3.2  Short-run estimation

Tables 3 and 4 list the key results from the short-run error correction models of equation (2), based on the long-run relationship of Model C in Table 2. The models presented in Table 3 uses the long-run relationship of Model C estimated using the OLS procedure, while the models presented in Table 4 uses the long-run relationship of Model C estimated using the S&W procedure.

For both short-run models, an initial equation ECM was estimated based on the general structure of equation (2), with contemporaneous as well as lagged differences of log income, log net non-financial wealth, log net financial wealth, unemployment rate, real 90-day interest rate, real net migrant transfers, real mortgage equity withdrawal, and log of consumption (lagged differences only). Due to the limited sample period, 2-lags as opposed to the 4-lag structure commonly used by other researchers was employed. The initial ECM equation therefore is represented by following form:

(3)    

In both initial ECMs, all the variables were not significant, including the error correction terms. The F-statistics for both initial ECMs were also not significant. This could be due to over-specification given the relatively small number of observations. Parsimonious ECMs were obtained using Hendry’s general to specific modelling approach. Lagged structures were preserved where appropriate to adequately model the dynamic effects.

In both the parsimonious ECMs, the overall models are significant. The error correction terms have the expected negative signs and are significant, implying that when consumption is below (above) the target consumption level as implied by the long-run relationship, consumption growth is faster (slower) than usual in the following quarter to close the gap. The error correction coefficients from the two parsimonious ECMs are also broadly similar. A negative percentage point consumption deviation from the long-run in the current quarter leads to an additional 0.22 percentage point increase in consumption growth in the following quarter according to ECM (A), and an additional 0.27 percentage point increase according to ECM (B).

Income growth was found to have a positive contemporaneous effect on consumption growth, but no lagged impact. McDermott (1990) and Corfield (1992) found similar contemporaneous effects in their short-run models. Households respond immediately to changes in their income growth, but the magnitude differs between ECM (A) and ECM (B). A 10% increase in income growth in the current quarter leads to a 2.9% increase in consumption in the same quarter according to ECM (A), but only a 1.7% increase according to ECM (B), although the income coefficient in ECM (B) does not appear to be significant.

Unlike the long-run equation, non-financial wealth was found to have contemporaneous as well as lagged effects on consumption growth. However, financial wealth was not found to have significant short-run influence although it was found to have a long-run effect on consumption. This suggests that while financial wealth influences consumption in the long-run, households react to changes in their non-financial wealth in the short-run. Because the non-financial wealth variable is essentially housing wealth, this implies that households react to changes in house prices over the short-term, but if their consumption levels exceed the long-run levels, then consumption growth needs to be curbed further out. A 10% rise in non-financial wealth leads to a 1.9% increase in consumption within a year according to ECM (A), and 2.1% increase according to ECM (B).

The change in the unemployment rate has a negative impact on consumption growth over two quarters, but no contemporaneous effects. This implies that uncertainty leads households to reduce their consumption but this adjustment does not take place immediately. The uncertainty impact is not large, with a 1 percentage point increase in the unemployment rate leading to between 0.0015% and 0.0024% decrease in consumption within a year, indicating perhaps that households smooth through periods of uncertainty.

Note: Normal t-values are reported in parentheses. Refer to Appendix B for more detailed diagnostic tests.

** Significant at the 1% level.

* Significant at the 5% level.

^ Significant at the 10% level.

Note: Normal t-values are reported in parentheses. Refer to Appendix B for more detailed diagnostic tests.

** Significant at the 1% level.

* Significant at the 5% level.

^ Significant at the 10% level.