Accuracy of USDA Forecasts of Corn Ending Stocks

The magnitude of marketing year ending stocks may be the single most important factor that summarizes the price implications of the U.S. corn balance sheet. While there are obvious limitations to the explanatory power of a single price factor in a market in which price is determined by the dynamics of a number of supply and demand factors, the corn market clearly reacts to the USDA forecast of marketing year ending stocks contained in monthly WASDE reports. Given the importance of those forecasts, it is useful to examine the accuracy of the forecasts. In addition there has been considerable discussion of the corn stocks estimates in the most recent2014年5月WASDE报告，许多分析人士认为，旧作物估计太低，有些人认为新作物估计太高。在这里，我们总结了市场营销年度预测的准确性，结束了1990 - 91年至2012 - 13年度营销年度美国玉米股票的股票，并讨论了2014年5月WASDE结束股票估计的影响。

WASDE Accuracy

The marketing year for U.S. corn extends from September of the year of harvest through August of the following calendar year. The USDA issues the first WASDE forecast of marketing year ending stocks in May before harvest. Those forecasts are updated monthly in WASDE reports released between the 9^Th和12^Th下一个月的每个月。营销年度结束股票的官方估计中包含USDA谷物股report released at the end of September in the year after harvest. As a result, there are 17 monthly USDA forecasts of marketing year ending stocks that begin in May of the year of harvest and ending in September of the year after harvest. Forecast errors are calculated as actual year ending stocks minus the forecast of those stocks so that a positive difference represents an under-estimate by the USDA and negative difference represents an over-estimate.

图1总结了使用Box和Whisker图的1990 - 91年至2012 - 13年第23年结束股票的每月预测的准确性。对于每个月，盒子代表25个预测错误的范围^Th到75^Th在23年期间所有错误的百分位数。也就是说，中间的50％的预测错误在盒子代表的范围内。50％的错误发生在盒子范围之外，晶须（线）表示在23年内每个月的最大预测错误。例如，5月在收获前发布的预测中，中间的50％的错误是在过度的4.75亿蒲式耳和3.01亿蒲式耳之间。最大的过度度量为12.68亿蒲式耳，最大不足的蒲式耳为13.71亿蒲式耳。可以预料的是，随着越来越多的有关农作物大小和消费的信息，预测趋于在预测周期中变得更加准确。尽管如此，在预测周期中很晚才遇到了令人惊讶的大错误。

除了一般的错误模式外，预测错误还有其他重要方面。结束股票预测的一个重要方面是预测是否没有偏见。也就是说，平均错误接近零吗？如图1所示，每月的预测错误不会以零为中心，而在17个月中15个月中，最终结束股票的趋势趋于低估。每月偏见的幅度通常很小，在17个月中的10个月中少于5000万蒲式耳。最大的偏见与收获年的8月（+8.5亿蒲式耳）和9月（+95亿蒲式耳）的预测有关。每月偏差计算在统计上都没有与零不同。因此，从统计的角度来看，所有月份的最终股票预测都是公正的。

预测错误的第二个重要方面是错误的根源。错误是玉米资产负债表其他元素预测中错误的结果。在整个预测周期中，在三个主要类别的饲料和剩余使用，食物，种子和工业用途以及出口中，预测玉米消耗可能存在错误。此外，由于USDA最终生产估算估计直到收获后才发布，因此可能从5月到12月收获后的五月到12月，可能会出现货物大小的预测错误。最后，自USDA 9月1日的股票估计量以来，在营销年初的股票估算中的估计错误可能是在收获年的五月份出现的，直到9月底才发布。进口的估计幅度也可能是终止股票预测的错误来源，但是由于这通常是少量，因此在此分析中不考虑。

