Global imbalances and household savings: The role of wealth

Abstract

Many claim that fluctuations in U.S. private savings help to create and to sustain global imbalances because of their influence on the current account deficit. To test this claim, this paper investigates the determinants of aggregate household savings using a panel of 18 developed countries for the period 1980–2005. We weave two strands of literature: the first strand from consumer theory, considering specifically the ‘wealth effect’, the second strand from aggregate private savings theory. The original contribution of this paper derives from the main explanatory variables of the household savings function: two measures of household wealth, the first a financial variable and the second a variable for tangible/housing stock. The salience of these variables has not been tested before. The model is then enriched with variables taken from the private savings literature. To find the best technique to estimate the long run savings function, unit root and cointegration tests are carried out, from which evidence of a cointegrating relationship is found. The group means FMOLS is used to estimate the model. The empirical evidence suggests effects consistent with theory: an increase in wealth negatively affects household savings. Furthermore, when important explanatory variables, such as government savings and population dependency ratios, are included in the model, tangible wealth becomes the only kind of wealth to (weakly and negatively) influence household savings in developed countries. In the U.S. however, wealth does not seem to affect household savings negatively, it seems instead that government savings and population changes better explain the decline of savings during the past two decades. This finding provides additional evidence on the issue of global imbalances, and suggests that the recent booms of the stock and the real estate markets should not be blamed for the decline in U.S. household and private savings.

PACS:

• E21
• C23

Notes

1 However, some researchers do not agree on the need for rebalancing actions, feeling that the system is perfectly capable of sustaining such imbalances even in the long run (for instance, Gray, 2004, writes about the possible exhaustion of the international role of the US dollar; Dooley, Folkerts-Landau, and Garber (2005), discuss the Bretton Woods II regime).

2 From a global perspective, Bernanke (2005), Summers (2004), and Truman (2004); from a US perspective, Roubini and Setser (2004), Faulkner-MacDonagh (2003); sometimes attempts to combine those two perspectives have been made, as in Terrones and Cardarelli (2005).

3 The usual justification for the use of private savings instead of household savings is that there is a perfectly negative correlation between household savings and corporate savings. Ferrucci and Miralles (2007, 11) argue as follows, “Although the offset is generally found to be less than one for one, the finding that household saving indeed reacts to corporate saving seems to be sufficiently well documented to allow focusing this study on aggregate private saving.” However, we tested the correlation between those two variables at a country level in our panel and, though we found a significant and negative correlation in 13 out of 18 countries, the coefficient was greater than −0.5 in only half of the countries. To concentrate on household savings, instead of limiting the study to private savings, was considered worthwhile.

4 Another part of the literature concentrates on studying more deeply the effects of single variables, which are claimed to be more important than others. Two notable examples are Heer and Suessmuth (2006), who find no evidence of a negative effect of inflation on savings in the U.S., claiming that the role of this variable should be revised at a theoretical level. Another is Bloom, Canning, and Graham (2003) who use a modified version of the life cycle model to study the effects of health and longevity.

5 Consider the following examples. For the USA, Skinner (1996) finds a large and significant effect of housing wealth on consumption using aggregate data, even if in some specifications (for instance, when the long term interest rate is included in the model) the significance disappears. More recently, Belsky and Prakken (2004) find evidence in favor of the housing wealth effect, while Poterba (2000) and Juster, Lupton, Smith, and Stafford (2006) concentrate on the financial wealth effect. Turning to panel studies, Case et al. (2005) study both the financial and the housing wealth effect for a panel of 14 countries, finding a significant effect for the latter only; Edison and Sløk (2002) study the financial wealth effect in seven different countries. See Paiella (2007) for an excellent survey on the evidence of wealth effects, both from aggregate data and from microdata.

6 To be precise, the differences are the following: household savings include savings of non-profit institutions serving the household sector in all the countries of the sample apart from Finland, France, Japan and New Zealand; most countries report household savings on a net basis (excluding consumption of fixed capital by households), the exceptions being Belgium, Denmark, Portugal, Spain and the United Kingdom.

7 The best source of data on tangible wealth is the table “Household Wealth and Indebtedness” usually present in the Statistical Annex of the OECD Economic Outlooks. It contains data on tangible wealth for 7 OECD countries since 1980, but for most of the series the data are not harmonized. For some additional countries, data for much shorter periods of time are available from national official sources.

8 This idea has been exploited by other authors too. For example, Slacalek (2006) uses similar data but claims to be able to construct a measure comparable to the financial wealth one in order to compare the magnitude of the two different wealth effects. However, we believe that in order to do this he makes certain strong assumptions that we prefer to avoid.

9 This is the usual proxy in the literature, but we feel that it poses some problems. For example, part of the credit to private sector enters the measure of net financial wealth. We would prefer to use a different, better variable, to measure liquidity constraints, but we are not aware of its existence.

10 We choose not to perform tests at single country level, since “it is well known that the traditional unit root tests (…) method involves the low power problem for nonstationary data” (Kim, Oh, and Jeong (2005), 75). Detailed results are available upon request.

11 However, we performed some estimations with those methods in order to have some benchmark results. The GMM results for a dynamic specification of model (2) are similar to the ones obtained with a fixed effects model, and become closer and closer as we limit the number of instruments. However, even with the minimum number of instruments possible, we cannot satisfy the requirement of having it less than or equal to the number of groups. That is, this estimator is not recommended for a panel with N = 18 and T = 26. With the PMG estimator we did not obtain clear results, probably for the same reasons.

12 See Carroll, Otsuka, and Slacalek (2006) for a critical point of view on the reliability of the use of cointegration in the analysis of wealth, consumption/savings and income.

13 “Overall, Pedroni (2004) suggests that the panel-rho statistic seems to be the most reliable when T is large enough; for small T, the parametric group-t statistic and the panel-t statistic appear to have the highest power, followed by the panel-rho statistic” (Maeso-Fernandez et al., 2004, p. 17). Moreover, “Pedroni (2004) reports Monte Carlo simulations indicating that the panel rho and group rho have power ranging from 0 to 20% for samples with T = 20, N-20. He finds that the variance-ratio tests consistently produced low power and large size distortions” (Bonham et al., 2004, p. 16).