Bridge over ocean
1 April 2014 CFA Institute Journal Review

Exchange Rate Predictability (Digest Summary)

  1. Heather K. Traficanti

To determine whether exchange rates are predictable, the author performs an analysis of research papers from the last 10 years that predicted exchange rate movements. She examines traditional predictors, single-equation linear models, and multivariate models and concludes that no model or methodology exists that can consistently forecast exchange rates. The random walk model without drift provides better out-of-sample predictions than other models do.

What’s Inside?

The author’s objective is to offer a critical survey of the literature that has predicted exchange rates in the last 10 years or more. She seeks to determine whether exchange rates are predictable and, if so, which predictors are the most useful in forecasting exchange rates. Furthermore, she examines whether the random walk model performs better than other fundamental-based models (i.e., the Meese–Rogoff puzzle) and what factors cause the different conclusions documented in the exchange rate literature.

How Is This Research Useful to Practitioners?

Economists and researchers in academia, forecasters and policymakers at central banks, and private practitioners in business are the target audience for this research. The research findings provide background information on a number of foreign exchange models, predictors, and methodologies commonly used by the industry to help accurately forecast exchange rates. The author summarizes predictors and economic fundamentals and provides an evaluation of each forecasting methodology. The evaluation is documented in each section of the literature review and in a table that indicates whether the forecast methodology resulted in successful predictive ability.

The author makes several important conclusions. First, for short horizons, the Taylor rule and net foreign asset fundamentals are the most successful predictors of exchange rates out of the traditional fundamental methods. Second, linear specifications are more successful than nonlinear models. Meese and Rogoff (Journal of International Economics 1983 and Exchange Rates and International Macroeconomics 1983) and other researchers have determined that the random walk without drift has the best predictive ability for exchange rates, but it does not consistently produce effective forecasts. Forecasting capability, when present, is an occasional and short-lived phenomenon.

By noting differences in results across studies, the author demonstrates that data transformations—including but not limited to filtering and seasonal adjustments—generally result in different research conclusions for the same model. In addition, the choices of benchmark, horizon, sample period, and forecast evaluation method are important considerations.

The author’s literature review and empirical evidence confirm some previous researchers’ conclusions regarding the predictive power of the random walk at short horizons. She disagrees with others who have concluded that nonlinear specifications are successful in forecasting foreign exchange rates; she maintains that fundamentals’ predictive power varies across time, market, model, and predictor. Meese and Rogoff have suggested that sampling error, model misspecification, and instabilities may explain the poor forecasting performance of the economic models.

How Did the Author Conduct This Research?

The author critically reviews each article, discusses the predictors that have been used to forecast nominal exchange rates, reviews each model, examines the data used in the research and the forecast evaluation methods, and finally, provides a summary of the conclusions obtained. She performs a critical review of traditional predictors, including interest rate differentials, price and inflation differentials, money and output differentials, productivity differentials, portfolio balance, Taylor rule fundamentals, external imbalance measures, and commodity prices.

She analyzes single-equation linear models, single-equation error correction models, nonlinear models, time-varying parameter models, and multivariate models. In addition, the author examines the following multivariate models: value-at-risk (VaR) models, Bayesian VaR models, factor models, vector error correction models, multivariate time-varying parameter models, Bayesian model averaging, and panel models.

Next, she critically reviews a number of data considerations, including end-of-sample versus filtered data and calibrated parameters, forecasted versus realized (or ex post) fundamentals, countries and samples, frequency, and data revisions. The author then reviews which forecast evaluation methods to choose. The forecast evaluation methods include the choice of benchmark model, forecast horizon, forecast methodology, forecast sample, and evaluation method (i.e., loss function and test statistic).

For the empirical analysis, she collects data on exchange rates (relative to the US dollar) and several economic fundamentals for 18 developed countries. The economic fundamental data are obtained from the IMF database via Datastream and Philip Lane’s website. The empirical evidence results in only a few instances when the economic predictor’s forecasting ability is stronger than that of the random walk benchmark.

Abstractor’s Viewpoint

The author effectively leverages prior research from notable publications and delves deeper into an analysis of exchange rate forecasting. Within each of the model specifications, she provides relevant background information, concept definitions, critical formulas, and the most important research findings. Her findings lead to more critical questions. For example, why does the predictability of exchange rate modes change over time, and is it possible to improve forecasts? Perhaps the author will decide to perform more research to help address these final questions. In the final analysis, currency rate movements are driven by so many changing political and economic factors that it seems impossible to make predictions. If anyone were to find such a tool, the market would use it and arbitrage would eliminate its benefit, as always.

We’re using cookies, but you can turn them off in Privacy Settings.  Otherwise, you are agreeing to our use of cookies.  Accepting cookies does not mean that we are collecting personal data. Learn more in our Privacy Policy.