Skip to main content

The retirement consumption puzzle in Japan


In this paper, I examine, using Japanese cross-section data, whether “the retirement consumption puzzle” exists, and if so, why. My results show that both the anticipated consumption during retirement of working households and the actual consumption during retirement of retired households are much lower than the actual consumption before retirement of working households. I find that the anticipated decline in consumption after retirement is due primarily to the anticipated decline in family size after retirement, but that it might also be due partly to other factors. These results suggest that consumption does indeed decline after retirement, but that this decline is largely due to factors that are consistent with the life cycle hypothesis.

This is a preview of subscription content, access via your institution.


  1. On the other hand, this result is contrary to the result of Hamermesh (1984), even though they use the same data source. This is because they use different estimation models.

  2. Haider and Stephens (2004) also use data on expectations. They used workers’ subjective beliefs about the timing of their retirement as an instrument for retirement in their analysis of data from the RHS and the HRS. They found that subjective retirement expectations are strong predictors of subsequent retirement decisions but still found a retirement consumption decline for workers who retire when expected.

  3. Hayashi (1985) is the seminal study of household consumption behavior that controls respondents’ expectations about both expenditure and income. He claims that because expectations are directly measured, there is no need to make a specific assumption about expectations of future income and expenditure. He uses data from the 1982 Survey of Family Consumption conducted by the Economic Planning Agency of the Japanese Government, in which families are interviewed every 3 months for 1 year, to test whether or not the permanent income hypothesis applies in Japan. He finds that the permanent income hypothesis applies to a large fraction of the population of wage earners.

  4. Restricting the sample to respondents who report their gender reduces the number of observations from 3,327 to 3,309; restricting the sample to respondents who are married reduces the number of observations further to 2,831; restricting the sample to respondents whose age is 40 or older reduces the number of observations further to 2,155 (1,241 working households, 625 retired households, and 289 other households), and restricting the sample to respondents for whom all of the necessary information is available reduces the number of observations further to 1,143 (609 working households and 534 retired households). In addition, I defined working households as a household in which all of the following three conditions apply: First, the respondent’s current age is equal to or less than his planned retirement age; second, neither the respondent nor his or her spouse is receiving social security benefits. Third, the respondent answers questions directed at those who are before retirement. I refer to households who do not meet all of these conditions as “retired households.”

  5. I also calculated the means and standard deviations of consumption (CLE and RLE) for both the subsample of working households whose head plans to retire within particular years and that of retired households whose head has retired within a particular number of years (for example, 10, 5, 3, and 2 years), and three patterns emerged: First, the amount of CLE in the case of working households whose head plans to retire within 2 to 5 years is higher than the average amount of CLE, whereas that of households whose head plans to retire within less than 2 years is lower than the average thereof; second, the amount of CLE in the case of retired households does not vary significantly by the number of years since retirement; and third, the closer one is to retirement, the higher is the amount of RLE.

  6. As I explained in Section 1, leisure-related expenses might increase after retirement if consumption and leisure are complements. In this case, respondents must have expected the decline in consumption arising from the decline in family size and/or work-related expenses to exceed the increase in consumption arising from the increase in leisure-related expenses.

  7. For example, Bernheim et al. (2001) calculated the change in the (ln) average consumption between the 2 years prior to retirement and the 2 years postretirement. The average change was −14%, whereas the median decline was −12%.

  8. The survey I used in my analysis asks “After the household head retires, with what kinds of income do you expect to finance your living expenses? Check all that apply.” The answers are as follows: social security, 79.1%; dissaving, 45.2%; employment income during retirement, 40.2%; insurance and private pensions, 38.3%; and lump-sum retirement payments and company pensions, 30.7%.

  9. Unfortunately, I cannot focus on the latter reason in this section because no information is available in the survey I used on the composition of consumption, although previous studies do focus on this reason.

  10. In this section, I did not confine the sample to respondents for whom all of the necessary information is available because the subsample of respondents who do not have any children is very small. As seen from Table 3, this has little impact on the average amounts of RLE and CLE (see Section 2.4).

  11. Restricting the sample to working households for whom both the amounts of CLE and RLE are available reduces the number of observations from 1,241 to 1,154.

