Tariffs and the Macroeconomy
Some Empirical Evidence

This paper examines the macroeconomic impact of tariffs. Existing theoretical models do not provide clear-cut predictions concerning the co-movement between unilateral tariff changes and a set of macroeconomic variables consisting of the real exchange rate, the trade balance, and the level of output. Three different data sets are found to be consistent with the hypothesis that tariffs have no statistically significant impact on the trade balance, the real exchange rate, or the level of output.


This paper examines the macroeconomic impact of tariffs. Existing theoretical models do not provide clear-cut predictions concerning the co-movement between unilateral tariff changes and a set of macroeconomic variables consisting of the real exchange rate, the trade balance, and the level of output. Three different data sets are found to be consistent with the hypothesis that tariffs have no statistically significant impact on the trade balance, the real exchange rate, or the level of output.

I. Introduction

This paper is concerned with the macroeconomic impact of tariffs. We emphasize two points. First, a review of the theoretical literature indicates that the macroeconomic effects of tariffs are ambiguous; there is no presumption that an increase in the tariff rate is likely, e.g., to raise output or to improve the trade balance. Second, there is little evidence that fluctuations in tariff rates have actually had significant effects on such important macroeconomic variables as the trade balance, the exchange rate, and the level of economic activity.

Our analysis is motivated in part by the lack of empirical research on the macroeconomic impact of tariffs; we are aware of almost no econometric analysis which addresses the matter. 1/ However, the subject is also of current policy concern. For instance, Branson (1987) suggests an across-the-board tariff increase to reduce the size of current U.S. trade deficits; 2/ proposals of trade restrictions against countries whose bilateral trade deficits with the United States exceed some critical value have also been widely discussed (e.g., the 1988 Gephardt Amendment). The infamous Smoot-Hawley Tariff Act was passed in part because of congressional beliefs that tariffs would be expansionary. 3/

In the next section, we review some popular theoretical models which have been used to analyze the macroeconomic effects of tariffs. Section III describes the data and the methodology used in the empirical analysis, while the main empirical results are presented in Section IV. The final section concludes.

II. Theoretical Overview

1. A Keynesian model

The usual textbook analysis of the macroeconomic effects of a tariff considers a small country, with fixed prices, flexible output, and a fixed exchange rate. 4/ Output is demand determined and world prices are given; the model is completely static. The domestic country produces good Y at price P (supply is perfectly elastic), but consumes both the domestic good and a foreign good. The price of the foreign good in domestic currency is eP*, where e is the exchange rate (defined as units of domestic currency per unit of foreign currency) and P* is the price in foreign currency of a unit of the foreign good. Imports are a function of relative prices, q = eP*/P, and real income in terms of domestic goods, Y, while foreign imports (our exports) are a function of relative prices only, foreign output being assumed constant (or rather unaffected by shocks emanating in the domestic country). The trade balance equals the excess of the value of exports over the value of imports. Measured in terms of the domestic good, the balance of trade, BT, may therefore be written as:


where M* denotes foreign imports and M denotes domestic imports from the rest of the world. Output equals the sum of domestic expenditure E(Y), and net exports:


Consider the imposition of an ad valorem tariff on imports at rate τ. The domestic relative price of imports rises to q(1+τ) while the world relative price is unchanged. If the tariff revenues are not redistributed (so that the government runs a budget surplus), then (2) becomes:


Net exports depend on the tariff rate because imports are a function of the internal relative price, q(1+τ), while exports are a function of the world relative price, q. The last term in (2′) reflects the negative demand for domestic goods by the government (i.e., the budget surplus).

Differentiating equation (2′) around an initial equilibrium of free and balanced trade yields:


where: α is the absolute value of the price elasticity of import demand; m is the marginal propensity to import; and 0 < 1-Ey < 1 is the marginal propensity to save.

Output rises if import demand is sufficiently price elastic (i.e., α exceeds unity) but falls otherwise. The intuition is straight forward. Because the tariff revenue is not redistributed, the tariff is a combination of an expenditure-switching policy and an expenditure-reducing policy. The expenditure-switching part serves to raise demand, and hence supply, of the domestic good. However, because the government does not redistribute the tariff revenue, there is in addition an income effect which tends to reduce demand for domestic goods. Only if import demand is sufficiently price elastic will the substitution effect dominate and the overall effect of the tariff be expansionary.

Differentiating the expression for the trade balance and using (3) yields:


A tariff switches domestic expenditure from foreign to domestic goods; the trade balance necessarily improves as a result. The improvement in the trade balance exceeds, falls short of, or is equal to the budget surplus, according to whether the tariff induces a rise, fall, or no change in output.

Several aspects of the analysis are noteworthy. First, the revenue-redistribution scheme adopted by the government will affect the magnitude, and possibly even the sign of the comparative static results. If the tariff proceeds are redistributed to consumers, then the only effect of the tariff is a pure substitution effect in favor of the domestic good, so that the output effect of a tariff is necessarily expansionary. At the same time, the trade balance effect will be lower (but still positive).

Second, the assumption that the exchange rate is fixed is crucial for the comparative static results. Suppose that the exchange rate adjusts to maintain external balance, so that BT(.) = 0. Then it can be shown 1/ that a tariff which would improve the trade balance, were the terms of trade to stay constant, instead leads to a real appreciation (improvement in the terms of trade) to maintain external balance. In this case, (2) reveals that the tariff has no effect on the level of output.

Third, the analysis above assumed that the level of expenditure, E(.), is independent of relative prices. In general, however, there is no particular reason for this to be the case. There are two conflicting forces at work. A deterioration in the terms of trade serves to reduce expenditure on relatively more expensive foreign goods and hence raise spending on domestic goods via a substitution effect. In addition, however, the deterioration in the terms of trade lowers real income and this tends to reduce spending on domestic goods. The net effect is therefore ambiguous but is thought by some (e.g., Laursen and Metzler (1950) and Mundell (1961)) to be dominated by the substitution effect.

