Regime Switches in Swedish Interest Rates

This paper examines the forecasting properties of endogenous regime switching models for Swedish interest rate volatility. To obtain a valid statistic for determining the exact number of states, a Monte Carlo procedure is employed. The results point out the ability of regime switching models to account for even very large degrees of leptokurtosis. I evaluate the prognoses with the Diebold-Mariano statistic versus a constant variance model, a random walk in volatility and an IGARCH process. The final specification a 3 state model with the lagged level and spread between a one monthand three month rate entering both the level and variance equations is never significantly worse than the benchmarks at any horizon. All benchmark processes perform significantly worse than our model at some horizon.