  12. The reasons why I use Phipps’s results are as follows: First, it is often said that scales based on demand behavior, estimated using a complete demand system and subjective estimators, seem more promising than other methods such as the Engel method or the Rothbarth method. Second, the complete demand system approach is formally grounded on economic theory, whereas subjective estimators are not based on any economic theory. Third, the estimates obtained by Phipps appear reasonable relative to others currently available in the literature [for example, Suruga’s (1995) estimated equivalence scales for two parents living with one child is about 1.15 to 1.30]. Fourth, Phipps calculates equivalence scales for two parents living with multiple children, which is suitable for my research.

    I do not use the Organisation for Economic Co-operation and Development (OECD) modified scale, which is one of the most commonly used scales along with the OECD Oxford scale and the square root scale because these scales are generally so steep that the welfare of large families is underestimated in comparison to that of small families [see Van Praag and Warnaar (1997)]. In addition, there are many careful studies that calculate equivalence scales using Japanese data, but few of them focus on the case of multiple children. The only exception is Nagase (2001) but she considers only two specific cases. Nagase’s (2001) estimated equivalence scales for two parents living with two children are about 1.47 (if one child has not yet started compulsory education and the other child is in elementary school) and 1.65 (if one child is in elementary school and the other child is in junior high school or above).

  13. Previous studies use the similar estimation models. For example, Banks et al. (1998) regress the change in log nondurable expenditure on the number of adults in excess of two, the real interest rate, the age of the head, the change in the logarithm of the survival probability, a dummy variable that equals one for those whose head is out of the labor market, and a dummy variable that equals one for those whose head is unemployed, whereas Bernheim et al. (2001) regress the change in log consumption on income replacement quartile, family size, marital status, a disability dummy, a female widower dummy, and a dummy variable for whether the household was working part-time for 3–4 years before full retirement.

  14. I also tried regressing consumption growth from before retirement to after retirement [RCONS2 = ln(RLE/ADJCLE)] on the rate of change of income after retirement (RINC), retirement span (RETSPAN), working span (WRKSPAN), and occupation (SLFEMPLY) using the variable ADJCLE instead of CLE (ADJCLE is RLE adjusted for family size using equivalence scales) and found that the coefficients of RINC and WRKSPAN are still positive and significant, which is consistent with the estimation results presented in Section 4.2.

  15. In Japan, annual income includes not only regular monthly income but also special income (especially seasonal bonuses). The amount of lump-sum retirement payments is calculated as a multiple of regular monthly income at retirement. The composition of annual income is as follows: annual income for private companies = regular monthly income × 12 + bonus + special income, and annual income for government = regular monthly income × 12 + family allowance + temporary allowance + end-of-the-year bonus + special end-of-the-year bonus + diligence allowance.

  16. I did not control for the “cohort effect” (Dicks-Mireaux and King 1984) for the following two reasons: First, it is not unrealistic to assume that not only young workers but also old workers are capable of learning new technologies. Second, in the case of Japan, using historical data to estimate the cohort effect will lead to substantial biases because the high growth period is included.

  17. I use five age groups (40–44, 45–49, 50–54, 55–59 and 60 above), three education groups (junior high school graduate, high school or junior college graduate, and college graduate), and six firm size groups (1–4 employees, 5–29 employees, 30–99 employees, 100–499 employees, more than 500 employees, and government workers—government workers and official workers). There is no information on lump-sum retirement payments in the case of firms with less than 30 employees in the “Survey on Retirement Allowance System and Payments.” Thus, I regressed the multiple of lump-sum retirement payments to regular monthly wages at retirement on firm size and used the predicted value from this regression.

  18. The “Survey on Retirement Allowance System and Payments” collects information only on males. Thus, I assumed that the male–female gap in lump-sum retirement payments is the same as the male–female gap in wages, as given in the “Wage Census” when calculating the lump-sum retirement payments of spouses working for private companies. In addition, I use the same method as in the case of household heads to calculate the lump-sum retirement payments of spouses who work for the government because it is often said that the male–female gap in wages and lump-sum retirement payments is smaller in the case of government workers than it is in the case of private company workers.


