This paper evaluates out-of-sample exchange rate forecasting with Purchasing Power Parity (PPP) and Taylor rule fundamentals for 9 OECD countries vis-à-vis the U.S. dollar over the period from 1973:Q1 to 2009:Q1 at short and long horizons. In contrast with previous work, which reports "forecasts" using revised data, I construct a quarterly real-time dataset that incorporates only the information available to market participants when the forecasts are made. Using bootstrapped outof-sample test statistics, the exchange rate model with Taylor rule fundamentals performs better at the one-quarter horizon and panel estimation is not able to improve its performance. The PPP model, however, forecasts better at the 16-quarter horizon and its performance increases in panel framework. The results are in accord with previous research on long-run PPP and Taylor rule models.A common problem with the papers discussed above is their reliance on ex-post revised data for the forecasting analysis. Although it seems obvious that out-of-sample exchange rate forecasting should be evaluated using real-time data, which reflects information available to market participants, it is still very rare in the exchange rate literature. Almost all existing studies on exchange rate forecasting exploit revised data which contains future information, due to revisions and additions of new data, that is not available to either policymakers or market participants. Out-of-sample forecast evaluations based on ex-post revised data yield misleading inference about the exchange rate models, and information problems of market agents are not accounted in the analysis. Meese and Rogoff (1983a) use both ex-post revised data and actual realized values of future explanatory variables to test the forecasting ability of structural models. As Rossi (2005) emphasizes, to forecast economic variables which are driven by persistent and permanent shocks, the econometrician might measure agent's probability distribution poorly by using actual realized values of future explanatory variables.To forecast exchange rates, which are primarily driven by expectations, real-time data would be more advantageous due to capturing the information set of market participants as closely as possible in contrast to ex-post revised data and actual realized values of future explanatory variables.The first paper to use real-time data to evaluate nominal exchange rate predictability is Faust, Rogers and Wright (2003). Examining the predictive ability of Mark's (1995) monetary model using real-time data for Japan, Germany, Switzerland and Canada vis-à-vis the U.S, they report that the models consistently perform better using real-time data than fully revised data. However, none of the models perform better than the random walk model. More recently, Molodtsova, Papell (2008, 2011) find evidence of predictability with Taylor rule fundamentals using real-time data for the Deutschmark/dollar and Euro/dollar exchange rates. Molodtsova, Nikolsko-Rzhevskyy, and Papell (2008) find evidence of out-...