Conclusion and Further Research Direction

This chapter has presented a method in which a hierarchical neural networks system can be used to model and predict the fluctuations of the Australian Interest Rate using Australian economic data.

The application of the proposed method to modelling and prediction of interest rate using Australian economic indicators is considered.

From simulation results, it was found that the hierarchical neural networks system is capable of making accurate predictions of the following quarter's interest rate. The results from the hierarchical neural networks were compared to a neural network that used all the indicators as inputs.

Long-term predictions for 6 months and 1 year from the current quarter were then undertaken with the hierarchical neural network systems proving to be more accurate in their predictions than the neural network systems. These results were found to be similar to those obtained when quarterly interest rates were predicted.

Having a time lag for some economic indicators may increase prediction accuracy. There are some indicators whose effect is not felt on the interest rate for a number of quarters, such as Consumer Price Index (Larrain, 1991). Delaying the indicator results in the system using the indicator when it has more effect on the interest rate. The accuracy of the hierarchical neural network system may also be increased if an indicator that fluctuates greatly between quarters is smoothed out using some form of moving average (such as 2-quarter, or 6-month, moving average). This would then remove any sudden peaks (or valleys) that the indicator may exhibit which could greatly affect the prediction accuracy.

Finally the structure of the hierarchical neural network system may affect the performance of the system. This is the subject of our further research.

0 0

Post a comment