Expenditure and disposable income trends of UK households: evidence from micro-data.

van de Ven, Justin

The view that there is an 'under-savings' problem in the UK domestic sector has had an important bearing on the contemporary agenda for policy reform. This paper considers the temporal variation of household expenditure and disposable income described by micro-data observed during the past four decades (1971 to 2009). I find a very close correspondence between the growth of expenditure and disposable income over the sample period, with the implication that households have not offset the well-documented decline in occupational pension provisions during the period through increased private saving out of disposable income.

Keywords: Under-saving; retirement; expenditure; disposable income

JEL Classifications: D12, D14, D31, D91

I. Introduction

The general perception that there is an 'under-savings' problem in the UK has motivated much discussion in the contemporary literature and provided an important impetus to the agenda for policy reform. Any judgement about the sufficiency of savings behaviour is, however, complicated by the diversity of incentives that bear upon intertemporal planning in practice. In this study I take a fresh look at the statistical evidence, by considering how age profiles of household expenditure and disposable income have varied between birth cohorts over a 39-year period. The results obtained suggest that households have not increased their savings out of disposable income to offset the decline in occupational pension provisions that has occurred during the past thirty years.

The perception that the national savings rate is too low is worrisome from a macroeconomic perspective because it suggests that investment may be insufficient to sustain prospective growth, or that--if adequate rates of investment are maintained--then domestic borrowing may pose a risk to creditworthiness. From a microeconomic perspective, under-saving is likely to result in insufficient finance to meet aspirations for prospective consumption, exposing the public sector to political economy risk. Disgruntled retirees, for example, may pressure the government to increase state sponsored benefits, putting the public finances in doubt.

It is, however, exceedingly difficult to determine whether there is an under-savings problem, both because of the complexity that is usually attached to identifying 'optimal' saving rates for any specific objective, and because of the diversity of competing objectives that characterise our economic environment. Increased saving may be required to sustain a particular level of growth on the one hand, while lower saving may be required to support a desired consumption stream on the other. And even if attention is limited to the household savings problem, then basic economic theory suggests that savings will respond to a range of issues that include expectations over future employment and investment prospects, the evolving policy environment, credit market conditions, individual demographic and financial circumstances, and the nature of preferences. We have little data to draw upon in relation to many of these factors (particularly where expectations are concerned), and some we can identify only indirectly (e.g. preferences).

Analyses that are concerned with the adequacy of savings have consequently tended to focus upon particular issues of concern, each on the basis of a discrete set of assumptions. One subject of concern in the literature has been the sustainability of national aggregates. Weale and coauthors, for example, focus upon the savings rate required to ensure that consumption grows at the same rate as income in steady-state equilibrium. (1) Comparisons between national saving and the implications drawn from a highly stylised macro-model that are reported by these authors suggest that the UK is under-saving by between 1 and 10 per cent of GDP. In a similar vein, a very large literature has explored the sustainability of public finances, following the seminal contribution of Auerbach and Kotlikoff (1987). (2)

Although these types of macro-level analyses provide useful insights into the long-run sustainability of national patterns of income and expenditure, they are insensitive to the motives that underlie savings. They do not, for example, distinguish between dis-saving that is the product of an undesirable behavioural distortion like myopia, from dis-saving that is an appropriate short-run response to altered expectations regarding the returns to capital. Most of the related literature has consequently taken a microeconomic approach.

Microeconomic studies of the adequacy of savings range from sophisticated applications of fundamental economic theory (e.g. Scholz et al., 2006), to relatively straight-forward accounting exercises based on observed age profiles of consumption and income (e.g. Khoman and Weale, 2006, 2008). Two studies that have had a particularly pronounced impact on the policy debate in the UK are those reported by the Pensions Commission (2004, 2005), and the subsequent work undertaken by the Department for Work and Pensions (2006). Both of these studies focus primarily upon the adequacy of existing pensions provisions. They find that, whereas increases in longevity suggest that larger pensions provisions would be prudent, public and private pensions taken together have been declining since the early 1980s. These findings provided the fundamental motivation for the extensive pension reforms that were enacted in 2007 and 2008, and continue to bear upon the associated policy debate.

