Long-Term Relationship: Interest & Exchange Rates - Fisher Effect Analysis, Summaries of Productivity and Time Management

Download Long-Term Relationship: Interest & Exchange Rates - Fisher Effect Analysis and more Summaries Productivity and Time Management in PDF only on Docsity! INTERNATIONAL FISHER EFFECT: A REEXAMINATION WITHIN THE CO-INTEGRATION AND DSUR FRAMEWORKS A THESIS SUBMITTED TO THE GRADUATE SCHOOL OF SOCIAL SCIENCES OF MIDDLE EAST TECHNICAL UNIVERSITY BY EDA ERSAN IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF BUSINESS ADMINISTRATION IN THE DEPARTMENT OF BUSINESS ADMINISTRATION DECEMBER 2008 Approval of the Graduate School of Social Sciences Prof. Dr. Sencer Ayata Director I certify that this thesis satisfies all the requirements as a thesis for the degree of Master of Business Administration.. Prof. Dr. Cengiz Erol Head of Department This is to certify that we have read this thesis and that in our opinion it is fully adequate, in scope and quality, as a thesis for the degree of Master of Business Administration. Assist. Prof. Dr. Seza Danışoğlu Supervisor Examining Committee Members Assoc. Prof. Dr. Nuray Güner (METU, BA) Assist. Prof. Dr. Seza Danışoğlu (METU, BA) Assoc. Prof. Dr. Burak Günalp (Hacettepe., ECON) v ÖZ ULUSLARARASI FISHER ETKISI: EŞBÜTÜNLEME VE DİNAMİK GÖRÜNÜŞTE İLİŞKİSİZ REGRESYON YÖNTEMLERİ DAHİLİNDE YENİDEN İNCELEME Ersan, Eda Yüksek Lisans, İşletme Bölümü Tez Yöneticisi: Yrd. Doç. Dr. Seza Danışoğlu Aralık 2008, 68 sayfa Uluslararası Fisher Etkisi (UFE), ülkeler arasındaki geçerli kurun bu ülkeler arasındaki faiz oranı diferansiyeliyle zıt yönde hareket etmesi gerektiğini savunan bir uluslararası finans teorisidir. Bu tezin amacı, ülkeler arasındaki nominal faiz oranlarındaki farklılıkların ve bu ülkelerin para birimleri arasındaki geçerli kur hareketinin uzun dönemde birlikte hareket etme eğiliminde olup olmadığının analiz edilmesidir. UFE’nin varlığı 1985:1 - 2007:12 dönemleri arasında G-5 ülkelerinde ve Türkiye’de test edilmiştir. Johansen eşbütünleme yöntemi ile uzun dönem ilişki tahmin edilmiş ve tüm ülke çiftleri için destekleyici kanıt bulunmuştur. Buna ilave olarak, ayrı ayrı modellenmiş denklemler, Dinamik Görünüşte İlişkisiz Regresyon (DGİR) yöntemiyle test edilmiştir. Türk para birimini içeren DGİR denklemleri, Türkiye lehine yüksek faiz oranlarının Türk Lirası’nda değer kaybına yol açacağına yönelik, UFE’ni destekleyici, kanıt sunmuştur. Bu etkinin şiddeti beklendiğinden daha küçük olmuş ve bu da ekonomide kur hareketlerini etkileyen, enflasyon oranı gibi, başka faktörlerin de olabileceğini göstermiştir. Anahtar Kelimeler: Uluslararası Fisher Etkisi, Dinamik Görünüşte İlişkisiz Regresyon, Eşbütünleme, Karşılanmamış Faiz Paritesi vi To My Dear Mother and Grandmother vii ACKNOWLEDGMENTS This study is accomplished with precious contributions of many people whom I am very pleased to thank for sharing both joy and sorrows of completion of this thesis. I would like to thank my thesis supervisor Assist. Prof. Seza Danışoğlu for her guidance, assistance, advises and criticisms during the long and difficult phases I faced to complete this study. Also I would like to express my gratefulness to other members of the examining committee, Assoc. Prof. Dr. Z. Nuray Güner and Assoc. Prof. Dr. Burak Günalp and for their valuable comments and suggestions. I owe a thank to my instructors at Bilkent University, especially to Prof. Dr. Hakan Berument for igniting and encouraging my interest in macro economy and for teaching me the fundamentals of econometrical analysis. I would like to appreciate the tolerance of my colleagues at Deloitte, during difficult and long working hours that we worked together on field assignments. I am grateful to my dear friends Mine and Serkan for their psychological support during the hard times I faced for the completion of my thesis. I would like to thank them for motivating and supporting me during this period. And finally I wish to express my deepest gratitude to my beloved family for their patience, endless support and for their belief in me. Without their encouragements this thesis would not have been completed. 1 CHAPTER 1 INTRODUCTION Investment is the commercial way of using money with an aim to obtain profit in return. A crucial point for investors is predicting the forthcoming returns of their investment. Investors usually have to compare different investments on the basis of their risks and return potential in order to make a choice from among different alternatives. When the decision is about an international investment opportunity, the comparison of investment returns becomes a lot more complicated. In addition to the risk and return comparisons, the investors also have to take into account the potential changes that they expect to occur in the exchange rate between the foreign currency and the domestic currency. In such a case, theories regarding the determination and interaction of interest and exchange rates come into the picture. Many economic theories have been developed to analyze how macroeconomic variables like interest and exchange rates move through time when exterior interventions to financial markets are minimized. These theories simply reveal that the interactions between international markets should be analyzed carefully concerning the expected responses of such macroeconomic variables. One of these theories is derived from the well-known Fisher Effect, which asserts real interest rates across countries are equalized when nominal interest rate differences are driven by the discrepancies in inflation rates across countries. In other words, according to the Generalized Fisher Effect, high inflation countries should bear higher interest rates, if perfect integration of capital markets is achieved by financial markets. When the main question is the international equilibriums, in addition to interest rates and inflation rates, movements of exchange rates are also 2 important. Many theories have been developed to explain the changes in exchange rates. The Purchasing Power Parity condition is one of these theories. The theory implies that the exchange rates will move to offset changes in inflation rate differentials. The rationale behind this parity theory is the equality of prices across countries. Another international market equilibrium condition is the International Fisher Effect, which can be defined as a combination of the Generalized Fisher Effect and the Purchasing Power Parity. Briefly, the theory asserts that the higher interest rate country’s currency is expected to depreciate until the real returns of investments are equalized across countries. The aim of this thesis is to analyze the responses of exchange rates to the differences in interest rates across countries. In an open market, changes in exchange rates will affect the future value of current investments. Therefore, this analysis is important for investors, as they are concerned about future earnings while making decisions about investments today. Theoretically, in order to analyze the relationship between interest rates and exchange rates, parity conditions are developed in the international finance literature. In line with the theory, this thesis evaluates the relationship between exchange rates and interest rates of six selected countries, namely the G-5 countries and Turkey. The theory is evaluated with selected countries for the period starting from January, 1985 and ending with December, 2007. It is expected to find empirical results in favor of the International Fisher Effect so that the higher/lower interest rate country’s currency is expected to depreciate/appreciate in order to equalize interest rates across countries. Since the financial data used are not stationary, the traditional regression models are not appropriate to test this relationship. Instead, the relationship between interest rate differentials and changes in exchange rates is analyzed within the co-integration framework. The Johansen co-integration method is applied to the data and at least one co-integration vector is found for each of the county pair, which is a result indicating that interest rate differentials and exchange rate movements drift together in the long-run. 3 Econometric