3.2  Short-run estimation (continued)

Figure 1 – Actual, fitted and residual values from parsimonious ECM (A)

Figure 2 – Actual, fitted and residual values from parsimonious ECM (B)

The change in migrant transfers was found to have a significant positive contemporaneous impact on consumption growth, with no lagged effects. Migrant transfers can be seen as a windfall gain to the economy from funds brought in by immigrants that are available for consumption. An additional $1,000 per capita increase in net migrant transfers would result in a 0.16% increase in consumption based on ECM (A), and a 0.11% increase based on ECM (B).

A change in mortgage equity withdrawal was found to have a positive one-quarter lagged effect on consumption growth, indicating that funds raised in excess of residential investment find their way into consumption. However, the impact on consumption is not large. An $1,000 increase in per capita mortgage equity withdrawal would result in between 0.008% to 0.01% increase in consumption, although the coefficients were not found to be significant in either parsimonious ECMs.

Interest rates were not found to have any significant short-run influence on consumption, a similar finding to Rae (1997). Cross-country studies done by Boone et al (1998), Boone et al (2001) and Bertaut (2002) found mixed evidence for the significance of the interest rate variable in the short-run consumption function. A priori, changes in the interest rate may be expected to have an impact on consumption, with consumption rising when interest rates fall and vice versa. A number of different formulations of the real interest rate were estimated, to determine whether misspecification of the interest rate variable may have been leading to the insignificant finding. Alternative formulations included deflating nominal interest rates by the consumers price index, rather than the consumption deflator, and deflating with a forward-looking measure of inflation.[5]

The only interest rate variable that showed up as significant was when the nominal rate was deflated with a forward looking measure of inflation. The fourth, fifth and sixth lags of the interest rate were significant, suggesting that the effect of changes in interest rates took 12-18 months to flow through to per capita consumption. However, the coefficients on the lags of the real interest rate effectively cancelled each other out, such that the net impact was close to zero. For this reason the interest rate variable was excluded from the parsimonious ECMs.

The fitted and residual values from the parsimonious ECMs are presented in Figures 1 and 2. The models fit actual quarterly consumption growth well apart from a two and a half year period from 1998:1 to 2000:2 when the model consistently under-fitted consumption growth. That period coincided with the “Asian crisis”, two severe droughts, and a falling New Zealand dollar. Surprisingly, consumption growth remained resilient during that period, indicating that the models may be missing a variable to capture the dynamic of that period.

Out-of-sample forecasts were constructed for both parsimonious ECMs. For comparability, the both models were re-estimated up to 1999:4 and out-of-sample forecasts produced eight quarters out form 2000:1 to 2001:4. Figure 3 presents the forecasting performance of both models compared with actual out-turns over that period.

The out-of-sample performances of both parsimonious ECMs are broadly similar over the eight quarters. Neither managed to predict the large decline in consumption growth in 2000:4, but both over-estimated consumption growth over most of 2001, although the forecast from ECM (B) for the 2001:2 quarter was closer to actual than the forecast from ECM (A). Table 5 presents some forecast evaluation statistics for the out-of-sample forecasts. ECM (B) had a smaller root mean squared error, indicating better out-of-sample forecast performance. The statistics indicate that systematic bias is not a major issue in both models, and they have the ability to replicate the variability of actual consumption growth.

Figure 3 – Out-of-sample forecasts of parsimonious ECMs: quarterly percentage change in per capita consumption growth

Note: Forecast evaluation is for the period 2000:1 to 2001:4.