In this paper I consider the relationship between temporal trends observed for household expenditure and disposable income, focussing upon identifying whether expenditure growth between cohorts has out-paced income growth during the past 40 years. Comparing age profiles of income and expenditure between birth cohorts reveals weak evidence that expenditure growth has increased relative to income growth, and that saving rates have fallen among households towards the bottom of the income distribution.

The study is organised as follows. Section 2 briefly describes the data upon which the analysis is based. Results are reported in Section 3, and implications of the results are discussed in the conclusion.

2. Data

The analysis is based upon micro-data drawn from the Family Expenditure Survey (FES, and its successors), which are reported at annual intervals for all years between 1971 and 2009. The FES provides detailed information regarding expenditure, income, and demographics for a random sample of UK households. The household level data that are reported by the FES were reconfigured to focus upon the family unit, defined here as a single adult or adult couple and their dependent children. Further details regarding specifics of the micro-data considered for analysis, including survey methodology and coding definitions, are provided in Appendix A.

The representative nature of the income and expenditure data reported by the FES has been the subject of much research. Significant studies in this vein include Atkinson and Micklewright (1983), Kemsley et al. (1980), and the edited volume by Banks and Johnson (1998). These studies have typically found that the FES under-reported measures of self-employment and investment income described by the National Accounts, and of expenditure on selected categories including tobacco. On the whole, however, there appeared to be a reasonably close correspondence between the FES and the National Accounts up until the early 1990s, for measures of both household income and expenditure.

However, the past fifteen years have seen disparities between aggregates reported in the National Accounts and those implied by the FES widen substantially; whereas the FES accounts for 91 per cent of aggregate household expenditure and disposable income reported in the National Accounts for 1971, it accounts for 71 per cent of expenditure and 77 per cent of income in 2009 (see Appendix B for further details). There is no consensus regarding the reasons underlying this decline.

In the absence of further detail regarding this issue, the current study is primarily based upon FES data that were proportionally adjusted to ensure that the aggregates implied by the survey population cross-sections match those reported by the National Accounts. Statistics that incorporate these adjustments tend to lower rates of saving, relative to the unadjusted data.

3. Trends described by micro-data

This section explores how age profiles of expenditure, disposable income, and savings have varied over time and across the income distribution. Section 3.1 discusses trends in expenditure by age and birth cohort, and Section 3.2 discusses associated statistics for disposable income and saving. Statistics that disaggregate by income quintile are reported in Section 3.3.

3.1 Expenditure

A broad summary of how family expenditure has varied by age and year of birth over the last four decades is displayed in figure 1. Panel A of this figure displays the well-known hump-shaped profile of expenditure over the life-course that has motivated much empirical research. Expenditure tends to rise early in life as liquidity constraints are relaxed and as consumption needs rise with relationship formation and child birth, peaking between ages 40 and 60, before falling away into retirement. Furthermore, expenditure tends to be higher at any given age among younger birth cohorts, who benefit from real wage growth relative to older generations. The growth rates implicit in the relative consumption profiles reported in panel A of figure 1 are returned to below.

Comparing the expenditure profiles reveals that the FES data describe the least variation between birth cohorts--in both absolute and relative terms--early in the life-course, which corresponds to the youngest birth cohorts in the sample. This observation suggests that the credit market liberalisation that took place during the 1990s in the UK has not resulted in considerable profligacy, at least amongst the young.

The disparity in family expenditure between birth cohorts increases between ages 20 and 30, is then approximately stable between ages 30 and 50, after which relative disparity tends to rise again while absolute disparity between birth cohorts tends to fall. Although the rise in the relative disparity between birth cohorts reported late in the life-course--and for older birth cohorts--can be attributed to a range of interrelated issues, three factors stand out.

[FIGURE 1 OMITTED]

First, is the maturing of the Beveridgean welfare state. This was introduced with the National Insurance Act 1946, so that the first generation to spend its entire working lifetime under the scheme was born in 1930. Second, is the maturing of Defined Benefit occupational pensions, which only became prevalent during the 1950s and 1960s, and offered generous (ex post) terms to their members. And third, is the higher fertility observed among older generations, which exaggerated their expenditure needs during peak child-rearing years. (3) These three factors, in conjunction with the general rise in affluence and longevity, played an important role in improving the retirement provisions held by recent retirees, relative to older generations.

