Temperatures across the U.S. Midwest have been very persistently below normal since the end of July, and in the past few weeks this anomaly has become slightly unfavorable for soybeans, leading to a slight reduction in the yields expected by CropProphet (see figures below). Nevertheless, soybean conditions remain very good overall, with CropProphet showing above-trend yield in every state except Indiana. According to CropProphet, the most likely U.S. yield is 48.5 bushels per acre, which would be the second highest on record after last year. Owing to record acreage, U.S. production will probably exceed the 2016 record, although there is still a greater than 25% chance that production will be below 4.3 billion bushels, as indicated by the confidence intervals in the third figure below.





The USDA estimates of U.S. soybean yield and production were revised upward again in the September 12 crop production report, and the CropProphet forecasts remain significantly lower; the difference in the U.S. production forecasts is 100 million bushels, which is similar to the difference at the time of the August 10 report. The CropProphet state-level forecasts have increased relative to USDA in Minnesota and Nebraska, where the USDA estimates were revised downward, but the CropProphet forecasts remain much lower than the USDA in eastern portions of the Midwest and from Iowa to Louisiana (see figure below).



While the CropProphet forecasts continue to suggest that the USDA estimates are too high, it is useful to note that U.S. soybean yields have consistently outperformed expectations in recent years, and it appears likely that the USDA is now attempting to correct for this bias. For example, the September USDA yield estimates have been too low in each of the past six years, and in the past four years the cross-validated CropProphet hindcasts were also too low (see figure below; note that the blue (USDA) and green (CropProphet) lines have been lower than the red line in recent years).



A plausible explanation for the recent low bias in the forecasts is that soybean seed technology may have undergone a sudden advance in around 2013. CropProphet and USDA both use a linear trend line to account for technological advances over time, and it is possible that the trend line may now be too low, leading to forecasts that fail to anticipate the high yields that can now be achieved. While the USDA may now be subjectively adjusting their forecasts upward to account for this issue, the CropProphet forecasts are entirely objective, which provides a major advantage in most situations; however, it may be possible to improve the CropProphet forecasts by accounting for sudden improvements in baseline soybean yields. Further investigation will be required, but if it is possible to identify these discontinuities in baseline yield, then the skill of the forecasts may improve significantly as the year-to-year deviations from the revised baseline would then correlate more closely with the weather and satellite predictors. A detailed examination of this issue will be part of the annual CropProphet upgrade process in preparation for the 2018 season.