Introduction
The twin deficits hypothesis claims that government budget deficits cause current account deficits. To understand how this is possible, a breakdown of the hypothesis is useful: firstly, the national account identity tells that the income of a country (GDP) comes from the monetary amount of private (household) consumption, the government’s consumption, and the investment. Given that a large part of economies is nowadays, open one adds to the consumption side the monetary value of exports and subtracts the imports, as they are goods and money which leave the economy. Very intuitively, a government budget deficit arises when the consumption is greater than the income. In other words, all else being equal, the investment/the savings of the country have declined and the government cannot fund the consumption with domestic means. The consequence of this is that the country must borrow from abroad who is not in the same situation to maintain consumption. This involves a current account deficit. This, however, can be attributed to the national accounting identity (Corsetti & Müller, 2006; lecture notes):
Y=C+I+G+EX-M
Current Account
Y-C+I+G=CA
If a fiscal deficit exists (left hand side is negative) implies that the current account also has to be negative (right hand side, i.e. exports are smaller than imports), which represents a current account deficit. Also, from the point of view of policy-making, this concept is relevant when the government decides the adopt measures (e.g, increasing taxes) which could increase the fiscal deficit.
The experience of United States in the 1980s could be regarded as evidence for this pattern: the current account deficit and the budget deficit showed positive correlation. However, the sign of the correlation changed after the 1990s.
In the empirical literature, the twin deficit concept has been analyzed more in depth: Bernheim tries to explain this pattern by regressing the current account balance on the current budget surplus and current and lagged values of GDP. He finds a positive relationship between the two deficits (the coefficient on the budget variable is 0.303 with standard deviation of 0.080). He also tests other specifications of the regression (including the government spending as a regressor, to control for the cyclical effects and government spending) and the same positive relationship (the coefficient being 0.366) maintains itself. Kim and Roubini employed a different econometric methodology (a VAR model) and their results were different: as a result of an expansionary fiscal policy shock, the current account improves (via an increase in private savings and decrease of investments). Lastly, Bussière, Fratzscher and Müller claim that productivity shocks and budget deficits are the main drivers behind movements in the current account. They relax the assumption of the Ricardian household behavior and test their claim with the standard intertemporal model.
Data
In order to test the hypothesis of the twin deficits for US, I have chosen the following time-series: real GDP, real private consumption, real government consumption, personal savings and the current account balance. In order to calculate the budget balance, one needs the data for the above macroeconomic indicators. In this setup, the Y from the formula above is the GDP of the country, the rest of the series is explicit. The series are reported on a quarterly basis, the real time-series are all reported in billions of chained 2009 dollars and the time span is from the first quarter of 1947 until the last quarter of 2014. In order to obtain real data for all the time-series, I will convert the nominal data to 2009 chained dollars, by dividing the nominal values to the GDP deflator with index 2009=100. Real-data is always preferred because it allows the analysis and results to bring up the true relationship between the variables, as opposed to the nominal data, which can be heavily influenced by the movements in prices and might lead to untrue statements.
Figure 1 shows the evolution of the budget and current account balances from first quarter of 1947 until the last quarter of 2014. Based on the FRED data samples, it appears that the deficits are not at all times positively correlated, on the contrary, they seem to be moving against each other: in the late 80s-early 90s, one can clearly see that while the budget balance was positive (budget surplus), the current account balance was suffering from a deficit. The same phenomenon happens at the end of the 2000s. Until the 70s however, the budget and current account balances seem to be evolving similarly (although not visible from the graphic due to the scaling, both balances were fluctuating around approximately [-10,10] billion dollars). If one restricts itself to this figure, one could conclude that in fact, the budget deficit does not cause or
Figure 1: Budget and current account deficits
Source: Author’s calculations
determinates the current account deficit.
After the visual inspection, the next step to check whether this relationship holds within the data sample is the correlation between the two time series. One approach is to compute the correlation matrix for the two time series. This can be done for all the observations, pooled together or also interesting could be dividing the series per administrations. Table 1 and 2 from below present the results.