A clearer picture of how family expenditure by age has changed over time is reported in figure 2. Panel A of the figure displays family expenditure at selected ages by year of birth of the eldest family member. Linear trend-lines and seven point moving averages for each series are also included in the figure to help identify whether there has been any appreciable variation in the rate of change of expenditure by birth year described by the FES data. This figure confirms the trend growth of family expenditure by year of birth, and the observation, made in regard to figure 1, that growth over the 39-year sample period is relatively anaemic at age 20, but rises strongly for people aged in their 30s and 40s. Each year by which birth is delayed tends to increase family expenditure at age 20 by 1.80 [pounds sterling] per week on average (2006 prices), rising to just over 11 [pounds sterling] per week by age 40.

The linear trend-lines, and moving averages that are reported for each series make clear that the growth of family expenditure by year of birth displays some convexity, rising at a faster rate later in the sample period. Panel B, which reports differences between the linear and moving average trend lines, is provided to clarify this point, where the horizontal axis has been respecified in terms of the sample period, rather than year of birth. The data series reported in Panel B of figure 2 reveal that family expenditure tended to rise more strongly from the mid-1990s for families between ages 25 and 40 than it had done previously. No equivalent effect is observed for households at age 20 (noted above), and the timing of the accelerated rise is somewhat earlier for older households, as displayed by the series reported for 45 year-olds.

[FIGURE 2 OMITTED]

Two important transitions took place in the UK during the early 1990s. On the one hand, the UK experienced an economic recession between 1990 and 1992, and on the other the 1990s corresponded with an important period of credit market liberalisation. The former of these effects is likely to be responsible for depressed expenditure growth early in the decade, and accelerated growth towards the end of the decade as the economy recovered from its slump. The latter effect is likely to have relaxed credit market restrictions, at least among some members of the population.

An alternative way of interpreting the variation displayed by the FES expenditure data is to calculate the growth rates that convert the various cohort specific age profiles to a comparable basis. Related statistics are displayed in figure 3, which reports the geometric growth rates that minimise the least squares variation between age specific expenditure reported for any given cohort, and the expenditure reported for cohorts born more than nine years earlier.

Figure 3 indicates a general downward trend by year of birth in the growth rates associated with cohort specific expenditure, reflecting a number of the observations made in the discussion set out above. An additional issue to note, however, is that the downward trend with age of birth reported for the growth rates of cohort expenditure is in part attributable to the limited time period of the sample. The growth rates reported for the cohort born in 1931, for example, were calculated by comparing cohort specific expenditure observed between ages 50 and 78 (inclusive). This contrasts with growth rates reported for the cohort born in 1976, which were calculated on expenditure data reported between ages 20 and 33. As noted above, expenditure variation between birth cohorts, and trend growth in particular, appears to be depressed early in life, and this consequently has an important bearing upon the temporal variation of the growth rates that are reported in figure 3.

Nevertheless, the temporal profiles of cohort expenditure growth reported in figure 3 cannot be entirely attributed to variation in the underlying age bands used for the associated calculations. The downward trend that is evident for older cohorts, for example, extends to cohorts born between 1945 and 1961, and the calculations for these cohorts refer entirely to consumption below state pension age and for whom maturation of the UK pensions system was complete. Furthermore, there is evidence of a slight increase in expenditure growth from the cohort born in 1960, which contrasts with the increasing weight placed on younger ages that underlies the associated calculations.

[FIGURE 3 OMITTED]

Discussion now turns to profiles of disposable income, and saving.

3.2 Disposable income and saving

The FES data describe very similar profiles for disposable income as they do for expenditure, which is made clear by comparing the profiles reported in figure 4 with those reported in figures 1 and 2. Like the expenditure profiles, the profiles of disposable income display a hump-shape, rising to peak about age 50, before falling away into retirement. The dispersion of income is relatively low at the beginning of the working lifetime, and rises in a profile similar to that reported for expenditure. Furthermore, the age profiles of income by birth cohort exhibit similar convexity to those reported for expenditure, suggesting that the increased rate of ascent observed for the expenditure profiles late in the sample period may be more attributable to the economic recovery following the recession of the early 1990s than to the coincident credit liberalisation.

