Exchange rate sensitivity of U.S. trade flows: evidence from industry data.

• Author: Bahmani-Oskooee, Mohsen

1. Introduction

   In an effort to boost employment, a country could stimulate its exports and discourage its imports and thereby improve its trade balance. One policy that has received a great deal of attention in the literature is currency devaluation. By making exports cheaper and imports expensive, devaluation is said to improve the trade balance. The only condition required is that the sum of import and export demand price elasticities exceed unity (that condition, known as the Marshall-Lerner condition, is derived under the assumption of perfectly elastic supply of trade). Most previous studies that attempted to assess the Marshall-Lerner condition relied on price elasticities that were obtained by estimating aggregate import and export demand functions. These studies provided mixed conclusions as far as the effectiveness of devaluation or depreciation is concerned. Examples include Houthakker and Magee (1969), Khan (1974), Goldstein and Khan (1976, 1978), Wilson and Takacs (1979), Haynes and Stone (1983a, 1983b), Warner and Krienin (1983), Bahmani-Oskooee (1986, 1998), and Bahmani-Oskooee and Niroomand (1998). The mixed conclusion could be related to aggregation bias. When aggregate trade data are employed in import and export demand functions, significant price elasticity with one trading partner could be more than offset by an insignificant price elasticity with another trading partner, yielding an insignificant price elasticity.

   Because of aggregation bias, another body of the literature has emerged in recent years that concentrates on using trade data at the bilateral level. Examples in this latter group include Rose and Yellen (1989), Cushman (1987, 1990), Summary (1989), Marquez (1990), Haynes, Hutchison, and Mikesell (1986), Eaton (1994), Bahmani-Oskooee and Brooks (1999), Nadenichek (2000), and Bahmani-Oskooee and Goswami (2004). Except Bahmani-Oskooee and Goswami (2004), all other studies in this second group have estimated bilateral trade elasticities between the United States and her major trading partners and concluded that the real bilateral exchange rate is a significant determinant of bilateral trade balance, at least in some cases. Bahmani-Oskooee and Goswami (2004), who considered the bilateral trade flows between Japan and her nine major trading partners, found that in most cases Japanese exports are not sensitive to the real bilateral exchange rate, but her imports are. These studies as well as those in the first group estimate price elasticities in demand under the assumption that supply is perfectly elastic and then evaluate the Marshall-Lerner condition involving own-price coefficients in demand. Thus, evidence in these studies is limited because it assumes perfectly elastic supply of trade for both exports and imports.

   Although there is additional room to expand the literature in the second group by considering the experiences of countries other than the United States and Japan, in this paper we would like to open another avenue of research by investigating the impact of real depreciation of the dollar on imports and exports of 66 American industries, a disaggregation by industry rather than by country. Disaggregation by industry will avoid problems associated with petroleum imports, as does disaggregation by country (Rose and Yellen 1989). (1) For this purpose, in Section 2 we outline the import and export demand functions for each commodity group along with the estimation method. In Section 3, we present the empirical results. Section 4 provides our summary and conclusion. Data definitions and sources are cited in an appendix.

2. The Models and the Method

   In formulating any import and export demand function, it is a common practice to relate the volume of imports and exports to a measure of income and relative prices. The main purpose is to obtain estimates of import and export demand elasticities so that we can better judge the effectiveness of currency devaluation in increasing a country's inpayments and reducing outpayments. One major limitation of these studies is that they have assumed a perfectly elastic supply. An exception is Haynes, Hutchison, and Mikesell (1986), who considered the bilateral trade between the United States and Japan by formulating the demand and supply equations. They then estimated not only the bilateral demand and supply models but also reduced-form models in which bilateral import and export values were directly related to real bilateral exchange rate in addition to other variables. The advantage of this direct method is that one could easily determine whether currency depreciation has favorable effects on a country's inpayments and outpayments. Furthermore, Bahmani-Oskooee and Goswami (2004), who considered Japan's experience with her nine largest trading partners, argued that because import and export prices are not available at the bilateral level, relating import and export values directly to real bilateral exchange rate is the only way to assess the impact of currency depreciation on inpayments and outpayments. At the industry level, because of lack of import and export prices, we also concentrate on nominal figures and try to investigate sensitivity of import and export values of each industry to a change in exchange rate. In doing so, we modify BahmaniOskooee and Goswami's (2004) models so that they conform to industry data. Thus, for each commodity group (or industry) we formulate the inpayments and outpayments functions by Equations 1 and 2 respectively (2):

