| Abstract: | Artificial neural networks were used to search for non-linear relations in high- frequency foreign exchange time series. Three years (1985-7) tick-by-tick bid prices for the Swiss franc to the US dollar exchange rate were used in this study as training data to specify predictive models for intra-day trading, which was then tested on the same exchange rate time series in the following year (1988). A simple trading rule was adopted to evaluate the models, which showed statistically significant trading profit under moderate transaction costs. In contrast, a standard linear model did not produce profit with the same training and test data and under the same trading rule and transaction cost assumption. This provides evidence for the non-linear nature of the foreign exchange time series under study. |
