Quarterly Grain Stocks - Soy

Quarterly Grain Stocks - Soy is the soybean line of Grain Stocks, a quarterly USDA inventory count. NASS (National Agricultural Statistics Service) publishes it, with estimates approved by its Agricultural Statistics Board. It covers about 56,000 farms except Hawaii and about 8,000 facilities in all states except five.

It measures US soybeans stored on farms and at commercial facilities on December 1, March 1, June 1 and September 1. It is a physical stock count in bushels, not a production forecast, price report or world supply table. September 1 stocks are old-crop ending stocks, not new supply, with no seasonal adjustment and no market thresholds.

Responses are collected around the 1st of March, June, September and December over about 15 days. Field offices may start up to two days early, with 4-5 days for editing and 5-6 days for board review. Results come at 12 noon Eastern at month end, except the December 1 edition in early January.

Each release appears with Prospective Plantings in March, Acreage in June, the Small Grains Summary in September and the Crop Production Annual Summary in January. Stocks are revisable next quarter and again the following January on late reports, errors or production revisions. They close after the five-year Census of Agriculture. Traders read results against trade expectations and the same date last year.

Soy stocks move with farm and commercial inventories plus implied quarterly use.

• On-farm stocks: grain held on farms for feed, seed, sale or government programs, the more volatile leg.
• Off-farm stocks: grain at elevators, warehouses, terminals, mills and processors, usually the larger share.
• Total stocks: on-farm plus off-farm soy in all positions, by state and nation, on the reference date.
• Indicated disappearance: quarterly use from stock change after exports, crushings and other recorded uses, not a direct survey.
• Production revision: prior-year output correction published with September stocks from the balance check.

In the September 30, 2024 release, September 1, 2024 stocks totaled 342 million bushels, up 29% from a year earlier. On-farm stocks rose 54% to 111.2 million bushels while off-farm stocks rose 20% to 230.8 million bushels. Indicated disappearance for June to August 2024 was 628 million bushels, up 18% from a year earlier.

The 2023 output was cut by 2.62 million bushels, with planted area unchanged at 83.6 million acres. Harvested area held at 82.3 million acres and yield at 50.6 bushels per acre. The total came below the average trade estimate near 351 million bushels, a lighter-than-expected bullish surprise. Released alongside the Small Grains Summary, corn rose 29% to 1.76 billion bushels and wheat 12% to 1.99 billion.

1. Sample: each quarter, draw list farms by multivariate probability-proportional-to-size sampling, keep a stocks panel to track change, and cover unlisted farms through the area frame. Weights are calibrated each quarter.
2. Impute: fill capacity first, then stocks from prior-stock ratios, stocks-to-capacity ratios or group means of at least five respondents, manually for the largest operations.
3. Expand: scale reports by sampling weights to estimate on-farm stocks by state. Ratio estimators use reported data only with adjusted weights.
4. Enumerate: census commercial storers and adjust for nonresponse using capacity coverage, with manual work on the largest strata.
5. Combine: add on-farm and off-farm state totals, set the national total first, then force states to sum to it. Check the balance sheet: beginning stocks plus production and imports minus recorded use must equal ending stocks.
6. Publish and check precision: publish implied quarterly disappearance from the identity. Total equals on-farm plus off-farm, and disappearance equals prior stocks plus production and imports minus current stocks. A simplified toy illustration: 7.1 million bushels in total from 2.1 on-farm plus 5 off-farm, with disappearance of 2 million. The US soybean target coefficient of variation is 2-5% by quarter, and soybean relative standard error is 5.2%, about ±10.4%.