U.S. fiscal policy and trade deficits: a broad perspective.

• Author: Mohammadi, Hassan

1. Introduction

   The emergence of large government budget deficits and large international trade deficits stands out as the dominant feature of U.S. macroeconomic behavior since the early 1980s. Economists trained in the traditional Keynesian theory have pointed to the budget deficit as the major cause of the trade deficit. According to the Mundell-Fleming model, changes in the budget deficit trigger changes in the real interest rate, the real exchange rate, and the level of real income, which in turn alter the trade balance. Others, believing in the real-world relevance of the Ricardian Equivalence Proposition, have argued that the deficits are not really twins, but merely distant cousins, if not entirely unrelated. If Ricardian Equivalence holds, shifts between taxes and borrowing have no effect on real interest rates or aggregate demand.

   Previous studies of the "twin deficits" have adopted a variety of approaches, both theoretical and empirical. Theoretical treatments run the gamut from the standard Mundell-Fleming model [18] to the Ricardian model [7] to the atheoretical [1; 2; 3; 4; 12; 14; 17]. Empirical approaches range from simple single-equation OLS [5; 6; 16] to two-stage least squares applied to a system of structural equations [18] to unconstrained VAR modeling [1; 2; 12; 17] to cointegration [3; 14]. The variables used to measure the budget and trade deficits vary across studies, as does the set of related variables included in the models. The form in which the data are utilized (e.g., as a ratio of GNP or as a first difference of levels) also differs across studies. Not surprisingly, a variety of results emerges. Table I provides a brief overview of a number of studies.

   The estimated effects of budget deficits on the trade balance range from the substantial [12] to statistically insignificant [7]. Abell [1; 2] finds the transmission mechanism to be primarily through interest and exchange rates, while Zietz and Pemberton [18] find that channel to be insignificant in comparison to the domestic absorption channel. Yet some studies do not even include domestic income as a separate variable in their equations, making cross-study comparisons difficult. Thus, it is impossible to draw with confidence any conclusions about the twin deficits relationship. This confusion of results leads us to examine systematically the relationship between the two deficits using a multivariate model.

   In this paper we report the results of a systematic analysis of the link between the deficits using a five-variable system for the flexible exchange rate period, 1973.I to 1991.IV. We begin with [TABULAR DATA FOR TABLE I OMITTED] a baseline model that includes the real total government budget surplus and the real current account balance as our measures of the budget and trade balances, as well as money growth, real income, and the real exchange rate. The variables chosen correspond to those in the Mundell-Fleming model, with money growth substituted for the real interest rate. We use the total government budget surplus rather than a narrower measure because it better measures the effect of the government budget on national saving.

   We test for multivariate cointegration in the five-variable system, and, finding it, we estimate a vector error-correction (VEC) model which we use to produce variance decompositions and impulse response functions. Since the results of such VEC estimations typically are sensitive to lag lengths and the ordering of variables, the effects of altering lag lengths and variable orders are considered. We also examine the effects of using different data transformations and of estimating a standard VAR model, which, although inappropriate, is found frequently in the literature. Finally, we examine the effects of substituting a real interest rate for money growth in the five-variable system.

   Having thoroughly investigated the properties of the baseline model, we then examine the robustness of our results by estimating the system using alternative measures of budget and trade balances. To conserve space, we present only a summary of these results; complete results are available on request.

   To cut to the conclusion, our results indicate that the effect of the budget balance on the trade balance, if any, is modest. Results vary according to the form of the data, the number of lags in the VEC (or VAR) model, the definitions of the two balances, and the orderings of the variables. Generally the variations follow predictable patterns.

   In section II, we present our baseline model and examine its robustness to changes in lag length and variable orderings. In section III, we compare the results from the baseline model to the results from alternatively specified systems. We end the paper with a summary of our results and a discussion of our conclusions.

2. Empirical Results: The Baseline Model

   The major aspect of our empirical work is the detailed attention we give to the univariate and multivariate time-series properties of the variables and to the specification of our autoregressive models. Recent developments in this area suggest that (i) models in levels that ignore the nonstationarity of individual series can lead to spurious regression results, and (ii) models in first differences are misspecified if the series are cointegrated and converge to stationary long-term equilibrium relationships. Thus, we begin our empirical work by testing for nonstationarity in the individual data series using the augmented Dickey-Fuller procedure. We find that the variables are nonstationary in levels but stationary in first differences. Next, we investigate the multivariate properties of the variables using the maximum likelihood procedure of Johansen [10] and Johansen and Juselius [11]. We find evidence in favor of at least one cointegrating vector. Thus, we proceed with our dynamic analysis by estimating VEC models and computing variance decompositions. Finally, we present results from some unconstrained VAR estimations, in order to ascertain the difference between VEC and (the more-common) VAR estimates. All statistical operations are performed using the RATS econometric package.

   Data

   The five variables included in the baseline models are the real U.S. current account balance, the real total government budget surplus, the growth rate of the M2 money supply, the log of real GDP, and the real exchange rate. The choice of variables was influenced by Eisner [5; 6] and Summers [16], and by Friedman [8] and Roberds and Whiteman [15], who show that over the last two decades M2 provides a more useful measure of the effect of monetary policy than does M1. The real exchange rate is the trade-weighted average exchange value of the U.S. dollar against currencies of the industrial countries, adjusted by the ratio of export-to-import price indexes. The real budget and trade surpluses are obtained by dividing the corresponding nominal magnitudes by the GDP deflator. All data are quarterly for the period 1973.I to 1991.IV and are from the Citibase data tape.

   Tests for Stationarity

   We used the...