William Verkooijen

Estimating structural exchange rate models by artificial neural networks

No theory of structural exchange rate determination has yet been found that performs well in prediction experiments. Only very seldom has the simple random walk model been significantly outperformed. Referring to three, sometimes highly nonlinear, monetary and nonmonetary structural exchange rate models, a feedforward artificial neural network specification is investigated to determine whether it improves the prediction performance of structural and random walk exchange rate models. A new test for univariate nonlinear cointegration is also derived. Important nonlinearities are not detected for monthly data of US dollar rates in Deutsche marks, Dutch guilders, British pounds and Japanese yens.

A neural network approach to long-run exchange rate prediction

In the economics literature on exchange rate determination no theory has yet been found that performs well in out-of-sample prediction experiments. Until today the simple random walk model has never been significantly outperformed. We have identified a set of fundamental long-run exchange rate models from literature that are well-known among economists. This paper investigates whether a neural network representation of these structural exchange rate models improves the out-of-sample prediction performance of the linear versions. Empirical results are reported in the case of the US dollar-Deutsche Mark exchange rate.

Connectionist projection pursuit regression

We present a novel regression method that combines projection pursuit regression with feed forward neural networks. The algorithm is presented and compared to standard neural network learning. Connectionist projection pursuit regression (CPPR) is applied to modelling the U.S. average dollar-Deutsch mark exchange rate movement using several economic indicators. The performance of CPPR is compared with the performances of other approaches to this problem.

Design of Neural Networks for Prediction and Classification in Economic Problems

Abstract In this paper results are presented of a study on economic prediction and classification with neural networks. Comparison is made between neural networks and linear modelling techniques, and in particular remarks are made to the problem of overfitting and the estimation of prediction errors in cases where the available data sets are relatively small. It is shown that selecting network parameters by k-fold cross-validation combined with weight decay training, is an effective remedy for those phenomena. The conclusions are illustrated in two cases: predicting the volume of new mortgages and the classification of bond ratings.