2.2  Data

The consumption variable c is the official New Zealand real private consumption series. The choice of income variable y is restricted by the lack of an official quarterly disposable income series. Interpolating the annual disposable income series into a quarterly series is not satisfactory as it is not possible to capture significant quarterly variations, limiting the ability of the interpolated series to explain variations in consumption. Hence, a derived after-tax labour income series, which includes wages and salaries and government transfer payments, was used. The literature does not favour the use of labour income over disposable income, but Rae (1997) argues that using disposable income can lead to the double counting of property income. Rae also argues that at a more practical level, disposable income includes the highly volatile entrepreneurial income component, which should be excluded. In the New Zealand case, farm income makes up a large proportion of entrepreneurial income, and perhaps has a more significant share of disposable income than most other developed countries, ranging from between 2% to 7%. However, the official farm income series is only available on an annual basis, and interpolation to a quarterly basis renders the series meaningless. The use of after-tax labour income in this paper for y remains a short-coming until such time when an official disposable income series becomes available.

There are two data sets for measuring household net wealth in New Zealand. The first is the Reserve Bank of New Zealand (RBNZ) data on household financial assets and liabilities, first reported by Thorp and Ung (2000). The second is the WestpacTrust Household Savings Indicators (HSI) measurement of household assets and liabilities. Both data sets provide comprehensive breakdown of household wealth, but the HSI is the only series available on a quarterly basis. The RBNZ data is annual. For this reason, the data from the HSI was used for the aggregate net wealth variable w, as well as the disaggregated net non-financial wealth nfw and net financial wealth nf variables.

The official Household Labour Force Survey (HLFS) measure of the unemployment rate, was used for unr, and the 90-day bank bill rate was used as the interest rate variable ir. The net migrant transfers series, migtr come from the balance of payments capital accounts, while the mortgage equity withdrawal series, mew is measured as the difference between the change in household claims against housing and the nominal value of residential investment.

Where relevant, all variables are quarterly, seasonally adjusted, and measured in real per capita terms. There is no clear preference in the literature on how the dependent variable of the consumption function is specified (aggregate, per capita, or as a percentage of income). For the New Zealand case, it is preferable to model consumption in per capita terms to account for the small population base, which tends to be influenced by large fluctuations in net migration.

The implicit price deflator for consumption is used to adjust all nominal variables into real variables. The full sample period is restricted to 1989:4 to 2002:1 because the quarterly household wealth data from the HSI is available only from 1989:4. While it is possible to interpolate the data using the RBNZ annual series, as Tan and Voss (2000) did for Australian data, the backdated data would be spurious as it will be difficult to take into account short-term asset movements in the run-up to the 1987 sharemarket crash, or short-term liability movements after financial liberalisation in the mid-1980s. The short sample period is an obvious limitation of this paper. But one advantage is the absence of structural breaks due to financial liberalisation, which is present in many other studies using longer sample periods. For further information on the data used in this paper, refer to Appendix A.

A necessary condition for variables to cointegrate in a long-run relationship like equation (1) is that they are integrated processes of order 1 or I(1), that is they are non-stationary. All variables were tested for stationarity using the standard Augmented Dickey Fuller (ADF) procedure. The appropriate lag length for the ADF test was chosen by minimising the Schwartz criteria. For the level and log-level variables, the chosen specification included an intercept in the equation, but no trend except for c and y. For the first difference variables, all the chosen specification included an intercept only.

The results are presented in Table 1. All the variables are I(1) except for the unemployment rate and mortgage equity withdrawal variables. There is evidence to suggest that the mortgage equity withdrawal variables could be stationary or I(0), and that the unemployment rate may be I(2). However, when the unemployment rate was tested over a longer time period (from 1985:4), it was found to be I(1) at the 10% level. Based on this, it was assumed that the unemployment rate is I(1).

** Stationarity at the 1% level (MacKinnon critical values).

* Stationarity at the 5% level (MacKinnon critical values).

Note: All variables were tested over the 1989:4 to 2002:1 period.