Forecasting price movements on asset markets is very difficult, especially at high frequencies. This is also true for exchange rates, where economists have been hard-pressed to come up with a theoretical or statistical model that can beat a random walk. And when they did, it did not hold up to the test of time. So what is the latest in this quest?
Mario Cerrato, John Crosby and Muhammad Kaleem point out the the statistical tests used to evaluate the forecasting performance and not relevant. Indeed, one does not care whether the mean square errors are low out of sample, or what the Sharpe ratio is. What really matters is how a portfolio managed using the forecasting model performs. And there, the news is good, one can beat the random walk. And this not even with a purely statistical model, but rather with a model that has some theoretical foundations. Very few of those are needed: money and GDP growth in both countries, and a time trend, the latter not being essential to beat the random walk. One can imagine that a more elaborate model could do even better.
But note that nobody here has claimed you can beat the market.