The impact of retirement on household consumption in Japan

Abstract

Using monthly data from the Japanese Family Income and Expenditure Survey, we examine the impact of retirement on household consumption. We find little evidence of an immediate change in consumption at retirement, on average, in Japan. However, we find a decrease in consumption at retirement for low income households that is concentrated in food and work-related consumption. The availability of substantial retirement bonuses to a large share of Japanese retirees may help smooth consumption at retirement. We find that those households that are more likely to receive such bonuses experience a short-run consumption increase at retirement. However, among households that are less likely to receive a retirement bonus, we find that consumption decreases at retirement.

Introduction

Driven in part by the aging of populations across the globe, there is widespread concern regarding the ability of households to adequately plan and save for retirement. A lack of planning for retirement as well as a lack of knowledge regarding one’s private pension benefits has been found in studies of households in the US (Gustman and Steinmeier, 1999, Lusardi, 2003, Chan and Stevens, 2008). The empirical evidence as to whether households are adequately saving for retirement yields conflicting findings. Studies that use simulation methods to compare optimal savings levels to those found in the actual US population indicate that savings are indeed adequate (Engen et al., 1999, Scholz et al., 2006). Studies that examine consumption changes at retirement find evidence that consumption systematically falls at retirement which, based on the Life-Cycle/Permanent Income Hypothesis (LCPIH), may indicate that households do not adequately save in advance of an expected reduction in income at retirement. This result has been found in the UK (Banks et al., 1998, Smith, 2006), the US (Bernheim et al., 2001, Fisher et al., 2008, Haider and Stephens, 2007), Italy (Miniaci et al., 2003, Battistin et al., 2009), and Germany (Schwerdt, 2005).

More recently, however, there is evidence of heterogeneity in the consumption response at retirement. For example, Smith (2006) only finds a response for those individuals who involuntarily retired while Battistin et al. (2009) find no effect once they control for changes in household composition at retirement. In addition, Aguiar and Hurst, 2005, Hurd and Rohwedder, 2003, Hurd and Rohwedder, 2008 provide evidence which suggests modifying the LCPIH to incorporate household production may help explain the drop in consumption at retirement. In a thorough review of this literature, Hurst (2008) notes that there is substantial heterogeneity in spending changes at retirement across consumption categories and across households. In this sense, more evidence is needed.

In this paper we investigate whether consumption falls at retirement in Japan. Specific features of the Japanese retirement system provide an interesting setting for examining retirement consumption changes. Many private employers in Japan institute a feature known as “Teinen” retirement, now typically at age 60, when employees must either leave their firm or are transferred to significantly lower paying jobs elsewhere within the company. However, many workers receive rather sizable bonuses upon reaching their Teinen retirement age which, depending upon the worker’s sector of employment, can average more than four times their annual income. In addition, retiring private sector workers were eligible to receive unemployment benefits, in addition to their public pension benefits, for up to one year following retirement prior to 1998. These additional sources of income can help households offset the large drop in monthly income at retirement and may lead to dramatically different retirement consumption changes in Japan relative to other developed countries.

We use data from the Japanese Family Income and Expenditure Survey (JFIES), which is a large monthly household panel survey that collects information on consumption and income for six consecutive months and thereby enables us to analyze within-household consumption changes at the exact month of retirement. As Blau, 2008, Hurd and Rohwedder, 2008 emphasize, using a true panel dataset is much more advantageous when identifying the impact of retirement on consumption, whereas prior studies typically use synthetic cohorts or cross-sectional surveys (e.g. Banks et al., 1998, Hurd and Rohwedder, 2003, Fisher et al., 2008, Wakabayashi, 2008).

