Score contribution per author:
α: calibrated so average coauthorship-adjusted count equals average raw count
A long-standing puzzle is the near-random-walk behavior of exchange rates. Recent literature has proposed models to forecast exchange rates at medium- and long-horizons. Such tests suffer from small-sample bias but inferring the true test distribution is difficult. We propose two approaches to address the problem. First, since economists are interested in the value of economic models versus purely statistical models, we propose a horse-race that pits the economic models not against the random walk, but against the forecasts from the level of the exchange rate. These economic models are challenged because the level of the exchange rate appears to be a more powerful predictor than “global risk” variables. We also propose a second more general but less powerful test. But with both tests we demonstrate using bootstraps that the random walk cannot be rejected, so the predictive power of the lagged exchange rate and many other variables is illusory.