[FIGURE 4 OMITTED]

Given the similarity between the profiles reported for expenditure and income, it is perhaps unsurprising that the temporal variations of savings rates exhibit more noise than they do any persistent temporal trend. After filtering out this noise, savings rates are observed to vary inversely with the economic cycle, falling during the booms of the 1980s and 2000s and rising with the recession in the early 1990s and the 2008 financial crisis (for which only two data points are available). Related statistics are displayed in figure 5, which reports differences between age specific savings rates and their respective averages taken over the sample period from 1971 to 2009.

The statistics reported in figure 5 suggest that household savings tend to respond similarly to changes in the economic cycle across the ages reported in the figure, with some evidence of heightened sensitivity among 60 year olds. The figure suggests a consistent and broad-based downward trend in savings rates throughout the economic boom of the early and mid-2000s. This observation gives some credence to the proposition that the duration and strength of the economic boom of the 2000s gave households the confidence to reduce savings in a way that was not observed at any other time during the 39-year sample period covered by the FES.

[FIGURE 5 OMITTED]

An alternative way of presenting this evidence is by contrasting the growth rates implicit in the cohort profiles of disposable income with those calculated for expenditure (discussed at the end of Section 3.1). Related statistics are displayed in figure 6. The statistics that are calculated on series adjusted to reflect the National Accounts aggregates suggest that expenditure growth was approximately matched by growth in disposable income for cohorts born to 1950. There is, however, some evidence to suggest that expenditure growth has exceeded disposable income growth for cohorts born after 1950. The greatest disparity between the two series is reported for the cohort born in 1968, for which consumption growth is reported to exceed income growth by 0.4 percentage points per annum (1.9 per cent c.f. 1.5 per cent). Interpretation of these results is discussed in Section 4.

[FIGURE 6 OMITTED]

3.3 Distributional variation

The population average statistics reported above indicate that family expenditure has tracked disposable income fairly closely over the 39-year sample period covered by the FES. They also suggest that savings have, on average, reacted more to the economic cycle than they have displayed any temporal trend over the period. This section explores these issues at a greater depth, on the basis of statistics disaggregated by family income quintile.

Figure 7 indicates that ratios of average family expenditure between the highest and lowest income quintiles exhibited a fairly consistent pattern across the age distribution during the 40-year sample period. The inequality of expenditure tended to recede during the 1970s, a trend that was reversed during the 1980s and early 1990s. From the early 1990s, there appears to have been some reversal of the increased age specific inequality of expenditure amongst retirees, which is not evident for families of working age. The five-year moving averages reported in the figure indicate that, in 1980, families in the highest income quintile spent between 3.0 and 4.1 times that of families in the lowest income quintile. This multiple rose to between 5.1 and 6.0 in 2007 for families aged between 25 and 55 reported in the figure, was slightly lower for families aged 75 (4.1), and considerably lower for families aged 65 (3.1).

[FIGURE 7 OMITTED]

The decline in inequality of expenditure amongst retirees observed from the mid-1990s coincides with the maturation of the welfare state in the UK, and the pensions system more generally, as discussed above. With regard to the other trends described in figure 7, the analysis presented in Section 3.2 highlights the underlying importance of coincident trends in disposable income, and it is to these trends that we now turn.

Figure 8 restates figure 7 with disposable income in place of family expenditure. Comparing these two figures reveals that the close relationship between income and expenditure that is discussed in Section 3.2 carries over to the distributional statistics that are reported here. The highest to lowest quintile ratios for both expenditure and income exhibit a slight downward trend during the 1970s, and rise relatively rapidly during the 1980s, before levelling off from the early 1990s. Furthermore, there is some evidence that the highest to lowest quintile ratios for households aged 65 and 75 tend to decline from the early 1990s, particularly relative to the wider population. These observations reinforce the impression that expenditure patterns and income patterns are largely coincident, and to the extent that income variation is subject to substantial exogenous drivers, (endogenous) family expenditure adjusts to accommodate.

The rise in inequality from the 1980s in the UK is well documented (see, e.g., Atkinson, 2007, and National Equality Panel, 2010), and the reasons for this rise are widely recognised as complex. Important factors identified in the literature as underlying drivers of the contemporary trend in income inequality include the growth of earnings inequality, the reduction in tax progressivity, and the growing disparities within demographic groups over a broad range of circumstances.

