This article investigates the statistical and economic implications of adaptive forecasting of exchange rates with panel data and alternative predictors. The candidate exchange rate predictors are drawn from (i) macroeconomic 'fundamentals', (ii) return/volatility of asset markets and (iii) cyclical and confidence indices. Exchange rate forecasts at various horizons are obtained from each of the potential predictors using single market, mean group and pooled estimates by means of rolling window and recursive forecasting schemes. The capabilities of single predictors and of adaptive techniques for combining the generated exchange rate forecasts are subsequently examined by means of statistical and economic performance measures. The forward premium and a predictor based on a Taylor rule yield the most promising forecasting results out of the macro 'fundamentals' considered. For recursive forecasting, confidence indices and volatility in-mean yield more accurate forecasts than most of the macro 'fundamentals'. Adaptive forecast combinations techniques improve forecasting precision and lead to better market timing than most single predictors at higher horizons.