Executive Summary

The United States Department of Agriculture’s (USDA) Crop Production reports in the fall have historically provided relatively accurate estimates of Kansas’ final corn production, with estimates generally becoming more accurate between August and November. However, that pattern has weakened over the past five years. The fall Crop Production reports are also considerably more accurate in predicting overall U.S. corn production, though the accuracy does not consistently improve between the August and November reports.

Between marketing years (MY) 2016/17 and 2025/26, the USDA’s corn production estimate for Kansas in its August, September, October and November reports were off by an average of 5.21%. The same estimate for the U.S. was 2.29%. This equates to a difference of roughly 37 million bushels in Kansas and 338 million bushels in the U.S.

The USDA’s estimates have been less accurate over the past five years, particularly in Kansas, where reports have been off by an average of 8.00% since MY 2021/22. This has been driven in large part by MY 2022/23, where the fall reports were off by over 15%. Each year from MY 2022/23–2025/26 has had at least one report off by 6.98% or higher. The overall U.S. estimates have seen the same trend, though substantially lessened, with reports being off by only 2.46% on average over the same time.

The final January estimates in MY 2025/26 caused significant consternation and criticism of the USDA by farmers. The August, September and November reports (the October report was cancelled due to the government shutdown) underestimated Kansas production by nearly 8% before the increased production figures in January. However, the magnitude of the MY 2025/26 revision was not anomalous. Instead, the reaction to the January report highlights the challenge created when a relatively large revision follows several consecutive years of greater-than-historical uncertainty in the USDA’s Kansas production estimates.

The fall USDA reports vary in the accuracy in the predictions of the final January production numbers. Historical revisions between the August and November reports do not provide a reliable indication of either the direction or magnitude of the final January revision. However, farmers can utilize this information to create a range of possible outcomes for the January report based on the relative historic accuracy of the reports.

Future research could and should be conducted on the impact of the USDA’s reports to swings in pricing and whether or not the market understands crop production realities better than the USDA.

Introduction: How Accurate are the Fall USDA Reports?

The USDA releases monthly crop production reports throughout the growing season, providing farmers and participants in crop markets information on expected crop size before the final annual estimates are released in January. These reports give important information affecting futures prices, basis expectations, marketing decisions, crop insurance price guarantees and more.

The accuracy of such information is important to end-users and row crop producers, but the USDA is burdened with a difficult task of predicting outcomes in the face of tremendous uncertainty. Crop production is inherently difficult to estimate, especially before harvest is completed. Yield conditions vary geographically, weather can change throughout the seasons, harvested acreage can differ from earlier expectations and more information becomes available throughout the crop cycle. Challenges related to collecting massive amounts of data also contribute to the difficulty of accurately estimating production. However, as additional information becomes available, the estimates should become more accurate as the growing season progresses.

In recent years, production estimates from the USDA have faced increased scrutiny from farmers, particularly following the January 2026 report, which substantially increased harvested acreage and corn production from earlier fall estimates. The reaction highlighted a broader challenge facing the USDA: maintaining producer confidence in estimates that depend, in part, on voluntary producer surveys and other data collection efforts. Though not necessarily caused by the public outcry, the USDA has since undergone efforts to find ways to make the survey and data collection process more transparent and easier to conduct, while establishing accurate reporting data as one of their top priorities.

Thus, this paper seeks to determine how accurate the fall USDA Crop Production reports are compared to the final January report. This analysis compares the USDA’s August, September, October and November Crop Production estimates with the January estimate for MY 2016/17 through MY 2025/26. Estimates for both Kansas and the U.S. are evaluated. The analysis also examines whether historic revisions during the fall provide information about subsequent revisions in January. Finally, the analysis seeks to provide farmers information on how to think about the final report of MY 2026-27 given the August and September reports.

Data and Methodology

All data used in this analysis came from the USDA’s August, September, October, November and January Crop Production reports from 2016 to 2026. The analysis analyzed harvested acres, yield and production, though this paper focuses solely on production as a derivative of harvested acres and yield.

The analysis treated the January report as a benchmark for actual production, though it should be noted that future analysis can, and at times does, revise the January report’s figures.

To determine how far away each report was from the January report, a percentage error was calculated. In this instance:

A positive percentage error indicated the report’s estimates were larger than January’s. A negative percentage error indicated the report’s estimates were smaller than January’s.

Using a simple percentage error calculation to determine the accuracy of the fall reports on average has one major flaw: large positive and negative percentages can offset each other. If a report has a percentage error of -10% for five of the 10 years and 10% for the other five years, the average percentage error over the 10 years is zero, which does not accurately reflect the reality of the data.

