Thursday, April 24, 2008

An empirical study on the Turkish J-Curve and the Marshall-Lerner Condition

The Turkish economy has suffered several shocks to its economy over the past 30 years. These shocks have led policymakers to multiple devaluations of the new Turkish lira as an attempt to bring the country out of its increasing current account deficit. The extent and frequency of these devaluations has not led to a surplus in Balance of Payments as expected. Possible explanations for the cause of a “perverse reaction” to devaluation are an unsatisfied Marshall-Lerner condition and a multiple devaluation shifting the J-curve to the right under the Kulkarni hypothesis.

Using simple Ordinary Least Squares regression methods, I will test the effect that a change in exchange rate has on the trade balance for the Turkey since 1980, during which several international trade reforms were implemented. I predict that I will find that the multiple devaluations of the Turkish Lira have led to an indefinitely negative trade balance, rather than a trade balance surplus as policymakers had desired. This perpetually negative trade balance is also potentially due to the consumers and investors’ expectations of devaluation.

Since 1947, Turkey has experienced a negative current account balance. In the 1950s, like most of the rest of the world, the Turkish Lira was fixed, in this case with the United States Dollar. Gross National Product growth was very unstable during this time period, due to large fluctuations in output. The currency appreciated 13% from the period of1953-1959 but did not help the trade balance, as exports fell from 7.6% of GDP in 1953 to 2.0% of GDP in 1958. The Turkish government faced a debt of approximately 1 billion USD in 1958, and devaluated the currency in August of the same year in an attempt to bring the economy out of a negative trade balance. The devaluation was partnered with more liberalized trade policies, though Turkey was still a much-closed economy.

In the 1960s, the currency was particularly overvalued due to inflationary pressures, but the time period was also associated with a reduction in the current account deficit as exports rose from15% of GDP in 1971 to 30% of GDP in 1972. The deficit reduction may have been a result of extensive import substitution policies, which also included a heavy emphasis on investment in manufactured and capital goods. However, producers still relied on importing raw materials for production. In 1973-1974, Turkey suffered from oil shocks that made oil more expensive to import and therefore increased the trade deficit, followed by a second oil shock in 1979. Also in the 1970s, Turkey’s domestic spending started to increase, leading to further terms of trade deterioration. Turkey’s inflation rate remained high, sometimes past 100%, but never reached hyperinflationary levels.

Turkey failed to negotiate a stabilization agreement with the International Monetary Fund in 1978 and 1979, but in 1980, policymakers turned what had been an import-substitution policy into an export-oriented growth policy. These trade reforms were led primarily by the Deputy Prime Minister Turgut Őzal, also the leader of a military regime that took over Turkey during a short time period of 1980-1983. Őzal’s reforms included export incentives such as tax rebates and favorable credit terms. Some restrictions on imports were lifted to encourage trade as well, such as quotas. The Turkish lira was considered overvalued and a devaluation policy was set in place. The country also moved towards a freer exchange rate system, though it was still relatively controlled by the government. Industrialization accompanied the movement of 1980, though it had been taking place along with the import-substitution policies already in place. Turkey began moving from a primarily agricultural exporter to an industrialized goods exporter. As a result of the 1980 reforms, foreign investors also became more confidence in the Turkish economy, which helped to turn the low foreign direct investment from $100 million USD in 1987 to $800 million in 1992.

When the military regime fell and parliamentary democracy returned in 1983, more currency devaluations were instituted. Even though previous devaluations were not particularly effective, policymakers in this time were much more optimistic about the demand elasticities of imports and exports, and thus determined that devaluation of the domestic currency would create a surplus in the balance of trade for Turkey. In 1987, Turkey experienced a mini-crisis where inflation was accelerating out of control, to the extent that real interest rates on deposits were becoming negative. Therefore, people fled from the devalued Turkish Lira and invested in dollar currencies. As a result of the excessive inflation, the Turkish government tightened monetary policy in February of the next year. The government successfully stabilized the currency but slowed growth in the process. The current account in 1988 and 1989 was in surplus temporarily. The 1990s began with a short period of high inflation and growth, creating an increasing current account deficit. In the early 1990s, fiscal policy in Turkey created higher inflation by increasing government spending, leading to an annual inflation rate of 73% in 1993. This further deteriorated the terms of trade and deteriorated the value of the Turkish lira. In 1991, the real GDP per capita was decreasing by 1% annually, along with rising inflation. The Turkish economy faced a mini-crisis in 1994 and the Turkish lira had depreciated by 76% against the US dollar (World Bank and OECD). From 1999-2001, the Turkish government worked with the International Monetary Fund to stabilize the economy by reducing inflation and restoring external balance. Unfortunately, this plan failed in2001 as it resulted in sharp devaluations of the lira, and did not help to control inflation or the trade balance.

