Page 350 - Special Topic Session (STS) - Volume 3

STS550 Angelia L. Grant et al.

Forecasting household consumption
components: A forecast combination approach
Angelia L. Grant, Liyi Pan, Tim Pidhirnyj, Heather Ruberl, Luke Willard

Abstract
This paper outlines a methodology for forecasting the components of
household final consumption expenditure, which is necessary in order to
forecast revenue collections from a number of different taxes. A forecast
combination approach using autoregressive models, regressions on relative
prices and the almost ideal demand system developed by Deaton and
Muellbauer (1980) is found to offer a more robust forecasting framework than
using one of the single models alone. In particular, the combination approach
outperforms the almost ideal demand system, which is currently used by the
Australian Treasury to forecast the components of consumption. The
combination framework takes advantage of models that account for the
persistence and longer-term trends experienced in a number of the
consumption components, as well as shifts caused by evident relative price
changes. A forecast combination framework is shown to be particularly useful
when forecasting over a three-year forecasting period.

1. Introduction
Forecasts for each of the expenditure components of nominal GDP are
important for forecasting tax revenue collections – different compositions
result in different tax revenue forecasts. A particularly important task is the
forecasting of the components of household final consumption expenditure.
This is because different components of consumption are subject to different
taxes. For example, alcohol, tobacco and fuel are subject to excise taxes, while
motor vehicles may be subject to the luxury car tax. A number of the
components of household final consumption expenditure – durables, other
goods, electricity and gas, and other service – are also subject to the goods and
services tax.
A wide variety of models can be used to forecast the components of
household consumption, with different models using different types of
information. Some models are good at accounting for the persistence and
longer-term trends experienced in a number of the consumption components,
while other models are better at taking into account shifts caused by relative
price changes. It is also the case that some models are better at forecasting
over shorter time horizons, while others are better over longer time horizons.
Under these circumstances, a forecast combination approach has a
number of advantages. It allows the use of information across a number of

339 | I S I W S C 2 0 1 9