Suppose that the expenditure function in (2) is replaced by the more general function E(q, Y), in which expenditure (in terms of domestic goods) depends on both relative prices and income, with Eq > 0 (the Laursen-Metzler assumption). If the exchange rate adjusts to maintain external balance, a tariff leads to an improvement in the terms of trade, i.e., to a fall in q, the effect of which is to raise real income and hence saving out of any given level of nominal income. The lower demand for domestic goods causes a fall in domestic production. 1/ This result (originally proved by Mundell (1961)) stands in sharp contrast to the case in which relative prices are fixed so that higher tariffs have an expansionary effect on output.

Further, a tariff may have contractionary effects even in the absence of a Laursen-Metzler effect on saving. In the Mundell-Fleming model of flexible exchange rates and perfect capital mobility, the interest rate is exogenous; therefore the imposition of a tariff cannot have any lasting effect on the value of nominal income. Redistributed tariff revenues raise disposable income; therefore income from production must fall when tariffs rise. Moreover, since the marginal propensity to spend is less than unity by assumption, the tariff engenders a trade deficit and corresponding capital account surplus. The intuition is simply that the tariff leads to an incipient interest rate differential which causes a real appreciation; the latter crowds out net exports. Eichengreen (1981) and Krugman (1982) have shown that, while these results are suggestive of the long-run effects of permanent tariffs, temporary tariffs in general have ambiguous effects on macro-economic variables of interest, at least in the short run.

Fifth, the model of equations (1) and (2) can be expanded to include repercussion effects due to changes in foreign output. Suppose that domestic exports, M* (.), depend on the level of foreign output, Y*, and relative prices, q, and that the domestic country is no longer “small” in the world economy so that changes in the tariff have effects on the level of foreign output. It is easily shown that the (domestic) output effect of a tariff is smaller in this case than in the original case in which Y* is exogenous. The intuition is simply that a tariff that increases domestic net exports has a negative effect on foreign output. The fall in Y* reduces demand for domestic exports so that the repercussion effect on Y is negative.

Finally, the analysis thus far has assumed that foreigners do not respond to the tariff. If the foreign country retaliates by raising its own tariff, the effects on output (both foreign and domestic), the trade balance, and the real exchange rate are in general ambiguous. 2/

To sum up, tariffs do not have clear-cut effects on the macro-economy in Keynesian models. Comparative static results depend critically on assumptions concerning inter alia fiscal policies (especially whether tariff revenue is redistributed); the exchange rate regime; the Laursen-Metzler effect; the degree of capital mobility; and repercussion and retaliation effects.

2. A monetary model

An alternative approach to the analysis of the macroeconomic effects of tariffs is provided by the monetary approach to the balance of payments; see Mussa (1974).

The monetary approach assumes that the economy’s long-run real equilibrium is given by the standard two-sector (Heckscher-Ohlin) model of international trade. In such a model, a tariff raises the internal relative price of import-competing goods and therefore leads to an increase in production and a fall in consumption of importables. Thus a tariff reduces the volume of imports. The tariff also increases consumption and reduces production of exportables, so that the volume of exports also declines. In fact, the main effect of a tariff in such a model is to reduce the volume of trade, leaving the trade balance unchanged. The interesting issue is therefore distributional; i.e., how does the tariff affect income distribution between the factors of production? 1/

Although the long-run equilibrium is characterized by the condition of balanced trade, the adjustment path toward this long-run equilibrium is characterized by changes in the economy’s net borrowing position. Mussa (1974) shows that the dynamic trade balance effects of the tariff are ambiguous. Consider the case in which the economy is completely specialized in the production of the export good. Suppose further that the government raises a (pre-existing) tariff on imports and redistributes the proceeds to consumers in a nondistortionary fashion. Finally, suppose that the demand for nominal money balances is proportional to the value of income in terms of the domestic good and that the monetary authorities hold the domestic component of the money supply fixed.

As long as the domestic economy is a price taker in world markets, the income of consumers (measured in terms of the domestic good) varies only according to the redistribution of the tariff proceeds. Income will rise, fall, or remain constant depending on whether the original tariff rate falls short of, exceeds, or is equal to that tariff rate which maximizes revenue (i.e., the product of the tariff rate and the initial value of imports). Consequently, the excess flow demand of money (i.e., the trade surplus) will increase if the initial tariff rate is initially below the revenue-maximizing rate. Succinctly, both (disposable) income and the trade balance may either rise or fall when the tariff is raised.

3. An optimizing model

A third approach to the analysis of the open economy has recently been developed. In this work, agents’ spending and saving decisions are viewed as solutions to the problem of maximizing an intertemporal utility function subject to lifetime solvency constraints. Razin and Svensson (1983) consider a model of a small open economy which produces and consumes two goods in each period, and faces fixed world prices and interest rates. Firms maximize profits subject to the economy’s endowment of productive factors and given technology. Consumers maximize lifetime utility subject to the constraint that the present value of their spending not exceed the present value of their income. From the point of view of the economy, this constraint is equivalent to the condition that the present value of the sum of the economy’s current and future trade surpluses equal the economy’s historically given external debt commitment.

Razin and Svensson emphasize that the effects of tariffs on saving, and therefore (ceteris paribus) the trade balance, depend crucially on the timing and expected duration of changes in the tariff rate. A temporary tariff will have very different effects on the trade balance from those of a tariff which is expected to be permanent. A temporary tariff raises the price of current consumption relative to future consumption. Agents will substitute consumption intertemporally (consuming less today and more in the future) by lending in the international capital market, i.e., by running a trade balance surplus. Thus, tariffs which are viewed as temporary lead to a trade surplus.

In contrast, a tariff which is expected to be in place permanently will not induce such an intertemporal substitution effect. In fact, if the initial equilibrium is stationary (in the sense that expenditure shares are constant through time), a permanent tariff will leave intertemporal consumption decisions, and hence the trade balance, completely unaffected.

The Razin-Svensson model has since been extended in a number of directions. Edwards (1989) and Ostry (1988) consider the issue of how tariffs affect the real exchange rate and, in turn, the trade balance. The dynamic behavior of the real exchange rate after a change in the tariff depends upon the ease with which agents substitute consumption within a period (i.e., the elasticity of substitution between tradables and nontradables in a given period) versus the degree of substitution in aggregate consumption across time periods (the intertemporal elasticity of substitution in consumption or the reciprocal of the coefficient of relative risk aversion). Depending on the parameter values, a temporary tariff may improve, worsen, or leave the trade balance unchanged.

