| Summary: | 碩士 === 中原大學 === 國際貿易研究所 === 95 === Most regime switching models give good in-sample fits to exchange rate data but are usually outperformed by random walks model when out-of-sample forecasts. Prior research on the ability of exchange rate models to forecast exchange rates relies on statistical measures of forecast accuracy, but lower forecast error does not necessarily imply higher profitability or economic value.
The aim of this study is to test for and model nonlinearities in the USD/GBP and USD/JPY exchange rate returns. We apply the STAR family models (STAR; STR) used by Ter�鱷virta and Anderson (1992); Granger and Ter�鱷virta (1993), to test nonlinearities of monetary fundamentals (MF) model and AR model, and measure the economic value of predicting USD/GBP and USD/JPY exchange rate returns. Using monthly data from 1990 to 2005. Our tests rejects the linearity hypothesis for the USD/GBP and USD/JPY exchange rate returns during the 1990s, are classified as logistic ST(A)R and as exponential ST(A)R respectively. ST(A)R models all provide better good in-sample fits than linear models. We also compare forecast error, market timing ability, and asset allocation performance of out-of-sample forecasts. ST(A)R models can not beat random walk model and linear models of USD/JPY exchange rate returns MAE or RMSE forecasting, but they can provide more market timing ability and Mean-Variance asset allocation performance than linear models of USD/GBP and USD/JPY exchange rate returns forecasts. Furthermore, STAR models are best market timing and asset allocation models, and also provided most efficiency portfolio. These findings confirm the economic value importance of accounting for the presence of regimes in exchange rate returns.
|
|---|
