We analyze the effect of parameter estimation error on the size of unconditional population level tests of predictive ability when they are implemented under a class of loss functions we refer to as ‘discrete functions’. The analysis is restricted to linear models in stationary variables. We...