Why is the U.S. current account deficit so large? Evidence from vector autoregressions.

I. Introduction

Most observers consider large U.S. current account deficits to have been the main contributor to world macroeconomic imbalance in the 1980s. The implications of continuing U.S. current account deficits of 100 to 200 billion dollars per year are troubling. Other countries must adjust to the scarcity of capital as the United States acquires a significant portion of the world's savings. At the same time U.S. companies seek protection from "unfair" foreign competition as foreign companies increase market share in the United States. Determining the source of the problem, and designing the correct solution, are therefore major objectives of economic research.

Policymakers and economists have suggested a number of reasons for the sudden jump in the current account deficit. Different schools of thought have therefore proposed different policy responses, ranging from doing nothing to radically altering the country's commercial policy. Debate has centered on the proposed policies, with little attention paid to comparing the explanatory power of the underlying hypotheses.

II. Theories of the Current Balance

In the National Income accounts, the counterpart to the current account deficit is an imbalance of domestic savings and domestic investment. Two explanations of the current account deficit start from this uncontroversial observation. Martin Feldstein |11; 12~ popularized the conventional view held by many economists, sometimes referred to as the problem of the "twin deficits". (See also the 1984 Economic Report of the President |5, 54-7~.) In Feldstein's view, the large U.S. government deficits attracted foreign investment to the United States, thus causing the current account to move into deficit. This view's policy recommendations are clear. If Congress and the President do not reduce the budget deficit, the U.S. will continue to absorb a large portion of the rest of the world's savings.(1) Since the capital account determines the current account, policies that do not address capital movements will not alter the current account deficit.

Supporters of the Reagan Administration's macroeconomic policy, including William Poole |23~, oppose this view. They believe that the savings-investment imbalance that caused the current account deficit reflected a shift in the demand for investment goods in the U.S., which occurred because of the 1981 investment tax cuts. (See also the 1985 Economic Report of the President |6, 102-106~.) According to this view the U.S. current account deficit does not present a policy problem. As the country accumulates debt to finance the current deficit, it accumulates capital and therefore the additional productive capacity to pay off the debt. Efforts to shrink the current deficit risk destroying the beneficial investment boom, so the government should take no action to reduce the current account deficit.

Economists and politicians who are apprehensive about U.S. international competitiveness believe that the current account balance reflects falling productivity in the U.S. compared to its trading partners. (See Lovett |20~ for a defense of this view, and Carvounis |4~ for a detailed discussion of what he calls the "structural approach".) This view lacks a solid basis in macroeconomic theory, because it does not address the source of the savings-investment imbalance. Supporters of the productivity explanation nevertheless argue that the U.S. must adopt policies to support export industries and improve productivity. These range from reforming the U.S. educational system to adopting an industrial policy and protection of high-technology or infant industries. Many of these policies would require a radical change in how U.S. citizens view the relationship between government and industry, and proponents of this view indeed argue that such attitudes must change. Proponents of the competitiveness hypothesis argue, therefore, that reducing the current account deficit may require drastic changes in American society.

Two hypotheses of the current account deficit focus on foreign exchange markets. One is the "safe haven" hypothesis, which suggests that U.S. assets became relatively less risky than foreign assets in the early 1980s. The resulting demand for U.S. assets drove up their price, (the international value of the dollar) and the high dollar then caused the current deficit. The second hypothesis assumes that the dollar's rise was not caused by market fundamentals, and that the expensive dollar caused the current account deficit. The "risk premium" in foreign exchange markets is a measure of both phenomena. The risk premium on home relative to foreign assets is the difference between the return on home assets and the return on foreign assets (where r is the log return on home assets, |r.sup.*~ is the log return on foreign assets, e is the log exchange rate, and e is the log expected exchange rate.

r - |r.sup.*~ - e + e = |rho~ (1)

