71 Extending Sales Forecast from Compressed Interest Rates with Wavelets and Neural Networks

This article extends our work on the forecasting of sales from compressed interest rates when the compression is done via the Discrete Wavelet Transform (DWT). Those results suggest that the forecast of sales from compressed interest rates using the DWT is often as good as and sometimes better than that from uncompressed interest rates. Further, the results show that the strength of the forecast is also dependent on the choice of wavelets used for the compression, as well as the scale of the compressed interest rates sequence. Our previous work considered the first ten members of the Daubechies family. This work expands on those results by comparing the performance in the forecasting of sales from compressed interest rates for the first twenty members of the Daubechies' family {db1, db2, …, db20}. Further, it compares the forecast results obtained from a linear prediction algorithm with that of a focus time-delay neural network (FTDNN).