Univariate vs. Multivariate Moving Window Spectral Analysis

摘要

The Univariate Moving Window Spectral method of time series analysis is compared with its Multivariate Moving Window Spectral Method (MWS) counterpart. We consider a tri-variate sales, price and income data set for an electrical energy market. The particular time period is for 1936 to 1978. This period is of interest because it includes 1972, the year in which the price of oil and its dependent price of electricity underwent a turning point from a downward annual trend to and upward annual trend. One might speculate that this change is due to a structural break. In this case, such a structural break may be due to a policy decision arising out of the formation of an oil cartel. Both the Univariate and Multivariate Moving Window Spectral methods forecast the turning point, suggesting that the turning point is due to the market economic forces and a better specification of the model by the Moving Spectral Method. In addition to the foregoing observations, we observe that the Multivariate MWS method produces smaller mean absolute percentage error forecasts.