all months.
• Long-lasting spikes coincided with bigger declines. In the three spikes where the crack stayed high for most of a year (2000, 2011 and 2022), the S&P 500 fell 17–22% within it, though in two the decline was over before the persistence was visible.
• The extreme readings are not clearly worse. Whether today’s 3× reading predicts a bad quarter depends on how the months are counted.

The crack spread is a benchmark gross margin: the price of a refined product less the price of the crude it is made from. It leaves out a refiner’s operating costs, so it is a measure of how tight the product market is rather than of anyone’s profit (EIA). Diesel’s matters more than gasoline’s because diesel moves trucks, trains, ships, tractors and construction equipment, and it is much harder to use less of. A record diesel margin works like a tax on everything that moves.

My hypothesis is that the cost reaches stock prices slowly, through shipping costs, goods prices and the rate path, which would fit a weak year more than a weak week. The data below show the timing; they cannot show that this is the reason for it.

What has happened to the price of diesel?

In 2025, a normal year, a gallon of diesel at an American pump cost $3.66 on average. In September through the 29th it cost $6.29, +72%. That understates the shock, because about $1.34 of every gallon is taxes, distribution and the retailer’s margin, which barely move. The fuel itself, crude plus the refining margin, rose +113%, more than double, and the margin alone +241%.

The margin is now bigger than the crude itself. Since 1986 the monthly diesel crack has topped the price of WTI in only six months, and three of them are this year’s. The others were April 2020, October 2022 and November 2022, when crude collapsed or Russia’s invasion squeezed diesel. September’s gap, $13.4 a barrel, is the widest on record; the largest before was $7.8.

That is why the shock reaches further than a gasoline spike would. Wholesale diesel, the price trucking companies, railroads, farms and builders buy at, is up +113% on 2025, more than double, and it feeds the cost of moving almost every physical good. The bill arrives with a lag, through fuel surcharges, freight contracts and the price of goods, which is why I look at what stocks did over the following months and not the following days.

A barrel of diesel is a barrel of crude plus the crack, exactly, so the gap between the lines below is the margin. Since February 2026, diesel has risen $103 a barrel, and 68% of that rise was the margin.

The benchmark matters. On September 29, 2026 the crack was $114 against WTI and $96 against Brent, and adding a Brent price to a WTI crack counts the gap between the two crudes twice.

Shipping is a separate pressure on crude buyers. BWET, an ETF holding near-dated freight futures on crude tanker routes, mostly very large tankers from the Middle East to Asia, closed on September 29, 2026 at 16.8 times its February average, against 2.8 times for the diesel crack and 1.6 times for Brent. It is a futures-based proxy whose returns can differ substantially from spot hire rates, and it covers crude tankers only; product tankers, which carry diesel, have their own rates. The daily series is on the oil tanker rates page.

Six months in, stocks have ignored it

The crack first doubled its three-year norm in March 2026. From that month’s last close the S&P 500 has risen 17.5% in 125 sessions without closing below it, and set a record on August 13, 2026. At the same point, that is ahead of eight of the nine past spikes, behind only 2003 (+18.5%). Meanwhile the crack kept climbing, from 2.33 times its norm in March to 3.19 in September.

How unusual is a $110 crack?

Nothing in the record compares in raw dollars. After inflation the previous peak, October 2022, was $106.7 in August 2026 dollars. September 2026’s CPI is not out yet, but prices would have to rise 3.8% in one month to pull September’s $110.7 below that.

What counts as normal has moved a lot, from a few dollars in the 1990s to $20 to $30 in the 2010s. So I measure a spike against the crack’s own recent history: the monthly average divided by its median of the prior three years, with twice the median as the line. September 2026 reads 3.19.

What did stocks do after every spike since 1989?

The median spike went nowhere for most of a year while the typical month gained +11.6%. The gap was 8.2 points at twelve months and 5.9 at 24: the second year was better, but it did not close the gap.

The first three months were already soft, a median of +0.3% against +3.1%, and the index was lower a year later after 1989, 2000, 2008 and 2022. The six- and twelve-month results hold when I move the line: across thresholds of 1.75 to 3 times the median and gaps of six to 24 months between episodes, the twelve-month median ran from −1.5% to +5.2% and the six-month from +0.0% to +2.7%, all below normal. At 24 months the versions ran from +14.4% to +22.6%, around the +20.9% norm.

What does the record say from here?

Those results start the clock at the spike. A reader today is six months past it. Measured from the last close six months after each past spike, the picture changes.

Three things stand out. The next year was closer to normal. Drawdowns from this point were ordinary. And the next-quarter result was unstable: under the default rule it was −3.6%, but across the twelve thresholds and gaps it ran from −3.8% to +4.3%, so I give it little weight as a timing signal. Most of the damage after past spikes came in their first year, and 2026 has already spent half of that year going up.

A strong first six months did not decide what came next. After 2003’s +19.2% start the following year added +9.5%; after 2000’s +2.6% it lost 15.3%. Nor did starting near a high: four spikes began within 4% of a record (1989, 1991, 1994 and 1996) and their next years ranged from −7.0% to +24.1%.

What if the crack keeps climbing?

Most past spikes did not last. In six of the nine, the crack fell by between 46% and 77% within six months. Crude was less obliging: WTI was higher six months later in four of them. The margin, more than the oil price, is what came back down.

Looking back, that split the outcomes. In 2000, 2011 and 2022 the crack stayed at twice its norm or more for most of the following year, and each of those years included a decline in the S&P 500 of 17% to 22%. This is a retrospective grouping: it uses the whole following year to decide which spikes persisted.

