Journal Issue

Hungary: Selected Issues

International Monetary Fund
Published Date:
May 1999
  • ShareShare
Show Summary Details

VII. The Trade Balance and the Real Exchange Rate83

A. Introduction

180. This chapter analyzes the relationship between the trade balance and the real exchange rate in Hungary. The next section discusses developments in the trade balance relative to domestic and foreign demand conditions, and measures of the real exchange rate, highlighting the few but major swings in Hungary’s external position during the 1990s. Sections C and D explore the possibility of estimating the magnitude and timing of the impact of the real exchange rate on the trade balance. Section E concludes.

B. Trade Balance Developments

181. Developments in the Hungarian trade balance during the 1990s, both in merchandise trade and in the balance of goods and nonfactor services (GNFS), are shown in Figure 20. The growing surplus on nonfactor services—primarily due to tourism—explains the widening divergence between these measures since the mid-1990s. This chapter focuses on the GNFS balance, as nonfactor services trade likely also depends on competitiveness. From a surplus of about 2 percent of GDP in the early 1990s, the GNFS balance deteriorated very sharply in late 1992-early 1993, and remained in a large (about 8 percent of GDP) deficit position in 1993 and 1994, Table 16. A strong turnaround in 1995 halved the GNFS trade deficit to 3.9 percent of GDP, with a further improvement to reach 1.2 percent of GDP by 1997.

Figure 20.Hungary—Trade Balance, 1990–98

(US$ millions per quarter)

Source: National Bank of Hungary.

Table 16.Hungary: Trade Balance 1991–97 1/
(In millions of U.S. dollars)
Goods and services, net790717-2,996-3,377-1,763-1,146-558
Exports, GNFS11,62113,33210,88610,67117,02819,18924,508
Imports, GNFS-10,831-12,615-13,882-14,048-18,791-20,335-25,066
Goods, net189-48-3,246-3,634-2,443-2,645-1,735
Nonfactor services, net6017652502576801,4991,177
Of which: Travel, net5605904425036591,2881,428
Other, net41175-192-24621211-251
(In percent of GDP)
Goods and services, net2.41.9-7.8-8.1-3.9-2.6-1.2
Exports, GNFS34.835.828.225.638.142.854.3
Imports, GNFS-32.4-33.9-36.0-33.7-42.1-45.3-55.5
Goods, net0.6-0.1-8.4-8.7-5.5-5.9-3.8
Non-factor services, net1.
Of which: Travel, net1.
Other, net0.10.5-0.5-
Sources: Data provided by the Hungarian authorities; and staff estimates.

Through 1995 in convertible currencies, from 1996 in both convertible and nonconvertible currencies.

Sources: Data provided by the Hungarian authorities; and staff estimates.

Through 1995 in convertible currencies, from 1996 in both convertible and nonconvertible currencies.

182. These trade balance swings reflect the influence of both demand and supply factors in Hungary, external demand conditions, and the changing relative prices of domestic and foreign goods and factors of production, or the real exchange rate. The top panel of Figure 21 illustrates the likely importance of domestic demand—using interpolated data on gross national expenditure—for both the trade balance deterioration in 1992–93 and the 1995 turnaround. However, it is notable that the demand expansion since 1996 was not associated with a renewed deterioration in the trade balance in 1997 or early 1998. This stability may reflect the expansion in external demand seen in the bottom panel of Figure 21, as measured by the deviation from trend of industrial production in Hungary’s six largest western trade partners. External demand appears to have also played a role in 1992–93, though the trade balance further weakened in 1994 despite the external recovery in that year.

Figure 21.Hungary: Trade Balance and Domestic and External Demand, 1990–98

(In millions of U.S. dollars)

Sources: National Bank of Hungary, Hungarian Central Statistical Office, and staff calculations.

183. The early 1990s saw rising real effective exchange rates (REER) on both a CPI and unit labor cost basis (ULC), Figure 22, indicating declining competitiveness ahead of the high deficits in 1993–94. There was a very sharp decline in the real exchange rate on a ULC basis through 1995, of 20 percent compared to the average in 1994, as labor productivity in manufacturing (gross production relative to employment) rose by 11 percent, but nominal labor costs increased by only 20 percent compared to a nominal effective exchange rate depreciation of 29 percent. The CPI basis REER also depreciated in 1995, by 3 percent. Thus the turnaround in the trade deficit through 1995 coincides with a significant real depreciation, suggestive of a quite rapid trade balance response to the real exchange rate. The depreciation of the ULC basis REER continued at a declining pace over 1996 to 1998, while the REER on a CPI basis reverted to its tendency to appreciate over time.

