| The United States agricultural sector is an integrated component of the international economy. Exchange rate forecasting is important for both agricultural policy makers and individual agribusiness traders. Thus, the purpose of this study is to construct alternative vector autoregressive forecasting models to be used to forecast the exchange rates of major agricultural trading partners. Then, using bias test, mean square error test, and market timing tests, the forecasts of dollar/yen, dollar/Canadian dollar, and dollar/D. mark exchange rates are evaluated to determine if the models yield precise information for individual users to incorporate into their decision making framework.; The estimation period starts in 1973:3 and extends through 1982:12. The out-of-sample forecasting time period used for analyzing the models is from 1983:1 through 1989:12. The forecasts are evaluated across one-, three-, six-, twelve-, and twenty-four-month forecast horizons.; The alternative VAR models include a full VAR, mixed VAR, and two Bayesian VAR models. The models are based on the monetary/asset model of the exchange rate determination theory developed by Driskill et al. The four different VAR models show the different methods used to restrict the lag length of VAR system. The full VAR model assumes that the lag length in each variable and equation are identical. The mixed VAR model is less restrictive than the full VAR model since it permits the different lag lengths on each variable included in the VAR system. Further, the Bayesian approach imposes prior information regarding the mean and standard deviation of the parameters into the full VAR model.; The important findings in this study are first, that all exchange rate forecasts are unbiased at shorter forecast horizons but biased at longer horizons. Second, the Bayesian VAR I model yields more accurate forecasts than a random walk process but lacks of market timing ability in the dollar/yen exchange rate. Third, the mixed VAR model of the dollar/Canadian dollar exchange rate has economic value across all forecast horizons. However, the model performs more accurately than a random walk process at shorter forecast horizons (three month and less) only. Finally, none of the models is accurate in forecasting either the size or the direction of the dollar/D. mark exchange rate. |
