The 15-Hour Week: Keynes’s Prediction Revisited

In 1930, Keynes opined that by 2030, people would work only 15 hours per week. As such, this prediction will not be realized. However, expected lifetime hours of leisure and home production in the UK rose by 58% between 1931 and 2011, rather more than Keynes would have expected. This reﬂects increases in life expectancy at older ages and much longer expected periods of retirement. Leisure in retirement contributes to high life satisfaction for the elderly, but building up savings to pay for it is a barrier to working only 15 hours per week.


INTRODUCTION
In 1930, John Maynard Keynes published a short essay entitled 'Economic possibilities for our grandchildren'. It is famous (notorious?) for its prediction that a hundred years hence, people would work for only 15 hours per week. Although this was conceived originally as an after-dinner talk for schoolboys and then as a public lecture rather than as a serious piece of scholarly research, it has a high profile. This was underlined by a collection of 14 papers commenting on the essay, written mostly by distinguished economists, including four Nobel Prize winners (Pecchi and Piga 2008).
In his own words, this is what Keynes said: 'I would predict that the standard of life in progressive countries a hundred years hence will be between 4 and 8 times as high as it is today' (Keynes 1930, p. 21). At that point, 'everybody will need to do some work if he is to be contented . . . a 15-hour week may put off the problem for a great while. For 3 hours a day is quite enough to satisfy the old Adam in most of us!' (Keynes 1930, p. 23). Keynes was foreseeing a future where consumption needs could be met with very little or no work, but where most men would not want to give up work completely. 1 There is a back-of-the-envelope flavour to this prediction, but even so, it has been much discussed.
As well as market work, Keynes also contemplated the future of home production and envisaged that this would take up a modest amount of time. He thought that: 'We shall do more things for ourselves than is usual with the rich today, only too glad to have small duties and tasks and routines' (Keynes 1930, p. 23). This would, however, leave man 'faced with his real, his permanent problem . . . how to occupy the leisure, which science and compound interest will have won for him' (Keynes 1930, p. 22). In other words, work-life balance would in future become unhealthy, and leisure time would be excessive. 2 The general reaction among economists is captured nicely by Pecchi and Piga (2008, p. 3): 'How could it be that a man of Keynes's intelligence, with a deep understanding of economics and society, could be so right in predicting a future of economic growth and improving living standards and so wrong in understanding the future trends of labor and leisure, consumption and saving?' The economists in their volume discuss several plausible hypotheses to explain the failure of work hours to fall as Keynes predicted. These include: that for work-leisure choices, substitution effects were bigger and income effects smaller than Keynes (implicitly) assumed; that new goods created new consumer wants, and that preferences have changed in response to advertising; and that consumer expenditures keep increasing to keep up with or outdo fellow citizens. 3 The data reported in Table 1 are typical of those that have been used to compare outcomes with Keynes's predictions and to suggest that he was right about growth but wrong about leisure. 4 The UK has experienced lower growth than most other advanced economies since 1931 but, nevertheless, real GDP per person in 2011 was a little over 5 times the 1931 level, that is, already within the 4-8 times band, so on that measure, Keynes's prediction about economic growth was well on track. Although average hours of market work per year for men fell by more than a quarter between 1931 and 2011, in that year they were still a long way above Keynes's prediction of 780 hours, and had decreased by just over 10% in the previous 30 years.
Another way to express Keynes's prediction is as an expected increase in the amount of leisure (and/or home production). Allowing a conventional 8 hours per day for sleep, there are 112 waking hours available per week. The work week averaged 48.1 hours (2500/52) in 1931, so this left 63.9 hours for leisure and home production. A work week of 15 hours would leave 97 hours per week, so Keynes was predicting an increase of 51.8% ((97 − 63.9)/63.9) in leisure/home production time. This is more than two and a half times the actual increase between 1931 and 2011, which was 19.9% ((76.6 − 63.9)/63.9) according to Table 1.
However, this takes account of only time use pre-retirement and does not include later life. This is an important omission in the context of increases in life expectancy and length of retirement since 1931 (Tables 2 and 3), as this paper shows. I calculate that over his lifetime, a 20-year-old male in 2011 could expect 267,959 hours of leisure and home production, compared with 169,152 for his 1931 counterpart, an increase of 58.4% (Table 4). On reasonable assumptions, this is more than Keynes would have predicted, as the paper also demonstrates. The main reason for his 'under-prediction' is that, contrary to the expectation of 1930s experts, life expectancy at older ages has increased considerably since the 1960s.