Today, Turkey depends heavily on imported oils and fuel, accounting for 20.5% of total imports in 2006.Significant oil shocks during the past decades have helped to further push Turkey into a current account deficit, as oil became more costly. Turkey is also a large importer of mechanical machinery, road vehicles, iron, and steel. Turkey’s major exports include clothing and textiles—accounting for 21.3% of total exports in 2006, whereas road vehicles, iron, steel and electronic machinery are also major exports. Turkey’s main trading partner is the EU, which accounts for 51% of her exports and 45% of her imports, mostly from Germany. Other trading partners include the United States, Middle Eastern countries, and South Asian nations.

The “Marshall-Lerner condition” was theorized in the early 1990s by Alfred Marshall and Abba Lerner to explain why currency devaluations may not lead to a favorable effect on the trade balance of a country. In simple economic theory, currency depreciation will make a domestic country’s goods cheaper to foreign countries, and thus increase exports and create a surplus on the domestic country’s current account. Under a free market system with floating exchange rates, an overvalued or undervalued currency will self-adjust itself via the market mechanism. However, the Marshall-Lerner condition proves that this will only occur if the sum of the demand elasticities of imports and exports are greater than one, and generally holds true for small open economies. This is because in order to have a favorable trade balance after devaluation, import and export demand must be able to effectively respond to the change in exchange rate.

Mohsen Bahmani-Oskooee in his 1998 studies on the Marshall-Lerner condition and J-curve hypothesis found that the Marshall-Lerner condition holds true in most of the 30 developed and developing countries that were tested (Bahmani-Oskooee 101-109). However, Bahmani-Oskooeealso found that in the case of India—always the exception—her elasticities of demand for imports and exports satisfied the Marshall-Lerner condition but the trade balance has deteriorated in the short and long term. Other studies have conclusively proven that the M-L condition continues to be important in determining if devaluation will have a positive effect on balance of trade, if no other factors counteract the positive effects of devaluation. One of these offsetting effects could be if there is a relative increase in income domestically compared to the income of the rest of the world, which would create a higher demand for imports and offset the devaluation effects on balance of trade.

The J-Curve hypothesis states that initially, devaluation will lead to a deterioration of trade balance because the quantity of imports will remain the same for a period of time, and will cost more to domestic consumers. There will have been existing contracts before the devaluation that must be carried through even with the devaluation. Imports will be more valuable as exports remain of approximately the same value. Goods prices do not adjust overnight, and during the period of time that prices are adjusting, there will be a negative trade balance. After producers and consumers adjust their demand to the new currency rates, exports will be more desirable to foreign nations and the trade deficit will turn into a surplus in the long–run. This is under the assumption that the Marshall-Lerner condition holds true. Junz and Rhomberg in their 1973 research study concluded that in order for trade balance to become positive, the lags in recognition, decision, delivery, and placement must be considered.

Studies have found mixed results testing the validity of the J-Curve hypothesis. In Andrew K. Rose’s study in 1990, there was found to be no association between exchange rate and trade balance for developing countries. Rose’s study was somewhat unique because most J-Curve empirical studies only focus on industrialized countries. Empirical evidence has show that the J-Curve hypothesis may hold true for countries such as Egypt, Indonesia, Greece, Korea, Japan, and Spain—that is, after a devaluation of domestic currency, each country experienced a short-term negative trade balance followed by a long-term positive trade balance (Kulkarni 52-63). Some studies that employ disaggregate data have found that the J-Curve may not be apparent if the positive effects of devaluation in one country is offset by negative effects to its trading partners (Halicioglu 3).

Similarly, expectations of devaluation have a considerable effect on whether there will be a positive reaction in exports and imports. If devaluation is unexpected, devaluation is seen to be highly effective, while the opposite is true as well. Sellers will look to make contracts in currencies that are expected to appreciate, so that they will gain more from a transaction, though buyers will want to engage in economic contracts under which the currency is expected to deteriorate, as their costs will be minimized. The theory of rational expectations, therefore, contributes to the hypothesis that multiple devaluations will lead to an increasingly negative trade balance.

