The paper proposes a forward premium regression with time-varying parameters.
Uniform inference indicates the extent of rejections of uncovered interest parity.
Time variation in the slope coefficient can be explained by fundamentals.
Model averaging outperforms individual models based on risk variables.