As such, this analysis uses Mean Absolute Percentage Error (MAPE). The MAPE is used to measure the average magnitude of estimate error regardless of direction.

To determine the general accuracy of each month’s report compared to the January report, averages across years were calculated. The analysis examines five different averages of the MAPEs: 10-year Average MAPE, Previous 5-Year MAPE, Recent 5-Year MAPE, 10-Year Trimmed MAPE, Recent 5-Year Trimmed MAPE.

The trimmed averages were calculated to reduce the influence of unusually large or small estimate errors on the results. For each month, the observation with the largest absolute percentage error and the observation with the smallest absolute percentage error were removed before calculating the average MAPE. This provides an additional measure of estimate accuracy that is less sensitive to extreme observations.

 

Kansas Reports Have Become Less Accurate Over the Last Five Years

Table 2 shows each fall report’s Kansas production estimate and compares it to the January benchmark for each marketing year from 2016/17 through 2025/26. Percentage errors show the difference and directional bias between the fall report and the January benchmark. The percentage errors vary considerably across years and report months, ranging from an underestimate of 9.24% in September 2025/26 to an overestimate of 23.05% in 2022/23.

Table 3 shows the simple averages and MAPE calculated across all marketing years for each month. A comparison of the 10-Year Average MAPE, the Previous 5-Year Average MAPE, and the Recent 5-Year Average MAPE draws an interesting observation. Over the full ten-year period, the November report was the most accurate of the fall reports with a MAPE of 4.72%, followed by October at 4.81%. As expected, accuracy generally improved as the season progressed — except for the September report — its MAPE slightly increased from 5.40% to 5.91%. The same is true for the Previous 5-Year Average MAPE. Here, the pattern is particularly clear, with MAPE declining in every successive report, falling from 3.83% in August to 1.84% in November.

This observation is in line with previous expectations that as the growing season progresses and new information becomes available, reports are generally more accurate over time.

However, the Recent 5-Year Average MAPE shows a differing trend. The deterioration in accuracy is substantial. Not only is there no improvement in accuracy over successive reports, but each month is materially less accurate. November MAPE increased from 1.84% during the first five years to 7.60% during the most recent five years. September MAPE increased from 2.46% to 9.37%, while October increased from 2.21% to 8.07%. Meanwhile, August becomes the most accurate with a MAPE of 6.98% after being the least accurate with a MAPE of 3.83%.

An argument could be made that MY 2022/23 was a particularly bad year for reporting. In MY 2022/23, the USDA’s August report estimated total harvested acres would reach 5.05 million acres with a yield of 123 bushels per acre for a total of 621.15 million bushels. The September report increased harvested acres by 100 thousand, where it remained throughout the fall. Yield estimates were reduced to 122 bushels per acre in September, then further reduced to 115 bushels per acre in October where they remained. The final January report then reduced total harvested acres to 4.44 million, substantially reducing total production to 510.6 million bushels. This was by far the largest adjustment the USDA made over the ten-year period, creating the largest percentage errors.

However, simply removing MY 2022/23 does not change the trend towards less accurate reporting. A second five-year comparison was calculated excluding MY 2022/23. In this comparison, MY 2020/21 was included to maintain a five-year sample, though October has only four observations as the October 2025/26 report was not published. This second comparison brings the MAPEs for August through November to 4.10%, 5.16%, 4.76% and 4.59%. Removing MY 2022/23 substantially reduces the magnitude of recent estimate errors but does not eliminate the broader pattern.

Trimming the most and least accurate figures out of the MAPEs for the whole ten-year period and the most recent five years still shows the reports have become less accurate in each reporting month.

Not only have the fall reports become less accurate over the last five years, but the historical pattern that suggests accuracy improving as the growing season progresses has also disappeared.

One other observation from Table 2 is the tendency across the 10-year period for the USDA to over- or underestimate the January benchmark in each reporting month. Over the 10-year period, USDA overestimated the January benchmark five times in August, six times in September, eight times in October and six times in November. Likewise, the USDA underestimated the January benchmark five times in the August report, four times in the September report, once in the October report, and four times in the November report. While the August report underestimates as much as it overestimates, the September, October and November reports are more likely to overestimate.

 

Kansas vs. U.S. Reports

Over the ten-year period, the fall Crop Production reports are considerably more accurate in predicting overall U.S. corn production than Kansas corn production. Table 4 shows each fall report’s U.S. production estimate and compares it to the January benchmark for each marketing year from 2016/17 through 2025/26. The percentage error varies considerably across years and report months, though not to the same extent as the Kansas figures, ranging from an underestimate of 8.07% to an overestimate of 6.20%. The variation does not have the same extremes as those in Kansas.