analysis in this thesis aims to examine the effectiveness of international markets to respond to the capital flows and the relationship between macroeconomic variables of developed countries among themselves and their interaction with the Turkish financial markets. The main reason for selecting the G- 5 countries is the integration of the capital markets in these countries. In general, the capital markets of developed countries are accepted to be integrated so that the basis for international parity conditions is established. In addition, governments in developed countries are less likely to intervene in financial markets. Currency restrictions and other governmental interventions affect the adjustment of exchange rates over time and inhibit capital integration. Therefore, testing for the existence of parity conditions has to be carried out in markets where such market imperfections are deemed to be at a minimum. The following chapter reviews the literature and explains the underlying theories about the relationship between exchange rate adjustments and interest rate differentials. In Chapter 3, the data used for the analyses and the research methodology used to test the relationship between exchange rate movements and interest rate differentials are described. In addition, the data periods and formation of sub- periods are explained. In Chapter 4, the results and analyses are presented. EVIEWS outputs are rearranged in tables and results obtained from the tests are interpreted. Finally, the main findings of the research and implications for further studies are summarized in Chapter 5. 6 rates constant over time. This conclusion is known as the Fisher hypothesis/effect and has been debated for several years in the literature. In his study, Fama (1975) questions the Fisher hypothesis by using rational expectations assumptions while pointing out the characteristics of an efficient market which uses all relevant information in setting prices. According to Fama, in an efficient market, if the inflation rate is to some extent predictable, there will be a relationship between nominal interest rate observed at a point of time and rate of inflation subsequently observed. In order to test his assertion, he tests the efficiency of the US Treasury bill (T-bill) market by using 1- to 6-month maturity T-bills and the CPI during the postwar period between January 1953 and July 1971. Results suggest the existence of a definite relationship between nominal interest rates and rate of inflation. During the sample period, the nominal interest rates summarize all the information about future inflation rates. This observation leads to the conclusion that the T-bill market seems to be efficient in the sense that short term interest rates are good predictors of inflation. Another major conclusion of the paper is that during this period, equilibrium expected real returns on T-bills are found to be constant. As a result of these findings, Fama concludes that the US T- bill market is efficient in setting 1- to 6-month nominal interest rates. In other words, the market correctly uses all information contained in the time series of past inflation rates to form expectations for future inflation rates. The Fisher hypothesis has been studied for many different time periods and contradictory results are reached in different studies. The Fisher effect has been found to be strong in some countries, for instance in US, Canada and UK, during the postwar period until the late 1970s. However, the same consistent relationship between interest rates and expected inflation is not observed in other countries. Hence, several studies have attempted to examine the reason behind why Fisher effect holds for some countries while it does not for others. Mishkin (1992) analyzes the reason for obtaining different results over different sample periods. He uses monthly data on inflation rates calculated from CPI series 7 and 1- to 12-month US T-bill rates during the postwar period between February 1964 and December 1986. The analysis provides no evidence for a short run Fisher effect during the postwar period. However, the existence of a long run Fisher effect implies that when inflation exhibits trends, there will be a strong correlation between inflation and interest rates. Therefore, the Fisher effect appears to be strong during the periods when interest and inflation rates both exhibit trends and he concludes that the Fisher effect is due to the existence of a stochastic trend between interest rates and inflation. Mishkin also shows that between 1979 and 1982 period and during the pre-World War II period, no Fisher Effect is observed in the US data due to the nonexistence of this stochastic trend. Hence, the evidence in the Mishkin study suggests that the validity of the Fisher effect depends heavily on the period considered and that the Fisher effect is most apparent in periods when there is strong evidence for stochastic trends. Other studies in the literature also suggest that country selection is another critical factor in the empirical testing of the Fisher effect. One of the recent multi-country Fisher effect studies is done by Berument and Jelassi (2002). In his study, Berument and Jelassi test the existence of a long run Fisher effect by taking into consideration the short-run dynamics of the interest rates for 12 developed and 14 developing countries. Treasury bill rates from these countries are used for the tests. Whenever T-bill rates are not available, the lending rate is used instead. The authors use monthly data in order to avoid the aggregation bias problem which can occur with annual data. The inflation rate is measured by the logarithmic first difference of the CPI. The strong form of the Fisher hypothesis is examined and it is concluded that the short-run responses of the nominal interest rate to expected inflation do not display a consistent pattern. For some developing countries, the short term adjustment of the nominal interest rate to expected inflation is more than proportional. In contrast, for developed countries, the short run adjustment of the nominal rate to expected inflation is always less than proportional. The empirical results suggest a point-for-point relationship between nominal interest rates and expected inflation for 16 out of 26 countries and the conclusion of the existence of the Fisher hypothesis holds more for developed countries than developing ones. 8 The Fisher effect is a hypothesis for domestic interest rates. An extended version of this hypothesis is the Generalized Fisher Effect (GFE) which considers the interactions between countries and states that real returns are equalized across countries through arbitrage. Within the context of the GFE, the nominal interest rate differential between two countries is equal to their anticipated inflation differential. e f e hfh ii ππ −=− (2) In Equation (2), hi and fi are home and foreign nominal interest rates, respectively, and e f e h ππ − denotes the anticipated inflation differential between the two countries. The theory suggests that the higher inflation rate country should bear higher interest rates relative to the lower interest rate country so that, in the absence of government intervention, capital flows towards the higher expected return country until expected real returns are equalized. Capital market integration is an important condition for the GFE since there should not be any restrictions on capital mobility so that capital is can flow freely across borders. In the literature, most empirical studies concerning the GFE have primarily focused on developed countries. Most of these studies find evidence that does not support the existence of a GFE. However, some studies are able to present supportive results when the GFE is analyzed over longer time periods. The study by Al- Khazali and Osamah (2004) test for the existence of the GFE by looking at the relationship between inflation and common stock returns in nine Asian countries: Australia, Hong Kong, Indonesia, Japan, South Korea, Malaysia, the Philippines, Taiwan, and Thailand. They use monthly short-term interest rates (Treasury bill rates or deposit rates) and monthly changes in the CPI as proxies for expected inflation in the individual countries between January 1980 and December 1994. 