When separate calculations are made for expected lifetime hours in home production and in leisure time, the result is a 37-48% increase in the latter between 1931 and 2011, which is a good deal less than Keynes might have supposed based on the information that was available in 1930. My central estimate is that a 20-year-old male in 2011 could expect over his lifetime   Sources: Derived using mortality estimates taken from cohort and period life tables in Case et al. (1962) and ONS (2018), together with labour force participation rates based on Johnson (1994) and OBR (2015).
44,496 hours of home production and 223,463 hours of leisure, compared with 12,183 and 156,969 hours, respectively, for his 1931 counterpart. Here, the unforeseen surprise is the large increase in married women's participation in market work, and an associated rise in home production by men. Finally, it is important to note that taking this lifecycle perspective gives an important insight as to why, although the total of lifetime hours in leisure and home production has risen by more than Keynes would have predicted, average hours of market work for those in employment have fallen by less than he suggested. Consumption during a longer expected period of retirement needs to be financed by the accumulation of assets during working life, and this is likely to have been underwritten by continuing to undertake a substantial amount of market work pre-retirement.

I. ESTIMATING EXPECTED LENGTH OF RETIREMENT
The method used to obtain estimates of the length of retirement is basically the same as that proposed by Lee (2001). Retirement duration is estimated for male workers at age 20 by calculating a weighted average of life expectancy at each age of retirement. The weight assigned to life expectancy at age x , denoted e x , is the probability of retiring at age x . In turn, this is the product of the following probabilities: the probability of remaining alive to   et al. (1962) and ONS (2018). age x (denoted S x ), the probability of remaining in the labour force until age x conditional on surviving to age x (denoted T x ), and the probability of retiring at age x conditional on remaining in the labour force at age x (denoted γ x ). Among the men who would retire between ages x and x + 1 (S x T x γ x ), the proportion of those who die is given by the mortality rate within the age interval ( 1 q x ). If the likelihood of retirement does not vary within the age interval, then half of these men would die before they leave the labour force. Therefore the probability of retirement between ages x and x + 1 is S x T x γ x 1 − (0.5 × 1 q x ) . If a man retires between ages x and x + 1, then his expected length of retirement is (e x + e x +1 ) /2. Assuming for simplicity that the earliest age at retirement is 50, this gives the expected length of retirement period (ELRP) at age 20 as where ρ 20−50 is the probability at age 20 of surviving until age 50. Both period and cohort estimates of ELRP are reported in Table 3. Period estimates are based on the assumption that a 20-year-old man estimates his retirement duration based on current mortality, with retirement rates remaining unchanged in future. Cohort estimates are based on the mortality and retirement rates that obtain as the 20-year-old passes through his life and are equivalent to an assumption of perfect foresight of future life expectancy and labour force participation probabilities. Except for those cohorts reaching age 20 a long time ago, estimates are based partly on future projections. Since, over time, life expectancy has been increasing and, at least until recently, fewer people have been working late in life, in general, the period estimates of ELRP may be considered as a lower bound, and the cohort estimates as an upper bound. The gap between the two has varied in the past and has described an inverted-U shape since 1931.
The data used to estimate ELRP were as follows. Life expectancies and survival rates prior to 1951 were taken from Case et al. (1962), who compiled the evidence of early English life tables that were constructed using information collected by the Registrar General. From 1951, life tables published by the Office for National Statistics (ONS) (2018) were used. These include projections through to 2068 such that cohort estimates for a 20-year-old are available up to 2011. Labour force participation rates from 1931 to 1981 are taken from Johnson (1994), who derived them from the Population Census. 5 Later years use tabulations from the Labour Force Survey and projections of future participation rates through to 2065 in Office for Budget Responsibility (OBR) (2015). Table 3 reports estimates for ELRP that reflect a large increase between 1931 and 2001. In the years since Keynes published his essay, from 1931 to 2011, the period and cohort estimates gained 11.71 and 8.04 years, respectively. The main reason for longer expected retirement is increased longevity. The period estimates for life expectancy at age 20 were 47.1 years in 1931 and 59.4 years in 2011, and at age 65 were 11.4 and 18.3 years, respectively. When Keynes wrote, a 20-year-old had a 70.4% chance of surviving to age 65, which became the conventional age at which men retired, whereas by 2011 this had risen to 87.1%. ELRP is estimated as 7.58 years using 2011 participation rates and 1931 mortality. Thus 8.79 out of the 11.71 years increase in ELRP between 1931 and 2011 resulted from improved life expectancy.