Van Wijnbergen (1987) has extended the Razin-Svensson framework to include contract-based real wage rigidities. If nominal wages are indexed to consumer prices (which depend on the prices of both domestic and foreign goods), then a small country which raises a tariff temporarily will experience an increase in the internal price of the foreign good, although the domestic price of the domestic good will not be changed. The tariff therefore leads to a higher real product wage; if employment is demand determined, unemployment will consequently rise. This (transient) unemployment reduces real income temporarily; agents smooth their consumption by borrowing in the international capital market, i.e., by running a trade deficit. Thus, with real wage rigidities, a temporary tariff may worsen the trade balance, in contrast to the Razin-Svensson result. A similar finding is obtained by Ostry (1988) without the assumption of real wage rigidities by considering the case in which the government raises an initially positive tariff.

To summarize, a careful reading of the theoretical literature does not lead to a clear conclusion about the effect of a tariff on, e.g., the trade balance or the level of output. The effect depends on the timing and expected duration of the tariff shock, on the behavior of real wages and exchange rates, and on the values of a variety of elasticities, as well as on institutional factors such as the degree of capital mobility and the exchange rate regime. Any presumption that tariffs tend, e.g., to improve the trade balance and raise output, must therefore be based on empirical rather than theoretical knowledge. With this in mind, we turn to the data.

III. Empirical Framework

1. Data

The tariff data which we use are available through the Census Bureau, and are seasonally adjusted. Data are available on the values of: duty-free imports, dutiable imports, and calculated duties. These variables are available on a monthly basis from 1967 through 1988 (when the data were discontinued). The data exist on a bilateral basis, so that the value of, e.g., dutiable U.S. imports from Japan is known, as are the relevant duties. We have collected data for bilateral trade between the United States and the six other countries in the “Group of Seven” (G-7): the United Kingdom, Canada, France, the Federal Republic of Germany, Italy, and Japan. 1/

Two measures of the tariff rate are immediately available from our data: tariff revenues divided by the value of dutiable imports (denoted τ1 below); and tariff revenues divided by the sum of both dutiable and nondutiable imports (denoted τ2). We focus on the latter measure below, but use both measures in our empirical work.

Both measures of the tariff rate show a persistent downward trend throughout the period (a result of the Kennedy and Tokyo GATT rounds). A typical tariff rate (τ2) for Japanese imports) is depicted in Figure 1. 1/ The tariff rates vary considerably from country to country as well as over time. For instance, in November 1988, the second measure of the tariff rate (τ2) ranged from 1 percent for Canadian imports to 5.7 percent for Italian imports; in January 1967, the comparable rates were 2.3 percent and 15.3 percent. 2/ The growth rates of the tariff rates have positive but low correlations across countries.

Figure 1:
Figure 1:

Japanese Tariff Rate

Citation: IMF Working Papers 1989, 070; 10.5089/9781451960266.001.A001

Both measures of the tariff rate are clearly imperfect proxies for the ideal variable, which is the effective marginal tax on imports. Two considerations of particular interest are: (a) substitution bias; and (b) bias from the effect of nontariff barriers (NTBs).

Substitution bias arises from the fact that a tax on a given product leads both foreign production and domestic consumption of that good to fall; goods with high tariffs tend not to be imported. 3/ Succinctly, the (observed) average tariff rate is not the marginal rate. Our tariff rate measures will therefore tend, ceteris paribus, to understate the actual tariff rate; consequently, the estimated tariff rate elasticity will tend to be biased upward, and the null hypothesis of no effect of the tariff rate will tend to be rejected too frequently. 4/

The second consideration is more subtle: the tariff rate may be a poor proxy for the actual level of protection. Protection of imported intermediates implies that the effective rate of protection may differ considerably from our tariff rate. More importantly, for a variety of (mainly political) reasons, NTBs have recently been used for protection instead of tariffs. Particularly over the last part of our sample, our proxies for the tariff rate will therefore understate the actual rate of protection. Because fluctuations in the tariff rate may be (negatively) correlated with the unobservable “NTB rate,” the former may not be statistically exogenous, at least during the latter portion of our sample. We attempt to address this difficulty by examining a variety of countries facing differing degrees of NTB protection, and also by examining the stability of our results over time.

Another potentially serious econometric problem results from the fact that changes in the U.S. tariff rate may be highly correlated with foreign tariff rates (data which, to the best of our knowledge, are not readily available). This could result either from foreign retaliation, or from coordinated rounds of GATT tariff reduction. We attempt to handle this issue by comparing our results with those obtained from the pre-GATT period.

Despite the imperfect nature of our data, we note in passing that the latent variable in question (“the tariff rate”) is more likely to be econometrically exogenous than most variables traditionally used by macroeconomists. Since tariff rates tend to be set either in GATT rounds or on the basis of microeconomic issues, the rates are not likely to be influenced by cyclical considerations. Broadly speaking, tariffs can be characterized as nondiscretionary fiscal policy. 1/

The other variables needed for estimation are more conventional and are fully described in the data appendix. They include measures of: foreign and domestic industrial production; the real bilateral exchange rate (CPIs are used as the price deflators); and the real trade balance. Trade flows are measured both at U.S. and at foreign borders; consequently, two measures of the trade balance are available (denoted by the superscripts “D” and “F,” respectively).

2. Methodology

We use a nonstructural methodology to examine the impact of fluctuations of tariff rates on four aggregate variables which theory indicates to be of particular interest. In particular, we estimate vector autoregressions (VARs), which model domestic and foreign output, the real balance of trade, and the real exchange rate as functions of lags of the (four) endogenous variables. We then test the hypothesis that (lags of) the tariff rate do not add statistically significant explanatory power to this system. Our data do not reject this hypothesis at traditional significance levels.