11 In a later study, Webster (1987) analyzes the PPP relation by using various industry data from the 1970s for US and UK. The evidence in this study does not support the PPP but indicate a high degree of sensitivity for the exchange rates in the face of inflation differentials. Although the results are not statistically significant, Webster concludes that the goods that have significant trading volume between countries are associated with more rapid price adjustments. For this reason, the author argues, the insignificant results should not be interpreted as an argument against the PPP. In a more recent study, Taylor and Taylor (2004) analyze the PPP debate in a wider perspective and summarize finding from previous studies. The PPP predicts that internationally traded goods should have the same price anywhere in the world once the price is expressed in a common currency. Authors mention that the term “purchasing power parity” was introduced after the World War I, during a period of high inflation rates in industrialized countries. Many empirical studies have been conducted to analyze the relationship between price levels in different countries and special indexes were calculated and published in order to compare the prices of similar/identical goods in different countries. One of the most famous price indexes is the Big Mac index. According to the values of this index, when expressed in common currency prices, an identical hamburger sells at different prices in different countries. The usual explanation for this observation is the inclusion of non-tradable input prices like wages and property rental costs in hamburger prices in different countries. Since these inputs cannot be traded internationally, their prices are determined locally and this may cause deviation from the law of one price. The studies that do not provide supporting evidence for the PPP typically highlight the importance of some of the irrational assumptions of the theory, such as zero transaction costs. In real world, the presence of transportation costs, taxes, tariffs and non-tariff barriers are likely to cause violations of the PPP. Besides, Taylor and Taylor point out that data selection is very important. Since the PPP is based on what 12 traded goods, it might be more reasonable to use the producer price index (PPI) rather than the consumer price index (CPI). The PPI tends to contain the prices of relatively more manufactured goods while the CPI might reflect the prices of relatively more non-tradable goods and this may make the PPI a better measure of price level changes within the context of PPP. Also, empirical evidence obtained from British and American data shows that price levels, expressed in common currency, tend to move together in the long run and the correlation between two national price levels is much greater with PPI compared to CPI. Another recent study about the PPP is carried out by Jacobson et al. (2008). This study analyzes the period from 1974 to 1999 and the countries UK, Germany, France and Italy. The results indicate that the theoretical strong PPP relationship does not hold for all of the analyzed European countries. However, the authors are able to find evidence of co-integration between nominal exchange rates and prices. Although the strong form of the PPP is rejected, the overall panel estimated co-integrated vector supports the theory. Authors conclude that this result may be a reasonably accurate approximation of how nominal exchange rates and price levels evolve over time. 2.4 The International Fisher Effect Hypothesis Throughout Sections 2.2 and 2.3, two key international finance relationships are analyzed. In this section, the literature about the International Fisher Effect and the parity conditions is reviewed together. The International Fisher Effect (IFE) is a theory which should be considered as a combination of the Purchasing Power Parity (PPP) and the Fisher Effect (FE). The Fisher theory simply argues that real interest rates across countries will be equal due to the possibility of arbitrage opportunities between financial markets which generally occurs in the form of capital flows. Real interest rate equality implies that the country with the higher interest rate should also have a higher inflation rate which, in turn, makes the real value of the country’s currency decrease over time. 13 It is important to note that two crucial assumptions are made for the IFE to hold. First, investors view foreign and domestic assets as perfect substitutes, and, therefore, no risk premium is postulated by investors. Second, capital markets are perfectly integrated with no regulatory and psychological barriers so that free flow of capital is achieved across countries. Figure 2.4.1 Figure 2.4.1 summarizes the relationship between the three theories. In the figure, e%∆ denotes the change in spot exchange rate, e f e h ππ − is the difference in inflation expectations between the home and foreign country, and e f e h ii − is the difference in nominal interest rates between the home and foreign country. Generalized Fisher Effect asserts that nominal interest rate differences are caused by differences in inflation expectations. Besides, if PPP holds, inflation differentials should be offset by exchange rate changes. In conclusion, the IFE hypothesis states that, if real interest rates are equal across countries, the interest rate differential between two countries is an unbiased predictor of the future changes in spot exchange rates. At this point, it should be noted that this does not mean that the interest rate differential is a precise estimator of exchange rates but e%∆ e f e h ππ − IFE FE PPP e f e h ii − 16 ),,(),,( FYTLTVDYTLTV = (8) The equality in Equation (8) predicts that real returns across countries will be equalized. Exchange rate movements will cancel out the interest rate differentials. In other words, the currency of the high interest rate country will depreciate through time to equalize the expected uncovered returns in all markets. As previously explained, the IFE is a combination of the Purchasing Power Parity and the Fisher Effect. This predicted adjustment of exchange rates to interest rate differentials has been questioned in various empirical studies from diverse perspectives. In his empirical study, Throop (1994) mentions the importance of financial market integration for the IFE hypothesis and analyzes the real interest rate equality across five industrialized countries during the period of international integration of financial markets. The author observes that the integration of financial markets has increased dramatically starting from the 1970s after the collapse of the Bretton Woods system1 and government imposed barriers to international flow of capital between industrialized countries were reduced and mostly eliminated by 1980s. This integration further caused a decrease in the ability of central banks to influence their own national macroeconomic parameters, such as the nominal interest rates. Throop tests the relationship between the real interest rates of US, Canada, Germany, UK and Japan during the period between 1981 and 1994. He explains the well-known Mundell Fleming Model2 (MFM), which is essentially an extension of the IS-LM model. The MFM asserts that under a flexible exchange rate system 1 Bretton Woods is an international economic agreement signed between 44 nations after World War II in 1944. Briefly, it was a system in which each country adopted a monetary policy of fixing the exchange rate of its currency in terms of the US dollar and gold. The fixed exchange rate system collapsed in early 1970s and a system of mostly floating exchange rates was adopted 2 The Mundell-Fleming Model is also known as the Unhold Trinity, which simply asserts that only two out of (1) free capital flow (2) fixed exchange rate (3) independent monetary policy, can be attained at the same time. It is impossible to have all three at the same time. 17 with static exchange rate expectations and perfect capital mobility, nominal rates are expected to be equalized continuously. Throop indicates that after the 1980s barriers to capital flow were