Table 5 shows the simple averages and MAPE calculated across all marketing years for each month. While the estimates do not necessarily get more accurate over successive reports, the estimates are considerably more accurate than the Kansas figures. Averaged across the four fall reports, Kansas production estimates had a 10-year MAPE of 5.21%, compared with 2.29% nationally. In percentage terms, the typical Kansas estimate error was more than twice as large as the national error.

Over the full 10-year period, September was the most accurate national report, with a MAPE of 1.96%, while October was the least accurate at 2.80%. November was more accurate than August, but only marginally so. Furthermore, while national accuracy has deteriorated in three of the four reporting months during the recent five-year period, the deterioration is substantially smaller and less consistent than in Kansas. September national estimates became more accurate, while increases in MAPE for August, October and November were considerably smaller than the corresponding increases in Kansas.

Overall, while there is some variation in the percentage error for U.S. crop production, it’s clear the USDA is significantly more accurate in estimating overall U.S. production compared to Kansas. One possible explanation is aggregation. State-level production is more exposed to localized weather and growing conditions, while over- and underestimates across individual states may partially offset one another in the national estimate. This analysis does not test the source of the difference, however.

 

The 2025/26 Report

The 2025/26 January report caused significant frustration amongst farmers in Kansas and the U.S. more broadly. As such, it’s worth looking at 2025/26 individually, then putting it into context of USDA’s historical revisions.

Table 6 shows the 2025/26 estimates for harvested acreage, yield, production and each report’s percentage error from the January benchmark, and the 10-Year Average MAPE.

In January, the USDA increased estimated U.S. harvested acreage by approximately 1.2 million acres and Kansas harvested acreage by 210 thousand from the November report. Both the Kansas and national yield were increased, and altogether, these revisions increased estimated production by approximately 268 million bushels nationally and 55.61 million bushels in Kansas. The adjustment came as a shock to the market, which had been trading on the previously lower numbers, losing 24 cents on the March 2026 futures.

Despite the market reaction, the November national production estimate was more accurate than the USDA’s typical November estimate over the previous decade. Its 1.58% error was below the 10-Year Average MAPE of 2.20%. Kansas, however, was slightly different. The November production estimate underestimated January by 5.90%, exceeding the historical November MAPE of 4.72%.

Estimate errors of this magnitude are not unprecedented. Kansas production estimates in November have differed from January by more than 5% in each of the last four marketing years. Meanwhile, the overall U.S. estimates were about as accurate as the 10-year average.

This distinction is important. The November 2025/26 estimate was not the least accurate estimate in the sample; for example, the MY 2022/23 was substantially further from the January benchmark. The key issue with the January 2025/26 report was the direction of the final revision. The 1.58% percentage error in November 2025/26 was the largest underestimate of production since 2021/22, when the USDA underestimated production by 7.84%. Likewise, the 5.90% percentage error in Kansas is the largest underestimate in the 10-year period. The magnitude of the error was not unprecedented, but the combination of magnitude and direction was.

Meanwhile, market participants had been trading with the lower November production estimate as one important source of information, making the upward January revision bearish new information. Farmers found frustration because of the direction of the change. When USDA suddenly announced a bearish report, the market lost value.

This observation leads to an interesting question of how the market reacts to USDA reporting, though that is outside the scope of this analysis.

 

Can Historical Revisions Predict the January Report?

Are there certain historical trends that when applied to current reports, could predict updated estimates from the USDA in the January report? To determine whether these historical revisions could improve upon USDA’s August estimates, USDA’s Kansas estimates were adjusted using four averaging methods: total average of available data, an average of the prior five years, trimmed historical average and a trimmed average of the prior five years. Each hypothetical estimate used only revisions that would have been historically available before that year’s estimate.

Beginning with MY 2021/22, each method was used to generate hypothetical estimates for harvested acreage, yield and production. These estimates were then compared with the January benchmark. Because the resulting errors are expressed in acres, bushels per acre and bushels rather than percentages, Mean Absolute Error (MAE) was used to compare estimate performance. Table 7 shows the difference of each hypothetical estimate from the January benchmark for each year.

Table 8 shows the MAE for each hypothetical estimate method as well as the MAE of the August report, which serves as a baseline estimate for comparison to the hypothetical estimate method.

Historical adjustments provided little consistent improvement over the unadjusted August estimates. Some estimate methods were able to get close in a particular year. For instance, the 5-Year Trimmed Average method in 2023/24 was only a little over 7 acres off from the January report. Likewise, both trimmed averages in 2021/22 were only one bushel per acre off in their yield estimate.

