Forecasting High Frequency Exchange Rates Using Cross-Bicorrelations

Abstract

The central motivation for this paper is to combine a multivariate nonlinearity test with a nonlinear time series forecasting model of the VAR-type which incorporates lags of the cross-products with the other exchange rate in the system using high frequency data. The paper draws together two sets of literature which have previously been considered entirely separately: namely time series tests for nonlinearity and multivariate exchange rate forecasting. The out-of-sample forecasting performance of the new technique using cross-bicorrelations is inferior to that of univariate time series models, in spite of significant in-sample cross-bicorrelation statistics, although it does seem to be able to predict the signs of the returns well in certain cases. A number of explanations for this apparent paradox are proposed.