Is there a retirement consumption puzzle in Japan? Evidence from a household panel dataset spanning several years

ABSTRACT

Using a unique long-run panel of Japanese households, this paper examines the changes in consumption at retirement (‘the retirement-consumption puzzle’). Our analysis shows that households’ expenditure does decline after the retirement of the household head and that changes in household composition at retirement cannot fully account for this decline. Changes in life-style/preferences after retirement also do not appear to explain a salient feature of the expenditure decline, namely, the strong correlation between the magnitudes of the expenditure decline and the income decline upon retirement. On the other hand, our finding that the expenditure decline is larger for households with smaller savings and/or that experienced a large unexpected income decline is broadly consistent with the standard LC/PIH augmented with unexpected shocks, while it does not rule out the possibility that there is a relatively small subset of households that are myopic and lack sufficient saving discipline.

KEYWORDS:

• Retirement
• household consumption
• life cycle/permanent income hypothesis

JEL CLASSIFICATION:

• D12
• E21
• J26

Acknowledgments

This paper forms part of our research at the Economic and Social Research Institute (ESRI) on household consumption in Japan. We would like to thank the Ministry of Agriculture, Forestry, and Fisheries for providing the micro-data from the Statistical Survey on Farm Management and Economy (SSFME). We are grateful to Daiji Kawaguchi, Midori Wakabayashi, Naohito Abe, and Takashi Unayama for their valuable comments on an earlier draft of this paper. Further, we would like to thank Kenji Umetani, Junya Hamaaki, Koichiro Iwamoto, Takeshi Niizeki, and other ESRI colleagues for their support. We also gratefully acknowledge financial support through the Grants-in-Aid for Scientific Research (C)24530231 and (A)23243046 from the Japan Society for the Promotion of Science (JSPS). The views expressed in this paper are those of the authors and do not represent those of the institutions the authors belong to.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ Notable exceptions are the studies by Bernheim et al. (2001) on US households and Smith (2006) on UK households using long-term panels spanning multiple years.

² Aguila et al. (2011) report that, in the United States, a substantial number of household heads aged between 50 and 74 reentered the labour market after temporarily leaving work. Therefore, it is not sufficient to identify whether an elderly person left their job and/or retired during a specific observation period, but also whether they stayed retired, necessitating a long-run dataset.

³ In Japan, a ‘regular employee’ is a full-time employee directly employed by his/her employer and whose employment is open-ended. Moreover, such employees are typically covered under public insurance systems such as unemployment and health care insurance systems and a retirement pension.

⁴ In addition to these variables, we include a dummy variable that takes one in the year during which a co-residing parent (of the household head or his/her spouse) passed away and zero in the other years in our regressions, since household expenditures typically spike (by about 2 million yen on average) in those years due to funeral-related expenses.

⁵ In fact, our own results indicate that no significant effect on consumption in the year of retirement can be observed.

⁶ The share of households that are asked to report expenditures by categories is only around 10 percent.

⁷ A leading explanation of the failure of the LC/PIH is liquidity constraints. However, liquidity constraints cannot explain the retirement consumption puzzle, since income decreases at the time of retirement.

⁸ Another potential factor may be the household head’s risk preferences. (We are grateful to an anonymous referee for pointing out this possibility). However, since the dataset we use in this study does not provide an appropriate variable to incorporate the household head’s risk preferences, this is an issue left for future research.

⁹ When we conduct the same regressions without the household demographic variables, the size of the estimated coefficients on the retirement dummies is not noticeably affected. This suggests that there is something beyond changes in family size that affects households’ consumption at their retirement.

¹⁰ To conserve space, Table 3 reports only the regression results when control households are included. We obtain broadly similar results when excluding the control households. The same applies to the other regression results presented below.

¹¹ The reason for using $\mathrm{\Delta }Pensio{n}_i$ is that we are focusing on households whose household members other than the household head (such as the wife or a co-residing parent) may already be receiving pension benefits. Therefore, $\mathrm{\Delta }Pensio{n}_i$ represents the change in pension benefits the household receives as a result of the retirement of the household head.

¹² The standard error for θ is 0.15 (p-value = 0.000).

¹³ Unfortunately, our results do not allow us to conclusively determine whether there are myopic (undisciplined) households, since the expenditure decline observed for households with savings and without a negative income shock may have resulted from changes in lifestyle, which our dataset does not allow us to properly control for.

Additional information

Funding

This work was supported by the Japan Society for the Promotion of Science (JSPS) [Grants-in-Aid for Scientific Research (A)23243046,Grants-in-Aid for Scientific Research (C)24530231].