For harvested acreage, the historical-average adjustment produced the lowest MAE at approximately 190,000 acres, compared with 192,000 acres for the August baseline — a difference of only about 2,000 acres. For yield, the unadjusted August estimate was the most accurate, with an MAE of 4.20 bushels per acre compared with 4.70 bushels per acre for the next-best method.

Similarly, he trimmed historical-average adjustment performed best for production, with an MAE of approximately 41.44 million bushels compared with 44.40 million bushels for the August baseline. While this represents an improvement of approximately 2.96 million bushels, the improvement was relatively modest and was not replicated across harvested acreage or yield.

Taken together, the results provide little evidence that historical August-to-January revision patterns can be used to systematically improve upon the USDA’s August estimates. No adjustment method consistently outperformed the original August report across the three metrics. Historical revisions may therefore be more useful for understanding the range of potential January outcomes than predicting the direction or magnitude of the January revision itself.

 

Implications

Rather than using historical revisions to predict a single January production figure, historical MAPE can be used to establish a possible range of outcomes consistent with USDA’s estimate accuracy. Table 9 shows the August and September Crop Production reports in 2026.

The USDA’s August 2026/27 Crop Production report estimated Kansas corn production at 831.60 million bushels, harvested off of 6.60 million acres for an average yield of 126.0 bushels per acre. Nationally, the USDA estimated production at approximately 16.01 billion bushels, harvested off 88.60 million acres for an average yield of 180.70 bushels per acre. The September report lowered these estimates to 792 million bushels harvested off 6.55 million acres for an average yield of 121 bushels per acre. Nationally, these numbers were also reduced to 15.800 billion bushels, 88.50 million acres, and 178.50 bushels per acre.

Table 10 applies the historical MAPE measures developed in this analysis to the current USDA production estimates for August and September.

Each range represents the current production estimate plus or minus the corresponding historical MAPE. These ranges are not forecasts of January production or confidence intervals. Instead, they illustrate the range around the current estimate implied by USDA’s historical estimate error.

Depending on which historical period and treatment of extreme observations is used, the historical-error ranges around USDA’s August Kansas estimate are relatively wide. Using the full ten-year MAPE of 5.40%, an August estimate of 831.60 million bushels implies a range of approximately 786.60 million to 876.50 million bushels. Using the Recent 5-Year Average MAPE of 6.98% widens that range to approximately 773.60 million to 889.60 million bushels.

The difference between the previous and recent five-year ranges illustrates the practical consequences of the deterioration in Kansas estimate accuracy identified earlier in this analysis. Applying the previous five-year MAPE produces a range approximately 63 million bushels wide. Applying the recent five-year MAPE produces a range approximately 116 million bushels wide.

The national ranges are considerably narrower and more consistent across the alternative historical measures. Using the 10-Year Average MAPE, the August U.S. estimate of 16.01 billion bushels produces a range of approximately 15.66 billion to 16.36 billion bushels. Even when the Recent 5-Year Average MAPE is applied, the range only widens to approximately 15.63 billion to 16.40 billion bushels. This further illustrates the greater uncertainty surrounding state-level Kansas production estimates compared with the national estimate.

These ranges should not be interpreted as evidence that January production will fall within any set of bounds. MAPE measures historical average estimate error rather than the maximum possible error, and individual years can fall outside the ranges shown. Further, the analysis in Section 7 found little evidence that historical revisions can reliably predict the direction of subsequent revisions. The ranges are therefore intentionally symmetrical around the USDA’s current estimate.

 

Conclusion

USDA’s fall Crop Production reports provide valuable information about expected corn production, but the results of this analysis demonstrate that they should not be interpreted as final production totals. This is particularly true in Kansas, where estimate errors are considerably larger than at the national level. USDA’s estimate accuracy has also deteriorated over time to the point where reports no longer become more accurate as the growing season progresses.

For farmers and other market participants, this uncertainty means that fall production estimates are best viewed as purely estimates subject to meaningful revision. This offers both risk and opportunity to market participants. Historical errors provide context for the uncertainty that remains around a fall estimate.

Importantly, historical revision patterns do not provide a reliable method for determining the direction or magnitude of USDA’s January report. The historical record cannot tell farmers or other participants whether the USDA’s current production estimates will be revised higher or lower in January. Historical errors are therefore more useful as a tool for risk management and scenario planning.

This analysis does not include the impacts of large inaccuracies on crop futures markets. Future research could be done to estimate how the market reacts to large revisions to previous estimates.