Big news in small samples

Abstract

Univariate time series regressions of the forex return on the forward\npremium generate mostly negative slope coefficients. Simple and refined\npanel estimation techniques yield slope estimates that are much closer to\nunity. We explain the two apparently opposing results by allowing for both\nadditive and multiplicative news. No arbitrage arguments imply that the\nmultiplicative news component must be identical across all exchange rates\nat a given point in time. Cross section estimates reveal that the movements\nin the multiplicative news component are so large that a negative\nslope coefficient for the post Bretton Woods time series regressions is not inprobable.