[FIGURE 8 OMITTED]

The similarities between the expenditure and income statistics that are discussed above provide little indication of whether there have been any strong trends in coincident savings behaviour when the population is disaggregated by income quintile. A detailed consideration of the savings rates of individual income quintiles revealed no temporal trends of particular note during the past 40 years for the vast majority of the population. Two broad population subgroups do, however, depart from this wider generalisation: families in the bottom disposable income quintile, and those at young ages. Associated statistics are reported in figure 9.

Panel A of figure 9 reports temporal variation of the average savings rates of families in the lowest disposable income quintile, by age and survey year. The statistics in this panel reveal that, with the exception of the series reported for 75 year olds (the oldest age reported in the figure), there is a distinct downward trend in savings rates among households in the bottom income quintile across all age groups. This downward trend is coincident with the credit liberalisation that occurred during the period, which may have helped low income people smooth their expenditure through rough patches.

[FIGURE 9 OMITTED]

Panel B of figure 9 indicates that the downward trend in savings rates with survey year is also evident for families in the highest income quintile at age 25 and slightly less so for families in the fourth income quintile. Given that, all else being equal, credit constraints tend to bite most at the extremes of the life course, the statistics reported here give added credence to the proposition that the additional access to credit offered by financial market liberalisation has affected the expenditure patterns of selected household types, depressing savings for some as a result.

4. Conclusion

The 39-year period covered by the household micro-data explored in this study provides an interesting snapshot of how domestic circumstances in the UK have changed. One of the key observations reported in this study is the close relationship that exists between the temporal evolution of family expenditure and disposable income during the past four decades, a correlation that is consistent with international evidence, and has generated much research (see, e.g., Attanasio and Webber, 2010). Another is that no trends in domestic saving out of disposable income are identified that are of sufficient magnitude to compensate for the decline in occupational pension provisions that has occurred in the UK since the 1980s (e.g. Pensions Commission, 2004). This last observation corroborates an important empirical motivation cited in relation to contemporary pensions policy reforms (DWP, 2006).

Family expenditure and disposable income profiles by year of birth both display real growth over the generations covered by the data, where the rates of growth tend to be relatively low early in the working lifetime. Real growth in both expenditure and disposable income appeared to accelerate from the mid-1990s for families of all ages. For the young, the timing of this acceleration is consistent with the end of the recession that occurred between 1990 and 1992, and with the coincident liberalisation of financial markets. For the old, it reflects the maturation of the Beveridgean welfare state and occupational pension arrangements more generally.

Matching (incomplete) age profiles of family expenditure and disposable income of any given birth cohort against profiles observed for older generations, suggests that rates of growth by year of birth have exhibited a downward trend during the past four decades. From highs in the region of 2.5 per cent per annum for those born early in the 1930s, growth rates fell to 1.8 per cent for expenditure and 1.5 per cent for disposable income for cohorts born in the mid-1960s. These growth rates appear to have stabilised for subsequent generations, with some evidence that they have increased slightly for those born during the 1970s.

The downward trend in cohort specific growth referred to above in part reflects the fact that the generation born in 1930 entered its working life (at age 16) following the conclusion of World War II, in contrast to older generations whose working lives will have been disrupted by at least one world war. It also reflects the incomplete nature of the data upon which the analysis is based, so that growth rates estimated for older generations are disproportionately affected by the maturation of the pensions system in the UK, while those of younger generations are influenced by the low growth that is identified for young ages.

Furthermore, the reliability of the statistics that are cited above is called into question by trend deterioration of the correspondence between the micro-data upon which the analysis is based, and aggregates reported in the National Accounts. Although the analysis reported here has been specified to reconcile these two data sources (via proportional adjustments to the micro-data), the underlying causes for the increasing disparity between the two remain unclear.

Bearing the above in mind, it is of note that--based upon the preferred data series--the estimated rates of growth implicit in expenditure patterns appear to be higher than rates of growth of coincident disposable income patterns for cohorts born from the mid-1950s. This observation has interesting implications for savings behaviour. If the income expectations of individuals from cohorts born after the mid-1950s are based upon the experience of preceding generations, then the observation that their expenditure profiles exhibit higher growth than their income profiles suggests four possibilities: 1) that younger generations are choosing to bring some of their lifetime expenditure forward, relative to older generations; 2) that younger generations expect some additional source of financing to make up for the difference between their expenditure and income growth; 3) that younger generations have unrealistic expectations; 4) that the growth statistics reported here fail to provide an accurate description of the practical reality measured over the longer term. There is some credence to the conjecture that all four of these possibilities have acted in parallel.

