Intelligent Engineering Systems through Artificial Neural Networks
75 Variable Compression of Interest Rates in the Forecasting of Aggregate Sales
Download citation file:
- Ris (Zotero)
- Reference Manager
In exploring the degree to which compressed interest rates are comparable to their uncompressed versions in predicting the future values of aggregate sales, it was found that they were at the four different compression levels examined in 1-step ahead prediction. These levels were 3.4%, 8.6%, 19.9%, and 27.7% of the DCT spectral components of T-bills interest rate. The types of models used in the forecasting regimes were 2-input nonlinear neural networks and robust multilinear regression. They produced statistically similar performance statistics in the medians of correlations, RMSEs, MAEs, and Theils at the 0.05 significance level. From the examinations of the figures and performance statistics, robust regression models appeared to produce better forecasts for the one month horizon while the neural network models seemed to perform better for the four and seven month horizons.