Methodology: dates, units and comparison windows
The inventory, crop-balance and curve calculations behind the commodity atlas, including weighting, missingness and ratio limits.
The atlas separates observed stocks, dated crop forecasts and hypothetical delivery prices. Its calculations are descriptive. Each uses a declared period, unit and denominator; none establishes a tradable signal, investment return or causal price forecast.
Inventory observations and their vintage
The retained EIA data contain every weekly period date from 2019 through 2024 for U.S. commercial crude oil excluding the Strategic Petroleum Reserve and Lower 48 working natural gas. Crude values remain in thousand barrels; gas values remain in billion cubic feet (Bcf). A period ending date is not a release timestamp. The files were retrieved for this edition and can incorporate later revisions. They do not recreate the information available to a trader on each historical date.
The source record identifies the official file, extraction, retrieval time and hashes. Crude rows were checked against the separate official history table. Gas uses the total Lower 48 column, not a sum silently assembled from mixed regional vintages. Exact historical inputs travel with the downloads so the published calculations do not change when an upstream series is revised.
Monthly means and five-year comparisons
For each calendar month, the lab takes the arithmetic mean of weekly observations whose period dates fall in that month. Four and five observation months have different raw counts. This is an equal-observation mean, not a daily time-weighted average. The comparison first calculates each of the five previous years’ same-month means, then gives those five means equal weight. Pooling every weekly observation would give some years more weight and would be a different method.
A comparison requires at least four observations in the selected month and in each of the five prior same-month samples. Otherwise the baseline and differences are withheld with a reason. The 2019–2024 extract therefore supports full five-prior-year comparisons for 2024. Earlier years remain inspectable without an invented history. The baseline minimum and maximum describe five monthly means, not a confidence interval or the full weekly range.
The absolute difference is selected mean minus baseline mean. The percentage difference divides that change by the baseline mean and multiplies by 100; a zero baseline leaves the percentage unavailable. Annual statistics describe the first and last observed dates within the year, their difference, and the dated extrema. They are not assumed calendar-year opening and closing stocks. None of these measures is EIA’s official weekly five-year comparison.
Two frozen corn forecast vintages
The corn study transcribes the U.S. corn table on printed page 12 of the May 10 and August 12, 2024 WASDE reports. Both columns refer to marketing year 2024/25. Ending stocks divided by total use, multiplied by 100, gives the stocks-to-use percentage. The denominator contains total domestic use and exports; ethanol is already a subset of food, seed and industrial use and must not be added again.
The reports publish rounded million-bushel figures. Component sums can differ from a reported total by one million bushels. The download preserves those published numbers and names the rounding difference. A comparison of forecasts measures a revision to an outlook, not the realised harvest, final consumption or a market surprise relative to traders’ expectations.
A cost illustration with signed prices
For a near price N and far price F, the absolute spread is F minus N. When both are positive, the percentage slope is (F/N − 1) × 100 and its simple annual scaling multiplies by 365/days. These are cross-sectional curve descriptions. They are not the return from closing and reopening a futures position, nor a forecast of price convergence.
The illustrative cost-based far value is N + N × funding_rate × days/365 + storage_per_day × days + insurance. Funding uses a decimal annual rate; insurance is a per-unit cost over the full horizon. That illustration requires a positive near price. Negative or zero near prices leave its funding, far value and residual unavailable; the absolute spread and storage cost remain valid. A positive near and nonpositive far price still permits the cost calculation but withholds percentage slopes.
The residual is the entered far price minus the assumed cost-based value. It does not measure convenience yield or an executable arbitrage opportunity. Real financing, storage access, quality, location, delivery, margins, fees and trading constraints are outside this calculation. The example accepts matching units supplied by the reader and does not verify a quoted contract or market price.
Source notes
Source dates and scope matter. Historical inventory observations are retained snapshots; forecasts and synthetic curve assumptions are labelled separately. No investment outcome or product coverage is implied.
Version 1 · Published 16 October 2026. Inspect the method · Explore the data and curves