(1951-2014)
[,budget deficit] [,current account deficit]
[budget deficit,] 1.0000000 0.4273751
[current account deficit,] 0.4273751 1.0000000
Democrats 1961-1969
[,budget deficit] [,current account deficit]
[budget deficit,] 1.00000000 0.03236284
[current account deficit,] 0.03236284 1.00000000
Republicans 1981-1983
[,budget deficit] [,current account deficit]
[budget deficit,] 1.0000000 0.0456551
[current account deficit,] 0.0456551 1.0000000
The entire data sample shows a positive correlation between the two deficits of 0.42, which is a relatively significant relationship. However, one cannot say the same about the correlation performed on the subsamples (based on regimes): the values of the correlation coefficients are 0.032 and 0.045, which are too small values to be considered significant. An explanation for these results could be firstly, econometrically: the sample for the both administrations is too small for the data to signalize a correlation. Given that the two correlation coefficients for the administration are very close in magnitude, one cannot draw any meaningful conclusion whether during one regime, the evidence for the correlation is striking. Under both regimes, the two deficits seem to have been uncorrelated.
Another potential approach to establish the relationship between the two series is to conduct a correlogram for the two time-series, which also considers the lagged values (as opposed to the results above that are restricted to contemporaneous correlation). Figure number 2 below serves
Figure 2: The correlation between the budget and current account balances
Source: Author’s calculations
this purpose. One can clearly notice that the highest point of correlation between the two time-series is at more than 4 lags behind and that the correlation is positive. This says that a budget
deficit is likely to cause a current account deficit approximately five quarters later (lag =-5). This result also backs up theory: until all the transmission mechanisms of the fiscal policies work at full power, one should not expect to see such a fast effect in the current account deficit: in reality, when a fiscal policy is enforced, its effects are only registered a period of time afterwards (could be quarter of year or longer, or 2-3 months). In fact, time-series analysis is more suitable to this research question because it captures the effect of time on the data, whereas cross-section does not and might give erroneous results.
As argued by many (among others Corsetti & Müller, 2006), there are other mechanisms that play a role in the evolution of the current account, after certains fiscal policies were adopted. For instance, the announcement of increased public debt (which will be offset by higher taxes) will determine the households to increase their private savings. Higher savings in the following periods means the government can fund the public debt through internal funds, without having to resource to international means (i.e., current account deficit).
Furthermore, if one considers the scenario of a fiscal expansion, then the outcome changes: a reduction in taxes means more disposable income, which increases the demand for consumption and a decrease in private savings. This decrease can be offset if the country can borrow foreign credit, i.e. a rise in the current account deficit (in order to maintain the same level of investment). Another way to maintain the level is via interest rates: the will rise in order to temper down the consumption, which is makes it more expensive for investors to continue borrowing domestic funds (this phenomenon is known as the “crowding-out of investments).
In order to have a better grasp at the relationship between budget deficit, current account and savings in the data, I have ran a regression in the following form:
Current Account=Intercept+β1 Budget deficit+ β2Private Savings
The R output follows:
lm(formula = ca$current.account ~ budget$balance + sav$Savings)
Residuals:
Min 1Q Median 3Q Max
-672.55085 1.37811 43.52668 95.47933 410.85259
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 33.36598614 23.11025596 1.44377 0.15023
budget$balance 0.08708663 0.15962867 0.54556 0.58593
sav$Savings -0.72159440 0.08403452 -8.58688 1.6777e-15 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Residual standard error: 185.2854 on 219 degrees of freedom
Multiple R-squared: 0.3885253, Adjusted R-squared: 0.382941
F-statistic: 69.57527 on 2 and 219 DF, p-value: < 2.2204e-16
Given the data sample, the coefficient on the budget deficit is not statistically significant, which would be a result against the twin deficit hypothesis. The negative and significant coefficient in front of the savings variable confirms one the mechanisms discussed above: if the savings decrease with 1 unit, the current account deficit will increase with 0.7 units (keeping everything else fixed). The R-squared is however 38%, which can be improved.