  • Ameriks A, Caplin A, Leahy J (2002) Retirement consumption: insights from a survey. NBER working papers no. 8735. National Bureau of Economic Research, Cambridge, MA

    Google Scholar 

  • Attanasio OP, Weber G (1993) Consumption growth, the interest rate and aggregation. Rev Econ Stud 60(3):631–649

    Article  Google Scholar 

  • Attanasio OP, Weber G (1995) Is consumption growth consistent with intertemporal optimization? Evidence from the consumer expenditure survey. J Polit Econ 103(6):1121–1157

    Article  Google Scholar 

  • Attanasio OP, Banks J, Meghirm C, Weber G (1999) Humps and bumps in lifetime consumption. J Bus Econ Stat 17(1):22–35

    Article  Google Scholar 

  • Banks J, Blundell R, Tanner S (1998) Is there a retirement-savings puzzle? Am Econ Rev 88(4):769–788

    Google Scholar 

  • Bernheim BD, Skinner J, Weinberg S (2001) What accounts for the variation in retirement wealth among U.S. households? Am Econ Rev 91(4):832–857

    Article  Google Scholar 

  • Deaton A (1992) Understanding consumption (Clarendon lectures in economics). Oxford University Press, Oxford

    Google Scholar 

  • Dicks-Mireaux L-DL, King MA (1984) Portfolio composition and pension wealth: an econometric study. In: Bodie Z, Shoven JB (eds) Financial aspects of the United States pension system. National Bureau of Economic Research project report. University of Chicago Press, Chicago, pp 399–435

    Google Scholar 

  • Gourinchas PO, Parker J (2002) Consumption over the life cycle. Econometrica 70(1):47–89

    Article  Google Scholar 

  • Hamermesh DS (1984) Consumption during retirement: the missing link in the life cycle. Rev Econ Stat 66(1):1–7

    Article  Google Scholar 

  • Haider S, Stephens M Jr (2004) Is there a retirement-consumption puzzle? Evidence using subjective retirement expectations. NBER working papers no. 10257. National Bureau of Economic Research, Cambridge, MA

    Google Scholar 

  • Hayashi F (1985) The permanent income hypothesis and consumption durability: analysis based on Japanese panel data. Q J Econ 100(4):1083–1113

    Article  Google Scholar 

  • Horioka CY, Okui M (1999) A. U.S.–Japan comparison of the importance and determinants of retirement saving. Econ Lett 65(3):365–371

    Article  Google Scholar 

  • Hubbard RG, Skinner J, Zeldes SP (1995) Precautionary saving and social insurance. J Polit Econ 103(2):360–399

    Article  Google Scholar 

  • Hurd M, Rohwedder S (2003) The retirement-consumption puzzle: anticipated and actual declines in spending at retirement. NBER working papers no. 9586. National Bureau of Economic Research, Cambridge, MA

    Google Scholar 

  • Kotlikoff LJ, Spivak A, Summers LH (1982) The adequacy of savings. Am Econ Rev 72(5):1056–1069

    Google Scholar 

  • McClements LD (1977) Equivalence scales for children. J Public Econ 8(2):191–210

    Article  Google Scholar 

  • Miniaci R, Monfardini C, Weber G (2002) Is there a retirement consumption puzzle in Italy? Mimeo, University of Padua, Italy

  • Mutoh H (1992) Nihon ni okeru Kosodate Kosuto to Kodomo Jyuyou (The cost of a child rearing and the demand of a child). Nihon Keizai Kenkyu 22(2):119–136

    Google Scholar 

  • Nagase A (2001) Kodomo Kosuto no Suikei: Kakei Oyobi Shisan Men Karano Bunseki (The estimation of the cost of a child: the analysis of the household side and the wealth side). Jinkougaku Kenkyu 28:1–15

    Google Scholar 

  • Phipps SA (1998) What is the income “cost of a child”? Exact equivalence scales for Canadian two-parent families. Rev Econ Stat 80(1):157–164

    Article  Google Scholar 

  • Robb AL, Burbidge JB (1989) Consumption, income and retirement. Can J Econ 22(3):522–542

    Article  Google Scholar 

  • Suruga T (1993) Estimation of equivalence scales using Japanese data. Econ Stud Quarterly 44(2):169–177

    Google Scholar 

  • Suruga T (1995) Equivalence scale niyoru Kodomo no Hiyou no Keisoku (Measurement of equivalence scale for children). Nihon Tokei Gakkaishi 25(3):223–331

    Google Scholar 

  • Suruga T, Nishimoto M (2001) Touka Shakudo to Kodomo no Hiyou-Shouhi Seikatsu ni kansuru Paneru Chousa wo Shiyoushite (Equivalence scale and the cost of a child-using Japanese panel survey of consumers). Kikan Kakei Keizai Kenkyuu 50:25–31

    Google Scholar 

  • Takagi S, Tokoyama Y, Kaneko N (2005) Touka Shakudo wo Mochiita Koureisetai no Seikatsu Suijun no Hyouka (The evaluation of living standard for retired households using equivalence scales). Kaikei Kensa Kenkyu 30:109–128