Professor Kishore G. Kulkarni tested the hypothesis that multiple devaluations will lead to a perpetual trade deficit for Ghana and found that subsequent devaluations of the Ghanaian Cedi led to a long period of deterioration of terms of trade in the period of1983-1989. In countries that experience only a single devaluation, the effects are more align with the traditional J-Curve hypothesis (Kulkarni 41-51).

There is limited research on the impact of devaluations on Turkey’s trade balance. The research that has been conducted either used bilateral or aggregate data, sometimes with conflicting results. The benefit of bilateral data is that it can show how one country’s trade balance affects another’s. Bilateral trade studies on the J-Curve also more often find positive effects of devaluation in the long-run. A 1997 study found that in the 1970s, there was no relationship between trade balance and exchange rates in Turkey and in the 1980s there existed a negative relationship (Brada et al.). Another study in 2001 discovered a lag of one year after devaluation for trade balance to begin to have a positive behavior, looking at long-run aggregate data (Kale). Independent studies by Bahmani-Oskooee and Rose found no relationship in Turkey between exchange rates and terms of trade. A1991 study on devaluations concluded that a higher exchange rate may be contractionary if it is anticipated (Agenor), while unanticipated devaluations have a greater chance of producing a positive effect on output and economic activity.

For testing the hypothesis of multiple devaluations leading to a perpetual trade deficit in Turkey, I have found trade statistics from Turkey’s 12 major trading partners—Austria, Belgium, Canada, Denmark, France, Germany, Holland, Italy, Japan, Switzerland, the United Kingdom, and the United States. These were Turkey’s most significant trading partners in 2005 according to International Monetary Fund statistics. I have used data from the IMF’s International Financial Statistics database, from World Bank statistics, and from the Organisation for Economic Co-operation and Development Surveys of Turkey. Due to the multiple devaluations chasing rising inflation in Turkey particularly from 1980 onwards, I will attempt to find a relationship between trade balance and exchange rates between the time period of 1980 and 2006.

If we assume that trade balance is a function of the domestic country’s GDP, the trading partner’s GDP, and the real exchange rate, we can assume a linear-logged model for the relationship. The variables are logged in order to make them unit-free. The Ordinary Least Squares static regression model that I will use to fit the data is as follows:

Log(TB)=a₀+β₁log(Y)+β₂log(avgY*)+β₃log((avgP*/P) (ER))+e

Where TB is the annual nominal difference between exports and imports in Turkey, a₀ is the autonomous annual change in trade balance, Y is Turkey’s annual nominal GDP, Y* is the averaged annual and nominal GDP for Turkey’s 12 major trading partners, P* is the averaged Consumer Price Index for Turkey’s trading partners, P is the domestic Consumer Price Index, and ER is the exchange rate expressed in terms of the Turkish Lira’s purchasing power when compared to the IMF’s Special Drawing Rights unit. The SDR is a weighted unit term in the IMF’s IFS to determine exchange rates—it is derived from a basket of currencies as an international reserves unit. The value e is a random error term. I have logged the variables to attempt to keep elasticities relatively constant and to control the large units of these variables.
The estimated value of the coefficient β₁ is expected to be negative, as Turkey’s imports are expected to increase with a rise in domestic income via the import function M=a₀+b₁Y, and therefore will result in a negative trade balance. The estimated value of the coefficient β₂ is expected to be positive, as a rise in foreign income will result in a positive trade balance due to increased exports. However, the estimates for β₁ and β₂may be negative if import-substitution goods are the cause of an increase GDP. The coefficient β₃ is of particular interest to this test because it will assist in determining whether exchange rate, adjusted for price levels, has a positive or negative effect on trade balance. If the partial derivative ∂/∂ER[TB] is negative in the short run, and positive in the long run, the J-Curve hypothesis holds true. If the estimated β₃ is negative in the long-run, the Kulkarni hypothesis that a series of devaluations leads to an indefinitely negative trade balance may be an accurate explanation. I expect the multiple devaluations of the Turkish Lira will cause the trade balance to be indefinitely negative.
I will also test to see if foreign direct investment is affected by Turkey’s currency deterioration. I am expecting that investment in Turkey will decrease with currency devaluation, particularly because of the instability of the Turkish economy and the expectations of further devaluations. I will hold interest rate constant during the test because I would like to isolate the changes in GDP and exchange rate to respective changes in exports and imports.