Figure 22.Hungary: Trade Balance and Relative Prices, 1990–98

(In millions of U.S. dollars)

Sources: National Bank of Hungary, OECD Analytical Databank, and staff calculations.

184. Evaluating the relative roles of domestic and external demand and the real exchange rate in these trade balance developments may provide information on how sensitive financial policies should be to future developments in the real exchange rate. The following analysis attempts therefore to more clearly identify the sources of these trade balance dynamics.

C. Model Specification

185. As Hungary is a small open economy, the standard approach to this question would be to take world prices as exogenous, and to estimate import demand and export supply equations, as Halpern and Székely (1992) did for the 1968–1989 period. However, the structural change in the composition of Hungary’s trade through the 1990s would likely result in biased estimates for the income and price elasticities. Trade has become highly focused on manufactured goods and machinery, which rose from 34.6 percent of exports in the first half of 1992 to 84.6 percent in the same period of 1998, primarily as FDI related investments came on stream. These exports require a high share of imported intermediate inputs, which coupled with the trade liberalization, saw gross export and import volumes expand much faster than GDP, so that exports rose to 54 percent of GDP in 1997 from 35 percent of GDP in 1991. A standard model of exports and imports would therefore likely find artificially high income elasticities, and other parameter estimates may also be biased.

186. One approach would be to attempt to control for structural change by including a proxy in the standard trade equations, and perhaps including exports in the import equation to proxy the demand for intermediate imports. The difficulty is to produce an adequate proxy for structural change in the absence of trade data classified by broad economic category, i.e. intermediate, consumption, and capital goods. This section instead develops a model that focuses directly on the trade balance, with the expectation that the effect of the structural change on export and import volumes will be largely offsetting.

187. The short-run supply of gross exports (x) is assumed to reflect production capacity as proxied by domestic potential GDP (yp), the relative profitability of export production (rx), and also the trend (tx) due to structural change in the composition of exports. The foreign output gap (yf - yfp) is included to capture short-run effects from foreign demand.84

Gross imports (m) reflect aggregate demand (yd), and the relative price of imports to domestic goods(rm), and also the trend due to structural change (tm):

The trade balance, in terms of constant export and import prices, is therefore:

188. This constant price measure of the merchandise trade balance moves very similarly to the current price measure, see the second panel of Figure 22, as Hungary’s terms of trade have been relatively stable. The trade balance is increasing in potential GDP but falling with aggregate demand, increasing in the foreign output gap, and depends on the relative prices for both export supply and import demand, which in this study are represented by the real effective exchange rates on both a ULC and CPI basis. The trends (that are not likely linear) due to structural change (tx and tm)are expected to offset each other to a large extent.

189. Equation (3) is similar to the framework of Rose (1991) and Rose and Yellen (1989), aside from the inclusion of potential GDP, which is consistent with modeling a small open economy. It focuses on the short–term role of the real exchange rate in the switching of expenditures between domestic and foreign goods, and in stimulating traded goods production by increasing utilization of existing capacity and shifting existing resources into the traded goods sector. The real exchange rate may also affect investment and therefore productive capacity, but analysis of these long–run effects of the real exchange rate would require a longer span of data, and a more complete model.

D. Estimates and Tests

190. Equation (3) was estimated on quarterly data for 1992:Q1 to 1998:Q2, with variable definitions provided in Appendix I. Estimation is by OLS, treating all the explanatory variables as exogenous to the trade balance, which is acceptable for foreign activity, and also for domestic demand and the real exchange rate as these variables enter with a lag. The unit root tests in Table 17 leave doubts regarding the order of integration of these variables, as might be expected in this short sample. The estimation assumes that all variables included are stationary, or stationary around a deterministic trend.