A VAR is a convenient representation which incorporates dynamic feedback from each variable included in the vector to every other variable. It is similar to a reduced-form model in that the regressors can be treated as exogenous; however, a VAR is based on fewer assumptions than a traditional structural model in that all variables are treated symmetrically. Since we are interested in the impact of tariffs on four macroeconomic variables (the trade balance, exchange rate and domestic and foreign output), it is appropriate to use a VAR to link these variables to the tariff rate.

The fact that a VAR does not deliver structural parameter estimates is both an advantage and a disadvantage. On the one hand, using an atheoretical statistical methodology implies that our inferences are not based on possibly fallacious auxiliary assumptions about the relevant data generation process. That is, our inferences are not restricted by the validity of secondary hypotheses concerning, e.g., the appropriate structural models of trade flows, the exchange rate, or output. On the other hand, our tests may be sensitive to the information set considered. 1/ We attempt to account for this problem by checking the robustness of our results extensively. Further, some loss of power is inherent in the estimation of a nonparsimonious VAR system. Bayesian priors are used in some of the estimates in an attempt to improve statistical efficiency.

It is important to state from the outset that our goal is not to test a specific structural model of the transmission mechanism between tariffs and the macroeconomy. Rather, we attempt to discover the existence of a stable relationship consistent with any of the theoretical models discussed in Section II.

3. Preliminary diagnostics

We begin our empirical analysis by examining the time series properties of our data. In particular, we test for unit roots in the univariate representations of the variables and a cointegrating relationship between the four key macroeconomic variables and the tariff rate. For a recent survey on unit roots and cointegration, see Stock and Watson (1988).

Many macroeconomic variables tend to drift over time in a manner that loosely resembles a random walk (technically, such variables are said to have a unit-root). For instance, consumption (like the stock market) can be predicted “well” by its current level; the growth rate of consumption (like the return on stocks) is random. Equivalently, innovations to the level of consumption do not tend to be reversed over time, i.e., shocks are permanent. If a number of such variables (e.g., consumption and income) tend to drift in a similar fashion so that they do not diverge “too much” over time, the series are said to be cointegrated; they share a common trend. If two (or more) variables each have a unit-root, but are not cointegrated, there is no steady state relationship linking the variables.

Testing for the presence of cointegration between the tariff and our remaining set of macroeconomic variables is important for two reasons. From an economic standpoint, cointegration is a prerequisite for the existence of a stable, long run relationship among our variables. Second, the presence (or not) of cointegration has important statistical implications for the form of the VARs; the absence of cointegration implies that a VAR can be consistently estimated if each of the variables is first-differenced.

The null hypothesis that each of the variables (the real trade balance, domestic and foreign real output, the real exchange rate, and the tariff rate) has a unit root cannot generally be rejected at conventional significance levels with standard tools. The Dickey-Fuller tests displayed in Table 1 have a constant and are augmented by four lags of the difference; the sample period is 1967:6 through 1988:10, for a sample size of 257 observations. A departure from the null hypothesis of a unit root at the 0.05 (0.01) level is indicated by an (two) asterisk(s). All variables except the trade balance are initially transformed by natural logarithms.

Table 1.

Augmented Dickey-Fuller Unit Root Tests

article image

The unit root tests are consistent with the presence of unit root nonstationarity in most of the variables. The exceptions are both U.S. and foreign measures of the bilateral U.S.-U.K. real trade balance. 1/ In addition, the German industrial production index does not appear to have a unit root (although this result is extremely sensitive to the sample size). The results for the tariff rates are particularly important, as they are inconsistent with the hypothesis that most tariff shocks are expected to be temporary in nature (a matter to which we shall return below). This accords both with common sense and with the results of Gardner and Kimbrough (1988). In the context of the intertemporal approach to the trade balance, the apparently permanent nature of tariff shocks also leads to the presumption that tariffs will have only minor effects (see Section II.3).

The primary questions of interest in this paper are multivariate rather than univariate in nature; we are interested in examining the impact of the tariff rate on a variety of variables, including domestic and foreign output, the real exchange rate, and the balance of trade. Given the presence of unit root nonstationarity in our variables taken one at a time, the existence of stable relationships between our variables depends on whether they are jointly cointegrated. A vector of variables is (most commonly) said to be cointegrated if each element of the vector has a single unit root, but some linear combination of the variables does not.

We tested for cointegration using the augmented Dickey-Fuller statistics recommended by Engle and Yoo (1987). These tests are augmented Dickey-Fuller tests for a unit root in the residual of the “cointegrating equation.” The latter is merely an OLS regression of one variable (e.g., the tariff rate) on the other variables (e.g., the real exchange rate, domestic and foreign output, and the real trade balance). The tests are augmented by four lags of the differenced residual, and contain a constant. The sample is again the 257 observations from 1967:6 through 1988:10. The tests are reported in Table 2 for both measures of the trade balance; the regressands are the tariff rates.

Table 2.

Augmented Dickey-Fuller Cointegration Tests

article image

The cointegration tests show relatively little evidence of cointegration between the relevant variables. 1/ The very weak signs of cointegration have also been confirmed in two additional ways: by cointegration tests of the “reverse” regressions; 2/ and by the tests for cointegration proposed by Johansen (1988). 3/ However, no method delivers strong evidence of cointegration. This finding is consistent with the hypothesis that there is no steady-state relationship between the tariff rate and the four macroeconomic variables of interest.

IV. Results

Given the relatively strong evidence of unit roots in our variables, and the weak evidence of cointegration, we estimate our VARs in differences, but include the residual from a cointegrating equation as a regressor to ensure consistent estimation. 4/ To check the robustness of our findings, we also estimate our VARs in levels.

Most of our empirical results arise from a VAR of the form:


where: L is the lag operator, Lzt=zt=1; Δ denotes the difference operator (1-L); Xt ′=(BTi, q, y, y*); BTi is one of the two measures of the real bilateral trade balance; q is the logarithm of the real bilateral exchange rate; y is the logarithm of the domestic (U.S.) industrial production index; τ* is the logarithm of the foreign industrial production index; τj is the logarithm of the jth measure of the bilateral tariff rate; u is the residual from a (cointegrating) regression of BTi on a constant, q, y, y*, and τj; and ε is a random disturbance term which represents omitted factors, assumed to be iid.

