Herding in the USDA International Baseline Projections

USDA’s annual Agricultural Baseline Projections contribute significantly to agricultural policy in the United States, and hence their accuracy is vital. The baseline projections present a neutral policy scenario assuming a specific macroeconomic situation and allow the analyses of alternative policies and their micro and macroeconomic impacts in the United States. We investigate the trends and heterogeneity in the incidence of bias in the USDA International Baseline Projection reports from 2002 to 2021. The evaluation of bias as it varies geographically, temporally, and by crop-variable allows us to make inferential judgments about the sources of this bias. First, we use the dynamic time warping algorithm to examine whether experts tend to group together the projections for certain crops across different countries, producing similar projection trends. We find that projection series for all countries in the sample are correlated with the United States in their trends. Second, we compute the bias in projections and decompose it by projection horizon. Third, we assess whether the bias is higher across crops or across countries with more substantial evidence for grouping behavior and find that for soybeans imports, soybeans ending stocks, and wheat area harvested, similarity in projection trends with the United States lowers the bias while for most other crop-variables it increases it. This suggests that the projections for our sample countries are unnecessarily made to follow similar trends to the United States projections which proves to be a bias inducing choice in most cases.