4 Conclusions
This paper presented two error-correction models of consumption for the primary purpose of forecasting consumption expenditure growth in New Zealand. An improvement over earlier New Zealand studies is the use of the WestpacTrust Household Savings Indicators (HSI) wealth data, allowing the wealth variable to be modelled in disaggregated form. In addition, two novel variables were included in the modelling to address the criticisms raised by Hendry et al. (1990). A mortgage equity withdrawal variable was used to proxy for increased liquidity by households able to borrow against the value of their housing assets. Although the mortgage equity withdrawal variable was not found to be significant, it did contribute to a higher overall model fit. The other variable used was net migrant transfers, to capture the effect of funds brought into the country by immigrants which are available for investment or consumption. The migrant transfers variable was found to have a significant contemporaneous effect on short-run consumption growth.
Non-financial wealth was found to have contemporaneous as well as lagged effects on consumption growth in the short-run, but no effect in the long-run. In contrast, financial wealth was not found to have any significant short-run influence although it had a long-run effect on consumption. The non-financial wealth variable is essentially capturing housing wealth, implying that households react to changes in house prices over the short-term. However, the long-run level of consumption depends on the level of households’ financial wealth.
The out-of-sample forecast performance of both models appear reasonable, with no evidence of major systematic bias. Although both models over-estimated consumption growth over most of 2001, the models had the ability to replicate the variability of consumption growth. ECM (B), the parsimonious model based on the Stock and Watson estimated long-run model, performed better than ECM (A), which was based on the ordinary least squared estimated long-run model, over the out-of-sample forecast period. However, ECM (B) lacks one observation point due to the Stock and Watson lead-lag structure.
There are two major shortcomings in this paper, both data related. While there is now a comprehensive breakdown of household wealth data available, the data series is relatively limited. For this reason, the period of analysis is restricted to the 1989:4 to 2002:1 period. The second shortcoming is the absence of a quarterly disposable income series. An after-tax labour income series was derived, which excludes the potentially important farm income. Data limitations are not unique to this paper, it is a common frustration for many empirical researchers seeking to analyse New Zealand data. Further research and model improvement include incorporating farm income into the income variable, a longer time period for estimation, and testing other variables that may explain the period where consumption growth was consistently higher than expected over the 1998:1 to 2001:2 period.
