Seasonality in Commodity Futures: What the Evidence Says About Natural Gas, Gasoline and Grains

Seasonality in commodity futures means that supply, demand and storage follow the calendar (harvests, heating seasons, driving seasons, fuel specifications), so spot prices and the shape of the futures curve show recurring yearly patterns. The evidence is clear that these patterns exist in prices and curves. It is much weaker that they create reliable returns for someone holding futures, because the market prices the expected season into the deferred contracts long before it arrives.

In short: natural gas is dearer in winter, summer-grade gasoline costs more than winter grade, and grain prices tend to be lowest around harvest. But a December natural gas future already trades at a premium to a September future, and an April RBOB contract already reflects summer-grade gasoline. What a trader earns from a futures position is the change between today’s price for that contract and its eventual settlement, and that change only has a seasonal pattern if the season turns out different from what the curve expected. Our test on 30 years of EIA natural gas futures data found no calendar month with statistically significant returns, and winter months were, on average, negative.

The basic shapes of futures curves are covered in contango and backwardation. The economic theory behind them (storage, convenience yield, inventories) is in futures term structure in depth. This piece looks at the seasonal case specifically.

Three kinds of seasonality, and why the difference matters

People mean three different things when they say a commodity is “seasonal”:

Spot seasonality. The physical price rises and falls through the year. Henry Hub gas in January versus April is the classic example.
Curve seasonality. On any given day, futures for different delivery months trade at predictable premiums and discounts to each other. A winter gas contract trades above a summer one, today, before anyone knows the weather.
Return seasonality. Holding a futures position over a particular calendar month earns a predictably positive or negative return.

Only the third is a trading edge, and the first two make it harder to find. If spot gas reliably rises into winter, and the futures curve reliably prices that rise, then a long position in the winter contract bought in summer earns nothing from the season itself. It earns only the surprise.

What the academic models say

Carsten Sørensen, “Modeling Seasonality in Agricultural Commodity Futures” (Journal of Futures Markets, 2002), models log commodity prices as “a sum of a deterministic seasonal component, a non-stationary state-variable, and a stationary state-variable,” estimated with a Kalman filter on weekly CBOT corn, soybean and wheat futures from 1972 to 1997. The seasonal component is built into the term structure of futures prices, and the paper reports evidence for the theory of storage: a negative relationship between inventories and convenience yields.

Svetlana Borovkova and Hélyette Geman, “Seasonal and Stochastic Effects in Commodity Forward Curves” (Review of Derivatives Research, 2006), make the same point for energy. They show that “the well-known cost-of-carry relationship is significantly improved by introducing a deterministic seasonal premium within the convenience yield.” Using IPE (now ICE) futures, they estimate that December UK natural gas futures traded on average at a 28% premium to the average forward price level, December electricity at 7%, heating oil at most 3%, and that Brent crude’s seasonal premium was not significantly different from zero. In other words, the futures curve carries the season, and the size of the seasonal premium tracks how hard the commodity is to store.

The return question is addressed by Matti Keloharju, Juhani Linnainmaa and Peter Nyberg, “Return Seasonalities” (Journal of Finance, 2016). Their abstract reports that strategies based on same-calendar-month historical returns work in stocks and that they “document similar return seasonalities in anomalies, commodities, and international stock market indices.” Their interpretation is that these seasonalities are tied to shared systematic factors rather than being a separate anomaly. That is evidence that calendar effects in returns can exist, but it is a cross-asset, portfolio-level result, not a statement that buying gas in a given month works.

Natural gas: the strongest seasonal market

Physical seasonality

Gas is injected into underground storage through the warmer months and withdrawn through the heating season. The EIA’s Weekly Natural Gas Storage Report for the week ending October 30, 2026 (released November 5, 2026) shows working gas in the Lower 48 at 3,801 Bcf, after a net injection of 77 Bcf that week. That was 109 Bcf below a year earlier and 54 Bcf above the 2021-2025 five-year average of 3,747 Bcf. Late-autumn storage relative to the five-year average is the single number gas traders watch most heading into winter. More on the contract and the market is in natural gas futures in depth.

We computed spot seasonality from EIA’s daily Henry Hub spot price, 1997-2025: for each year, the log of each month’s average price minus that year’s mean of those twelve logs, then averaged across years.

Month	Jan	Feb	Mar	Apr	May	Jun	Jul	Aug	Sep	Oct	Nov	Dec
Deviation from yearly average	+5.6%	-0.5%	-6.5%	-4.9%	-1.4%	+1.4%	-1.5%	-1.4%	-1.2%	+2.1%	+2.2%	+5.8%

The pattern is real (winter high, shoulder-season low) but noisy. January’s average spot price exceeded July’s in only 18 of 29 years (62%). Weather, production growth, and in recent years LNG exports, can swamp the calendar.