The hypothesis of interest to us is that the tariff rate has no effect on any of the four macroeconomic variables of interest, i.e., Ф(L) = 0.

We include 24 (monthly) lags in our basic VAR results. While standard likelihood ratio tests indicate that the systems could be reduced to 12th-order systems, we are wary of an excessive reduction in the lag length, given the existing literature which stresses the long lag length of price variables in international trade flow equations, e.g., Goldstein and Khan (1985). Allowing for two years’ worth of lags sets our sample period of estimation at 1969:2 through 1988:9. There are no indications that more than two years’ worth of lags is warranted.

The relevant test statistics appear in Table 3. Under the null hypothesis that the tariff rate does not affect any of the variables of interest, the test statistics are distributed as chi-squares, with 96 degrees of freedom.

Table 3.

Chi-Square Tests of the Joint Impact of Tariff Rates

article image

The data are manifestly unable to reject the hypothesis that the tariff rate has not had a stable and statistically discernible joint effect on the four variables of primary interest.

While the evidence presented in Table 3 presents the impact of the tariff rate on the system of four endogenous variables, it is of interest to examine the impact of the tariff rate on each of the four variables of interest. Indeed, the negative results of Table 3 could conceal a strong and consistent impact of the tariff rate on (say) one of the variables of interest, or a variety of important but transient effects.

Table 4 presents some evidence relevant to this issue. F-tests are tabulated for the hypothesis that 24 lags of the (difference of the) bilateral tariff rate have no joint impact on the macroeconomic variable in question, once the effects of (24) lags of all four macroeconomic variables have been partialled out (these F-tests are the single equation analogues to the system wide chi-square tests of Table 3). Also tabulated are the point estimates of the cumulative impact of the 24 lags of the tariff rate. Test statistics which are significantly different from zero at the 0.05 significance level (thereby rejecting the null hypothesis of no effect of the tariff rate) are denoted by an asterisk. None of the test statistics is significant at the 0.01 level. The tests were computed using the U.S. measure of the trade balance and the τ2 measure of the tariff rate.

Table 4.

Hypothesis Tests of Φ = 0 by Variable

article image
* denotes significance at the 0.05 level

There is little evidence that (lags of) the tariff rate have a consistent but transitory impact on the macroeconomic variables, or that the tariff rate tends to have a noticeably stronger impact on some of the macroeconomic variables than on others. The F-tests do not typically reject the hypothesis that 24 monthly lags of the tariff rate jointly have no significant impact on each of the four variables of interest. The cumulative impact of the tariff lags varies considerably by country; that is, the tariff rate does not appear to affect any of the variables in a uniform way across countries. Increases in the tariff rate are associated with worsening of the bilateral trade balances vis-à-vis France, Italy, and Japan, but with improvements for the other countries; increases are also associated with appreciations for France, Germany, and Japan. Virtually none of the cumulative point estimates are statistically distinguishable from zero. The economic size of the tariff rate coefficients is generally reasonable, but varies widely by both variable and country. The (cumulative) exchange rate elasticities are typically less than unity, while the output elasticities tend to be much smaller.

1. Robustness

We have extensively checked the sensitivity of the results of Tables 3 and 4 to a variety of economic and statistical perturbations of our basic methodology. In this section of the paper, we describe briefly these checks for robustness. However, there is no evidence that our results are affected by changes in the way in which our system is estimated.

We estimated our VAR systems with both more (36) and less (12) lags to test the sensitivity of the results. However, the results do not seem to be affected by the lag length. 1/

We have argued that the tariff rate can be treated as an exogenous variable. If this is appropriate, the reduced form of the structural system which leads to (5) would include the contemporaneous values of the tariff rate, as well as lagged values. However, when the contemporaneous values of the tariff rate are added to our system, the results are not changed.

Theory suggests that the effects of tariffs may depend on the exchange rate regime. We have changed our period of estimation in a variety of ways to ensure that our results do not depend on the fact that our sample spans fixed and flexible exchange rate regimes. Our results are quite insensitive to the exact choice of a sample period (e.g., test statistics are similar when only the post-Bretton Woods regime of floating rates is used in estimation).

Estimating our system in (logarithms of) levels instead of differences does not lead to different results.

Using the PPI instead of the CPI does not change any results. Aslo, using employment in place of industrial production does not affect the results. 1/

As stated in the previous section, the theoretical effects of a tariff depend on whether the tariff is expected to be temporary or permanent. We used the Beveridge and Nelson (1981) technique to decompose the tariff into temporary and stationary components. This technique makes an important and potentially restrictive identifying assumption, namely, that the two unobserved components are perfectly correlated; for a survey of related issues, see Stock and Watson (1988). 2/ Negative results are again obtained when we estimate the system in levels and replace the actual tariff rate with either the temporary or permanent component of the tariff rate.

Our results do not change when the effects of the real U.S. federal budget deficit are taken into account. 3/ Thus, controlling for possible effects of the revenue redistribution scheme does not seem to make a difference.

Finally, we have used a variety of Bayesian priors in an attempt to improve the precision of the fit of our system by using non-sample information. Our priors take the form of maintaining that the VAR system can be well approximated as a first-order univariate autoregression. That is, we use non-sample information to specify zero coefficients and moderate standard deviations for the coefficients of the lagged non-dependent variables. However, our use of Bayesian techniques does not affect the finding that the tariff rate has no discernible impact on the variables of interest. This negative result does not depend on the exact number of lags used in the system.

2. Aggregate results

Our negative results are characteristic not only of bilateral trade between the United States and other G-7 countries; they also hold true for aggregate trade flows between the United States and its trading partners collectively.

We have estimated our system with aggregate variables, using the U.S. real net trade balance; an effective exchange rate in place of the bilateral rate; and global indices of the CPI, industrial production, and the tariff rates, in place of their bilateral counterparts. The methodology used on the bilateral data yields similar results when applied to aggregate data. In particular, two years’ worth of lags of the aggregate tariff rate have no statistically discernible impact on the four variables of interest. This result is true of both measures of the tariff rate, and is insensitive with respect to the lag length, estimation in levels or differences, and the exact sample period chosen.