II. ESTIMATING EXPECTED LIFETIME HOURS OF MARKET WORK AND OF HOME PRODUCTION AND LEISURE
Having made estimates of ELRP, and using estimates of annual hours worked (AHW) from Table 1 The limitation of these formulae is that they do not capture phased moves into retirement where workers initially scale down their hours before retiring completely. However, since only a small proportion of lifetime expected market work hours are post-65, and there were shorter hours at these ages both in 1931 and in 2011, any bias in comparisons of expected lifetime leisure/home production between these two years is likely to be very small.  Table 4 reports a substantial change in the expected proportions of a worker's lifetime spent in work and leisure/home production. Whereas in 1931, with then current mortality and labour force participation rates, 38.5% of expected waking hours would be spent in market work, by 2011 that had fallen to 22.8%. According to these estimates, a 20-year-old male in 2011 could expect 267,959 lifetime hours of leisure or home production compared with 169,152 hours for his counterpart in 1931, an increase of 58.4%.
The increase in lifetime hours of expected leisure and home production has some interesting features. First, a relatively large part of this increase, 68,433 hours or 69.3% of the total, accrues during retirement, that is, towards the end of life when health may be Economica deteriorating. The increase in expected leisure or home production time pre-retirement is only 21.4%, compared with 58.4% overall (Table 4). Second, over half of the 35,414 increase in expected pre-retirement leisure/home production hours is in the form of vacation time, which rose from an average of 2.2 weeks per year in 1931 to 5.8 weeks in 2011 (Ward et al. 2018), so 16.2% of these hours were on vacation in 2011, compared with 7.4% in 1931. A rising share of leisure time being taken as vacations is perhaps not surprising when the very high valence rating of 'vacation' is recognized. 7 III. ESTIMATING LIFETIME HOURS OF MARKET WORK, HOME PRODUCTION AND LEISURE It is possible to make estimates of expected hours spent in home production and leisure, respectively, for pre-retirement men based on time-use diaries, but only for snapshot years beginning in the 1970s. A summary of key findings for males aged 20-59 is reported in Table 5 for 1974 and 2014. It is reasonable to regard 'core domestic work' and 'other unpaid work' as home production, and 'uncommitted time' as leisure time. 8 A notable feature of these estimates is the rise in men's time in core domestic work between 1974 and 2014. Postretirement, we might suppose that the total amount of home production time per household increases modestly, and the division between men and women is a little more equal. In view of the data collected by Ramey and Francis (2009) for the USA, I assume that in retirement compared with pre-retirement, male hours of home production increase by 1. Making an estimate of pre-retirement home production time for men in 1931 is difficult, and can be no more than an educated guess. It seems highly likely that it would be lower than in 1974. An obvious reason for this is the large increase in married women in employment, from 10% in 1931 to 49% (many part-time) in 1971. Time-use evidence for later periods suggests that more market hours worked by women is associated with increased home production time for their male partners.
This pattern is reflected in the experience of the USA. In 1930, for the 25-54 years age group, men spent 5.8 hours per week in home production and 44.7 hours in market work, whereas for women the hours were 45.9 and 10.1, respectively. In 1970, men worked 12.0 hours per week in home production and 40.1 hours in paid work, while for women the hours were 39. 1 and15.7, respectively. In round numbers, between 1930 and1970, roughly 6 hours of market work switched from men to women, and a similar number of hours of home production went the other way. With the US experience in mind, it is reasonable to assume that in the UK, on average, men pre-retirement devoted half as many hours per year to home production in 1931 as in 1971, 249.4 compared with 498.7. 9 Clearly, this is somewhat questionable, so I also report estimates based on this ratio being a quarter and three-quarters, 124.7 and 374.0 hours per year, respectively. 10 This methodology produces the results reported for all men in Table 6. Again, there is a substantial increase in the amount of expected lifetime leisure, which goes up by 42.4% in the central estimate (1931 (2)), from 156,969 hours in 1931 to 223,463 hours in 2011, or by 37.2% and 47.9% in the alternative calculations. It is striking, however, that expected lifetime hours in home production are estimated to increase markedly-from 12,183 to 44,496 in the central estimate.