eliminated for the majority of industrialized countries and in such an environment only the currency risk was left for similar investments in different countries. Therefore, according to author, the expected change in the value of the currency is the source of the differences between the real rates of return on similar assets in different countries. Throop’s results indicate no causal linkage between the US and foreign real rates during the period between 1981 and 1994. This evidence is interpreted by the author to imply that central banks have been able to influence their domestic interest rates quite independently from the influence of interest rates abroad. In another study Juntilla (2001) extends the traditional Fisher equation through international direction by introducing foreign interest rates and exchange rates into the standard Fisher equation. He first tests the traditional Fisher hypothesis for Finland and the results do not support the existence of a Fisher effect for monthly Finnish data for the period between 1987 and 1996. Second, by taking the international dependencies into account, the author tests the dependencies between the Finnish interest rates and rates from its close trade partners, Germany and US. Results of the tests provide supportive evidence for a positive long run relationship between nominal interest rates and inflation in Finland, and, moreover, tests of the augmented version of the Fisher equation indicate that the Finnish money market is not independent of the markets of those countries with which Finland has a high trading volume. Mishkin (1984) analyzes whether real rates are equal across countries. Data from US, Canada, UK, France, West Germany, Netherland and Switzerland are used for the period between 1967 and 1979. Mishkin tests for this equality by using the international parity conditions. UIP and PPP hypotheses are tested independently and jointly, where the joint testing of the parity conditions means testing for the equality of real rates. When individually tested, neither the PPP nor the UIP are rejected. However, when these hypotheses are tested jointly, the null hypotheses what 18 are rejected which indicates that the joint results are not supportive of real rate equality across countries. Therefore, the basic conclusion of Mishkin is the same as that of Throop’s (1994). The equality of real rates is statistically rejected across countries, which leaves the possibility for central banks to control their interest rates independent from other countries. Mishkin’s explanation for the rejection of equal real rates is that the underlying assumptions of the theory may not hold in the real world. For example, the marginal tax rates on interest payments might differ across countries. Also, assuming no transaction costs is unrealistic. According to Mishkin, real rates can differ across countries because risk premiums asked in the forward exchange market and securities denominated in different currencies are not perfect substitutes of each other. The empirical results and their interpretation are often contradictory when the equality of real rates is analyzed across countries. Although Mishkin (1984) and Throop (1994) have rejected real rate equality, contradictory results are also present in the literature. For instance, Roll (1979) argues that in efficient capital markets, expected real interest rates should be equalized across borders with the assumption of perfect integration and the strict presence of the PPP. It is important to mention that in his hypothesis, Roll assumes homogeneous goods/assets which are perfect substitutes of each other and they are traded with no costs across countries with perfect capital mobility. In a more recent study, Minford and Peel (2007) question the findings of Roll (1979). Overall, their empirical findings are not supportive of Roll’s hypothesis of real rate equality. According to Minford and Peel, rejection of equality does not imply that the asset markets are inefficient but rather that the theoretical conditions that have to be met and the underlying assumptions should be evaluated. Wu (1999) examines the relationship between the exchange rate and interest rate differentials for Japan and Germany against the US for the period between 1974 and 1996. In order to test for a long run relationship, Johansen’s co-integration test 21 interest rates. The results are supportive of the uncovered interest parity within this high frequency data. . Interestingly, if the data frequency is extended for even a few hours, it is no longer possible to find evidence of the uncovered interest rate parity condition. Most of the IFE studies are conducted on data from industrialized countries. It may be plausible to expect that this relationship may not hold in emerging countries with high levels of inflation. In these countries, it may take longer for the interest rates to adjust to unanticipated inflation. As a result of the uncertainty, high inflation countries tend to invest more in inflation forecasts, and, therefore, may have a greater incentive to incorporate inflationary expectations in required returns and this might cause the exchange rate adjustments to take longer. In addition, emerging economies might be affected more from developments in industrialized countries. Within this context, Özmen and Gökcan (2004) test the validity of the PPP and the UIP by using Turkish and US data for the period between 1986 and 1999. Findings are consistent with some of the earlier studies which reject the pure parity conditions. The evidence in this study suggests that neither the PPP nor the UIP can be valid by itself for the Turkish data. According to the authors, disequilibrium in international commodity markets may affect international asset markets. In open economies, policy makers should consider the importance of the interaction of PPP and UIP while setting targets for the exchange rates. In another study which uses the Turkish data, Saatçioğlu and Korap (2007) investigate the empirical validity of the uncovered interest rate parity. Econometrical results obtained from the Johansen co-integration tests are in favor of the UIP hypothesis in the long run for the Turkish economy. Positive interest rate differentials which are in favor of domestic interest rates cause nearly a one-to- one increase in the expected exchange rates. Just like Özmen and Gökcan (2004), Saatçioğlu and Korap mention that the policymakers should consider spot exchange rate movements while setting targets in open economies. 22 2.5 Conclusion The International Fisher Effect (IFE) asserts that differences in nominal interest rates between two countries would determine the movement of the nominal exchange rate between their currencies. As explained in Sections 2.2 and 2.3, the IFE should be considered as a combination of the two well-known international finance theories: the Fisher Hypothesis and the Purchasing Power Parity. The PPP suggests that there is a one- to-one relationship between the inflation differential of two countries and the percentage change in the spot exchange rate of their currencies over time. The IFE suggests that there is a relationship between the interest rate differential of two countries and the percentage change in the spot exchange rate over time. The IFE is based on nominal interest rate differentials which are, in return, influenced by expected inflation. Thus, the IFE is closely related to the PPP. Typically, the inflation expectation is higher in countries that have higher interest rates and this is expected to cause the depreciation of such a country’s currency against the currencies of lower interest rate countries. Throughout the literature, it is generally argued that evidence that supports the existence of the IFE can be found only if the underlying assumptions are met in the markets. The crucial assumptions of the IFE can be summarized as follows: • Perfect mobility of capital without any regulation or restriction by government institutions, indicating that trade barriers are eliminated. • Investors are risk neutral, therefore, no risk premium is asked by them. • Zero transaction costs, with no psychological barriers and transportation costs, so that investors are indifferent between countries. 