Did Japanese consumers become more prudent during 1998–1999? Evidence from household-level data

Abstract

This paper explores empirically whether Japanese consumers became more prudent in the second half of the 1990s, a decade in which Japan registered historically low economic growth. Employing the methodology developed by Dynan (1993), this study uses micro-level data from the Family Savings Survey and the Family Income and Expenditure Survey to estimate the coefficient of prudence for Japanese households in the second half of the 1990s. The estimates reveal that the coefficient of prudence is positive and statistically significant in the 1998–1999 period. The obtained value for the coefficient of prudence is four, which is much higher than that estimated for US households (not significantly different from zero) or UK households (around 2). The estimated coefficient for young households is higher still, which is consistent with simulation studies conducted by Gourinchas & Parker (2002) showing that precaution is the most important saving motive for younger households.

Keywords:

• Precautionary saving
• coefficient of prudence
• Euler equation
• household data

Acknowledgment

An earlier version of this paper was published as part of Hori & Shimizutani (2002). The authors are grateful to the Ministry of Internal Affairs and Communications for providing the micro-data from the FIES, and would also like to thank the anonymous referees for their very constructive comments. The views expressed in this paper are the authors' own and do not necessarily represent those of any of the organizations to which they belong.

Notes

¹The survey allowed for multiple answers so that the sum of responses for each choice exceeds 100%.

²In addition to prudence coefficients, several studies estimated a saving function that includes demographic variables as well as several proxies for uncertainty among the independent variables. Several studies found that the precautionary motive is large (Dardanoni, 1991; Kazarosian, 1997) while others came to the conclusion that it is weak (Guiso et al., 1992; Lussardi, 1998). For Japan, Ogawa (1991) confirmed that greater income risk leads to a rise in the personal saving rate. Ishihara & Doi (2003) showed that the risk of unemployment increased the savings/GDP ratio. See Murata (2003) for a survey of recent empirical studies on precautionary saving in Japan.

³This description closely follows Dynan (1993).

⁴The FIES began to cover agricultural and fishery households in July 1999. Hayashi (1997, Chapter 5) used the FIES in his test of the permanent income hypothesis. Hori et al. (2002) also used the FIES to examine the effect of the shopping coupons program on household consumption.

⁵The FSS was merged with the FIES in 2001.

⁶Our approach to calculating annual household consumption opens the door to substantial measurement error, but unfortunately, this is the only way consumption data can be calculated using the FIES and the FSS. Moreover, while Dynan (1993) used non-durable consumption to estimate the coefficient of prudence, our data do not allow us to exclude durable consumption.

⁷On a national accounts basis, the average annual growth rate of real private consumption (excluding imputed rents) for the 1995–1999 period was 0.28%. The rate of change for individual years is as follows: 1995–1996: 2.5%; 1996–1997: 0.5%; 1997–1998: –1.0%; 1998–1999: –0.6%.

⁸The number of households removed from the sample due to this criterion is very small.

⁹We did not include the occupation of a household head as an instrument since the occupation is likely to be endogenous, as suggested by a referee.

¹⁰Recent empirical studies on consumption often assume a felicity function with a preference shifter (see, for example, Attanasio & Low, 2001; Gourinchas & Parker, 2002). The shifter is assumed to depend on household characteristics. Dynan (1993) used the age of a household head as a representative demographic variable and included this when estimating the Euler equation.

¹¹We omit the results for the case where the head of household is not employed because the sample size is too small to obtain reliable statistical results.

¹²An alternative hypothesis to explain the increase in the prudence coefficient, pointed out by a referee of this journal, is that the risk exposure of households increased in the 1998–1999 period. For instance, it may be labor income uncertainty, not the structural prudence coefficient, that jumped up in this period. However, looking at employment and household income statistics for the second half of the 1990s, we find that, according to national accounts data, both labor compensation and the number of employees already started to decline in the 1997–1998 period and did not register a substantial further deterioration in the 1998–1999 period. As a robustness check, we pooled all data and ran the same regression with time dummies indicating each period in order to see whether the time dummy for 1998–1999 was significant. We found that the coefficients on the time dummies were not significant.