Assuming covered interest parity, (r - |r.sup.*~ - f + e = 0), where f is the forward exchange rate), the risk premium is equal to the difference between the forward exchange rate and the expected future exchange rate. If expectations are rational, (the expectational error |epsilon~ has zero mean), the difference between the forward rate and the actual future exchange rate is an unbiased estimate of the risk premium:

|f.sub.t~ - |e.sub.t+1~ = ||rho~.sub.t~ + ||epsilon~.sub.t+1~. (2)

||rho~.sub.t~ + ||epsilon~.sub.t+1~ is the forward bias which economists have attempted to explain for the last decade. (Hodrick |17~ and Boothe and Longworth |3~ have surveyed this extensive literature. Most researchers interpret the empirical results of forward bias tests as pointing to a time-varying, non-zero risk premium. The source of the risk, however, has not been established.

I constructed a measure of the risk premium using monthly forward data averaged over each quarter for the U.S. dollar exchange rate of four currencies: the Canadian dollar, the German mark, the Japanese yen, and the British pound. This average risk premium was -0.0014 during the 1970s and -0.0005 during the 1980s, although neither mean was significantly different from zero. The average risk premium rose over the period, while the "safe haven" hypothesis requires that it should have fallen. (Note that the term Pt is the riskiness of foreign assets relative to home assets. The safe haven hypothesis is therefore an unlikely explanation of the current account deficit.

The ||rho~.sub.t~ term may not be a risk premium. Frankel and Froot |13~ used survey data to replicate efficient market tests. They claimed that the "risk premium" reflects consistent expectational errors by traders and investors. The "risk premium" therefore represents the extent to which foreign exchange markets are not efficient or do not reflect fundamentals. Williamson, |25; 26~ among others, has blamed the expensive dollar and subsequent current account deficit on foreign exchange market conditions which do not reflect economic fundamentals. Williamson therefore argues that the world economy should adopt a more carefully managed float, including sterilized government intervention in foreign exchange markets to reduce current account imbalances.

The relationship between risk premiums and international trade has not been explored, despite the large literature on exchange market efficiency. I will use the "risk premium" to measure possible inefficiency in foreign exchange markets. If Williamson is correct, the "risk premium" representing market inefficiency might have caused the current account deficit. (The risk premium did not rise in the 1980s, so the safe haven hypothesis cannot be correct, and need not be formally tested.)

III. Previous Empirical Tests

Previous research has examined only the Feldstein's hypothesis that the current account deficit was caused by large Federal government deficits. Researchers have approached the problem of testing the hypothesis in several ways. Some economists compare the "twin deficits" relationship among several countries. Others estimate models for the United States over the postwar period to describe the relationship. One paper uses a variety of sophisticated techniques for examining the correlation between the two variables adjusted for the business cycle. All find some relationship between the current account and the government budget deficit.

Bernheim |2~, Laney |19~, and Kearney and Monadjemi |18~ used panel data to examine the relationship between government deficits and current account deficits in a number of countries over the postwar period. Bernheim found that about 30 per cent of the government deficit is translated into a budget deficit in the United States, Canada, the United Kingdom, West Germany, and Mexico. Laney estimated the relationship between the two variables for 59 countries. He found that the relationship was stronger in developing countries and in the smaller developed countries. Laney estimated that 12 per cent of the U.S. government deficit was translated into a reduced current account during the period 1948-82. Kearney and Monadjemi estimated VARs for Australia, the United Kingdom, Canada, France, Germany, Ireland, Italy, and the United States. They found a temporary change (from one year in Ireland to five years in Germany) from a debt-financed rise in government spending but also found feedback from the current account to government spending in most of the countries.

Darrat |7~ and Abell |1~ estimated vector autoregressions (VARs) for the United States alone. Darrat calculated Granger-causality tests for the twin deficits. He determined that feedback existed during the period of floating exchange rates between the government deficit and the trade balance. Other variables, including the money stock, inflation, and interest rates, do not appear to have Granger-caused the trade balance after 1973. Abell estimated a constrained system including the trade balance, the Federal government deficit, the interest rate, a trade-weighted exchange rate, U.S. income, and the money stock. He found that the Federal deficit influenced the trade balance indirectly through the interest rate and exchange rate, as theory predicts.