Finally, I will test the Marshall-Lerner condition and find a value for elasticities of demand for exports and imports in Turkey during the time period of 1980-2006. I will do this by testing the import elasticity function M=a₀+β₁((P*/P)ER) where a₀ is the level of autonomous imports, P* is the foreign country’s price level, P is the domestic country’s price level, β₁ is the amount that imports are expected to change with respect to a one unit change in adjusted exchange rate, Y is the domestic income, and β₂ is the amount that imports are expected to change with respect to Y.

β₁ is expected to be negative because as exchange rate increases (exchange rate here is expressed as domestic currency units/foreign currency units), the domestic currency loses value and thus imports will be expected to decline, as they are more expensive for domestic consumers. β₂is expected to be a positive value, because as the domestic country’s income increases, they will be able to purchase more imports. Similarly, I will test an export elasticity function where X=a₀+β₁((P*/p)ER)+β₂Y*, the only difference in independent variables being the foreign country’s GDP as the value of Y*. The coefficient β₁is expected to be positive as prices become more appealing to foreign consumers. The coefficient β₂ is expected to be positive as well, because as foreign country’s income increases, their demand of the domestic country’s goods will increase as well, leading to a greater value for exports.

The statistical analysis package that was used for this test was SAS, Version 9.1. Data used, program details and output are attached to the end of this paper. I am using aggregate data rather than bilateral trade between Turkey and her trading partners.

The results from testing the a regression expressing trade balance in terms of foreign and domestic income and exchange rates, produced the following results:

[6.22824][.09131][.73519][.44] [s.e]

The estimated β value shows that as Turkey’s GDP increases, the estimated, averaged value of trade balance decreases. This was expected, as Turkey’s income decreases, that there should be a positively increasing demand for imports and therefore a deterioration of the terms of trade. The positive estimated value of β₂ is negative, which was unexpected. This means that as a foreign country’s GDP increases, the trade balance deteriorates even further. This may be due to import-substitution goods in other countries or the sensitivity of unit measurements in the data. The estimated value for the coefficient β₃ is positive, which means that in the long-run, devaluation of the Turkish Lira has a positive effect on trade balance according to the statistics that I have obtained.

To test the accuracy of these regressions, I conducted a 95% confidence interval for the estimated value of β₃. I found that the estimated value of β₃ for the Trade Balance equation was between .77986 and 2.58562. This allows me to state with 95% certainty that the true value for β₃ is positive, and therefore exchange rate deterioration has a positive effect on trade balance in the long-run. The adjusted R-squared value obtained from this regression was .7883, out of 1, and helps to explain how well the data fits the regression model.

When I isolated the years 1980-1990 and performed the same test, I derived the following regression:

Log(TB)=-35.82520 -.28647Y-.11370Y*+2.30724((P*/P)Log(ER))+e
[15.47301][.54657][.2.57417][.94041] [s.e.]

The negative value of b₁ shows that as domestic GDP increases, the trade balance deteriorates, which is in line with traditional international economic theory, as imports are expected to increase as domestic income rises. The value of b₂ is negative, which was unexpected, because as foreign income increases, they will demand more of Turkey’s products and exports should increase. The positive value of b₃ is of particular interest because it is of positive value and a greater value than the b₃for the long-term regression. Running a confidence interval for the value of b₃ in this short-term regression, I can say with 95% certainty that the true value of β₃ lies between .377518686 and .4.23696132. Therefore, it can be concluded that the derivative of a change in exchange rate with respect to trade balance is positive in the short-term. That the value of b₃ is greater than in the previous regression assumes that the trade balance was even more responsive to devaluation in the 1980s than over the long-term. Therefore, the devaluations and policy reforms of 1980 were relatively effective in turning the negative trade balance around. However, this does not support the theory of a J-curve, as the partial derivative ∂/∂ER[TB] is positive in the short and long terms.

The analysis of 1990-2006 shows similar results to the long-term regression model, but with a positive value for Y*, more aligned to traditional economic theory that exports should increase as foreign GDP also increases. The value for ∂/∂ER[TB] is again positive at 1.88340.

When testing the foreign direct investment changes with respect to exchange rate, the data obtained is estimated in the following regression:

Log(FDI)=65.83022-4.16918((P*/P) (ER))+e
[40.39278][2.92040] [s.e.]

