Which continuous-time model is most appropriate for exchange rates?

This paper attempts to realistically model the underlying exchange rate data generating process. We ask what types of diffusion or jump features are most appropriate. The most plausible model for 1-minute data features Brownian motion and Poisson jumps but not infinite activity jumps. Modeling periodic volatility is necessary to accurately identify the frequency of jump occurrences and their locations. We propose a two-stage method to capture the effects of these periodic volatility patterns. Simulations show that microstructure noise does not significantly impair the statistical tests for jumps and diffusion behavior.>