A further notable feature of the estimates in Table 5 is that between 1974 and 2014, the traditional relativity between the market work hours of highly educated men and those with little education was overturned. In 2011, the former performed 331.6 minutes of market work per day, compared with 318.2 minutes per day for the latter. Indeed, time spent in market work in 2014 by highly educated men had increased compared with 40 years earlier.
As highlighted by Gershuny and Fisher (2014), in the early 20th century, the best-educated men were the leisure class during their careers, but this was no longer the case a century later.
However, the life expectancy of highly educated men in 2014 greatly exceeded that for men of low educational attainment, as reported in Table 7, with the implication that they could look forward to 5.5 more years of retirement. In turn, this meant that at age 20, they could expect about 40,000 more hours of leisure over their lifetime, even though they would expect to work longer hours prior to retirement.   (2018), and labour force participation rates from Chandler and Tetlow (2014).

IV. A LIFETIME EQUIVALENT OF KEYNES'S PREDICTION
Keynes did not discuss the future of work over the whole lifecycle, nor did he ever predict what life expectancy or retirement behaviour might be a hundred years after he wrote. It is possible, however, to make some informed guesses at what he might have assumed, and thus to make an estimate of lifetime expected hours of leisure and home production in 2030 from a 1930 Keynesian perspective. The demographic future was a high-profile issue in the 1930s in the context of worries that the future was one of a declining population. The best-known and much-publicized predictions were made by Charles (1935). She offered three variants, of which her estimate (b), which took account of increases in future life expectancy, is the most appropriate. Her views were typical of the time; she foresaw steady decreases in mortality below age 70 for the next 30 years, at which point a minimum level would have been reached, but she expected mortality over the age of 70 to remain at the 1935 level. In 1965, the maximum life expectancy at birth of 68.3 years would have been attained. This is very close to the 68.1 years in the life table for 1961 (Table 1) In his essay, Keynes seemed to think that in future, home production would not take up much of men's time. This would fit with a world in which few married women worked outside the home. It is most unlikely that Keynes would have foreseen the large increase after the Second World War in market work by married women, which became an important reason for men to spend more time in home production. The married women's labour force participation rate flatlined at about 10% from 1901 to 1931 (Hatton and Bailey 2001). Married women going out to work was seen widely as an unfortunate consequence of 'need' or a social problem (McCarthy 2016). On this view, it would surely have withered away in Keynes's future land of plenty. I shall assume that Keynes would have projected that men's hours in home production would remain at the 1931 level.
By 1930, the concept of occupational pensions, which would become widespread after the Second World War, had arrived for the lucky few. These schemes typically had a normal retirement age, and by the 1930s this was generally 65 years for men. This was also established as the state pension age in 1928, and as Hannah (1986, p. 128) put it: 'The age of 65 as the pension age for men had long been considered ideal'. By the 1950s, it was possible for middle-class men to retire completely at 65 and enjoy a comfortable retirement, which most chose to do. I shall assume for my lifetime equivalent of Keynes's prediction that age-specific labour force participation rates were as in 1931 up to age 65, with AHW = 780, after which age no one works.
If a lifetime equivalent of Keynes's prediction is based on the life table of 1961 and full retirement at age 65 for those still in work, then expected lifetime hours of leisure or home production for a 20-year-old man in 2011 are 258,720, and expected hours of market work are 35,233. The increase in lifetime leisure and home production time compared with 1931 is 53.0%. This is less than the 58.4% increase that actually accrued between 1931 and 2011. So, contrary to conventional wisdom, it seems that actual increases in the expected time available for leisure and home production taken together have exceeded anything Keynes might have predicted. Keynes was right to predict a large increase in leisure and home production time, but failed to see that much of this would be enjoyed in retirement. Keynes did not put his prediction in terms of lifetime experience, but if he had done so, then he would surely have under-rather than over-predicted the expected increase for a 20-year-old man in 2011 compared with his predecessor of 1930. The change in life expectancy since the 1930s would be a very nice surprise.