Zietz and Pemberton |27~ estimated a standard structural model of the U.S. macroeconomy. They use the model to conduct policy simulations. In their model, had the budget deficit been kept to 28 billion dollars during the 1980s, net exports would have been about 40 billion dollars higher in 1986. However, this is less than half of the trade deficit of the period. The authors conclude that microeconomic factors may be responsible for at least a significant part of the trade deficit.

Miller and Russek |21~ examined the correlation between the fiscal deficit and the trade deficit under a wide variety of specifications and decompositions of the data. They found significant correlations between the deficits decomposed into secular trend and cyclical components, but could not reject the null hypothesis that the two variables are not cointegrated.

None of the above authors has compared their results with tests of other hypotheses. Below I estimate parallel models for each of four hypotheses--the Feldstein hypothesis, the investment hypothesis, the productivity hypothesis, and the risk premium hypothesis.

IV. Empirical Tests

I use four variables to represent the causal agent for each of the four hypotheses. The variables are the Federal government surplus, gross domestic investment, U.S. relative to foreign productivity, and the estimated risk premium (||rho~.sub.t~ + ||epsilon~.sub.t~ + 1). Investment, the Federal budget surplus, and the current balance are measured as a percentage of GNP. (See the appendix for the precise calculations for each variable and data sources.) All estimates use quarterly data for the period 1974-1988 except for systems including the risk premium, which are estimated over the period 1974-1987. Comparing the behavior of these systems will show the relative ability of each of the hypothesized causes to explain the current account deficit of the 1980s.

Christopher Sim's |24~ VAR method is a convenient estimation technique, because it imposes little theoretical structure on the estimates. VARs, however, are efficient only in the absence of cointegration. First 1 show test the four variables for cointegration with the current account. Then I present comparative VAR estimates.

I estimate bivariate VARs because the interrelationship between explanatory variables is not the focus of this research. Instead, 1 want to show which variables can be empirically eliminated as possible explanations, and to determine which best explains what happened to the current account. The strategy also allows for a simple, explicit comparison of the empirical performance of the alternative hypotheses. The variables tested here cover several important explanations offered for the phenomenon; the results indicate which of the explanations fits the data well.(2)

Table I. Tests for Unit Roots
      Current   Budget                                  Risk
      Account   Surplus   Investment   Productivity   Premium
DF      -0.82    -0.76       -0.50         -4.2(**)    -6.0(**)
ADF     -1.06    -0.74       -0.43         -3.2(o)     -2.0
Notes:
**Reject null hypothesis of a unit root at 99% level.
oReject null hypothesis of a unit at 90% level.
DF - Dickey Fuller Test.
ADF - Augmented Dickey Fuller Test with four lags.
Table II. Tests of Cointegration
      Budget                                 Risk
      Surplus   Investment   Productivity   Premium
DW      0.32        0.06         0.12          0.24
DF     -1.52       -0.79        -1.52         -1.53
ADF    -1.59       -1.02        -1.78         -0.36
Notes:
DW: Durbin-Watson Statistic.
DF: t-statistic from Dickey-Fuller Regression.
ADF: t-statistic from augmented Dickey-Fuller Rgression with
four lags.

