Forecasting Exchange Rate Volatility: The Case of the Czech Republic, Hungary and Poland

We study various models for forecasting one-day forward volatility of the exchange rates of the Czech koruna, Hungarian forint and Polish zloty against the euro. We used high-frequency data to calculate realized volatility. We found that our benchmark model, the heterogeneous autoregressive (HAR) model of Corsi (2009) is rarely out-performed, even if we extend the standard HAR model by including signed jumps or substituting continuous and jump components, or if we allow the autoregressive parameter of the HAR model to vary with the estimated degree of the measurement error (Bollerslev et al., 2016). Our results suggest that the preferred forecasting strategy is to average univariate forecasts, as these combination forecasts offer improvements upon the benchmark (CZK/EUR, PLZ/EUR) or do not lead to worse forecasts (HUF/EUR). Extensions of the HAR models with regional and global exchange rate volatilities and multivariate HAR models which also model covariance between exchange rates (Baruník and Èech, 2016) have usually performed worse than the benchmark. Therefore, our study offers little evidence of volatility spillovers, an exception is spillovers from USD/EUR to CZK/EUR and PLZ/EUR and from HUF/EUR to CZK/EUR and from CHF/EUR to PLZ/EUR.