3. Annual U.S.-U.K. evidence

We now confirm the relevance of our monthly results by using a long span of historical evidence on U.S.-U.K. trade flows. We test and cannot reject the hypothesis that our system is not affected by the tariff rate when annual data from 1889 to 1970 are used in place of the monthly postwar data.

There are a number of advantages to using long historical series. NTBs are likely to be less of a problem during the pre-GATT era. Further, the stochastic nature of foreign tariff rates may have changed dramatically as a result of GATT, which coordinates tariff reductions internationally. If tariffs have the same observable effects on macro-economic variables before and after GATT, it is less likely that GATT is responsible for our negative results. Finally, many economic time series appear to be more variable in the prewar period (although this has been the source of much recent debate). Tariff rates were certainly both high and volatile during the period in question. Figure 2 provides a plot of both measures of the tariff rate against time.

Figure 2:
Figure 2:

Historical Tariff Rates

Citation: IMF Working Papers 1989, 070; 10.5089/9781451960266.001.A001

Data on bilateral trade flows between the United States and a variety of its trading partners are available on an historical basis from the early 19th century to the present. Measures of real output and prices for both the United Kingdom and the United States are available from the late 19th century to the present, as is the bilateral exchange rate (consistent data for other countries does not appear to be available). However, to the best of our knowledge, no data on bilateral tariff rates is available until 1967. We therefore use measures of the aggregate U.S. tariff rate in place of the unknown U.S.-U.K. bilateral tariff rate. Clearly, these measures will be poor proxies for the relevant bilateral tariff rate if U.S.-U.K. trade differed significantly in composition from aggregate U.S. trade 1/ or if the U.S. tariff structure discriminated against the United Kingdom. During the period in question, U.S. imports from the United Kingdom averaged 10 percent of total U.S. imports (the maximum was 24 percent), while 23 percent of all U.S. exports were sold in the United Kingdom (at the beginning of the sample over half of all U.S. exports went to the United Kingdom).

We employ the same techniques as were used on our monthly data. In particular, we estimate VARs, treating real bilateral trade balances, U.S. and U.K. real output, and the bilateral real exchange rate as the variables of interest. The VARs are estimated from 1882 through 1970, with two (annual) lags, a constant term, and the residual from a cointegrating equation. We also check the robustness of our results by: estimating the equations in levels; increasing the lag length to three years; and changing the sample size. As in Table 3, we test the hypothesis that the tariff rate has no significant impact on the system; under the null hypothesis, the test statistics are distributed as chi-squares, with tabulated degrees of freedom. The actual test statistics (together with their marginal significance levels in parentheses) are reported in Table 5.

Table 5.

Historical Chi-Square Tests of the Impact of Tariff Rates

article image

In no case is the null hypothesis of no impact of the tariff rate on our system rejected at the 0.05 significance level. Further, the impact of the tariff rate on each of the dependent variables appears to be small. Finally, the result is robust with respect to changes in the number of lags, the levels/differences transformation, and, most importantly, to changes in the sample size. Our sample period covers a number of different exchange rate and fiscal regimes, as well as differing degrees of capital mobility. As summarized above, theory suggests that results may be sensitive to such factors. However, our results are in fact insensitive to the choice of sample period. For instance, when the postwar (GATT) period is excluded, the tariff rate still does not have any discernible impact. The latter evidence seems inconsistent with the view that the coordinated tariff reductions fostered by GATT are wholly responsible for our negative monthly findings.

4. Evidence from a panel of data

As a final check, we now test our results on a third data set. This data set is a panel of annual data spanning 38 countries from 1978 through 1985. Again we find that our null hypothesis of no significant impact of the tariff rate on our system of macroeconomic variables cannot be rejected.

The data is taken from Government Finance Statistics (GFS) and International Financial Statistics (IFS), both IMF publications. We collect data on both “import duties” 1/ and the more narrowly defined “customs duties” 2/ from GFS; IFS data is used for: imports, exports, nominal and real output, and the real effective exchange rate. Our choice of countries 3/ and sample period was dictated by data availability. Our sample includes a number of developing countries that use tariff revenues as a major source of government revenue.

The equations that we estimate take the form:


where x is now a vector of three variables (the real exchange rate, the real trade balance, and real domestic output), and all other notation remain unchanged.

We note that (5’) is expressed in growth rates, so that country-specific “fixed-effect” intercepts are consistent with the setup. We are forced to drop the cointegrating residual because of the time series size of the sample, although results are not changed if (5’) is estimated in levels.

We estimate (5’) and use a likelihood ratio statistic to test the hypothesis Ф(L)=0. When two lags of the regressors are included, the test statistics indicate that the hypothesis cannot be rejected even at the 50 percent significance level. The same result is true of each of the variables taken one by one; tariffs do not have noticeable effects on any of the variables of interest. These results are robust with respect to: choice of tariff rate measure; choice of lag length and sample period; addition of time-specific dummies; addition of a measure of global output; and estimation in levels. In other words, there is no evidence that tariffs have a discernible impact on real exchange rates, output, and trade balances, even after pooling data across both countries and time.

V. Summary and Conclusion

Economic theory does not deliver strong implications about the macroeconomic effects of a tariff. Depending upon the nature of the economy, a given change in the tariff rate is consistent with a wide range of fluctuations in the trade balance, output, and the exchange rate, including no effect at all.

In actual fact, we are unable to isolate statistically significant effects of the tariff rate on bilateral trade flows, real exchange rates, or output, either domestically or abroad. These results appear to be quite robust statistically.

However, we certainly do not wish to claim that tariffs do not have potentially important effects on economic welfare. Trade theory predicts that the primary impact of a tariff is to shrink the volume of international trade, rather than, e.g., the balance, of trade. While we are not concerned in this paper with modeling the volume of international trade, Rose (1989) has found that tariffs are in fact strongly negatively correlated with the volume of international trade, using a panel of 12 countries over 35 years.

Our inability to isolate significant macroeconomic effects of a tariff is consistent with a number of hypotheses. For instance, tariffs on imported intermediates or NTBs may render the effective rate of protection small and stable, despite fluctuations in measured tariff rates. Alternatively, the behavior of foreign tariff rates may be responsible for our results. “Pricing to market,” which results in low pass-through of exchange rate changes to import prices may also account for negligible effects of tariffs, if firms treat exchange rate and tariff changes symmetrically.