Another specification that could be the benchmark is simply estimating how much /if the budget deficit causes the current account deficit:
Current Account=Intercept+α1 Budget deficit
R Output:
lm(formula = ca$current.account ~ budget$balance)
Residuals:
Min 1Q Median 3Q Max
-782.07514 -35.25668 103.45489 132.89108 235.46174
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -125.1775165 16.0328952 -7.80754 2.3509e-13 ***
budget$balance 0.9797574 0.1397340 7.01159 2.8575e-11 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Residual standard error: 213.7307 on 220 degrees of freedom
Multiple R-squared: 0.1826495, Adjusted R-squared: 0.1789343
F-statistic: 49.16238 on 1 and 220 DF, p-value: 2.8575e-11
Although the coefficient is significant at 5% confidence level, its magnitude is however unlikely: it would mean that they almost move in a 1:1 ratio. Furthermore, the R-squared is very low, which is not enough for a meaningful interpretation.
Another variable of interest would be investments (private). To see how this affects the current account, I have considered the following specification:
Current Account=Intercept+γ1 Budget deficit+ γ2Investments
R Output:
lm(formula = ca$current.account ~ invest$Inv + budget$balance)
Residuals:
Min 1Q Median 3Q Max
-312.719545 -45.645353 3.470151 63.558556 220.918122
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 203.955534472 13.396328156 15.22473 < 2.22e-16 ***
invest$Inv -0.272136580 0.009309805 -29.23118 < 2.22e-16 ***
budget$balance 0.184488015 0.068861006 2.67914 0.0079416 **
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Residual standard error: 96.75758 on 219 degrees of freedom
Multiple R-squared: 0.83325, Adjusted R-squared: 0.8317272
F-statistic: 547.1717 on 2 and 219 DF, p-value: < 2.2204e-16
Conclusion
The data sample on which the analysis was performed does not present clear evidence for the twin deficits hypothesis. This fact is supported by the graphics as well as by the results of the regressions. In the specification in which investment is considered as an explanatory variable, there is a positive correlation of the budget deficit with the current account, deficit. Nonetheless, the magnitude is not high enough.
An interesting result however, is the correlogram that shows that there might be a significant correlation, but its effects are lagged with 4-5 quarters. Another econometrical approach (time-series analysis – VAR or SEM) would be more appropriate for this kind of data. Moreover, to obtain better results, the data sample should be as big as possible. In this direction, panel data analysis could provide other insights.
Overall, due to general equilibrium, all variables/indicators change at the same time, which makes it difficult to pin down which variable causes which. One way to get around this would be impulse response functions. For that kind of analysis, one needs a general equilibrium model setup.
Works Cited
Bartolini, L., & Lahiri, A. (2006). Twin Deficits, Twenty Years Later. Current Issues in Economics and Finance, Vol. 12, No. 7 , 1-7.
Bernheim, B. D. (1988). Budget Deficits and the Balance of Trade. In Tax Policy and the Economy: Volume 2 (pp. 1-32). MIT Press.
Bussière, M., Fratzscher, M., & Müller, G. J. (2010). Productivity shocks, budget deficits and the current account. Journal of International Money and Finance , 1562-1579.
Corsetti, G., & Müller, G. J. (2006). Twin Deficits: Squaring Theory, Evidence and Common Sense. Economic Policy . Vienna.
Coughlin, C. C., & Pollard, P. S. (2011, May). What Drives Large. Retrieved March 6, 2016, from International Economic Trends: https://research.stlouisfed.org/datatrends/pdfs/iet/20010501/cover.pdf
Federal Reserve Bank of San Francisco. (2005, July 22). Understanding the Twin Deficits: New Approaches, New Results. Retrieved March 5, 2016, from FRBSF Economic Letter: http://www.frbsf.org/economic-research/publications/economic-letter/2005/july/understanding-the-twin-deficits-new-approaches-new-results/
Federal Reserve Bank of St. Louis. (n.d.). FRED Economic Data. Retrieved March 6, 2016, from http://research.stlouisfed.org/fred2/
Kim, S., & Roubini, N. (2008). Twin deficit or twin divergence? Fiscal policy, current account, and real exchange rate in the U.S. Journal of International Economics , 362-383.
PennState Eberly College of Science. (n.d.). Cross Correlation Functions and Lagged Regressions. Retrieved March 8, 2016, from Applied Time Series Analysis: https://onlinecourses.science.psu.edu/stat510/node/74
