Recent research has pointed to a reduction in predictive content in several agricultural futures markets. We investigate short-run forecast in the soybean futures market complex to identify predictive content and the sources of forecast errors. A non-parametric local linear regression framework is first applied to investigate biasness, and to guide the specification of parametric regime-switching models in which we perform statistical testing. To identify effects of risk premiums, we use a nonlinear realized volatility framework. Our non-parametric and parametric findings indicate nonlinearities in efficiency and risk premiums are present. Depending on the level of futures prices, thresholds or regimes of predictive performance exist. Evidence of market exuberance/pessimism emerges as many of the pricing errors occur at the extremes of the price distribution. Our research demonstrates that failure to account for these non-linear relationships can distort our understanding of market effectiveness. Finally, we show that the use of higher frequency data can be useful in identifying the presence and magnitude of risk premiums. This finding may make uncovering risk premiums in agricultural commodity markets more tractable in the future.