   Ln [VX.sub.i,t] = a + b Ln [Y.sub.w,t] + c Ln [RE.sub.t]+[[epsilon].sub.t] (1) Ln [VM.sub.i,t] = d + e Ln [Y.sub.us,t] + f Ln [RE.sub.1] + [[mu].sub.t], (2)

   where [VX.sub.i,t] is industry i's exports (in dollars), which is assumed to depend on world income, [Y.sub.w], and real effective value of the dollar, RE. In Equation 2, [VM.sub.i,t] is the value of imports by industry i, which is assumed to depend on U.S. income, [Y.sub.US], and real effective value of the dollar. Although estimates of coefficients assigned to income variables in both equations are expected to be positive, estimate of c in Equation 1 is expected to be negative, and that of f in Equation 2 positive. These expected signs of c and fare based on the definition of RE, which is defined as number of units of foreign currencies per dollar. (3,4) Note that Equation 1 is a reduced-form equation that is derived from a supply-and-demand model in which the export supply of good i by the United States is perfectly elastic, whereas the rest of the world demand for good i depends on rest of the world income ([Ysub.w]) and the real effective exchange rate (RE). Similarly, the reduced form Equation 2 is derived from a supply and demand model in which the supply of good i by the rest of the world is assumed to be perfectly elastic, yet the demand by the United States for the same good is assumed to depend on the U.S. income ([Y.sub.US]) and the real effective exchange rate (RE). Thus, except the real exchange rate, other supply factors are excluded from the specifications that could render the empirical results somewhat biased.

   Equations 1 and 2 are long-run relationships among the variables of interest. The advances in econometric literature dictate that in estimating the long-run relations we must incorporate the shortrun dynamics into the estimation procedure. This could be done by specifying Equations 1 and 2 in an error-correction modeling format. Following Pesaran, Shin, and Smith (2001) and their new method of Autoregressive Distributed Lag (ARDL) approach to cointegration analysis, we rewrite Equations 1 and 2 in an error-correction modeling format as in Equations 3 and 4:

   [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (3)

   [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (4)

   Note that Equations 3 and 4 differ from standard distributed lag models in that they include a linear combination of the lagged level of all variables, normally referred to as an error-correction term. The first step in estimating Equations 3 and 4 is to decide whether to retain the lagged level of variables in Equations 3 and 4. Pesaran, Shin, and Smith (2001) propose using the standard F-test with new critical values that they tabulate. The advantage of this approach is that there is no need for preunit root testing. Based on the assumption that all variables could be I(1) or I(0) or some I(1) and some I(0), Pesaran, Shin, and Smith (2001) provide an upper and a lower bound of critical values. If our calculated F-statistic turns out to be significant (higher than the upper bound), the lagged level variables are to be retained in Equations 3 and 4, which is an indication of cointegration among the variables. Once the first stage is settled, we estimate Equations 3 and 4 by employing a standard criterion to select the optimum lag length of each first differenced variable. Although the short-run effects of devaluation are judged by coefficient estimates of [DELTA]Ln [RE.sub.t-k], the long-run effects are judged by estimates of d and h that are normalized on the estimates of b and f in Equations 3 and 4, respectively.

3. Empirical Results

   Monthly import and export data from 66 industries in the United States (SITC Commodity Groupings) over the January 1991-August 2002 period are employed in estimating error-correction models 3 and 4. The first step in applying Peseran, Shin, and Smith (2001) ARDL technique is to carry out the F-test to determine whether we are justified in retaining the lagged-level variables. As mentioned above, this amounts to testing the null hypothesis of no cointegration, i.e., b=c=d=0 in Equation 3 and f=g=h=0 in Equation 4 against the alternative of b [not equal to] c [not equal to] d [not equal to] 0 in Equation 3 and f [not equal to] g [not equal to] h [not equal to] 0 in Equation 4 by using the familiar F-test with new critical values. Bahmani-Oskooee and Brooks (1999) have demonstrated that the results of the...