However, given that three very different data sets yield similar results, we are most attracted to the conclusion that the important effects of the tariff are not macroeconomic, but instead are distributional in nature (as documented by, e.g., Hufbauer et al. (1986). We conclude that there is little evidence that tariffs have important effects on the macroeconomic variables considered in this paper, and believe that further research on tariffs should focus on microeconomic aspects.

Appendix: Description of the Data

The data and programs used in this project are available upon request. The data has been carefully checked through a variety of diagnostic procedures, including descriptive statistics on the levels and differences of the data, and plotting the levels and differences. Numerous errors in the IFS data have been corrected.

Most of the relevant monthly data was collected from the IMF’s International Financial Statistics data base. The bilateral variables (IFS mnemonics) are as follows: period average bilateral (US$ PFX) exchange rates (rf and rh); industrial production index (66..c); CPI (64); PPI (63); and employment (67, 67..c, 67ey and 67eyc). The additional aggregate variables are: the U.S. MERM effective exchange rate (amx); the U.S. net merchandise trade balance (70-71); the global CPI (001..64x); and the industrial country industrial production index (110..66).

The monthly nominal bilateral trade data (all measured in U.S. dollars) is taken from the IMF’s Direction of Trade data base; mnemonics for exports and imports are given by xy…ZDz where “x” represents the country doing the trade; “z” represents the trading partner; and “y” is 71 for imports, 70 for exports. It is interesting and distressing that the data for, e.g., German exports to the United States are quite different from U.S. imports from Germany. The two variables are highly correlated in levels, but their growth rates have only a low (and, in the case of the United Kingdom, negative) correlation.

The monthly tariff data is available from Highlights of U.S. Export and Import Trade (FT 990), published by the U.S. Department of Commerce, Bureau of the Census. The data is taken from Section B, “Imports for Consumption--World Area and Country of Origin” (the exact table number varies over time). “Imports for consumption” measures total merchandise cleared through customs, either because it directly enters consumption channels, or because it is withdrawn for consumption from warehouses under customs’ custody. The data is available on a “customs value basis,” which represents the price actually paid for merchandise when sold for exportation to the United States, excluding U.S. import duties, freight, insurance, and other charges incurred in bringing the merchandise to the United States. Relationships between buyers and sellers should not influence the customs value.

Much of the annual data is available in Historical Statistics of the United States. This includes both measures of the tariff rate, bilateral imports and exports, and real GNP. The U.K. measure of real output is spliced from a variety of series, mostly taken from the Abstract of British Historical Statistics. Jeff Frenkel kindly provided us with the bilateral exchange rate and both U.K. and U.S. net national product price deflators (the data is mostly taken from Friedman and Schwartz). Further documentation is available along with the data.


  • Beveridge, S., and C. R. Nelson (1981), “A New Approach to Decomposition of Economic Time Series into Permanent and Temporary Components,” Journal of Monetary Economics 7, 15174.

    • Search Google Scholar
    • Export Citation
  • Boyer, R.S. (1977), “Commercial Policy under Alternative Exchange Rate Regimes,” Canadian Journal of Economics X-2, 21832.

  • Branson, W.H. (1987), “Comments on ‘Macroeconomics and Protection’,” U.S. Trade Policies in a Changing World Economy (Stern, ed.) (MIT, Cambridge).

    • Search Google Scholar
    • Export Citation
  • Chan, K. (1978), “The Employment Effects of Tariffs under a Free Exchange Rate RegimeJournal of International Economics 8, 41424.

    • Search Google Scholar
    • Export Citation
  • Dornbusch, R. (1980), Open Economy Macroeconomics (Basic, NY).

  • Dornbusch, R. (1987), “External Balance Correction: Depreciation or Protection?Brookings Papers on Economic Activity 1, 24969.

  • Edwards, S. (1989), “Tariffs, Terms of Trade and the Real Exchange Rate,” Economica, forthcoming.

  • Eichengreen, B. (1981), “A Dynamic Model of Tariffs, Output, and Employment under Flexible Exchange Rates,” Journal of International Economics 11, 34159.

    • Search Google Scholar
    • Export Citation
  • Engle, R. F., and Yoo (1987), “Forecasting and Testing in Co-Integrated Systems,” Journal of Econometrics 35, 14359.

  • Gardner, G. W., and K.P. Kimbrough (1989), “The Behavior of U.S. Tariff Rates,” American Economic Review 791, 21118.

  • Goldstein, M., and M. S. Khan (1985), “income and Price Effects in Foreign Trade,” Handbook of International Economics (R. W. Jones and P.B. Kenen, eds.), North Holland.

    • Search Google Scholar
    • Export Citation
  • Helkie, W., A. J. Hughes Hallett, G.J. Hutson and J. Marquez (1989), “Protectionism and the U.S. Trade Deficit,” CEPR DP. No. 286.

  • Hufbauer, G.C., D.T. Berliner, and K.A. Elliott (1986), Trade Protection in the United States (IIE, Washington).

  • Johansen, S. (1988), “Statistical Analysis of Cointegration VectorsJournal of Economic Dynamics and Control, 12-2/3, 23154.

  • Kindleberger, C. (1986), The World in Depression (University of California, Berkeley).

  • Krugman, P. (1982), “The Macroeconomics of Protection with a Floating Exchange Rate,” Monetary Regimes and Protectionism (Carnegie-Rochester Series on Public Policy Volume 16, K. Brunner and A. Meltzer, eds.), 14182.

    • Search Google Scholar
    • Export Citation
  • Laursen, S., and L.A. Metzler (1950), “Flexible Exchange Rates and the Theory of Employment,” Review of Economics and Statistics 32, 28199.

    • Search Google Scholar
    • Export Citation
  • Mann, C. (1987), “Protection and Retaliation,” Brookings Papers on Economic Activity 1, 31135.

  • Mitchell, B. R., and P. Deane (1962), Abstract of British Historical Statistics, (CUP, NY).

