Banking on currency forecasts: How predictable is change in money?

Nonlinear prediction of exchange rates with monetary fundamentals

“…Mark (1995) finds that at the 3-and 4-year horizons, a flexibleprice monetary model, produces lower RMSE than the RW for three out of four currencies that he investigates, and for the Deutsche mark at the 4-year horizon, his model can beat the RW at the 5% significance level. Using data for five currencies, Chinn and Meese (1995) report that at the 3-year horizon, one of the three structural models they estimate produces a significantly lower RMSE than the RW for the Japanese yen, although results for the other four currencies remain largely negative. 1 Other studies on exchange rate unpredictability include Diebold et al (1994), who report that incorporating the cointegration relationship among exchange rates as documented by Bollerslev (1989, 1994) does not appear to improve upon the RW in out-of-sample forecast.…”

Learning to forecast the exchange rate: Two competing approaches

“…We now substitute equation (8) and (10) into equation (11) and the latter into equation (2). This yields:…”

“…Frankel and Froot (1987) find an expected half-life for deviations from the fundamental proxied by PPP of around 3 years. Similarly, Mark (1995) and Chinn and Meese (1995) demonstrate that the models incorporating a set of fundamentals have some statistically significant power over the horizon of 3 years. These results are not inconsistent with traders' expectations.…”

“…In Section 7, we set a higher value of the gain, 7 For the detailed description of the traders' forecasts see Cheung and Chinn (2001) and Cheung et al (2004). 8 The empirical evidence suggests that the convergence of market rate to the fundamental rate may take longer than one year. Frankel and Froot (1987) find an expected half-life for deviations from the fundamental proxied by PPP of around 3 years.…”

Macro Approaches to Foreign Exchange Determination

“…In the earliest implementation of this approach, Nelson Mark (1995) used a calibrated flexible price monetary model to perform out-of-sample predictions of the dollar-deutschemark rate, and finds substantial improvement for the monetary model's performance vis-a-vis the random walk. Chinn and Meese (1995) examine a broader number of models including the flexible price, the Hooper-Morton, and augmented monetary, and imposing the cointegrating vector in an error correction framework, find that a random walk can be outperformed, in a statistically significant sense, at longer horizons of two to three years. This set of results is intuitive.…”