Curve seasonality

Now look at the futures curve. Using EIA’s daily NYMEX natural gas futures series (contracts 1 through 4, through April 2024), we averaged the log spread between the fourth and first nearby contracts by the month in which it was observed, 2000-2024:

Observed in	Jan	Mar	May	Jul	Aug	Sep	Oct	Nov	Dec
Contract 4 vs contract 1	-3.1%	+6.9%	+4.5%	+9.3%	+15.4%	+17.1%	+10.0%	0.0%	-6.9%

In August and September, the fourth contract is a winter month and the first is an autumn month, and the curve sits about 15% to 17% higher out to winter. That premium is the season, priced in advance. Buying the winter contract in September means paying for the winter already.

Return seasonality: our test

The fair test is: if you held a single natural gas futures contract for one calendar month, did some months pay consistently? We built single-contract monthly returns from EIA’s series (the second nearby contract at one month-end, valued as the first nearby at the next, so no roll is mixed in) for 362 months from February 1994 to March 2024. Log returns, no costs:

Holding month	Average return	t-statistic	Positive months
January	-5.1%	-1.60	37%
February	-2.1%	-0.89	48%
March	+1.1%	0.45	58%
April	+1.4%	0.83	53%
May	-0.8%	-0.42	47%
June	-1.2%	-0.46	43%
July	-3.4%	-1.18	37%
August	-1.8%	-0.56	50%
September	-0.8%	-0.29	47%
October	-1.4%	-0.79	43%
November	-5.7%	-1.79	30%
December	-5.0%	-1.41	37%

What this shows:

No month is statistically significant at the 5% level (that would need a t-statistic of about 2). With 12 independent tests at that threshold, chance alone would produce at least one “significant” month 46% of the time (1 minus 0.95 to the 12th power), so even one would not have meant much.
Winter months were the weakest, not the strongest. November, December and January averaged about -5% each. Spot gas is highest in winter, but futures bought just before winter were, on average, priced above where they settled. This is consistent with the curve over-pricing winter risk on average, though the effect is not statistically reliable either.
The average month lost 2.1%, reflecting the persistent contango in gas futures.
Patterns flip between periods. March averaged +6.4% in 1994-2008 and -3.9% in 2009-2024. December averaged +0.2% and then -10.2%. A pattern that changes sign is not a pattern you can bank on.

When seasonal spreads go wrong: Amaranth, 2006

The most famous seasonal trade in futures history was in natural gas calendar spreads. In the US Senate Permanent Subcommittee on Investigations hearing of June 25, 2007 (record S. Hrg. 110-235), Senator Carl Levin stated that in prior years January gas futures had cost “$1 to $1.50 more than futures contracts delivering natural gas in October,” but that in 2006 the January contract exceeded October by $4, and that Amaranth Advisors at times held 75% of the outstanding November 2006 contracts and 60% of the January and March 2007 contracts. Hilary Till’s review of the Senate staff report summarises the outcome: in September 2006 Amaranth “lost about 65% of its $9.2 billion in assets in a little over a week due to U.S. natural-gas futures positions.” The full story is in famous futures blowups.

The arithmetic shows why. A NYMEX natural gas contract is 10,000 MMBtu, so a January-October spread moves $10,000 per contract for every $1 per MMBtu. A trader long 1,000 January and short 1,000 October spreads at a $4 difference who sees the spread fall back to its historical $1.50 loses 2.50 x $10,000 x 1,000 = $25 million. Amaranth’s positions were on the order of 100 times that size. A seasonal spread is a bet on the size of a premium the market has already priced, and the premium can be bid far above its norm by one large trader, then collapse.

Gasoline and diesel: seasonality written into the rules

The vapor pressure calendar

Gasoline seasonality is partly regulatory. Under EPA rules (40 CFR 1090.80), the summer volatility season runs from May 1 to September 15 for refiners, terminals and most of the supply chain, and from June 1 to September 15 for retailers. Summer gasoline must meet a lower Reid vapor pressure (RVP), which is more expensive to make. The EIA explains how this hits futures: the underlying of the RBOB front-month contract must meet the lower-RVP summer specification, so “starting March 1, RBOB futures prices tend to increase several cents from February averages,” and “on September 1, the underlying commodity of the RBOB front month contract reverts back to a higher RVP specification.” For distillate, the EIA notes that crack spreads “are usually highest from October to February,” driven by harvest diesel demand and heating.

