The forward puzzle is traditionally explained as the presence of a covariance-risk premium, market friction or limits to arbitrage. Recently, Liu and Sercu, working on intra-ERM rates for the DEM, presented evidence consistent with career risk considerations: portfolio managers shun assets with danger signals. In this paper, we test the external validity of this finding: we compare floating rates to band regimes, and strong base currencies to weak ones. We find that both the exchange rate regime and base currency strength influence the evidence on various theories: floating and strong intra-ERM rates weakly support market friction or limit-to-arbitrage theories, while the HKD and weak intra-ERM strongly support the career risk effect. We also decompose forward premium into a short-term filtered component and a long memory trend. The filtered component is good at recognising danger signals or ‘extreme’ observation effects for band regime rates and weak floaters, while the trend works best for strong floaters. Lastly, the filtered premium provides the best fit, consistent with the idea that it has a closer link to expectations than the trend component.