4  Exploring the Cohort Patterns

Section 3 reported on the pattern of coefficients for the birth year cohorts. A range of these estimates for different quantiles and with different sets of controls was reported in Figure 2. They display a distinct V-shaped pattern. In short, saving rates appeared to differ significantly for different cohorts.

Figure 6 – Cohort effects with different definitions of consumption and saving

The cohort coefficients are estimated with reference to the 1910-14 birth cohort. Those coefficients for cohorts 3, 4 and 5 are all significantly lower than the reference cohort. In contrast, the coefficients for cohorts 9 through 14 are all significantly greater. In other words, those born from 1920 to the mid-1930s have demonstrably lower lifetime saving rates, while those born after 1950 have significantly higher rates of saving.

At first glance this result may seem surprising. Anecdotal evidence might have suggested that those born in the early inter-war period would have been conditioned by wars and the Great Depression, which could have lead to higher saving rates, at the least that part of saving driven by a precautionary motive. In contrast, the post WWII cohorts facing greater economic growth and security, together with liberalised financial markets after 1986, might have been expected to have displayed greater profligacy, and have had lower, not higher, rates of saving.

While these effects may be responsible for some influence on the estimated coefficients, clearly some other forces have operated to override them and produce a ‘V’ rather than an ‘inverted-V’ shaped pattern of saving by birth cohorts.

It is important to explore further the cohort pattern of saving. As cohorts with different patterns of saving move through their life cycle, they may influence the aggregate pattern of saving. The cohorts with significantly lower saving rates were aged between 45 and 60 in 1980, and entering their peak saving years. This is precisely the time that aggregate household saving was observed to start declining (Claus and Scobie, 2002).

Attanasio (1998) finds a similar ‘V-shaped’ pattern of cohort saving behaviour. He notes that the lower saving rates of the group aged in their 40s and 50s in the 1980s “are those mainly responsible for the decline in aggregate saving…because those cohorts were in the part of their lifecycle when saving are highest” (p.604). While he adds that even in for the USA where the data are much more consistent, it is not possible to precisely match the aggregate and micro level data. Nevertheless, the estimated cohort effects “explain a substantial part of the decline in the aggregate saving rate”.

Attanasio continues, noting that:

The main deficiency of the analysis is its failure to explain why those particular cohorts did not save ‘enough’. A plausible hypothesis, that I have not tested explicitly, is that the negative cohort effects for the middle cohorts are linked to increases in social security entitlements that the same cohorts have enjoyed. (p.604)

Attanasio (1998) posited that more generous public pensions might have explained the different lifetime saving patterns of different cohorts. If over an individual’s lifetime there is increasingly generous provision by the state for public pensions, it seems entirely plausible that this would shape expectations about the level of the publicly subsidised pension that one might receive. Those expectations could then in turn influence an individual’s decisions about the optimal allocation of consumption over their lifetime. Knowing (or at least predicting) that there would be a generous state pension to underpin consumption levels after retirement may lead to higher consumption prior to retirement from a given level of lifetime wealth. The consequence would be that the lifetime saving rate would be lower than in the absence of the public pension scheme, or with a less generous scheme.

If this were the case one would expect to see lower saving rates among those cohorts whose expectations were for the receipt of a more generous pension and higher rates among those who expected to receive lower real pension payments.

The impact of the social and economic environment over the lifetime of an individual (household) could be called the direct effect. There is also the indirect effect transmitted through family and those close to a person. These are the values and norms that are transmitted to them by their parents and others, and which in turn were formed by the conditions prevailing in an earlier time and shaping the values of their parents. In that sense, the behaviour patterns observed at any one time are a function of the prevailing climate, the expectations of the future climate, together with the effects of all previous environments. Those in the most immediate past could be expected to have the greatest effect, with the impact tailing off the further back we go[25]. We have not tried to explicitly allow for these indirect influences of past conditions, which might shape the behaviour of a particular cohort.

