3  Empirical Results

3.1  Long-run estimation

Table 2 lists the key results from the long-run equation (1). The log of consumption was regressed against the log of income and the log of net wealth (Model A) using OLS, shown in column 2 of Table 2. The existence of a cointegrating relationship was not supported by the ADF test. However, because the test has low power in small sample sizes, it is assumed that a cointegrating relationship exists.

Since the estimated standard errors and the associated t-values of the estimated coefficients from a cointegrating regression are not valid, Stock and Watson’s (S&W) procedure of including leads and lags to estimate valid t-values for the estimated coefficients was used. In the interests of preserving the degree of freedoms, two leads and two lags were used. The results, presented in column 3 of Table 2, suggest that while the income variable is significant, the net wealth variable is not.

Note: The sample period for the OLS procedure is 1989:4 to 2002:1. Normal t-values are reported in parentheses. The sample period for the Stock and Watson procedure is 1990:3 to 2001:3. Adjusted t-values are reported in parentheses. The leads and lags variables are not reported.

* Significant at the 5% level. ** Significant at the 1% level.

Equation (1) was re-estimated, this time with the wealth variable disaggregated into net non-financial and net financial wealth (Model B). The results from the OLS and S&W estimation of Model B are presented in columns 4 and 5 of Table 2. Again, the adjusted t-values from the S&W estimation suggest that the income variable is significant, but neither of the two wealth variables are. Another regression was run, this time excluding the net non-financial wealth variable (Model C). Columns 6 and 7 of Table 2 show the results of the OLS and S&W estimation of Model C. The income variable remains consistently significant at the 1% level, but this time the net non-financial wealth variable is significant at the 5% level.[4]

The coefficients of the log-levels of income and non-financial wealth can be interpreted as long-run elasticities of consumption. Based on Model C, the estimated long-run elasticity of income is significantly larger than that for wealth at between 0.84 (OLS estimate) and 0.86 (S&W estimate). For net financial wealth, the elasticity is between 0.19 and 0.23. The income elasticity estimates are much higher than those found by McDermott (1990), Corfield (1992) and Rae (1997), but the wealth elasticity was broadly similar. For income elasticities, McDermott’s ranged from 0.463 for non-durables to 0.783 for services, and Corfield’s ranged from 0.153 for services to 0.683 for durables, but Rae found small and seemingly insignificant income elasticity estimates for aggregate consumption. For financial wealth elasticities, McDermott and Corfield found estimates of 0.214 and 0.205 respectively for non-durables (based on M3 variable), and Corfield also had an elasticity of 0.393 for services. Rae found a very strong net wealth elasticity at 0.66. Although the non-financial wealth variable was found to be insignificant, McDermott and Corfield found elasticities ranging from 0.139 to 0.317 (based on house price variable).

Differences in the coefficients between the estimated model and previous New Zealand research is not surprising, due to different estimated time periods and choice of variables. Ludvigson and Steindel (1999) found that the trend relationship linking consumption, wealth and labour income exhibits some instability. Poterba (2000) argues that the marginal propensity to consume out of wealth may vary over time due to shifts in consumer preferences over the wealth composition.

The finding that net non-financial wealth (essentially housing wealth) is not significant deserves comment, since it has become generally accepted that changes in house prices has an influence on consumption in New Zealand. Intuitively, the effect of housing wealth on consumption is not immediately obvious, since housing is illiquid and incurs significant transactions costs to liquidate. Even though households experience an increase in overall net wealth from rising house prices, it is also possible that households may not feel wealthier since their implicit rental costs have also increased. While it is possible that homeowners can borrow against the higher value of their housing for consumption (mortgage equity withdrawal), at the same time, an offset occurs from prospective home-buyers needing to save more for higher deposits by reducing their consumption. The empirical evidence on the role of housing wealth in determining consumption is mixed for major industrialised countries (Girouard and Blöndal, 2001).