Tests for Cointegration

Two variables are said to be cointegrated if both are integrated of order d, but a linear combination of the variables is integrated of order less than d. Engle and Granger |10~ have pointed out that VAR models may be misspecified if the two variables are cointegrated. Engle and Granger show that, when X and Y are cointegrated, the correct model of the relationship between X and Y is an error correction model of the form:

|Mathematical Expression Omitted~

where ||epsilon~.sub.t~ is the error term from the regression

|Mathematical Expression Omitted~

A VAR in first differences would look the same as equation (3), but without the y||epsilon~.sub.t~ term. The estimated VAR system will fail to detect the effect of the cointegration (the e term), and therefore may miss the causal relationship between the variables. A VAR in levels will not be efficiently estimated because the cointegration constraint is not included in the estimation. Granger and Engle showed that the VAR estimates for levels are not asymptotically biased, however |10, 259~. The properties of the moving average representation in levels or differences are unknown in the presence of cointegration. In general, VAR models which do not take cointegration into account may fail to detect causality where it exists |15, 2041. Miller and Russek |22~ discuss a practical example of this problem.

The first step in cointegration analysis is to test the series for integration. If the series are not integrated (do not have unit roots) then they cannot be cointegrated. Table I shows two standard TABULAR DATA OMITTED tests for unit roots devised by Dickey and Fuller |8~. The null hypothesis of these tests is that there is at least one unit root (i.e., the series is non-stationary).

The Dickey-Fuller tests suggest that we can reject the null hypothesis of a unit root for the relative productivity and risk premium series. However the augmented tests suggest that there may be a unit root in the risk premium series, and (at a lower confidence level) in the productivity series. Cointegration may, therefore exist between any of the hypothesized causal variables and the current account.

Testing for cointegration is based on estimating the cointegrating regression, equation (4) above, as described by Granger and Newbold |16~. Table II shows three tests for cointegration of the four hypothesized causal variables with the current account(3). One test examines the Durbin-Watson statistic under the null hypothesis of serial correlation (where p = 1). The other two tests are the Dickey-Fuller test and the augmented Dickey-Fuller test for the residual series from the cointegrating regression. (Critical values are given in Engle and Granger). The null hypothesis of no cointegration cannot be rejected for any of the four variables, in any of the tests. The results confirm those of Miller and Russek |21~. The VAR estimates presented below are therefore efficient estimates of the relationship between these variables.

Bivariate VARs

The bivariate VARs provide the basic comparison between the four hypotheses. Each system includes two variables: the current account balance and one of the four proposed causal variables. I estimated the systems with quarterly data, and four or eight lags. The results presented below are those for the four lag systems. The results with eight lags are almost identical.

Granger-Sims Causality Tests. Table III shows the F-statistics for exclusion of the past lags of each variable in each equation of the four separate systems. The F-statistics can be interpreted as tests of Granger-Sims causality. Only the federal surplus Granger-causes the current account. None of the other variables passes this simple test for the period of floating exchange rates. Granger-causality tests are therefore consistent only with the hypothesis that the current account deficit has been caused by the Federal government deficit.

The current account does not Granger-cause any of the hypothesized causal variables, indicating that all four variables are exogenous to the current account. My results, unlike Darrat's and Kearney and Monadjemi 's, do not indicate the presence of any feedback from the current account (Darrat used the trade account) to the federal budget. The Feldstein hypothesis relies on savings-investment relationships, so the current account, not the trade deficit, should be used to test the hypothesis.

Impulse Response Functions(4). The Granger causality tests do not indicate the size of the impact of each hypothesized causal variable on the current account. Figure 1 shows the impulse response functions from a one standard deviation shock in each of the four hypothesized variables. The figure shows how much the current account would change one to 16 quarters after the shock. The impulse response functions show that the impact of the budget is reasonably large in the estimated system. A one standard deviation rise in the federal budget surplus (about 0.7 percent of GNP over this period) causes the current surplus to rise by almost 0.4 per cent of GNP after about 10 quarters. Investment, which has the next largest impact, appears to affect the current account perversely: a one standard deviation rise in investment would cause the current surplus to rise where theory implies a fall. The small effect of the risk premium indicates that it is unlikely to have caused the substantial change in the current account in the 1980s. A one standard deviation shock to relative productivity has only a small effect, and the effect is negative for only a few quarters. This fits well with economic theory, which suggests that any impact of a productivity shock exists only in the short run. In the long run, the exchange rate adjusts to offset the shock.