We find little evidence that consumption immediately decreases at retirement, on average, in Japan although monthly income decreases sharply when households exit the labor force. We also find that consumption falls when workers exit the labor force due to unemployment which suggests that data quality issues do not preclude us from finding a response at retirement. Our finding that consumption does not fall at retirement in Japan is consistent with the results of Wakabayashi (2008) who finds, using data from a cross-sectional survey, that expected consumption changes at retirement can be explained by life-cycle factors including expected changes in household composition at retirement. Since Japan has a relatively high personal savings rate, we address the possibility that these results are driven primarily by asset accumulation by separately estimating the model for above and below median income households. We find that consumption decreases at retirement for the below median income households although this reduction is limited to food and work-related consumption. This finding is consistent with results in the prior literature surveyed in Hurst (2008).

An important institutional difference between Japan and most other countries is the generous bonuses distributed by firms at Teinen retirement. Battistin et al. (2009) note that Italian households also receive large retirement bonuses and, similar to our full sample results for Japan, find that the consumption of these households does not fall at retirement (after controlling for contemporaneous family size changes). We present evidence that the receipt and magnitude of bonuses in Japan depends upon one’s sector of employment. Public sector and large private firm employees are more likely to receive these bonuses and, conditional upon receipt, collect larger bonuses. We find that workers in these sectors experience a short-run increase in consumption at retirement which is surprising given that the amount and receipt of these bonuses are known in advance to retirees. We also find that small private firm employees, who are less likely to receive bonuses and also earn smaller bonuses, reduce their consumption at retirement as has been found in other countries. Thus, while Japanese households do not, on average, exhibit significant consumption decreases at retirement, the estimated responses differ substantially across employment sectors.

One caveat for our estimates is that retirement is treated as exogenous in our analysis. Since the JFIES is a monthly panel dataset, we cannot use age as an instrument for retirement as has been done in prior papers (e.g., Aguiar and Hurst, 2005) unless we were to ignore the panel aspect of the JFIES. While Haider and Stephens (2007) have demonstrated potential concerns with using age as an instrument, they also find similar results in the US for both OLS and 2SLS estimates which use retirement expectations as an instrument. The widespread use of Teinen retirement by Japanese employers, however, means that most employee retirements in Japan are very predictable. However, since we cannot clearly define which employees are retiring at a time that they had previously anticipated, our findings must be interpreted with this endogeneity concern in mind.

It is also important to note that the JFIES covers a six-month sample period for each household which is a much shorter time interval in comparison to other panel datasets used in this literature such as the Panel Study of Income Dynamics, Health and Retirement Study, and the British Household Panel Survey (e.g., see Bernheim et al., 2001, Smith, 2006, Haider and Stephens, 2007, Blau, 2008, Hurd and Rohwedder, 2008). An advantage of the JFIES relative these other panel datasets is that the monthly data allow us to observe the immediate consumption response to retirement rather than examining the response over a one or two year window between survey waves. To the extent that other events which occur between survey waves also influence consumption decisions, the immediate consumption response we estimate can be directly tied to the household’s transition into retirement.

A limitation of following households for only six months, however, is that the impact of poor planning on consumption may not become evident for many months or even years after retirement. For example, Bernheim et al. (2001) find that in addition to a consumption decline one to two years after retirement, US households experience an additional drop in consumption three to four years after retiring. Since we are using a six month sample period, we cannot rule out that consumption subsequently declines across all employment sectors after our sample period ends. We view our approach as testing a weak form of the LCPIH which is that the retirement consumption response occurs immediately upon leaving the labor force. Thus, it is difficult to directly compare our short-run estimates to the prior literature which examines longer periods following retirement. If consumption continues to fall during the years following retirement as it does in the United States (Hamermesh, 1984, Bernheim et al., 2001), our findings for the short-run consumption response can be considered as a lower bound for the long-run impact of retirement on consumption in Japan.

The remainder of the paper is set out as follows. The next section discusses a number of aspects of the Japanese retirement benefit system including Teinen retirement. The following section discusses the data used in this paper, the Japanese Family Income and Expenditure Survey. Section four presents the results of our analysis and the final section concludes.