23 Time period under analysis is also important because the nature of the exchange rate regimes is crucial for the IFE hypothesis. For example, under a fixed or pegged float, exchange rates are not allowed to move freely. Rather, they are either set fixed or are allowed to fluctuate within a pre-announced band. Since exchange rates are not determined freely in the financial markets, adjustments necessary for the parity conditions to hold cannot occur. The level of financial development in a country is another important factor which effects capital mobility. Since institutional development is complete in developed countries, it is less likely for the authorities of these countries to restrict mobility of capital. The literature also presents evidence that political risk and psychological factors play an important role in the determination of exchange rates. Moreover, it is widely argued that emerging markets are affected more from developments in industrialized countries. The empirical evidence about the relationship between interest rate differentials and exchange rates is inconclusive since there are many studies that support or oppose the International Fisher Effect hypothesis. The reasons for these different results can be summarized as below: 1) The macro-economic characteristics of the period under analysis are important. Exchange rate targeting policies enforced by regulatory bodies vary through time. The external influences in the market are examples of violation of the assumptions mentioned above. 2) Financial development and capital market integration of countries under analysis are also important factors that affect the results of the empirical studies. Literature shows that close trading partners have more influence on each other’s financial markets. 3) The type and frequency of data used and other variables (such as current accounts, inflation differentials etc.) that are included in regression analyses are also important factors. 26 3.2 Data 3.2.1 Exchange Rate Data The theoretical definition of an exchange rate is the price of one currency against another currency. The exchange rate data used in this study are taken from the databases of relevant country central banks on the last business day of each month. Prior to 1999, Deutsche Mark and French Franc are used for Germany and France, respectively. Since Germany and France are member states of the European Union (EU), the official currency of these two countries became Euro as of January 1, 1999. For this reason, starting in 1999, Euro is used as the currency for Germany and France. Data on the historical values of the exchange rate between French Franc and Japanese Yen could not be found; therefore, this exchange rate is excluded from the analysis. 3.2.2 Interest Rate Data In the finance literature, Treasury bills are accepted as the risk free investment tools, and, therefore, it is assumed that Treasury bills in different countries are perfect substitutes of each other. All interest rates, except for the Turkish interest rate, are taken from the International Finance Statistics (IFS) database. For countries other than Turkey, interest rates used are the 90-day Treasury bill rates for the period between January, 1985 and December, 2007. For Turkey, the compound interest rates of Treasury bills that are traded on the İstanbul Stock Exchange (ISE) and that have a remaining maturity closest to 90 days are used. The Turkish treasury bill data, covering the period 1995:1 to 1999:12 are obtained from the ISE monthly bulletins. For the period 2000:1 to 2007:12, the compound interest rates are downloaded from the electronic database of the ISE. Since the interest rate 27 data for Turkey goes back to only 1995, instead of 1985, in order to avoid two short sub periods, Turkey is included in the analyses starting in January, 1999. As explained in Section 3.2.1, starting in January, 1999, Euro is used as the currency for Germany and France. For this reason, in addition to the individual interest rates of Germany and France, the 90-day LIBOR rate for Euro, which is published by the British Bankers’ Association4, is also used for the German and French interest rate as an alternative test. 3.2.3 The Sample Period Monthly data are used for both exchange rates and interest rates. The interval of analysis covers the period from January, 1985 to December, 2007. The analyses are performed over three sub-periods: (1) January, 1985 to December, 2007 (2) January, 1985 to December, 1998 and (3) January, 1999 and December, 2007. The rationale behind the three sub-periods is as follows: The first sub period data set includes three countries: USA, UK and Japan. The data of these three countries do not include any structural breaks, and, therefore, this period allows for testing the hypotheses in the study over the longest time period possible. After January, 1999, due to the adaptation of Euro as the official currency by the EU countries, a structural break exists in the data. The first period is between January, 1985 and December, 1998 and this period is uses the German Mark and the French Franc. The second period is between January, 1999 and December, 2007 and this period is analyzed with Euro. In addition, the January, 1999 - December, 2007 period is analyzed with two different interest rates. First, 4 For detailed information please refer to website at www.bba.org.uk 28 individual interest rates of Germany and France are used. As an alternative test, the 90-day LIBOR for Euro is used instead of the German and French interest rates. Throughout the thesis, the following abbreviations are used for countries and their corresponding currencies: Country Currency Turkey TR Turkish Lira TL Japan JP Japanese Yen Yen France FR French Franc FF United Kingdom UK Sterling Sterling Germany GER Deutsche Mark DM United States of America USA United States Dollar USD Figure 3.2.3.1 TR (TL) GER (DM) FR (FF) JP (YEN) USA (USD) UK (STERLING) DM&FF (Euro) TR- (TL) - - - x x x x GER-(DM) - xx xx xx xx - FR-(FF) - NA xx xx - JP-(YEN) - xxx xxx x USA-(USD) - xxx x UK-(STERLING) - x Euro (DM&FF) - Figure 3.2.3.2 – Analyzed County Pairs Figure 3.2.3.2 summarizes the country pairs analyzed for exchange rate and interest rate differentials. “NA” denotes the unavailability of data between Japan and France. Regression equations are estimated over the three sub periods: X: January, 1999 - December, 2007; XX: January, 1985 - December, 1998; XXX: January, 1985 – December, 2007. 31 non-stationary process. If exchange rates ( )te and interest rate differentials tfh ii )( − are proven to be non-stationary, it would be difficult to represent the relationship between the past and futures values of these variables within a simple algebraic model. If two such random variables are regressed on one another, the results can be misleading in that conventional significance tests may tend to indicate a relationship with a high R2 and a low Durbin-Watson statistic even though no true relationship exists between the variables5. Such cases create the so- called spurious regressions in which two independent and unrelated time series are found to be related. In this thesis, two types of unit root tests are applied. The Dickey-Fuller test, developed by D. A. Dickey and W. A. Fuller in the 1970s, is the most common test used to determine whether a unit root is present in an autoregressive model. First, the Augmented Dickey Fuller (ADF) statistic, which is developed by E. Said and A. Dickey (1984), is used to test the unit root characteristics of interest rate differentials and exchange rates. Second, the Kwiatowski, Phillips, Schmidt, and, Shin (KPSS) test is performed for the same purpose. 3.3.1.1 Augmented Dickey-Fuller (ADF) Test Unit Root test distributions are affected by the inclusion of deterministic terms such as dummy variables, constant terms or time trends. Therefore, according to the nature of variables that are being tested, different test regressions should be used. For non-trending economic and financial series, a regression model with a constant term and without a trend would be appropriate. In this setting, the interest rate differentials and exchange rates can be modeled in the following manner in order to test whether these two series have a unit root, and, thus, are non-stationary: 5 A low value for the Durbin-Watson statistic means that there is a positive correlation between tested variables. 