V. Conclusion

In this paper I have given four explanations of the U.S. current account deficit the opportunity, via bivariate VARs, to explain the current account. Despite the general nature of the tests--no economic model is imposed on the data, and the bivariate VARs eliminate competing variables--only the Federal budget deficit explains the evolution of the current account. Feldstein's twin deficits hypothesis proved to be superior by every measure of the performance of the VARs. The other three possible causal variables (investment, relative productivity, and the risk premium) cannot explain how the current account changed over time, even under the favorable circumstances of being estimated in a bivariate vector autoregression. Economists and other observers enjoy showing that conventional economic wisdom can be wrong. Conventional wisdom appears to be correct, however, in explaining of the current account deficit of the 1980s.

The United States must reduce the Federal budget deficit to eliminate its current account deficit. Policies aimed at competitiveness will have no impact, although such policies may be desirable to maintain favorable long-run terms of trade for the U.S. Policymakers should not expect sterilized intervention or reforms of the foreign exchange market to affect the U.S. current account. Doing nothing, as advocates of the investment hypothesis recommend, appears dangerous. Because the country is using the current account deficit to finance government spending (mostly consumption) instead of investment, the U.S. will face a lower standard of living in the future as it pays off the debt accumulated to pay for today's consumption. Adjustment to paying off the debt in the future is likely to be politically and economically costly. There is no real alternative to the major problem of world macroeconomic disequilibrium than closing the U.S. budget deficit.

Appendix: Description of the Data

Federal Deficit

The Federal government deficit (national accounts basis) as a percentage of GNP. Source: Department of Commerce.

Investment

Gross investment (including net foreign investment) as a percentage of GNP. Source: Department of Commerce.

Relative Productivity

An index of GNP per employed person in the U.S. divided by an index of GNP per employed person abroad. GNP per person abroad is the geometric average of indexes for Germany and Japan. Sources: Department of Commerce for U.S. data. OECD for German data and Japanese GNP. Bank of Japan for Japanese employment.

Risk Premium

Monthly spot and 30 day forward exchange rates for the Canadian dollar, German mark, Japanese yen and British pounds in U.S. dollars. I constructed the risk premium by taking the log difference between the previous month's forward rate and the current spot rate (|f.sub.t~ - |e.sub.t~ + 1), then calculating the simple average over all four currencies. Quarterly data are averages for the three months of the quarter. Source: International Monetary Market Yearbooks.

Current Account Balance

Current account balance (national accounts basis) as a percentage of GNP. Source: Department of Commerce.

Note: Some researchers (e.g., Eisner |9~) prefer to include the state and local government budget surplus in the government deficit. Gramlich |14~ points out that much of the state and local budget surplus is pension investment which is matched by equal future liabilities. More important, from the point of view of Federal policymakers, is that state and local surplus spending is autonomous; like private savings and investment, it cannot be directly determined by Federal policymakers. The Federal government controls only its own budget, and so the Federal budget deficit is the appropriate measure of Federal budget policy.

I used RATS (version 3.02 for the Macintosh) for all estimation presented in this paper.

1. International investors may change their portfolio preferences, or decide they have accumulated enough U.S. assets. The capital surplus and resulting current deficit will then disappear, but U.S. interest rates must then rise substantially so that the budget deficit crowds out domestic investment rather than exports.

2. Other explanations of the rise in the current account deficit exist; for example, some observers have pointed to relative income growth rates in the United States and in its trading partners.

3. Engle and Granger compare the power of seven such tests. They conclude that the tests I use here, particularly the augmented Dickey-Fuller test, are the most accurate and powerful.

4. Variance decompositions strongly confirm the pattern of causality described in the paper. They are available from the author upon request.

References

1. Abell, John, "Twin Deficits during the 1980's: An Empirical Investigation." Journal of Macroeconomics, Winter 1990, 81-96.

