Real time data and exchange rate models

Abstract

It is well known that macroeconomic data are revised from time to time. Recently more and more researchers have noticed that data choice may affect empirical results since we usually have more than one value for a particular economic variable. Considerable results have been made in real time data studies. The main purpose of this paper is to investigate how much data type would affect the in-sample-fit performance and out-of-sample forecast performance of flexible-price monetary model of exchange rate determination. In chapter two, related literatures on monetary model and real time data is reviewed. Several East Asian countries are chosen to test the theoretical monetary model in chapter three. The testing strategy is based on the MacDonald and Taylor (1993) paper. Evidence from chapter three shows that the theoretical flexible-price monetary model does not hold for any country which is in contrast with supportive results found by previous literatures employing cointegration tests. Chapter four, chapter five and chapter six try to find out how much data type affects in-sample-fit performance and out-of-sample forecast performance of a model respectively. Results show that real time data generates more volatile fundamentals than current vintage data does for all countries and regions. For error correction specification of the monetary model, real-time-vintage data has the highest out-of-sample predictive power in terms of the three criterions. For first difference specification, current-vintage data outperforms the other types of data in terms of the three criterions. We also find that first difference specification can generate better out-of-sample forecasts than error correction specification of the monetary model in general. No previous literatures have worked on how data type affects in-sample-fit performance of exchange rate models. Compare with papers identifying data choice matters for exchange rate forecasts, this paper tests more model specifications, examines more data type and uses most recent Clark-West statistical test to evaluate empirical performances.