RDP 2007-11: Global Imbalances and the Global Saving Glut – A Panel Data Assessment 5. Results

Table 2 presents the parameter estimates of our model with various lags of the dependent variable. If no lags are included the model has significant first-order autocorrelation; it is only when two lags are introduced that the regression displays no significant second-order autocorrelation.[16] Gruber and Kamin (2007) note that the lag can be motivated by habit persistence and find it to be significant at the 10 per cent level despite using five-year average data. It may also reflect adjustment costs and uncertainty (Kent and Cashin 2003).

Table 2: Estimation Results
Dependent variable is the current account to GDP ratio
CA lags
Including one
CA lag
Including two
CA lags
(CA/GDP)it–1   0.40
(CA/GDP)it–2     −0.24***
crisisit 0.51
crisisit–1 2.67**
crisisit–2 2.07*
crisisit–3 −0.56
crisisit–4 −0.29
dependencyratioit −46.14***
fiscalbalanceit 0.35***
termsoftradeit 0.11***
termsoftradeit–1 0.12***
financialdeepnessit −1.01
growthforecastsit −1.70
institutionsit −4.88***
institutionsit*growthit 0.22*
ongoingcrisisit*growthit 0.16
financialdeepnessit*growthit 0.05
R2 0.38 0.21 0.40
Number of observations 444 444 443
Wooldridge test for autocorrelation (p-value) 0.00    
Instruments   (CA/GDP)it–2 Δ(CA/GDP)it–3
Arellano-Bond test (p-value) for second-order autocorrelation   0.00 0.76
Notes: Robust standard errors are shown in brackets. ***, ** and * denote significance at the 1, 5 and 10 per cent levels respectively. Fixed effects are not reported.

Focusing on the model with two lags of the dependent variable, we find that financial crises appear to have a significant and positive effect on the current account balance (as a per cent of GDP), which is around 2.2 percentage points higher in the year after the crisis.[17] The effect of the crisis is short-lived – the crisis dummies for the third and fourth years after its onset are insignificant at the 10 per cent level. It is surprising that these effects do not last longer, given that after the Asian crisis, the building of ‘war chests’ of foreign reserves appears ongoing (see Figure 1). Nevertheless, the finding that financial crises do boost the current account balance is consistent with the global saving glut hypothesis, and contrasts with the findings of Gruber and Kamin (2007), whose models only found a positive relationship when the financial crisis dummy variable was interacted with a term to capture the openness of the economy.

Deeper financial markets appear to attract capital inflows and allow countries to run a lower current account balance than otherwise. This may reflect the greater ability of these nations to produce suitable financial assets (as suggested by Caballero et al 2007) or, more generally, they make a country a more attractive destination for foreign capital. However, the coefficient on stock market turnover (financial deepness), is not significantly different from zero at the 10 per cent level, which could, in part, reflect collinearity with the term that interacts this variable with the growth forecasts.[18] A possible economic rationale for the insignificance of the stock market turnover coefficient is that the institutional quality variable – which is highly significant and has the expected negative sign – is capturing the ability to deliver suitable financial assets, as outlined in Caballero et al. If so, this would suggest that the onset of a financial crisis might not be associated with a sharp reversal in perceived financial depth, as the institutions variable tends not to vary much around a crisis (Figure 3).[19] This may reflect both the infrequency with which the institutions variable is measured and the fact that, by construction, it places considerable weight upon objective (statistical) indicators of institutions and therefore it may not capture perceived institutional quality very well. It is possible that prior to the east Asian crisis, for example, investors were overly optimistic about the quality of prevailing institutions.

As we have not included a variable to explicitly capture the stage of development, one might have expected the institutions variable to do so. However, the negative coefficient for the institutions variable runs counter to the development hypothesis.

The fiscal balance is estimated to be an important determinant of the current account; a 1 percentage point reduction in the fiscal balance (relative to GDP) ratio is associated with an immediate 0.3 percentage point decrease in the current account balance (as a per cent of GDP). This estimate is approximately three times the value obtained by Gruber and Kamin (2007), but is within the range of estimates reported by Chinn and Ito (2007).[20]

Demographic factors are estimated to have a significant impact on a nation's current account, with a 1 percentage point rise in the dependency ratio lowering the current account balance by around 0.5 percentage points of GDP. The direction of this effect is consistent with expectations, as the savings rate for dependents is likely to be less than that for the working-age population.

Bernanke (2005) argues that the increase in oil prices since the late 1990s was a factor contributing to nations in Africa and the Middle East shifting to be net lenders of capital. Our estimates broadly support this: growth in the terms of trade of 1 per cent in a year is estimated to increase a nation's current account balance by around 0.2 percentage points of GDP by the end of the following year.

Better prospects for future growth appear to be associated with a significant decline in the current account balance. Ignoring the interaction terms, a 1 percentage point rise in the relative growth forecast is estimated to decrease the current account by around 1.6 percentage points of GDP. The interaction terms are generally insignificant. (These terms attempt to allow for the possibility that growth may create demand for financial assets, but if domestic financial markets are unable to provide such assets then this would stimulate capital outflows.) The exception is the interaction between growth forecasts and the quality of institutions, which suggests, for example, that the marginal effect of an increase in the growth forecasts on the current account balance is more negative for nations with below-average institutions. We do not consider this strong evidence against the hypotheses of Caballero et al (2007), as it is difficult to capture the interactions between the supply and demand for financial assets in our model.


Second-order autocorrelation is problematic as it implies that the residuals in Equation (1) have first-order autocorrelation and that our instruments are inappropriate. The Arellano-Bond test for autocorrelation is implemented in Stata with the abar command (Roodman 2004). [16]

In the discussion of the impact of a change in the explanatory variables, we assume that it does not alter the GDP-weighted average. This, obviously, is a poor approximation for a large country, such as the United States, for which any given change will have less impact (compared to a small country) as it will tend to lift the GDP-weighted average. [17]

The magnitude of the coefficient remains broadly unchanged when the interaction term is omitted, but the standard error decreases considerably. [18]

On the flipside, the institutions measure may not capture improvements that might be undertaken post-crisis in order to attract funds (including from the IMF). [19]

Chinn and Ito raise the possibility that the fiscal balance is endogenous. When we instrument the change in the fiscal balance with itself lagged two years, the coefficient drops to 0.14 and is insignificant. [20]