The credit liberalisation of the 1990s may have enabled younger generations to bring their expenditure forward during the life-course. This conjecture is supported by the observation that savings rates have shown a sustained downward trend among households in the lowest income quintile and the young, particularly since the 1990s.

Sustained growth in context of low inflation and low unemployment during the decade prior to the financial crisis of 2008 may have provided the basis for expecting that the favourable economic environment would continue into the indefinite future. Not only would this raise hopes regarding the lifetime resources available to younger generations, but it would also dampen precautionary motivation to save.

The upsurge in savings observed since the 2008 crisis suggests that households have heavily revised their expectations regarding the future.

The upsurge in saving also suggests that the growth rates for expenditure and disposable income that are reported here may alter substantially if they are recalculated on data for years beyond 2009. Furthermore, the finding that expenditure growth has declined relative to growth of disposable income that is reported here has not been subject to rigorous statistical testing, and is shown to be sensitive to the adjustments made to reflect National Accounts aggregates.

In any event, it can be concluded that temporal variation of expenditure has tracked reasonably closely that of disposable income over the 39-year period covered by the FES data. If some of the income growth that has been observed during the past two decades is interpreted as compensation for foregone occupational pension rights--consistent with the well-documented decline in occupational pension provisions that has occurred since the early 1980s (4)--then the observation that savings rates out of disposable income have not increased over the same period does suggest that younger generations are saving less than older generations. Whether this indicates under-saving, or an efficient reallocation of consumption over the life-course, remains an open question.

APPENDIX A. HOUSEHOLD HICRO-DATA

The Family Expenditure Survey (FES) was introduced in 1957, and reports detailed information regarding demographics, income, and expenditure for a cross-sectional sample of households in the United Kingdom. Although the FES was superseded by the Expenditure and Food Survey (EFS) in 2001, and by the Living Costs and Food Survey (LCFS) in 2008, the basic structure of the survey regarding the characteristics of concern in this study has remained largely intact from 1971. Reference throughout this paper is consequently made to the FES whenever discussing the time-series data provided by these three data sources (the FES, EFS, and LCFS). The survey is collected on a continuous basis and reported at annual intervals. (5)

The unit of analysis in the survey is the household, with households selected at random from the Post Office's list of addresses (for Great Britain, excluding the Scottish Isles and the Isles of Scilly; Northern Ireland is sampled through the Valuations and Lands Agency list), and participation being voluntary. One of the key advantages of the FES is that it reports data for both household income and consumption. All individuals aged 16 and over in participating households are asked to keep a diary of expenditure covering a two-week period, with children aged 7 to 15 also being asked to keep a simplified diary since 1998. Regular expenditure, demographic, and income data are recorded at a household interview, and retrospective information is collected on expenditure of selected large and infrequent purchases.

The representative nature of the FES for the UK population is imperfect. People in institutions--including retirement homes, the military, and prison--are omitted from the survey, as are people with no fixed address (the homeless). Furthermore, the voluntary nature of the survey typically obtains a response rate of between 50 and 60 per cent, and participation has been found in past surveys to be distributed non-uniformly across the population. Foster (1996), for example, compares the characteristics of households responding to the 1991 FES with information derived from the 1991 Census, and finds that response was lower than average in Greater London, higher in rural areas, and that the response rate tended to increase with the age of the household reference person. Low response rates were also found for ethnic minorities, the lower educated, self-employed, and the manual social class. (6)

A.1 Units for analysis

A household is defined by the FES as "a group of people living at the same address with common housekeeping that is sharing household expenses such as food and bills, or sharing a living room." (7) This definition of a household can consequently include a group of disparate adults sharing the same residence who may otherwise keep their finances quite separate from one another. Given the focus of the current analysis, the data reported by the FES were disaggregated to focus upon the family unit--comprised of a single adult or adult couple and their dependent children. This was achieved by analysing the demographics of each household reported by the FES to identify discrete family units, and reallocating measures of household income to each unit where necessary. (8) In undertaking this analysis, it was necessary to exclude any households comprised of:

* family units not including the reference person/head of household, but including dependent children;

* partner couples that did not include the reference person/head of household;

* more than five family units.