32 ttfhtfh uiiii +−+=− −11 )()( δα (1) In Equation (1), tfh ii )( − is the interest rate differential between home and foreign country at time t, α is the constant term, 1δ is the coefficient that represents the relationship between the current and one-period lagged values of the interest rate differential, and tu is the error term. ttt uee ++= −11δα (2) In Equation (2), te is the exchange rate between home and foreign country at time t,α is the constant term, 1δ is the coefficient that represents the relationship between the current and one-period lagged values of the exchange rate, and tu is the error term. In the following discussions, for notational parsimony, the interest rate differential and the exchange rate variables are both replaced with ty and the unit root testing steps are explained using the shorter notation: ttt uyy ++= −11δα (3) In Equation (3), tu is the white-noise error term with a zero mean and a constant variance: WN (0, 2σ ). In order to test the relationship between time t and t-1, 1−ty is subtracted from both sides of Equation (3): ttt uyy ++=∆ −1 * 1δα (4) In Equation (4), 1−−=∆ ttt yyy and 11 * 1 −=δδ . The existence of a unit root is tested with the following null hypothesis: 0: * 10 =δH 33 If the absolute value of the calculated ADF statistic is smaller than the critical value at the specified significance level, then the test statistic fails to reject the null hypothesis. Failing to reject the null means that the time series has the characteristics of a non-stationary series which is integrated of order 1 [I(1)]. This result also indicates that the time series tested has a unit root. The ADF unit root test is carried out for interest rate differentials as modeled in Equation (1). The log-difference command is used in E-Views™ for exchange rates to test the presence of unit root for changes in exchange rates. The mathematical logic of the log-difference is presented in Equations (5) and (6). )log()log()log( 1−−= tt eeed (5) 1 1 1 1 )( log)log()log( − − − − − ≈      =− t tt t t tt e ee e e ee (6) In Equation (6), 1 1)( − −− t tt e ee indicates the percent changes in spot exchange rates. 3.3.1.2 Kwiatowski Phillips Schmidt Shin (KPSS) Test The KPSS Test, which is proposed by Kwiatkowski et al. (1992), is an alternative analysis to test the presence of unit root in time series. The test differs from other unit root tests in a way that the analyzed series, ty , is assumed to be stationary, I(0), under the null hypothesis. )0(~ IyH to = For this reason, in order to confirm unit root by the KPSS test, the calculated statistics should be higher than the critical value. Therefore, contrary to the ADF test, the null hypothesis should be rejected to conclude for the presence of unit root in ty ~ I(1). 36 Equation (8) should be equal to 1. Thus, if the null hypothesis of β =1 cannot be rejected, then there is evidence in the time series which supports the unbiasedness hypothesis of IFE which argues that interest rate differentials are unbiased predictors of changes in the future exchange rates. After testing for cointegration between countries on a pair-by-pair basis, the same set of countries are also examined for the existence of cointegration on a panel basis. For this purpose, first, the panel of country series needs to be tested for the existence of a unit root. The following section explains this methodology. 3.3.3 Panel Unit Root Tests In Sections 3.3.1 and 3.3.2, country pairs are modeled individually and regression equations are tested separately for the unbiasedness hypothesis. In this section, the unit root hypothesis explained in Section 3.3.1 is tested by using panel data. It is widely agreed in the literature that panel-based unit root tests have higher power than individual unit root tests. The presence of cross-sections generates multiple series out of a single series and the unit root procedure is applied to panel data. Panel estimation brings additional information by evaluating cross-sectional dependencies between individually constructed regressions. In order to search for a cointegration relationship in the entire panel of countries, first the panel needs to be tested for the existence of a unit root. Panel unit root testing is carried out by estimating the following equation within the SUR (Seemingly Unrelated Regressions) framework: it p k kitijiitiiit yTyy ξδλαγ +∆+++=∆ ∑ = −− 1 1 (9) 37 In Equation (9), the p values are determined from the univariate ADF test results for each variable in the equation. Also, Ni ......,,2,1= where N is the number of cross sectional equations and Tt ......,,2,1= where T is the number of time series observations. The null hypothesis for the presence of a panel unit root is 0:0 =iH α . If the absolute value of the calculated ADF statistic is smaller than the critical value at the specified significance level, then the test statistic fails to reject the null hypothesis. In addition to the calculated test statistics, p-values can also be used to interpret the results of the tests. Failing to reject the null confirms the presence of a panel unit root. Once the panel unit root tests are completed, if supportive unit root results are found in panel data, the next step is to examine the cointegration relationship within the panel of countries. Alternatively, if the panel unit root is not confirmed, further panel cointegration testing is not performed. The following section describes the procedure for estimating the IFE equations for different countries as a system in order to test whether additional information is available across equations. 3.3.4 The DSUR Estimation In this section, individually modeled regressions are estimated as a system within the DSUR (Dynamic Seemingly Unrelated Regression) framework in order to examine the cross sectional dependence between the selected countries. The parametric DSUR estimator, proposed by Mark et al. (2005), is an econometric method to test for multivariate cointegration across countries and it tests for long- run cross sectional dependence in equilibrium errors. DSUR is applicable for panels where the number of cross sectional equations ( )N is substantially smaller than the number of time series observations ( )T . 38 Cross sectional panel data analysis incorporates additional information into the estimations by evaluating cross sectional dependencies between individually constructed regressions. One requirement of this analysis is strict exogeneity which means that the error term in each equation at a given point in time must be uncorrelated with the independent variable (the interest rate differentials) from all equations in the system. This may not be a very realistic assumption in this study since unexpected macroeconomic shocks in one country may affect the other countries considered within the system and this means that the forecasted error in each of the equations will be correlated with the interest rate differentials in the other equations. The SUR technique was first proposed by A. Zellner (1962) as a method for analyzing a system of equations simultaneously while allowing the examination of error correlations between equations. In other words, this model is designed to simultaneously test for the correlation relationship between the error terms within a system of independent equations. In this thesis, the unbiasedness hypothesis is tested within the DSUR framework based on Mark et al.’s (2005) methodology. In the first step, the exchange rate change over time is modeled as a function of the interest rate differential between the country pairs: ( ) ntitifhii t nti iie ++ +−+=∆ ,,, εβα (10) In Equation (10), t ntie +∆ , is the change in the spot exchange rate for country pair i between time t and t+n, tifh ii ,)( − is the interest rate differential for country pair i at time t, and the other variables are defined as before. For notational simplicity, the dependent and independent variables of Equation (10) are replaced by ity and * itx respectively and each equation Ni ,......,1= has