Table 17.Hungary: Augmented Dickey-Fuller Unit Root Tests
Sample: 1991:Q2 to I998:Q2, 28 observations.
No Deterministic Trend IncludedDeterministic Trend Included
VariableBeta 1/t-ratioBeta 1/t-ratioBeta 1/t-ratioBeta 1/t-ratioConclusion on the Order of Integration
GNFS-0.18-1.65-0.22-1.72-0.21-1.62-0.28-1.57I(0), based on the
First difference-0.96-3.38 *-1.24-2.69-0.97-3.33-1.28-2.69deviation of the beta estimates from 0.
In (R_ULC)0.010.13-0.01-0.25-0.28-3.59 *-0.40-3.39I(0) with trend.
First difference-1.13-4.82 **-0.57-2.05-1.38-5.53 **-0.71-1.93
ln(IPF/IPF_T)-0.15-3.93 **-0.12-2.17-0.12-1.99-0.14-1.89I(0)
First difference-0.32-3.01 *-0.38-2.13-0.36-3.34-0.45-2.39
ln(GDE/GDP_T)-0.11-1.52-0.25-2.69-0.15-1.88-0.36-2.90Unit root not rejected,
First difference-0.53-2.30-0.67-2.07-0.53-2.26-0.68-2.04but beta estimates and theory suggest I(0).
Critical values from MacKinnon (1991):
1 percent (**)-3.65-3.68-4.27-4.32
5 percent (*)-2.96-2.97-3.56-3.58

Beta (β) is the estimated coefficient on the lagged level of the variable (y) being tested for a unit root:

Δyt = α + µt + βyt-1 + Σi=1 to n γjΔyt-i + εt

Where n is the number of lags in the ADF(n) test. Under the null hypothesis of a unit root, beta equals zero.

Beta (β) is the estimated coefficient on the lagged level of the variable (y) being tested for a unit root:

Δyt = α + µt + βyt-1 + Σi=1 to n γjΔyt-i + εt

Where n is the number of lags in the ADF(n) test. Under the null hypothesis of a unit root, beta equals zero.

191. A specification search did not find significant effects from the REER on a CPI basis, and the better performing ULC basis measure is calculated using manufacturing gross production rather than manufacturing value added. The restriction that the long-run effect of aggregate demand and potential GDP be equal but opposite sign (αx = -αm) was accepted (F(1,19)=2.76, with a p-value of 0.113). The dynamics of the real exchange rate and the other variables were derived by testing down from a more general specification. The preferred estimation results are (t—statistics):

GNFS =1.321- 0.702 ln(GDE/GDP_T)-1+ 0.345 ln(IPF/IPF_T)
- 0.245 LRU4-1

- 0.0061 Trend

- 0.331 GNFS-3

R2 = 0.886S.E.E. = 1.33 percentDW = 2.30
Sample:1992:Q1 to 1998:Q2, 6 variables and 26 observations

Diagnostic tests [p-value]:

AutocorrelationF(1,19) = 0.507 [0.485]
HeteroskedasticityF(10,9) = 0.773 [0.655]
NormalityChi2 = 1.458 [0.482]
Functional formF(1,19) = 1.329 [0.263]

192. Graphical analysis of the equation is provided in (Figure 23). The equation passes all the diagnostic tests, including the Chow tests for stability reported in (Figure 24). However, the equation performed poorly when re-estimated until 1997:Q2 and projected out-of-sample over 1997:Q3 to 1998:Q2, with systematic under-predictions of the trade balance. Mismeasurement of potential GDP likely accounts for part of this result, as potential GDP probably rose significantly more strongly that the 3 percent calculated using the Hodrick-Prescott filter over these four quarters.

Figure 23.GNFS Trade Balance Equation

Figure 24.Chow-tests of Structural Stability

193. The long-run elasticities with respect to the explanatory variables are the appropriate coefficients multiplied by 1/(1+0.331) or 0.75, due to the inclusion of a lagged dependent variable with a negative coefficient.85 Thus an increase in gross national expenditure relative to trend GDP by 1 percent is estimated to increase the trade deficit by 0.53 percent of GDP. A rise in foreign industrial production by 1 percent relative to trend results in an estimated 0.26 percent of GDP long-run rise in the trade balance. The long-run elasticity of the GNFS balance with respect to the real exchange rate is estimated to be 0.18, so a 5.4 percent real appreciation would be required to weaken the trade balance by 1 percent of GDP. Note that the standard error on this estimate would imply an 80 percent confidence interval for this elasticity of 0.08 to 0.29, or 12.1 percent to 3.5 percent in terms of the required appreciation for a 1 percent of GDP impact on the trade balance.86

194. The equation displays a modest short–run overshooting in its dynamics. After a 1 percent real exchange rate shock, the peak response of the trade balance occurs 4 quarters later, at 0.225 percent of GDP on an annualized basis, before reverting to the long-run response of 0.18 percent after 8–9 quarters.