Furthermore, households that are recorded as having self-employed members are also omitted from the analysis, given the perceived inaccuracy associated with reported self-employed income.

A.2 Coding

Benefit unit identification

A unique benefit unit identifier, bu, was constructed for each family unit identified within each household reported by the FES. bu was set to 1 for the family unit that includes the reference adult (including their spouse and dependent children to age 17), and set to progressively higher numbers for each family unit identified thereafter. The age of each benefit unit was set equal to that of the eldest member of the unit.

Consumption

The FES reports three measures of aggregate household consumption: the ONS definition, p550tp, the National Accounts definition (I), p560tp, and the National Accounts definition (II), p600t. These statistics differ mainly in that the ONS definition includes expenditure on various taxes, charges, and fines (e.g. Council Tax, Stamp Duty, motoring fines), and unrequited domestic transfers that are not included in the National Accounts definitions. The analysis is based primarily upon the ONS definition, except where direct comparisons are made with measures of aggregate consumption reported in the National Accounts.

The consumption codes p550tp and p560tp are reported only from 1991, and p600t from 2001. Otherwise, total household expenditure was subject to two coding changes, reported as p378 to 1986, and p508 from 1987 to 1991. The definitions of total expenditure prior to 1991 differ from p550 primarily in that they do not include data derived through retrospective recall.

Disposable income

The FES has reported total household disposable income under the code p389 since 1978. Prior to 1978, it was reported under code p399. Family disposable income is allocated within households in proportion to each unit's share of total household consumption.

APPENDIX B. COMPARING MICRO-DATA AGAINST THE NATIONAL ACCOUNTS

The scale of the growing disparities between the FES and National Accounts aggregates is displayed in figure B1. This figure indicates that there has been a consistent downward trend in the proportions of the National Accounts aggregates that are accounted for by the FES during the 39-year sample period. From highs of just over 90 per cent for both household expenditure and disposable income, the ratios of FES to National Accounts aggregates have fallen to 71 and 76 per cent respectively, with some evidence that the rate of descent has accelerated during the past decade.

It is notable that the steepest decline in the ratio reported for expenditure (ONS definition) occurs from 1997/98, which corresponds to the year in which population weights were first supplied with the FES. There is also some evidence that the population sizes implicit in the weights that are supplied with the FES have fallen, relative to those reported by the National Accounts. The downward trends described in figure 10 cannot, however, be attributed to the FES sample weights, as is made clear by ratios of per capita averages that are reported in the figure, and is easily verified by reproducing the analysis ignoring the supplied weights altogether. (9)

Furthermore, disaggregating expenditure into its constituent sub-categories does not help to clarify the reasons underlying the growing disparity observed between the micro-data and national aggregates, with falls identified over almost all individual categories of goods and services (described previously for durable consumption by Attanasio et al., 2006). And the matter is further complicated by the observation that the measures of aggregate consumption reported by the FES that are directly referred to in calculating the National Accounts (denoted 'NA definition' in figure B1) show a more pronounced downward trend than those associated with the traditional 'ONS definition' of expenditure.

[FIGURE B1 OMITTED]

As the underlying causes for the trends described in figure B1 are not yet understood, it remains unclear what--if anything--should be done about them. In the current context, our concern with savings focusses attention upon variation of income relative to expenditure. In this regard, the coincident downward trend observed for both income and expenditure suggests that, whatever the underlying causes for the growing disparities between the FES and the National Accounts, the associated implications for savings rates are likely to be less pronounced than for either expenditure or income taken in isolation.

REFERENCES

Atkinson, A.B. (2007), "The distribution of top incomes in the United Kingdom 1908-2000', in Atkinson, A.B. and Piketty, T. (eds), Top Incomes of the Twentieth Century, Oxford, Oxford University Press.

Atkinson, A.B. and Micklewright, J. (1983), 'On the reliability of income data in the Family Expenditure Survey 1970-1977', Journal of the Royal Statistical Society, Series A, 146, pp. 33-61.

