The year 1983 was a turning point in studies on the prediction of the exchange rate. Two great economists R.A.Meese and K.S.Rogoff have published a surprising study. In it they showed the difficulties of all models built so far to "explain" the behavior of the exchange rate outside the sample period, which, in itself, was not very new. The worse was to come: a purely mathematical model and without any theoretical support , called "random walk ", seemed to dominate - in terms of forecast - the sophisticated models built with the "fundamentals"! This fact has been called the "Meese-Rogoff puzzle." It has stimulated
much of the research of the following thirty years.
After an analysis of more than 150 theoretical and empirical articles, the author, Barbara Rossi, summarizes her observations in five conclusions:
much of the research of the following thirty years.
After an analysis of more than 150 theoretical and empirical articles, the author, Barbara Rossi, summarizes her observations in five conclusions:
1) the successful forecast of the exchange rate, in a range outside the used sample, depends on the choice of the predictor. Although there is disagreement, in the literature, empirical evidences generally are not favorable to traditional predictors (interest rate, productivity, prices, GDP and currency). On the other hand, the estimate of the exchange rate between two countries that coordinate their interest rates under the same criterion (Taylor rule) seems more promising;
2) usually linear models seem to be more successful. In linear models of only one equation, the choice of the predictor is more important than the specification of the model itself;
3) the transformation of informational data (trend extraction, use of filters, seasonal corrections) can have an effect on the predictive ability of the model. This does not appear to depend on the frequency of the information, but it is sensitive to the choice of the country;
4) empirical results show that the choice of the model that will control the predictive ability, the horizon in which it will be made, the sample period and the method of evaluation are very important. The random walk without drift is the benchmark that must be overcome, and
5) finally, the author says: "In general, some factors (Taylor rule fundamentals and net foreign asset) reveal some ability of predictability in a shorter horizon, and other (monetary fundamentals, especially in panel models) a predictive skill at long-term horizons. She continues: "none of the predictors, models or tests shows systematic empirical support for all models, countries and time horizons. Typically, when the predictability appears, it does so occasionally for some countries and for a short period of time", to conclude: "Thus, the puzzle proposed by Meese and Rogoff, in 1983, does not seem to have been convincingly overcome so far.”
I am not qualified to assert, but I suspect that the use of Taylor rule fundamentals in the construction of exchange rates between two countries can be a victim of "petitio principii". If the hypothesis is true, it further increases the poverty of results of 30 years of active research on the predictability of the exchange rate, of which some economists seem so secure at expense of their credibility and profession.
Antonio Delfim Neto
Valor Economico
Antonio Delfim Netto is an economist and professor. He graduated, in 1951, from University of São Paulo’s School of Economics and Business (FEA). He was finance secretary for the state of São Paulo, finance minister, agriculture minister, planning minister and Brazil’s ambassador do France
