Bootstrapping and Jackknifing Neural Networks for Noisy Financial Time Series

In this paper we introduce resampling techniques to a multi-layer feed-forward neural network model for noisy financial time series in order to obtain more reliable interval forecasts of the time series along with a large amount of statistical information associated with the observed data. In particular, we develop two new grouped jackknife learning algorithms from cross-validation back-propagation learning as well as two new bootstrap cross-validation learning algorithms inspired by the parametric and nonparametric modelling strategy to be used on the neural network model selected from pre-tests. Our applicationis in forecasting the spot Canada/US foreign exchange rate, using the daily data from January 2, 1984 to October 1, 1996 and exploiting the existence of a stable transmission link between the spot rate and the short-term interest rate spread.