If, however, lifetime hours of home production and leisure time are estimated separately, then leisure time increased by considerably less than Keynes would have predicted between 1931 and 2011-by 42.4% compared with 56.5%. 11 This reflects a big rise in hours devoted to home production, which it is reasonable to suppose that Keynes would not have anticipated. Another striking difference between the historical experience and a Keynes-type prediction that shows up in Table 8 is that 36.7% of expected lifetime leisure occurs post-retirement in the experience, but only 11.5% is post-retirement in the prediction.
A notable feature of these estimates is that the percentage of lifetime hours devoted to leisure is a good deal lower in 2011 than Keynes would have projected, even though total leisure time has increased substantially. As reported in Table 9, whereas Keynes might have expected that a 20-year-old man in 2011 would use 83.6% of his waking hours on leisure, the period estimate for 2011 is 64.4%. Keynes may well have thought 42.4% more leisure occupying 64.4% of time a better outcome than 56.5% more leisure taking up 83.6% of time, since it might allay the fears that he had of excessive leisure, and strike him as a better work-life balance.

V. A LIFECYCLE VIEW
It is helpful to analyse decisions about how much to work in the context of a lifecycle framework in which their intertemporal aspect is recognized. A stylized example taught to economics students would see a phase of borrowing when young, accumulating assets while middle-aged, and dis-saving during retirement. It might be expected that there is consumption smoothing over time, and the central prediction is that optimization will be achieved by equating the marginal utility of consumption over time. The accumulation of assets preretirement, which reflects both work effort and savings behaviour, matters for living standards during retirement. A further aspect of a lifecycle perspective that is worth noting is that the payoff to investment in human capital when young is increased by working more subsequently, especially when wages increase significantly with experience, as is the case for cognitive workers (Lagakos et al. 2018). We might expect that decisions about education and work effort are mutually reinforcing, and that the volumes of each are positively correlated, a pattern that is seen clearly in recent data (Trostel and Walker 2006).
The much greater accumulation of assets by the average person at the end compared with the beginning of the 20th century underpins the notion of being able to finance an enjoyable retirement, as Hannah (1986) notes. These assets often included occupational pension and housing wealth as well as an entitlement to a state pension. A recent study found households adopting a three-pronged asset accumulation strategy based on home ownership, pensions and savings to provide financial security as welfare benefits have become less generous (Hillig 2019). The number of persons in receipt of payment of an occupational pension, which were based typically on defined benefit schemes, rose from 0.2 million in 1936 to 9.2 million in 2011 (Hannah 1986;ONS 2020). The number of dwellings that were owneroccupied rose from 23. 4% in 19184% in to 34.1% in 19394% in and 65.0% in 20114% in , having peaked at 69.8% in 20014% in (Holmans 2005Department for Communities and Local Government 2013). In 2011, house ownership was normal for the 'traditional working class' as well as the middle classes and the elite (Savage et al. 2013).