  • Mundell, R.A. (1961), “Flexible Exchange Rates and Employment Policy,” Canadian Journal of Economics 27, 50917.

  • Mussa, M. (1974), “A Monetary Approach to Balance of Payments Analysis,” Journal of Money, Credit, and Banking 6, 33351.

  • Ostry, J.D. (1988), “intertemporal Optimizing Models of Small and Large Open Economies with Nontradable Goods,” unpublished Ph.D. dissertation.

    • Search Google Scholar
    • Export Citation
  • Razin, A., and L. Svensson (1983), “Trade Taxes and the Current Account,” Economics Letters, 131, 5557.

  • Rose, A., (1989), “Why has Trade Grown Faster Than Income?unpublished.

  • Stock, J. H., and M. W. Watson (1988), “Variable Trends in Economic Time Series,” Journal of Economic Perspectives, 23, 14774.

    • Search Google Scholar
    • Export Citation
  • Van Wijnbergen, S. (1987), “Tariffs, Employment, and the Current Account,” International Economic Review 283, 691706.

  • U.S. Department of Commerce, Bureau of the Census (1975), Historical Statistics of the United States, (GPO, Washington).


Mr. Rose is Professor of Economics, University of California, Berkeley. This paper draws in part on Mr. Ostry’s Ph.D. dissertation submitted to the Department of Economics, University of Chicago. We thank Haydon Merkle and Mattie Halsey of the Foreign Trade Division of the Bureau of the Census, and Jeff Frankel for assistance with the data; Kellett Hannah for computer support; Eduardo Borensztein, Peter Garber, Charles Kindleberger, Cathy Mann, and seminar articipants at the Board of Governors of the Federal Reserve System for comments; and Robert Flood and Doug Purvis for discussions.


Helkie et al. (1988) use simulation techniques on a large macroeconomic model and conclude that protectionist policies are in effective in reducing trade imbalances while avoiding recession.


See, e.g., Mann (1987). Kindleberger (1986) argues that the macroeconomic effects of Smoot-Hawley were in fact contractionary.


The exposition that follows is drawn from Dornbusch (1980, pp: 65-66) although a similar analysis may be found in other texts.


We assume in what follows that the government runs a balanced budget.


The presumption of a contractionary effect is strengthened when money is introduced into the model, since the redistributed tariff revenue creates an additional demand for money, requiring a fall in income from production to clear the money market (see Chan (1978), Eichengreen (1981) and Krugman (1982)).


Krugman (1982) argues that, even in those circumstances in which a tariff raises output and improves the terms of trade when other countries are passive, “symmetric retaliation” will result in lower output and unchanged terms of trade.


Standard trade theory shows that the tariff will tend to benefit the factor used intensively in the import-competing sector.


We choose the G-7 countries for intrinsic interest, noting that they account for over half of U.S. imports (both dutiable and duty-free) and tariff revenues during the sample in question. However, the bilateral tariff data exist for other countries, and it would be interesting to extend the results to, e.g., developing countries.


There does not appear to be any fundamental explanation of the apparent outlier in August 1978.


The volatilities of the growth rates of the tariff measures also vary noticeably, both across countries and tariff rate measures.


If there is incomplete specialization, there will be additional effects on domestic production and foreign consumption which may also contribute to substitution bias.


As our empirical work below indicates that the null hypothesis cannot generally be rejected, this bias strengthens our results, so long as standard errors are not substantially biased.


The data indicate that tariff revenues do not seem to be redistributed. In particular, the tariff rate, and, to a much smaller degree, tariff revenues, have positive but small correlations with the federal budget surplus.


This statement is true of all tests of Granger “Causality.”


In all cases, the hypotheses that imports and exports separately, as well as the nominal trade balance, have unit roots cannot be rejected at traditional significance levels.


Using the nominal trade balance in place of the real trade balance does not change results.


There is no reason for the tariff rate to be used as the sole regressand; the “reverse” regression with, e.g., the trade balance as the dependent variable (and the remaining four variables as regressors), can also be used as the cointegrating equation. We have calculated the augmented Dickey-Fuller cointegration tests for all 96 (six countries x two measures of trade balance x two measures of tariff rate x four alternative regressands) reverse regressions. Almost uniformly (in 92 out of 96 cases), they are consistent with the hypothesis that there is no cointegration between the trade balance, the exchange rate, domestic and foreign output, and the tariff rate.


The Johansen tests--which are valid under more general conditions than the standard Dickey-Fuller tests--indicate that there is one and possibly two cointegrating vectors in the five-variable system. This result is robust to various measures of the trade balance and the tariff rate.


The results do not change if the cointegrating residual is dropped.


The only exception occurs when the lag length is reduced to one year and the U.S. measure of trade with Italy is used, in which case (lags of) τ2 are statistically significant.


The relevant data is available for three countries: Canada, Germany, and Japan.


To implement the Beveridge and Nelson methodology, we assumed that (the logs of the) tariff rates follow IMA (1,6) processes, univariate models which appear to fit the data reasonably well.


Use of national accounts data necessitates estimation at the quarterly frequency.


In principle, the hypothesis that U.S.-U.K. trade is similar in composition to aggregate U.S. trade is testable on the basis of existing data. In particular, bilateral data exist at the annual frequency on an historical basis from Foreign Commerce and Navigation of the United States, but only at the commodity level.


All levies collected on goods because they enter the country.


Duties levied under the customs tariff schedule and annexes, but excluding consular fees, tonnage charges, statistical. taxes, fiscal duties, and other taxes.


The countries included are (listed by IFS code): the United Kingdom, Austria, Denmark, the Federal Republic of Germany, Italy, the Netherlands, Norway, Sweden, Switzerland, Canada, Japan, Finland, Greece, Iceland, Ireland, Malta, Portugal, Spain, Australia, New Zealand, Colombia, Costa Rica, the Dominican Republic, Nicaragua, Paraguay, Venezuela, Guyana, Cyprus, Nepal, the Philippines, Burundi, Cameroon, Zaïre, Malawi, Morocco, Uganda, Zambia, and Fiji.

Tariffs and the Macroeconomy: Some Empirical Evidence
Author: International Monetary Fund