a triangular representation as follows: 41 Finally, the DSUR estimator with a known covariance matrix ηηΩ can be formulated as follows: [ ]         Ω              Ω      =      − − − = ∑ t t t tt T t t t dsur dsur Y Z X ZX Z X 1 1 1 1 . ηηηηφ β (20) In Equation (20), all variables are defined as before. The null hypothesis to test for detecting panel cointegration requires an equality of the β coefficients across equations: NHo ββ == ........: 1 In their study, Mark et al. compare the performances of DSUR, SSE, System DOLS and DOLS estimators in an environment where the cointegrating vector exhibits heterogeneity across equations. They are able to show that DSUR exhibits moderate to strong efficiency advantages compared to the other estimators. The relative efficiency of DSUR is proven to increase for larger values of N (number of cross-sectional equations) and T (number of time periods). The advantage of the DSUR methodology over DOLS is also mentioned in a recent study by J. Kim (2007). In this study, real exchange rates and real interest rate differentials are examined within the DSUR framework. DSUR provides more precise estimates compared to DOLS by incorporating the long run cross sectional correlation in the equilibrium errors. DOLS method only includes leads and lags of the first difference regressors from the own equation but not from cross equations. The DSUR method corrects the endogeneity in equation i by incorporating leads and lags of not only the first difference of the regressors of equation i but also regressors of all other equations in the system. 42 Mark et al. state that Equations (11) and (12) exhibit an endogeneity problem in the regression estimates. In order to control for the endogeneity among the variables, Mark et al. suggest that both lead and lag values of the independent variable should be included in the regressions. However, there is no standard method for the choice of the number of leads and lags ( −+ ijij pp and ). Mark et al. state that generally an ad hoc rule is used to determine the proper −+ ijij pp and in empirical studies. According to this rule, p is set based on the number of time series observations such that 1=p for 50=T ; 2=p for 100=T ; 3=p for 300=T . The same rule is adopted in this thesis and three leads and three lags are used for the DSUR estimations. 3.4 Summary The methodologies described in this chapter all have the purpose of examining the effect of interest rate differences on the behavior of exchange rates. The sample consists of six countries: Turkey and the G-5 countries of United States, United Kingdom, Japan, France and Germany. For interest rate data, Treasury bill interest rates are collected since it is assumed that Treasury bills are risk free investment vehicles and investors view Treasury bills of different countries as perfect substitutes of each other. In this chapter, a four-step methodology is constructed to test whether the interest rate differentials are unbiased predictors of the future changes in exchange rates. These four steps and their purposes can be summarized as follows: 1) ADF and KPSS unit root tests are applied to exchange rate and interest rate differentials to determine whether the data exhibit non-stationarity. 2) Johansen cointegration test is applied to those series that are shown to be non-stationary. Cointegration tests are performed to test whether residuals 43 of linearly combined exchange rates and interest rate differentials are stationary over time. Stationary residuals are an indication of the existence of a long run relationship between exchange and interest rate differentials. 3) Individually modeled regressions are tested together to analyze cross- sectional dependencies. Panel data for exchange and interest rate differentials are examined for the existence of a panel unit root. If evidence of a panel unit root is found, then a panel cointegration test is performed in order to utilize the additional information that is assumed to be contained in the panel data. 4) As a last step, individually modeled regressions are estimated as a system by applying the Dynamic SUR method proposed by Mark et al. (2005). Three leads and three lags of the independent variable are used in each equation within the system. , The four-step procedure described above is applied to monthly interest and exchange rate value over three sub periods: (1) January, 1985 to December, 2007 for USA, Japan and UK; (2) January, 1985 December, 1998 for Germany, France, USA, Japan and UK; and, (3) January, 1999 to December, 2007 for Germany, France, USA, Japan, UK and Turkey. The next chapter presents the results of the analysis. 46 Table 4.2.3(A) - Unit Root Test Statistics for the 1999-2007 Period Calculated Test Statistics Exchange Rate Differentials ADF Test (3) KPSS Test (3) Currency: Euro-Turkey -5.4834† 0.5407 Euro-USA -5.7050† 0.3969 Euro-UK -6.1231† 0.2210† Euro-Japan -5.5369† 0.3862 Turkey-Japan -5.8154† 0.7939 Turkey-UK -5.1146† 0.6825 Turkey-USA -4.9407† 0.9071 Critical Values: 1% -4.0495 0.739 5% -3.4540 0.463 10% -3.1526 0.347 Interest Rate Differentials ADF Test (3) KPSS Test (3) Country pair: Turkey-France -3.1715 1.9146 Turkey-Germany -3.1636 1.9085 USA-France -1.8814 0.5221 USA-Germany -1.5757 0.5159 UK-France -1.8823 0.4966 UK-Germany -2.0746 0.4196 Japan-Germany -1.7856 0.8297 Japan-France NA NA Turkey-Japan -3.1118 1.9201 Turkey-UK -3.1263 1.9096 Turkey-USA -3.1416 1.9038 Critical Values: 1% -4.0486 0.739 5% -3.4536 0.463 10% -3.1524 0.347 47 Table 4.2.3(B) - Unit Root Test Statistics for the 1999-2007 Period Calculated Test Statistics Exchange Rate Differentials Interest Rate Differentials ADF Test (3) KPSS Test (3) ADF Test (3) KPSS Test (3) Country Pair: Euro (LIBOR)-Turkey -5.4834† 0.5408 -3.1733 1.9153 Euro (LIBOR)-USA -5.7050† 0.3969 -1.7909 0.4845 Euro (LIBOR)-UK -6.1231† 0.2210† -1.7864 0.4179 Euro (LIBOR)-JP -5.5370† 0.3863 -1.9137 0.7373 Turkey-UK -5.1147† 0.6826 -3.1264 1.9096 Turkey-USA -4.9408† 0.9072 -3.1416 1.9038 Turkey-JP -5.8155† 0.7940 -3.1118 1.9201 Critical Values: 1% -4.0495 0.739 -4.0486 0.739 5% -3.4540 0.463 -3.4536 0.463 10% -3.1526 0.347 -3.1524 0.347 ADF Test (p) and KPSS Test (p) denote the number of lags. The ADF is calculated with trend and intercept. The MacKinnon (1996) critical values for ADF and asymptotic critical values of KPSS at 10%, 5% and 1% significance levels are provided in the bottom three rows of the tables. NA notes that the Japan-France exchange rate could not be found Superscript † denotes rejection of the null hypothesis at all significance levels for the ADF Test and † denotes failure to reject null hypothesis for the KPSS Test. Tables 4.2.1, 4.2.2 and 4.2.3 report the test results for the presence of a unit root for the three sub periods. For the 1999-2007 period, Table 4.2.3(A) presents the results obtained with the German and French interest rates while Table 4.2.3(B) presents the results with the 3-month Euro LIBOR. Numbers presented in tables are the calculated test statistics and critical values are provided at the bottom of each table. As explained detail in the methodology chapter, two test statistics are presented for each time series: the ADF statistics which tests the unit root null and the KPSS statistics which tests the null of trend stationary. As a result of this difference in the way the null hypotheses are formed in each test, the ADF null hypothesis should be failed to be rejected and the KPSS null should be rejected in order to verify the presence of a unit root in the series. 