When the crack faded, four of six years saw only a 5–8% decline. The exceptions are the other way stocks got hurt: in 1989 and 2008 the margin eased, but crude surged as Iraq invaded Kuwait in one case, and the economy was already in recession in the other.

The catch is timing. Persistence only becomes visible once the crack has spent six months at 2× or more, and in two of the three cases the damage had mostly been done by then. The 2000 spike qualified in November 2000, after a 13.9% decline, and the next year returned −11.2%. The 2011 spike qualified in November 2011, after a 19.4% decline, and the next year returned +13.0%. The 2022 spike qualified in September 2022, after a 21.8% decline, and the next year returned +18.0%. Only 2000, with the dot-com bust under way, kept falling.

This year is different on both counts. Six months in, the crack is +48% from its March level and at 3.19 times its norm, higher at this point than any past spike (the most was 2.96). It has been at 2× or more in four of the six months since, so the earliest it could qualify is November 2026. And the S&P 500’s largest decline since March has been 4.5%, from June 2, 2026 to June 10, 2026, without a decline comparable to those episodes.

Persistent diesel stress would make me more cautious about earnings and inflation. The three historical cases show that long energy shocks can accompany substantial equity losses. They cannot tell me whether a 20% decline is more likely from today’s starting point.

Is a 3× reading worse?

September 2026’s 3.19 is well above most spikes’ first readings. The answer depends on how the past is counted. Taking every month at three times its median or more, 19 months since 1989, the next three months had a median of −3.0%. Taking each of the five separate stretches once, it was +3.1%, about normal. Long shocks get repeated weight in the first count. Leaving out the 2000 stretch moves the monthly median to −0.3%; leaving out 2022, to −3.4%.

I read that as no reliable extra signal from the level itself. The 3× months were followed by deeper six-month drawdowns (11.2% median), which is consistent with the rest of the study, but the short-term return result rests on a few long episodes.

Five spikes, five endings

The other four spikes, in 1989, 1991, 1994 and 2011, are in the bar chart above and the study download. One of them, December 1989, came seven months before the 1990 recession began.

Where I come out

I would not sell because diesel’s margin hit a record. The weak stretch after past spikes was mostly their first year, and this market has already spent half of it rising to new highs. Measured from the spike, that first year carried a median six-month drawdown of 10.2% against 7.6%. Measured from six months in, where we are now, the next year was closer to normal and so were the drawdowns. The next-quarter result changes with the episode definition, so I would give it little weight as a timing signal.

The part I would watch is whether the shock fades. Four of the six spikes that faded saw only single-digit drawdowns. The three that persisted coincided with declines of 17–22%, although in two of them the decline was over by the time the persistence was visible. This one is already more persistent than most, without a decline comparable to those episodes. If the crack is still near 3× its norm this winter, I would be more cautious about earnings and inflation, without treating that history as a forecast of a 20% fall.

That is a thin edge on nine episodes, and the uncomfortable part is how much depends on what happens to diesel itself. Recessions began after the 1989 and 2000 spikes, the 2008 spike came inside one, and 2022’s bear market needed none, because inflation and rates did the work. A margin this high feeds both, and the longer it stays near three times its norm, the more weight I would give the weak first-year record over this year’s strength.

What would change my read

Toward an ordinary year: the monthly diesel crack against its prior three-year median, shown on the crack spread page, falling back below 2×, about $69 a barrel at today’s median, with the S&P 500 holding above its March close of 6,528.5.

Toward a weaker one: the same monthly measure holding near 3× into winter, when heating oil competes for the same barrels, while the four-week average of distillate demand drops below its level a year earlier and distillate inventories fall further below year-earlier levels. That would strengthen my concern that expensive diesel is weighing on activity.

How we checked it

• The crack. New York Harbor ultra-low sulfur diesel in dollars a gallon, times 42, less WTI crude, from June 2006; New York Harbor No. 2 heating oil before that. In the years both series exist, the two cracks have a correlation of 0.99 and differ by a median of $3.78. Monthly figures average the daily prices; September 2026 covers 20 sessions through September 29, 2026.
• Inflation. Real values are in August 2026 dollars, deflated by CPI-U month by month. No CPI is carried forward, so September 2026 is compared in raw dollars until its CPI is published.
• A spike. A month whose average crack is at least 2 times its median over the prior 36 months. A new episode needs 12 months without a qualifying month. The 2026 episode began in March 2026 and is kept out of every historical median.
• Returns. S&P 500 price return, without dividends, from the last close of the qualifying month, the first day its monthly average was known, over 63, 126, 252 and 504 trading sessions. “Six months in” starts from the last close of the sixth month after. Baselines use every month-end from January 1989 with a complete window.
• Persistence. A spike persisted when the monthly crack was at 2× its prior three-year median or more in at least six of the twelve months after its first month. The record splits with no cases in between: 7 to 12 months against 2 or fewer.
• Drawdown. The largest peak-to-trough decline in closing prices inside the window, with the starting close as the first peak.
• Recession dates. NBER business-cycle peaks: July 1990, March 2001, December 2007 and February 2020. FRED’s USREC series flags the months after each peak.
• Robustness. Thresholds of 1.75, 2, 2.5 and 3 times the median and gaps of 6, 12 and 24 months, from the spike and from six months in. Every episode list and its returns are in the study download.
• Small sample. Nine completed spikes show what happened, and several overlapped with other shocks. They are too few to give a probability.
• Frozen evidence. Every number on this page comes from the snapshot of October 4, 2026.