    Google Scholar 

  • Van Praag BMS, Warnaar MF (1997) The cost of children and the use of demographic variables in consumer demand. In: Rosenzweig MR, Stark O (eds) Handbook of population and family economics. Elsevier, Amsterdam, The Netherlands, pp 241–263

    Google Scholar 

  • Wakabayashi M (2001) Retirement saving in Japan: with emphasis on the impact of social security and retirement payments. J Jpn Int Econ 15(2):131–159

    Article  Google Scholar 

Download references


I am greatly indebted to Charles Yuji Horioka for his kind advice, comments, and encouragement throughout the process of writing this paper. I would also like to thank Daiji Kawaguchi, Miki Kohara, Wataru Kureishi, Colin McKenzie, Mayumi Nishimoto, Fumio Ohtake, Kei Sakata, Terukazu Suruga, Keiko Tamada, Jun Tomioka, Junmin Wan, the members of Professor Horioka’s graduate seminar participants of the Japanese Economic Association meeting of October 2003, and especially two anonymous referees for their helpful comments and discussions. This paper was prepared under the supervision of Charles Yuji Horioka, professor at the Institute of Social and Economic Research, Osaka University and former special guest research officer of the Institute for Posts and Telecommunications Policy. I am grateful to the Institute for Posts and Telecommunications Policy for allowing me to use these data. I am indebted to the Ministry of Education, Culture, Sports, Science, and Technology of the Japanese Government for the grant-in-aid for scientific research no. 18330068.

Author information

Authors and Affiliations


Corresponding author

Correspondence to Midori Wakabayashi.

Additional information

Responsible Editor: Junsen Zhang



The estimation of lump-sum retirement payments

In Japan, the amount of the lump-sum retirement payments of salaried workers is calculated as a multiple of their regular monthly income at retirement.Footnote 15 Hence, I first estimated permanent income at retirement using Dicks-Mireaux and King’s (1984) method and then calculated lump-sum retirement payments from my estimate of permanent income at retirement.Footnote 16

I calculated salaried workers’ permanent income at retirement as follows: First, because there is no information on the amount of permanent income at retirement, I regress the logarithm of current earnings (CI) on dummy variables pertaining to age, education, and firm size for each sex. Footnote 17 Second, I assume that one half of the residual from this regression is an unobservable individual-specific effect. Finally, I estimated the permanent income at retirement by calculating the fitted value from the earnings equation with the retirement age dummy substituted for the actual age dummy, adding the unobservable individual-specific effect to it, and taking the exponential of it.

Next, I calculated salaried workers’ lump-sum retirement payments. First, I explain how I estimated the lump-sum retirement payments of salaried workers working for private companies. I first converted permanent income at retirement from a yearly basis to a monthly basis by using data on the ratio of regular monthly wages to annual income from the 1996 “Wage Census (Chingin Sensasu)” conducted by the Policy Planning and Research Department, Minister’s Secretariat, Ministry of Labour of the Government of Japan. I then estimated the amount of lump-sum retirement payments by multiplying my estimate of regular monthly wages by the multiple of lump-sum retirement payments to regular monthly wages at retirement (by firm size) taken from the “Survey on Retirement Allowance System and Payments” (this survey is conducted only every 3 or 4 years, so I used the survey closest to the year to which my data pertain). Next, I explain how I estimated the lump-sum retirement payments of salaried workers working for the government. In the case of Japan, there is no information on the wages of government workers unlike in the case of salaried workers working for private companies. Thus, I used the 1997 “Wage Manual for Government Workers (Koumuin Kyuuyo Binran)” to calculate the regular monthly wages of government workers from their annual income and estimate the amount of their lump-sum retirement payments from my estimate of their regular monthly wages at retirement.Footnote 18 Finally, I assume that the lump-sum retirement payments of self-employed workers, part-time workers, and housewives is equal to zero because, as I explained earlier, lump-sum retirement payments are paid only to salaried workers working for a private company or for the government.

Rights and permissions

Reprints and Permissions

About this article

Cite this article

Wakabayashi, M. The retirement consumption puzzle in Japan. J Popul Econ 21, 983–1005 (2008).

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI:


  • Retirement consumption puzzle
  • Life cycle model
  • Japanese consumption behavior

JEL Classification

  • D12
  • D91
  • E21