195. While most parameters in the estimated equation appear plausible, there remains an unexplained trend with a relatively small effect of negative 0.46 percent of GDP per annum. This may reflect many factors including the omission of relevant variables, statistical problems in the included variables—e.g., the trend in the REER on a ULC basis may be biased down by the overstatement of industrial production growth, or the trend measure of GDP may be a poor approximation—along with the residual effects of structural change.

E. Conclusions

196. The discussion in section B suggested that the real exchange rate might have a significant role in Hungary’s trade balance, in addition to the more obvious effects of domestic and foreign demand conditions. An econometric estimate of the size and timing of the real exchange rate effects was made, that attempted to avoid biases due to the radical changes in Hungary’s trade structure during the 1990s. This estimate suggests that a real exchange rate appreciation of 1 percent would reduce the trade balance by 0.18 percent of GDP, with this effect coming through within the year following the shock. For example, this estimate suggests that the real depreciation in 1995 of 11 percent accounts for about 2 percent of GDP of the 5.5 percent of GDP improvement in the trade balance in 1996 over 1994.

197. This estimate of the real exchange rate effect was subject to significant uncertainty, as should be expected from the short sample available. Further research along these lines would benefit from improved data, in particular, actual rather than interpolated quarterly GDP and expenditure data. More sophisticated measures of potential GDP than the Hodrick-Prescott filter, or other indicators of supply-side developments in Hungary, also appear to be essential to understanding the strength of Hungary’s trade balance in 1997 and early 1998.


    HalpernLászló and IstvánSzékely (1992) “Export Supply and Import Demand in Hungary (An econometric analysis for 1968–1989),Discussion Paper No. 620Centre for Economic Policy Research.

    MacKinnonJames (1991) “Critical values for cointegration tests,”inEngleR. F. and GrangerC. W. J. (eds.) Long-Run Economic RelationshipsOxford University Press.

    RoseAndrew (1991) “The Role of Exchange Rates in a Popular Model of International Trade: Does the ‘Marshall-Lerner’ Condition Hold?,Journal of International Economics30 301316.

    RoseAndrew and JanetYellen (1989) “Is there a J-curve?,Journal of Monetary Economics245368.

APPENDIX I: Variable Definitions

Trade price data from the OECD Analytical Data Bank (ADB) was used to calculate the merchandise trade balance with constant U.S. dollar prices for merchandise exports and imports, as at 1990’Q1.


Where XG and MG are unadjusted export and import values from the National Bank of Hungary (NBH) and PXGD and PMGD are the ADB data on export and import prices in U.S. dollar terms, extrapolated from 1996’Q4 using data from the World Economic Outlook. This data was seasonally adjusted, and added to seasonally adjusted data for the nonfactor services balance (NFS), and converted to an annualized ratio to GDP using interpolated quarterly GDP in U.S. dollar terms:


R_ULC =Real effective exchange rate, ULC in manufacturing. Source: NBH
R_CPI =Real effective exchange rate, CPI basis. Source: NBH

The REER indices were averaged over 4 quarters to distribute their impact over time:

LRC=log(R_CPI)LRC4 = [LRC + LRC-1 + LRC-2 + LRC-3]/4
LRU= log(R_ULC)LRU4 = [LRU +LRU-1 + LRU-2 + LRU-3]/4
GDE =Gross domestic expenditure, cubic spline from annual index calculated from cumulative real growth rates.
GDP_T =Trend in real GDP, where quarterly GDP data were calculated by cubic spline, and the trend uses a Hodrick-Prescott filter with a smoothing parameter of 1600.
IPF =Foreign industrial production: trade weighted (percent) index of Germany (49.7), Austria (17.6), Italy (14.4), United States (7.1), France (5.9), and the U.K. (5.4).
IPF_T =Trend in foreign industrial production, using the Hodrick-Prescott filter with a smoothing parameter of 1600.

Prepared by Craig Beaumont.

In the long-run, exports are determined by domestic supply, but due to customer-specific production and some price stickiness in exports, foreign demand fluctuations will likely affect export volumes in the short-run.

The negative sign of the lagged dependent variable provides further assurance that the data are not non–stationary.

The Student’s t–distribution with 20 degrees of freedom is used.

Other Resources Citing This Publication