Long-term forecasts about commodity market indicators play an important role in informing policy and investment decisions by governments and market participants. The USDA’s baseline reports are the primary source of long-term information for the US farm sector, but recent research shows that these projections do not stay informative for longer horizons. We examine whether the accuracy of long-term forecasts can be improved using deep learning methods. We first formulate a supervised learning problem and set benchmarks for forecast accuracy. We then train a set of deep neural networks on a training sample and measure their performance against the benchmark model on a test sample using a walkforward validation strategy. We find that while the USDA baseline projections perform better for the shorter horizon, the performance of the deep neural networks improves for the longer forecast horizons. The findings may help future revisions of the forecasting process.