The negative coefficient of b₁tells us that as exchange rate deteriorates, the foreign direct investment will also decrease. As the domestic currency is devaluated and is cheaper in real terms, there will be less investment and less confidence in the currency, thus making direct investments deteriorate as well. It seems reasonable that the number of devaluations of the Turkish Lira have created negative expectations for the rational investor, and even if exchange rates were to appreciate, investment would most likely be slow to recover because of the history and expectations of devaluation. The overall values of foreign direct investment have increased steadily for the years 1980 to 2006 (see Figure 2), showing an increasing confidence in the Turkish economy, despite the series of devaluations of the Turkish lira.

From testing the Marshall-Lerner condition, the import elasticity function that was used for the test gave the following result:


This function shows that as the price level changes by one unit value, the volume of imports are expected to decrease by .06581. When fitting the data to the export elasticity of demand function, I found the following equation:


This function tells us that the coefficient for the percent change in price with respect to the quantity exports is .04226. When this is added to the coefficient for the percent change in price with respect to the quantity of imports, the absolute values for the elasticities add up to .10807. Since this value is not greater than 1, the Marshall-Lerner condition is unsatisfied and according to theory, the devaluation of currency will not lead to a long-term trade surplus. Although short-term elasticities were not analyzed, it should be assumed that the long-run elasticities of demand for imports and exports are more accurate than short-term, as elasticity is dependent upon time and the availability of substitute goods. However, this is not to say that the elasticity of demand for imports and exports were not larger during the time of devaluations, such as the sharp devaluations of 1980, 1994, and 2001.

Through this primitive study, it may be assumed that the trade deficit for Turkey has been narrowing throughout the past 27 years. Devaluation of the Turkish lira has been shown to have a positive relationship between trade balance, even though the Marshall-Lerner condition was not satisfied. If the Marshall-Lerner condition was in fact satisfied, perhaps Turkey might have been able to free itself from a perpetually negative trade deficit. It may also be concluded that investors are less likely to be interested in the Turkish economy if it is devaluating its currency. Expectations of devaluation are also likely to decrease the confidence in the Turkish lira and in investment in the country—as shown by the larger value of ∂/∂ER[TB] when devaluations were effective in 1980, rather than a smaller value of ∂/∂ER[TB] when the devaluations could have been expected and therefore had a lesser positive effect on trade balance. However, because of the results of the f-test performed on the regression, there may be other facts affecting trade balance that should be considered.

The Kulkarni hypothesis holds loosely for these results. The trade balance was perpetually negative, perhaps due to the subsequent devaluations of the lira as Kulkarni suggests, but devaluations had a positive effect on trade balance in the long-run.

Units Measurements Used:
GDP For all countries= Billions/National Currency
CPI=Unit Measure
ER=Purchasing power of Turkish Liras/SDR (market rate, end of period)
Exports and Imports=Millions/USD

Ardic, Oya Pinar. “Output, The Real Exchange Rate, and The Crises in Turkey”. Accessed 16 April 2008. MPRA, online source. .
Balance of Payments Statistics Part I: Country Tables. International Monetary Fund Publication: Washington DC, 2005
Hale, William. The political and economic development of Modern Turkey. St. Martin’s Press: New York, 1981.
Halicioglu, Ferda. “The Bilateral J-Curve: Turkey Versus her 13 Trading Partners”. Accessed 3 April 2008. MPRA, online source. .
Husted, Steven and Michael Melvin. International Economics: 7thEdition. New York: 2007.
Kimbugwe, Hassan. “The bilateral J-Curve between Turkey and her 9 Trading Partners”. Accessed 22 April 2008. MPRA, online source. .
Kulkarni, Kishore G. Readings in International Economics. Serials Publications: New Delhi, 2006.
Macdonald, Ronald and Mark P. Taylor. Exchange Rates and Open Economy Macroeconomics. Basil Blackwell: Cambridge, 1989.
OECD Economic Surveys 1994-1995: Turkey. Organisation for Economic Co- Operation and Development, 1995.
Onis, Ziya and James Riedel. Economic Crises and Long-Term Crowth in Turkey. The World Bank: Washington, DC, 1993.
Rose, Andrew K. “Exchange Rates and Trade Balance: Some Evidence from Developing Countries”. Economics Letters, Volume 34, pp 271-275. 1990.

No comments: