4  Data Construction Techniques and Measured Growth Rates

Section 3 examined different techniques to estimate an average growth rate over a particular period. Figure 4 highlighted that when these different techniques were applied to a common data series, the OLS approach could result in average growth estimates that looked quite different to those obtained from the other approaches. That is, the choice of average growth rate estimation technique can be quite important. Another factor that must be borne in mind is what dataset to use when calculating average growth rates over a period. For example, there exist several potential series for New Zealand’s real GDP per capita and these series differ in the length of their coverage and how real GDP per capita has been measured[13].

This section illustrates the differences in New Zealand’s growth rate in real GDP per capita when different data sets are used. Six data sets are used in this illustration. Their details are shown in Table 5.

The New Zealand series described in this table are presented in Appendix B.

For each of these series New Zealand’s average growth rate over every possible ten-year window has been calculated using the average annual growth rate (AAGR) approach outlined in Section 3. As the series are of different lengths the number of possible windows for each series also differs. Figure 5 shows the average growth rate for each ten-year period plotted against the window endpoint.[16] This means that a value for, say, 1990 represents the average growth rate over the period 1980 to 1990. Likewise, a value for 1970 represents the average growth rate over the period 1960 to 1970. Figure 5 illustrates the variation between the average growth rate for a ten-year period when different data sets are used.

One feature of Figure 5 is that there appears to be more variability in the growth rates obtained using different data sets earlier on in the sample. This is particularly the case if one omits the preliminary Penn World Table data that is yet to be finalised. However, it is worth pointing out that as the graph displays the average growth rate for a ten-year period even small differences would result in material differences in real GDP per capita over time. For example, even when we exclude the preliminary Penn World Table data, the average growth rate for the period ending 1992 (ie 1982 to 1992) differs between data sources by up to a bit under 0.5% per annum. If actual performance was to differ by this much then real GDP per capita would have increased by nearly 5% more over the ten year period when comparing performance for this period under the highest and lowest growth estimates.

Figure 5 – A comparison of New Zealand’s growth rates obtained from different data sets (NZ growth rates measured over different 10 year windows)

Note: The average annual growth rate method was used to obtain the average growth rates presented in this figure.

Figure 6 – A comparison of New Zealand’s growth rates obtained from different data sets (NZ growth rates measured over different 10 year windows)

Note: The average annual growth rate method was used to obtain the average growth rates presented in this figure.

Vertical axis displays annual average growth rate (%)

Horizontal axis displays year of window endpoint

Figure 6 illustrates the growth rates for each series shown in Figure 5 separately (using consistent scales). Casual observation of the Maddison and PWT plots in Figure 6 could suggest a downward trend over time in New Zealand’s average growth rate. However such a conclusion is likely to be misleading due to changes over time in the way in which real GDP for New Zealand has been measured.

4.1  Changes in the way New Zealand real GDP is measured

Over time there have been efforts to improve the way real GDP is measured. Unfortunately, however, these improvements mean that real GDP series constructed on a consistent basis and covering a long historical time period are not available. Consequently long time series of annual GDP data either tend to include data constructed in several different ways or require a considerable proportion of the series to be based on estimated rather than measured values.

Statistics New Zealand released upgraded national accounts at the end of 2000 and in mid 2001. These introduced a number of important changes, including moving from a fixed weight to a chain linked calculation of constant price (real) data, the adoption of the international accounting standard, System of National Accounts 1993 (SNA93), and the Australia New Zealand Standard Industrial Classification (ANZSIC). Real GDP figures back as far as the June quarter of 1987 are available on this consistent basis. These new SNA93 chain linked series are now New Zealand’s official data series and replace the previous official series that was based on a different accounting standard called System of National Accounts 1968 (SNA68). The previous official series was also a fixed weight rather than chain weighted series (more details of what this means are provided below). The previous official fixed weight series was available from September 1977.[17] What this means is that real GDP series that provide estimates of New Zealand’s real GDP for time intervals that include 1977 are likely to include data that is constructed in several different ways. When this is the case, estimates of economic performance for different sub periods are probably not strictly comparable and unfortunately there is no easy way around this.