Attanasio, O.P., Battistin, E. and Leicester, A. (2006), 'From micro to macro, from poor to rich: consumption and income in the UK and the US', Paper presented at the "Consumption, Income, and the Well-Being of Families and Children" Conference, Washington, D.C., May.

Attanasio, O.P. and Weber, G. (2010), 'Consumption and saving: models of intertemporal allocation and their implications for public policy', Journal of Economic literature, 48, pp. 693-751.

Auerbach, A.J., Gokhale, J. and Kotlikoff, L.U. (1994), 'Generational accounting: a meaningful way to evaluate fiscal policy', Journal of Economic Perspectives, 8, pp. 73-94.

Auerbach, A.J. and Kotlikoff, L.U. (1987), Dynamic Fiscal Policy, Cambridge, Cambridge University Press.

Banks, J. and Johnson, P. (1998), How Reliable is the Family Expenditure Survey, London, Institute for Fiscal Studies.

Department for Work and Pensions (DWP) (2006), Security in Retirement: Towards a New Pensions System, Norwich, The Stationery Office.

Foster, K. (1996), 'A comparison of the Census characteristics of respondents and non-respondents to the 1996 FES', ONS Survey Methodology Bulletin, 38.

Kemsley, W., Redpath, R. and Holmes, M. (1980), Family Expenditure Survey Handbook, London, OPCS, Social Survey Division.

Khoman, E. and Weale, M. (2006), 'The UK savings gap', National Institute Economic Review, 198, pp. 97-111.

--(2008), 'Are we living beyond our means? A comparison of France, Italy, Spain and the United Kingdom', National Institute Discussion Paper No. 311.

Leicester, A., O'Dea, C. and Oldfield, Z. (2009), The Expenditure Experience of Older Households, London, Institute for Fiscal Studies.

McCarthy, D., Sefton, J. and Weale, M. (2011), 'Generational accounts for the United Kingdom', National Institute Discussion Paper No. 377.

National Equality Panel (2010), An Anatomy of Economic Inequality in the UK, London, Government Equalities Office.

Pensions Commission (2004), Pensions: Challenges and Choices, The First Report of the Pensions Commission, Norwich, The Stationery Office.

--(2005), A New Pension Settlement for the Twenty-First Century, The Second Report of the Pensions Commission, Norwich, The Stationery Office.

Pomerantz, O. and Weale, M. (2005), 'Are we saving enough? The macroeconomics of the savings gap', National Institute Economic Review, 191, pp. 79-93.

Scholz, J. K., Seshadri, A. and Khitatrakun, S. (2006), 'Are Americans saving "optimally" for retirement?', Journal of Political Economy, 114, pp. 607-43.

Weale, M. (2004), 'National saving and the stability and growth pact', National Institute Discussion Paper No. 246.

NOTES

(1) Weale (2004) and Pomerantz and Weale (2005).

(2) See also Auerbach, Gokhale and Kotlikoff (1994), and McCarthy, et al. (2011) for Generational Accounts recently produced for the UK.

(3) The total fertility rate in England and Wales fell from an average of 2.45 between 1945 and 1973, to 1.76 from 1974 (source, ONS statistics).

(4) See, for example, Pensions Commission (2004, 2005), and DWP (2006).

(5) The reporting period was changed from financial years (starting in April) to calendar years in 2006.

(6) This issue is partially corrected for by weights that have been supplied with the FES since 1999.

(7) The definition prior to 2002 required both shared expenses and a shared living room, rather than either of these as described under the harmonised definition that has been applied since that year.

(8) The reallocation of consumption was undertaken by first allocating individual specific measures of consumption reported by the survey, and then dividing pooled measures of household consumption on a pro rata basis by the number of members per family in the household.

(9) The FES provides household weights that are designed to correct for sample bias from 1997/98. Weights can be generated prior to this date by comparing the population structure reported by the FES to alternative data sources, as in Leicester et al. (2009). Unpublished analysis conducted at the IFS suggests that undertaking such adjustments has no appreciable impact on the statistics reported here.

Justin van de Ven, National Institute of Economic and Social Research. e-mail: jvandeven@niesr.ac.uk. I am most grateful to Martin Weale and Cormac O'Dea for helpful advice regarding comparisons between the micro-data and the National Accounts. The usual disclaimer applies.