Dilemmas in model selection in time series analysis
No Thumbnail Available
Date
2018-06
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
KNUST
Abstract
This study seeks to resolve two important dilemmas in model selection in time
series analysis. These are to compare the performance of the graphical and
the information criterion methods in selecting the true model. In addition, Yu
et al. (2005) relative precision performance stability was modi_ed. For the
graphical and information criterion comparison, dataset from ARIMA models
were simulated. Also the cocoa production and rainfall datasets in Ghana were
used to validate the modified relative precision performance stability of Yu et al.
(2005). It was observed from the study that, in comparison to the performance
of the graphical method and the Akaike information criterion (AIC) in selecting
the ARIMA models, the information criterion performs better than the graphical
method. Also, in verifying for the size of the evaluation sets in forecasting,
whether to select a single model or combine the models of di_erent models, our
findings showed that the size of the evaluation sets may not influence the decision
of selecting or combining since 97% of the decisions were to combine the models.
In addition to that, though there was a modification on the computations of the
relative prediction performance stability formerly utilized by [Yu et al., 2005],
the decision rule still remains the same. Hence, whether the use of the mean
or median on different the size of evaluation sets and interval, the combining
strategy still outperforms.
Description
A thesis submitted to the department of mathematics, Kwame Nkrumah University of Science and Technology inpartial fulfillment of the requirement for the degree of M.phil Mathematical Statistics