Time Series Forecasting for Demand of Vehicles
Student Name: Krishna Murugan
Certain organization was interested in predicting the demand for their
freight vehicles in the European market for transportation of goods.
Past monthly data on demand was available for this purpose. To enahance
the forecasting model observations were collected on several monthly
economic indicator series as well, which include variables like consumer
price index, producere price index, production volume index, retail series
index, volume of trade, share price index, commercial bank lending rates,
employment index, enerrgy figures etc. Next several approaches were
considered for building an econometric forecasting model. First a simple
ARIMA model was built for the vehicle demand series with no co-variate.
Next a co-integration model was developed for the vehicle demand series
along with the individual covariates as they were observed. Finally
the independent series were factor analyzed, after a preliminary prinicipal
component analsis, to remove multicollinearity and dimension reduction
(which interestingly brought the oriniginal list of some ten odd independent
variable to only two), and then a cointegration model for the vehicle
demand series was built with the two factor score series. The performance
of these models was then compared using AIC and their prediction accuracy
in out of the sample period. The co-integration model with the factor scores
turned out to be the one with the best performance, while the simple ARIMA
model of the vehicle demand series was the worst.