When nontraded goods prices are accounted for consistently and genuine stock data on bilateral foreign asset holdings is employed, a modified sticky-price exchange rate model by far outperforms the benchmark random walk-model in empirically forecasting the D-mark/dollar parity out of sample. Superior forecast performance holds both over long horizons and from the first step. Extending the sample back to the Bretton Woods period leaves the model's parameters and its performance virtually unaffected. By implication, the explanatory variables of the model show a pattern of exchange rate regime-dependent volatility that is similar to that of the real exchange rate itself.