A lifecycle model with consumption smoothing predicts that an expectation of a longer period of retirement implies that individuals will aim to enter retirement with a larger stock of assets. This implies that an increase in ELRP will encourage working more and/or saving a higher fraction of income pre-retirement. Increases in life expectancy in the second half of the 20th century tended to mean longer retirement for many men, given the norm or even the contractual obligation to retire at age 65. 12 Life expectancy for men at age 65 rose from 11.4 years in 1931 to 18.3 years in 2011. Other things equal, this implies that to maintain a given level of consumption, the amount of assets required at retirement period was 1.6 times larger in 2011. If this is to be achieved by accumulation of housing wealth, then there is a strong incentive to move up the housing ladder rapidly by working more (Banks et al. 2017;Bottazzi et al. 2010). For those with occupational pensions, longer life post-65 acts as a disincentive in defined contribution or career-average defined benefit schemes to reduce working hours prior to retirement. 13 For those schemes where the defined benefit is based on final salary, it raises the cost in terms of foregone earnings and pension payments of a reduction in work hours. 14 The possibility of accumulating sufficient assets to enjoy retirement rather than endure a life of poverty seems to have paid off for many older people in recent times. Given a greater capacity to finance retirement and much better pension arrangements, Hannah (1986, p. 124) argues that: 'Voluntary retirement is . . . a luxury good whose incidence would be expected to grow'. Costa (1998) points to the lower price of recreational goods and their increased variety as reasons why retirement became a lot more enjoyable in the later 20th century. 15 Other economic and social historians have emphasized that many elderly people at this point had the option of a fairly comfortable retirement, that living standards in old age had improved markedly compared with the mid-20th century, both in absolute terms and relative to those in employment, and that older people often described themselves as active, involved and happy (Johnson 1989;Thane 2000). This last observation is consistent with the evidence of a large volume of happiness studies that report relatively high levels of subjective wellbeing in later life. In 2011-15, UK responses to the survey question 'Overall, how satisfied are you with your life nowadays?' on a scale of 0 to 10 averaged 7.8 at age 65, 7.85 at age 70, and 7.7 at age 85-higher than at any age between 20 and 65 (Blanchflower and Oswald 2019). 16 Overall, it seems plausible that increases in life expectancy and the associated lengthening of the retirement period not only have meant much increased time spent in non-market work and leisure in later life, but also have acted to discourage dramatic reductions in market work hours in youth and middle age. This provides a further important reason, not previously stressed in the literature, why Keynes's 15-hour week prediction was incorrect.

VI. CONCLUSIONS
It is generally agreed that Keynes's bold prediction of a 15-hour week by 2030 will not be realized. If this is taken to be 780 hours per year of market work, then it compares with 1840 hours in 2011. Between 1931 and 2011, annual hours of work fell by a little over a quarter, whereas Keynes predicted a fall of two-thirds by 2030. An alternative way to frame Keynes's prediction is in terms of an increase of 52% in time available for leisure or home production per week. Not surprisingly, this also seems wide of the mark since up to 2011, the actual increase was a little under 20%.
It is really important to consider the implications of increasing life expectancy. Partly, this is because changes in leisure and no-market work while in the labour force are not the whole story. If the increase in expected leisure and non-market work time is considered on a lifetime basis, then a very different picture is revealed. For a 20-year-old man, expected years of retirement have increased from 4.66 years in 1931 to 16.37 years in 2011, mainly as a result of improved life expectancy. The implication is that expected lifetime hours of leisure and home production taken together rose by 58.4% between 1931 and 2011. However, time spent in home production rose considerably over this period, with the implication that my central estimate of the increase in expected lifetime leisure rose by rather less, namely 42.4%.
It is also valuable to think about the choice of many hours of market work when young with a lifecycle perspective that takes account of the need to finance the additional years of retirement that come with greater longevity. Much of the additional leisure/home production time that a 20-year-old man can now expect will accrue in retirement. This should not be considered 'low value'. The evidence suggests that on average, persons aged 65 and over have a high level of life satisfaction nowadays. However, enjoying a comfortable retirement is facilitated by accumulating assets earlier in life, and this has mitigated against larger reductions in annual hours of market work.
Keynes did not make a prediction on a lifetime basis, but if he had, then it would surely have assumed that life expectancy, especially for the elderly, would have risen by much less than actually was the case. Using conventional 1930s demographers' assumptions plus a 15-hour week and universal retirement at age 65, his prediction for the increase in expected lifetime leisure and home production work hours would have been 53.0% rather than the 58.4% that transpired. Viewed from this perspective, contrary to conventional wisdom, the outcome has exceeded Keynes's expectations. However, for leisure itself, I think Keynes's expectation of the increase would probably have been around 56.5%, which exceeds comfortably the 42.4% actually achieved. ACKNOWLEDGMENTS Earlier versions of this paper were presented to a workshop at the University of Groningen, to a seminar at the University of Sussex, and to a conference at the British Academy; on each occasion, participants made helpful comments. I am also grateful to Paul Johnson and Leandro Prados de la Escosura for useful suggestions, and to Richard Disney and Daniel Sgroi for their advice. Jonathan Gershuny gave me expert help with time-use data. Two anonymous referees made excellent criticisms of an earlier version of the paper. The usual disclaimer applies.