To proceed further with this hypothesis, it is necessary to posit some mechanism of how and when expectations are formed. Clearly this is a complex process, and one that would reflect the person’s perception of their economic and social environment. In addition, the experiences of their parents and that of their childhood could well condition their perceptions and the need for saving. The environment prevailing during their working lives will affect their labour market experience and earnings (the ability to save) while the provision of social services (health, education, housing, pensions) and welfare (sickness, disability, unemployment, family and single mother benefits) will influence the need to save.

Other factors including capital gains on housing, real interest rates, rates of income growth, access to credit and life expectancy could all be expected to impinge on the decision to allocate income to current or future consumption. In addition, the desire to make bequests may also influence the saving rate.

In short there are many possible indicators that might affect the decisions of individuals with regard to their saving rate. Some of these will operate throughout their working lives. Given that the peak saving years are typically between ages 45 and 60, an individual’s perceptions and expectations during this period, would arguably have a significant influence on their saving behaviour.

Essentially there are two steps in the argument: the first, that different cohorts operated in different environments; and the second that these different environments shaped the responses of different cohorts, particularly in the present case, with respect to their household saving behaviour[26].

Thomson (1991) documents changes that would support the first of these steps. He argues that over “the last 50 years welfare states have been very uneven in the benefits they provide for successive generations, that is for people born in different decades” (p.1). “The prizes and penalties of living in a welfare state are distributed more on the basis of birth date than of need, justice or desert. In New Zealand the big winners in this have been the ‘welfare generation’ – those born between about 1920 and 1945” (1991, p.1).

Testing these hypotheses is clearly a challenge on at least four counts. First we have little theoretical guidance as to how expectations are formed; in particular what relative weight should be assigned to each of the three critical periods: the experience of the previous generations particularly parents and grandparents which through norms and values could be expected to shape the saving behaviour of a particular cohort; the conditions prevailing during their working life and in particular applying during the peak saving years; and their expectations throughout their working lives about the level and eligibility for state support of retirement income.

Second, we need long time series, arguably covering the last 70 to 100 years to provide a quantitative assessment of the different policy environments enjoyed by different cohorts. Third, we only have 14 observations of the “dependent” variable. ie, the cohort dummies that relate lifetime saving behaviour to year of birth[27], meaning the scope for any statistical tests is limited; and fourth, as the working lives and saving periods of the later cohorts extend into the future, some forecasts of future conditions will be involved in order to compare their behaviour with that of individuals from the older cohorts who are either retired or dead.

In what follows we “test” the hypotheses in a very elementary way. Basically we look at a series of indicators for which we can obtain at least partial data. Often we focus only on selected years or periods that might be “typical or representative”, or occur at that time of peak saving[28]. In effect we are conducting in a loose manner a non-parametric sign test as an initial step. Are changes in the indicators broadly consistent with the hypothesis that the savings patterns of different birth cohorts could have been influenced by the economic and social policy environment prevailing at key points in the lifetimes? The objective is to make a preliminary foray to establish whether the patterns of some key variables that arguably affect the saving decisions of individuals are consistent with the hypothesis that the cohort differences reflect the external environment. In particular, we examine both labour market indicators (the ability to save) and some measures of public pensions and welfare (affecting the incentive to save).

Table 5 summarises the working life (assumed to be 40 years from age 20) and the peak earning years (assumed to be from ages 45 to 60) for each of the key birth year cohorts selected for this section. The first cohort (1910-14) is the reference cohort in the sense that the regression coefficients for cohort presented in Section 3 are referenced to this cohort, which has a value of zero. The lifetime saving rate of the adjoining cohort (1915-19) is not significantly different. The next three cohorts covering birth years from 1920 to 1934 are the group that typically have demonstrated significantly lower lifetime saving rates, while the last two (covering 1950 to 1959) are representative of those showing a significantly greater level of lifetime saving rate.