2. Bernheim, B. Douglas. "Budget Deficits and the Balance of Trade," in Tax Policy and the Economy, edited by L. Summers. Cambridge: National Bureau of Economic Research, 1988.

3. Boothe, Paul, and David Longworth, "Foreign Exchange Market Efficiency Tests: Implications of Recent Empirical Findings." Journal of International Money and Finance, June 1986. 135-52.

4. Carvounis, Chris. The United States Trade Deficit of the 1980s: Origins, Meanings. and Policy Responses. New York: Quorum Books, 1987.

5. Council of Economic Advisors. Economic Report of the President. 1984. Washington: U.S. Government Printing Office, 1984.

6. Council of Economic Advisors. Economic Report of the President, 1985. Washington: U.S. Government Printing Office, 1985.

7. Darrat, Ali, "Have Large Budget Deficits Caused Rising Trade Deficits?" Southern Economic Journal, April 1988, 879-87.

8. Dickey, David, and Wayne Fuller, "Distribution of the Estimators for Autoregressive Time Series with Unit Root." Journal of the American Statistical Association, June 1979, 427-31.

9. Eisner, Robert, "Budget Deficits: Rhetoric and Reality." Journal of Economic Perspectives, Spring 1989, 73-93.

10. Engle, Robert, and C. W. J. Granger, "Co-Integration and Error Correction: Representation, Estimation, and Testing." Econometrica, March 1987, 251-76.

11. Feldstein, Martin, "American Economic Policy and the World Economy." Foreign Affairs, Summer, 1985, 995-1008.

12. -----, "Correcting the Trade Deficit." Foreign Affairs, Spring 1987. 795-806.

13. Frankel, Jeffrey, and Kenneth Froot, "Forward Exchange Bias: Is It an Exchange Risk Premium?" Quarterly Journal of Economics, February 1989, 139-61.

14. Gramlich, Edward, "Budget Deficits and National Saving: Are Politicians Exogenous?" Journal of Economic Perspectives, Spring 1989, 23-35.

15. Granger, Clive, "Some Recent Developments in a Concept of Causality." Journal of Econometrics, September/October 1988, 199-211.

16. Granger, C. W. J., and Paul Newbold. Forecasting Economic Time Series. Orlando: Academic Press. 1986.

17. Hodrick, Robert. The Empirical Evidence on the Efficiency of Forward and Futures Foreign Exchange Markets. Chur: Harwood Academic Publishers, 1987.

18. Kearney, Colm, and Monadjemi, Mehdi, "Fiscal Policy and Current Account Performance: International Evidence on the Twin Deficits." Journal of Macroeconomics, Spring 1990, 197- 218.

19. Laney, Leroy, "The Strong Dollar, the Current Account, and Federal Deficits: Cause and Effect." Federal Reserve Bank of Dallas Economic Review. January 1984, 1-14.

20. Lovett, William, "Solving the U.S. Trade Deficit and Competitiveness Problem." Journal of Economic Issues, June 1988, 459-67.

21. Miller, Stephen, and Frank Russek, "Are the Twin Deficits Really Related?" Contemporary Policy Issues. October 1989, 91-115.

22. -----, "Co-Integration and Error Correction Models: The Temporal Causality between Government Taxes and Spending." Southern Economic Journal, July 1990, 221-29.

23. Poole, William. "Overview," in The U.S. Dollar--Recent Developments, Outlook, and Policy Options. Kansas City: Federal Reserve Bank of Kansas City, 1985, pp. 235-40.

24. Sims, Christopher, "Macroeconomics and Realty." Econometrica, January 1980, 1-48.

25. Williamson, John. The Exchange Rate System. Washington: Institute for International Economics. 1985.

26. -----, "The Case for Roughly Stabilizing the Dollar." American Economic Review, May 1989, 41-5.

27. Zietz, Joachim, and Donald Pemberton, "The U.S. Budget and Trade Deficits: A Simultaneous Equation Model." Southern Economic Journal, July 1990, 23-34.