What the crack data shows

From EIA’s front-month futures (2010-2023), average monthly cracks in dollars per barrel:

	Jan	Mar	Apr	May	Aug	Sep	Oct	Dec
RBOB crack	14.43	23.96	24.01	24.71	22.22	16.48	15.27	13.84
ULSD crack	24.27	23.80	23.81	22.93	26.16	27.03	29.05	26.00

The front-month gasoline crack was higher in April than in December in 14 out of 14 years. That looks like a perfect seasonal signal, but most of it is the contract specification changing, not a market inefficiency. In December, the April RBOB contract already trades at a summer-grade price.

The spot market, where no contract roll is involved, is less tidy. Using EIA’s New York Harbor conventional gasoline spot price against WTI Cushing, 2010-2025, the May-July average gasoline crack beat the November-January average in 11 of 16 years. The gap ranged from minus $1.23 per barrel (2020) to plus $24.76 (2022). Refinery utilization adds another layer: EIA’s weekly data for 2015-2025 shows average utilization of 85.4% in February and 86.0% in October (maintenance seasons) against 93.2% in August. Crack spread mechanics in detail are in crack and crush spreads in depth.

Grains: harvest, storage and reports

For grains and oilseeds the season is the crop calendar. CME Group’s Soybean Crush Reference Guide states that “soybean prices are typically lowest at harvest and trend higher during the year as storage, interest and insurance costs accumulate over time,” and that November and December futures “often reflect the market’s perception of conditions in the new soybean crop year.” Read carefully, that is a statement about carry: the rise from harvest to the following summer is mostly the cost of storing grain, which the futures curve prices as contango between new-crop and old-crop months. A futures buyer does not collect it; the storage owner does.

Two features make grain seasonality different from energy:

Old-crop versus new-crop. The July and September soybean and corn contracts price the old crop, and the November soybean and December corn contracts price the new one. The spread between them depends on stocks: tight old-crop inventories can put the curve into steep backwardation, which is the theory of storage at work, consistent with Sørensen’s finding on inventories and convenience yields.
Scheduled information shocks. USDA reports concentrate risk into known dates: the monthly World Agricultural Supply and Demand Estimates, the NASS Prospective Plantings report at the end of March, and the Acreage and quarterly Grain Stocks reports at the end of June. Growing-season weather in July and August, when corn pollinates and soybeans set pods, adds a seasonal pattern to volatility. Models with seasonal stochastic volatility, such as Martin Richter and Carsten Sørensen’s work on soybean futures and options, exist precisely because option prices reflect that the summer is riskier, which is a different claim from summer futures returns being predictable. The crop calendar itself is covered in agricultural futures explained.

Why seasonal “edges” usually disappoint

The curve prices the season. Borovkova and Geman’s 28% December premium for UK gas is the clearest example: the season is in the price.
Multiple testing. Twelve months times dozens of commodities produces hundreds of tests. Some will look significant by chance. Our gas table, with zero significant months, is the honest base case.
Regime change. Seasonal patterns depend on how a market is supplied. Shale gas, LNG exports, ethanol blending, renewable diesel demand for soybean oil, and changes to fuel specifications have all shifted seasonal relationships over the past two decades. Our March gas result flipping sign between 1994-2008 and 2009-2024 is a small example.
Weather dominates. A warm winter or a drought can overwhelm any average pattern in a single year, and the losses from a wrong-way seasonal position concentrate in exactly those years.
Crowding. Widely known seasonal spreads attract the same capital at the same time, which can push premiums above their norm, as the Senate record describes for winter gas in 2006.

What hedgers do with seasonality

For commercial users, seasonality is a planning fact, not a forecast. A gas utility buys winter supply in the summer at the curve’s winter premium because it needs the gas, not because it expects to profit. A refiner hedges its summer gasoline crack in the spring at a price that already includes summer-grade costs. A grain elevator earns the carry between harvest and spring by storing grain and selling deferred futures. In each case, the participant uses the futures curve to lock in the seasonal shape, which is the opposite of betting that the shape is wrong.

The short version

Commodity prices have real seasons: Henry Hub spot averaged about 6% above its yearly level in December and January and 5% to 6.5% below it in March and April over 1997-2025, and gasoline cracks rise into summer under EPA vapor pressure rules. But futures curves price those seasons in advance, as Sørensen and Borovkova and Geman model, and as the EIA curve data shows (a winter premium of about 15% to 17% visible each September). Our test of 30 years of single-contract gas futures returns found no month with significant returns, and the winter months were the weakest. Seasonal spreads can also be pushed far from normal, as Amaranth’s 2006 collapse showed. Seasonality is essential knowledge for hedgers and for reading curves, but the evidence does not support it as a dependable source of trading profits. Nothing here is a recommendation to trade it.