3.3  Adding Conditioning Variables (continued)

Figure 2 – Cohort Effects with Different Sets of Controls

When the basic model of five-year birth-cohorts, is augmented with demographic, employment, family type and tenure status variables, we obtain a reasonably robust description of the underlying data. We therefore use the predictions from this model in to illustrate the shape of the age profile in median household saving rates and to show how that profile has shifted up and down across birth cohorts (Figure 3).

Figure 3 – Smoothed mean savings rate by Cohort

The typical age profile for the average saving rate is somewhat hump shaped with a peak around age 57 but does not become negative at older ages. While the hump shape is consistent with the lifecycle hypothesis, the apparent increase in the saving rate beyond age 70 is not.[24] The downward shift in the saving profile for earlier born cohorts, up until the fifth oldest one, is also apparent (earlier born cohorts are shown by the start and end points for their graph occurring at an older age, which is the age at the time of the first survey in 1984).

In Figure 2 the pattern of cohort effects is more apparent at the mean and 75^th percentile of the distribution of saving rates than it is at median and 25^th percentile. The quantile regression at the median is based on least absolute deviations of the residuals, rather than least squares, and so is less sensitive to the presence of outliers. To investigate whether the pattern of cohort effects is just due to some of the extreme values of saving rates that are present in the data various “trimmed samples” that removed extreme values of saving rates were used. In all cases, the models include the demographic, employment, family type and tenure status variables. The results again showed that the median saving rate falls from the earliest born cohorts until those born in 1930-34 and then rises across the cohorts born in later years. Hence, the pattern is the same as for the mean saving rates, although the rise in saving rates for the most recently born cohorts is not as marked.

The predictions from this quantile regression model give the smoothed median saving rates in Figure 4. The pattern is similar to that for the mean saving rate, except that median saving rates are everywhere lower so that there is negative saving at the start of the lifecycle and around age 65, and the downward shifts in the profile when moving from later to earlier birth cohorts are smaller.

Figure 4 – Smoothed median savings rate by Cohort: quantile regression

Figure 5 plots the cohort intercepts for each of the models estimated with reduced samples and for the model of mean saving rates estimated on the full sample. Although the magnitude of the cohort effects vary as the sample or estimation method is altered, the relative ranking of cohorts does not change. In all cases, saving rates fall from earlier to later born cohorts between the 1910-14 and 1930-34 cohorts and then the pattern reverses with later born cohorts exhibiting higher saving rates. These patterns seem sufficiently robust to warrant further investigation.

Figure 5 – Cohort effects with different estimation samples

3.4  Alternative Definitions of Saving

As noted in the introduction, consumption expenditures were adjusted to remove those items reasonably viewed as “investment” so as to give a truer picture of the underlying saving rate. In this section, we assess whether the cohort effects that we have found previously are sensitive to the re- inclusion of some of these items in the household consumption variable.

The cohort effects estimated under these different definitions of consumption and saving are plotted in Figure 6. These graphs illustrate the robustness of the relative cohort effects and especially the location of lower saving rates amongst those born ca. 1925-1939. The pattern of cohort effects in Figure 6 is consistent with our other sensitivity checks, including trimming the sample (Figure 5) and controlling for within-cell heterogeneity (Figure 2). Hence, we are confident that this lower saving rate for those born ca. 1925-1939 is a genuine feature of the saving behaviour of New Zealand households rather than just some artefact of the data or of our econometric procedures. The remaining task is to explain this cohort pattern of saving rates.