How reliable are initial carryout forecasts for US crops?

The US Department of Agriculture just released initial supply and demand outlooks for the upcoming marketing year, though many market participants take these numbers, especially those for ending stocks, with a grain of salt. This is a valid approach given that the year-end inventory for 2018/19 will not be known for another 17 months. But when it comes to US supplies, USDA's initial corn forecast has been a much better approximation to the final figure in recent years than it used to be.
Soyabeans may also be following a more promising trend as USDA seems to be better incorporating the recent boost in global consumption. But the wheat numbers are still a problem, and their effect trickles all the way through the global balance sheet. In its May 10 report, USDA projected 2018/19 US corn carryout down 23 percent on the year, soyabeans down 22 percent, and wheat down 11 percent.
USDA has placed 2018/19 US corn ending stocks at 1.682 billion bushels, the lowest initial forecast in seven years. Barring drastic crop losses this summer, the actual number is likely to end up in that vicinity based on the recent track record. This is in stark contrast to the sharp swings in the many years prior. USDA struggled to predict domestic corn use with the onset of the Renewable Fuel Standard (RFS) in the mid-2000s, which mandated the use of ethanol and other renewable fuels in gasoline. Harvest losses between 2010 and 2012 continued to skew the numbers.
Starting in 2014, USDA seemingly found a relatively successful method to predict corn demand against the expected supply, even as harvests have grown much larger than originally forecasted. But the success may partially owe to the lack of widespread weather problems during this time. Larger crops as of late have likely contributed to the too-conservative nature of early export outlooks, though USDA may have adjusted for that this year. Some 2.1 billion bushels of corn are slated for export in 2018/19, down 6 percent from the current year estimate, but still the agency's largest initial export forecast in a decade.
In recent years, market participants have become well aware of the typical declining trend with US soyabean ending stock projections throughout the marketing year, especially toward the end. But this is not expected to happen in 2017/18 to nearly the degree as in prior years. USDA's carryout estimate topped out at 555 million bushels in March, and of the 15 analysts polled by Reuters two weeks ago, none of them believe the final number will land below 490 million.
If the analysts are correct, 2017/18 will become the fourth year since 1996/97 in which soyabean ending stocks came in higher than initially predicted. The 2017/18 marketing year will make it four years in a row that USDA has underestimated US wheat carryout in its initial forecast. Overly aggressive usage assumptions have been the main culprit.
USDA has been overdoing wheat consumption for longer than four years. In the past 14 years, total domestic use was smaller than originally predicted in all but one year, 2012/13, when corn output was slashed by drought. In terms of global ending stocks, USDA's initial projections have recently been off from the final volume mostly due to misses in the United States.
When omitting China, the last five initial May predictions have been for global wheat carryout to fall on the year, and each year to a wider degree than the last. But supplies rose in those first two years, and fell only slightly in the second two, and US increases are largely to blame.
This appears to be linked with USDA's recent trend of gravely underestimating Black Sea exports at the start, which have been stealing US business in recent years due to the increasing availability and attractive pricing. US exports have come in lower than originally expected in three of the last four years, reflecting strong competition among suppliers in the global wheat market.