图2总结了对年终股票的预测错误来源的分析。对于五个资产负债表项目中的每一个，在预测表中的预测误差与预测表中的预测错误之间的预测错误之间进行了简单的相关性（+1完美正相关； -1个完美的负相关）。如前所述，最高的相关性是生产预测错误，导出预测错误以及饲料和残留预测错误。正如预期的那样，生产误差预测在预测周期的早期占主导地位，而在预测周期中，进料和剩余预测的错误占主导地位。乍一看，除食物，种子和工业外，所有资产负债表类别的正相关模式似乎很奇怪。这实际上是有道理的，因为在收获之前，生产和使用错误将往往会呈正相关。如果做出大量的预测，则将倾向于进行更大的使用预测，反之亦然。一旦确定作物的大小，就可以切断这种联系。

2014年5月WASDE

对年终股票的预测错误的分析可以应用于最新的预测May 9, 2014 USDA WASDE报告。这份报告包含第一个预测for the 2014-15 marketing year and year ending stocks were projected at 1.726 billion bushels. Historical forecast errors suggest that there is a 50 percent probability that actual year ending stocks will be between 1.251 billion and 2.027 billion bushels. There has been a slight bias towards over-estimating year ending stocks in this report in the past, with an average error of -42 million bushels. In addition, the May forecast this year includes a forecast of record high average yields and a year-over-year decline in consumption of U.S. corn. This combination suggests that actual year ending stocks will be smaller than the May forecast.

对于2013-14的营销年份，预计结束股票为11.46亿蒲式耳。历史预测错误表明，实际结束股票在11.21亿至12.67亿蒲式耳之间有50％的概率。过去，这份报告中的股票估计不足一年，平均错误为4,900万蒲式耳。这种偏见似乎支持了市场对USDA过高估计的营销年度出口的期望，因此估计了年终的股票。但是，我们先前的分析表明，在营销年底，预测一年结束股票的错误与饲料和剩余使用的预测中的错误相关。

我们进一步检查了5月WASDE预测（收获后）结束股票的错误，通过检查4月至5月的变更的关系与5月预测的错误有关。也就是说，我们计算了5月在5月预测中的变化条件下的预测误差。这些结果总结在图3中，其中在水平轴上绘制了从4月到5月（收获之后）结束股票的预测的变化，并且在垂直轴上绘制了5月预测中的误差。高于水平轴的观察值的优势验证了USDA在5月结束股票的不足年份的趋势。但是预测可能与这些股票的过度估计有关。虽然合身不是很好，但这种关系很有趣。这种关系支持了这样一个概念，即美国农业部倾向于“平滑”在预测周期结束股票的年度预测的变化。这并不奇怪，因为先前已经报道了类似的结果美国农业部玉米生产预测。

The forecast of 2013-14 marketing year ending stocks declined by 185 million bushels from April to May this year. Reading directly from the line fit in Figure 3, the change suggests that the current forecast of year ending stocks is 139 million bushels too high. The fit is not good enough to generate a lot of confidence, but that result is very different than current market expectations that year ending stocks will exceed the current USDA forecast. A large over-estimate of year ending stocks would imply that the forecast of feed and residual use of corn is too low.

Implications

我们总结WASDE马的预测的准确性rketing year ending stocks of U.S. corn for the 1990-91 through 2012-13 marketing years and discuss the implications for the May 2014 WASDE ending stocks estimates. Three main implications emerge. First, WASDE ending stocks projections for U.S. corn across the forecasting cycle are basically unbiased, having only a slight tendency towards under-estimation of final ending stocks. Second, the first WASDE estimate of ending stocks for U.S. corn, released in May before harvest, has a large potential range of errors. This is sensible since the ending stocks forecast at this early point in the forecasting cycle reflects both production and usage errors. Third, there is a clear tendency for the USDA to "smooth" the changes in projections of year ending stocks later in the forecast cycle. This tendency suggests that the WASDE ending stocks forecast for the 2013-14 marketing year released earlier this month (1.146 billion bushels) may be over-estimated. Intriguingly, this is the reverse of current market expectations that year ending stocks for 2013-14 will exceed the current USDA forecast.

Issued by达雷尔好and斯科特·欧文（Scott Irwin）
Department of Agricultural and Consumer Economics
伊利诺伊大学