Exchange Rate Predictability

This paper re-assesses the out-of-sample forecasting performance of the monetary exchange rate model with different specifications on a panel of 18 OECD countries.In view of the deficien-cies in the past literature,we build the heterogeneous panel models with multifactor error structures which allow for analysis of heterogeneous panels subject to cross-sectional dependence and miss-ing variable bias(that leads to endogeneity),and select the appropriate research samples,so that we can produce more convincing inferences and improve the predictive ability of monetary model predictors.We test and compare the predictive abilities of different specifications of monetary model in different forecast horizons,on all samples and different sub-samples,using different eval-uation methods so that we can fully demonstrate the superiority of the econometric model used in this paper.Our empirical results show that monetary model predictors can generally perform better in forecasting exchange rates in all horizons than the random walk if we can make accurate estimates of the relative money supplies and the relative real outputs.However,if we only use the current and past information,monetary model predictors can still perform better in forecasting exchange rates in the short and medium horizons but can not perform significantly better in the long horizon.Our findings also suggest that heterogeneities exist between different panel units and these hetero-geneities are mainly reflected in the different marginal effects of the common effects on different countries.Our empirical study provides a new answer with new evidences to the "Meese and Rogoff Puzzle".Besides,our findings can also provide reference for developing trading strategies in the foreign exchange market.