Climate change could induce severe reductions in agricultural yields, with global consequences in economics and food security. Understanding the inter-sectoral dynamics thus represents an important but difficult challenge, where accurate and high-resolution climate information is extremely valuable. Many holistic analyses employ climate projections that have been bias-corrected against an observational dataset and downscaled to a finer resolution. However, these procedures are inherently uncertain. In this work, we use the NASA Earth Exchange Global Daily Downscaled Projections (NEX-GDDP) dataset as well as the "raw" Coupled Model Intercomparison Project Phase 5 (CMIP5) outputs to investigate uncertainty in corn yield hindcasts and projections in the United States. The effects of temperature and precipitation on corn yields are modeled via the Schlenker and Roberts yield regression. We measure the skill of both the raw ensemble as well as the bias-corrected and downscaled ensemble in reproducing historical yield records, and project yields through the year 2100 to examine inter-model and inter-ensemble differences. Our results raise important questions about the utility of downscaled and bias-corrected climate information, particularly when employed in impact studies.