Time Aggregation and the RESET Test: Implications for Exchange Rate Modeling

Abstract

This study focuses on the effects of time aggregation on the performance of Ramsey’s RESET test in testing linearity, which may be especially relevant to exchange rates. The results of a Monte Carlo investigation across different levels of time aggregation, autoregressive parameter values, and sample sizes indicate that the wider the level of time aggregation, the greater the possibility of accepting the spurious hypothesis of linearity, reflecting lower sensitivity to misspecification detection itself based upon the autoregressive parameter and sample size. The data sampled systematically possess a better initial advantage in power compared to time-aggregated data over brief spans; their relative performance converges as the span increases. The empirical analysis of the exchange rates confirms such results. It shows that, with an increase in the sample size, relations among the exchange rates generally become more linear. For some levels of aggregation, it gives a ratio value that has persistently been lower, indicating that under such conditions, the modeling has been apt. These results bring out the crucial importance of considering the effects of time aggregation while conducting the RESET test, pointing to caution in result interpretation, especially for time-aggregated data. One can judiciously choose levels of aggregation, use cautious interpretation of test results, and employ complementary methods to be secure about model specification and correct estimation of economic relationships. These insights also generalize the wider analysis of time series, pointing out general problems of time aggregation and its implications for statistical criteria of interdependence of economic variables.