48 In the tables, the calculated test statistics for interest rate differentials are smaller than the critical values for the ADF tests, and, therefore, the results verify the unit root for interest rate differentials for all country pairs in all sub-periods. Similarly, the calculated test statistics are greater than the critical values for the KPSS tests, and, therefore, these results also confirm the existence of a unit root. Unit root test results for exchange rate differentials are more diverse than those for interest rate differentials. ADF and KPSS tests results are inconsistent with each other in that ADF calculated test statistics are larger than the critical values for ADF tests. Therefore, unit root of exchange rate differentials cannot be verified with ADF test in all periods. When the KPSS test results are analyzed, it is seen that calculated test statistics for exchange rates of Euro-UK during the 1999-2007 period, Japan-Germany, UK- France, UK-Germany exchange rates during the 1985-1998 period, and the UK- Japan and USA-UK exchange rates during the 1985-2007 period are smaller than the critical values. These results indicate that the null hypothesis of stationarity cannot be rejected for these currencies, which further implies the rejection of a unit root in these particular exchange rate series. Since both the ADF and KPSS tests fail to verify unit root in the exchange rate differentials for these currencies these country pairs are not tested further in the cointegration framework. Figure 4.2.1 presents those country pairs that are tested for cointegration in each of the sub periods. The next section presents the cointegration results. 1985-2007 1985-1998 1999-2007 (A) 1999-2007 (B) USA-Japan France-Germany Turkey-Japan Turkey-Japan Germany-USA Turkey-USA Turkey-USA France-USA Turkey-UK Turkey-UK France-Turkey LIBOR-Turkey Germany-Turkey Germany-Japan LIBOR-Japan France-USA LIBOR-USA Germany-USA Figure 4.2.1 – Co-integration Country Pairs 51 Tables 4.3.1, 4.3.2, 4.3.3(A) and 4.3.3(B) report the summary of Johansen cointegration test results. Trace and Max-Eigen statistics for the null hypotheses of no cointegration (r = 0) are greater than the critical values for all of the country pairs in all sub periods. Therefore, the null hypothesis of no cointegration is strongly rejected. These results confirm the presence of at least one cointegrating vector for the analyzed country pairs. The evidence is summarized in Figure 4.3.1. One Cointegrating Vector Two Cointegrating Vectors Period: 1985-2007 Japan-USA X Period: 1985-1998 France-Germany X France-USA X Germany-USA X Period (A): 1999-2007 Turkey-France X Turkey-Germany X France-USA X Germany-USA X Japan-Germany X Turkey-Japan X Turkey-UK X Turkey-USA X Period (B): 1999-2007 EURO-Turkey X EURO-USA X EURO-Japan X Turkey-Japan X Turkey-UK X Turkey-USA X Figure 4.3.1- Summary of Co-integration Test Results 52 For all of the analyzed country pairs, at least one co-integrating vector is found. Presence of at least one co-integrating vector indicates that the residuals of linearly combined exchange rate and interest rate differentials are stationary over time. These results support the IFE theory which asserts that exchange rate movements and interest rate differentials drift together in long run. The following section presents the panel unit root test results. 4.4 Panel Unit Root Results Table 4.4.1- Panel Unit Root Results 1985-2007 6 Exchange Rate Differentials Interest Rate Differentials Method: Levin, Lin & Chu (LLC) -9.5587 (0.0000) 0.4201 (0.6628)* Breitung t-stat -3.3879 (0.0000) -1.6649 (0.0480) Im, Pesaran and Shin (IPS) -12.6329 (0.0000) 0.2040 (0.4192)* ADF - Fisher Chi-square 138.7770 (0.0000) 5.0105 (0.5425)* PP - Fisher Chi-square 401.0370 (0.0000) 6.2579 (0.3949)* Table 4.4.2- Panel Unit Root Results 1985-1998 7 Exchange Rate Differentials Interest Rate Differentials Method: Levin, Lin & Chu (LLC) -10.4372 (0.0000) 0.9660 (0.8330)* Breitung t-stat -4.7366 (0.0000) -0.6021 (0.2735)* Im, Pesaran and Shin (IPS) -12.1905 (0.0000) 1.2601 (0.8962)* ADF - Fisher Chi-square 152.449 (0.0000) 8.6282 (0.7343)* PP - Fisher Chi-square 482.761 (0.0000) 9.0989 (0.6945)* 6 Three cross sections are tested within the panel framework for both exchange and interest rate differentials: Japan-USA, UK-Japan and USA-UK. 7 Six cross sections are tested within the panel framework for both exchange rate and interest rate differentials: France-Germany, France-USA, Germany-USA, Japan-Germany, UK-France and UK- Germany. 53 Table 4.4.3(A)-Panel Unit Root Results 1999-20078 Exchange Rate Differentials Interest Rate Differentials Method: Levin, Lin & Chu (LLC) -13.3976 (0.0000) -1.2095 (0.1132)* Breitung t-stat -5.6382 (0.0000) -1.6948 (0.0451) Im, Pesaran and Shin (IPS) -11.2895 (0.0000) -1.3339 (0.0911)* ADF - Fisher Chi-square 136.617 (0.0000) 26.8465 (0.1396)* PP - Fisher Chi-square 336.536 (0.0000) 124.466 (0.0000) Table 4.4.3(B)-Panel Unit Root Results 1999-20079 Exchange Rate Differentials Interest Rate Differentials Method: Levin, Lin & Chu (LLC) -13.3976 (0.0000) -0.9845 (0.1624)* Breitung t-stat -5.6382 (0.0000) -1.3727 (0.0849)* Im, Pesaran and Shin (IPS) -11.2895 (0.0000) -1.4046 (0.0801)* ADF - Fisher Chi-square 136.617 (0.0000) 20.4789 (0.1158)* PP - Fisher Chi-square 336.536 (0.0000) 97.5005 (0.0000) Statistics are estimated with both intercept and trend. Numbers in parentheses are the calculated p-values. The null hypothesis for the LLC and Breitung tests is a unit root with a common unit root process. The null hypothesis for the IPS, ADF and PP tests is a unit root with individual unit root process. The specified number of lags is determined in the same manner as the univariate ADF tests. The probabilities for the Fisher tests are computed using an asymptotic Chi-square distribution. All other tests assume asymptotic normality. * denotes that the null of panel unit root is failed to be rejected at the 5% significance level. 8 For exchange rate differentials, seven cross sections are tested within the panel framework: Euro- USD, Yen-Euro; TL-Euro, Sterling-Euro, TL-Japan, TL-Sterling, TL-USD. For interest rate differentials, ten cross sections are tested within the panel framework: France-USA, Germany-USA, Japan-Germany, Turkey-France, Turkey-Germany, UK-France, UK-Germany, Turkey-Japan, Turkey-UK and Turkey-USA. 9 Seven cross sections are tested within the panel framework for exchange rate differentials: Euro- USD, Yen-Euro, TL-Euro, Sterling-Euro, TL-Yen, TL-Sterling, TL-USD. Also, seven cross sections are tested within the panel framework for the interest rate differentials: LIBOR-USA, LIBOR-Japan, LIBOR-Turkey, LIBOR- UK, Turkey-Japan, Turkey-UK and Turkey-USA. 56 Table - 4.5.2 - DSUR Results for 1985-1998 1985-1998 β Estimated Coefficient P- value Β Estimated Coefficient P- value Coefficient Coefficient Constant 0.000088 0.9058 Constant (0.003362) 0.1984 FFGE** 0.003365 0.0393 FFUS 0.005845 0.1078 FFGE (1)** (0.003839) 0.0189 FFUS (1) (0.003343) 0.3583 FFGE(-1) (0.000492) 0.7629 FFUS(-1) (0.002514) 0.4871 FFGE (2) 0.000916 0.5707 FFUS (2) 0.000131 0.9708 FFGE(-2) 0.002944 0.0675 FFUS(-2) 0.002746 0.4405 FFGE (3) (0.000479) 0.6452 FFUS (3) (0.000201) 0.9295 FFGE(-3)** (0.002706) 0.0090 FFUS(-3) (0.002725) 0.2302 Constant 0.002396 0.4385 Constant (0.001384) 0.4611 JPGE 0.001482 0.8716 UKFF 0.003168 0.1464 JPGE (1) 0.008096 0.3765 UKFF (1) (0.002055) 0.3467 JPGE(-1) (0.005268) 0.5642 UKFF(-1) 0.000055 0.9796 JPGE (2) 0.009545 0.2971 UKFF (2) (0.001064) 0.6235 JPGE(-2) (0.005222) 0.5691 UKFF(-2) 0.001562 0.4652 JPGE (3) (0.007985) 0.2109 UKFF (3) 0.000690 0.6210 JPGE(-3) 0.000589 0.9262 UKFF(-3) (0.002250) 0.1047 Constant (0.003700) 0.1360 Constant (0.001410) 0.5380 GEUS 0.005396 0.2443 UKGE 0.003390 0.1797 GEUS (1) (0.004060) 0.3801 UKGE (1) (0.002720) 0.2848 GEUS(-1) (0.000970) 0.8350 UKGE(-1) 0.000269 0.9149 GEUS (2) 0.002004 0.6643 UKGE (2) (0.000910) 0.7176 GEUS(-2) 0.001505 0.7430 UKGE(-2) 0.001579 0.5224 GEUS (3) (0.001380) 0.6555 UKGE (3) 0.000640 0.7049 GEUS(-3) (0.002700) 0.3739 UKGE(-3) (0.002350) 0.1525 57 Table - 4.5.3(A) - DSUR Results for 1999-2007 1999-2007(A) β Coefficient Estimated Coefficient P- value β Coefficient Estimated Coefficient P- value Constant (0.002797) 0.2527 Constant (0.002964) 0.2198 FFUS (0.004487) 0.4800 GEUS (0.007689) 0.1167 FFUS (1) (0.002272) 0.7078 GEUS (1) (0.001712) 0.7278 FFUS(-1) 0.003272 0.6050 GEUS(-1) 0.006666 0.1729 FFUS (2) 0.002889 0.6194 GEUS (2) 0.003213 0.4945 FFUS(-2) (0.001320) 0.8381 GEUS(-2) (0.000021) 0.9965 FFUS (3) (0.000104) 0.9811 GEUS (3) (0.000232) 0.9494 FFUS(-3) 0.002379 0.6205 GEUS(-3) 0.000397 0.9165 Constant 0.003217 0.6160 Constant 0.004564 0.4373 JPGE (0.003075) 0.7509 TRFF** (0.000688) 0.0000 JPGE (1) 0.003767 0.6999 TRFF (1) (0.000018) 0.9088 JPGE(-1) (0.002375) 0.8091 TRFF(-1) 0.000114 0.4875 JPGE (2) 0.012985 0.1732 TRFF (2) 0.000144 0.3678 JPGE(-2) (0.004877) 0.6213 TRFF(-2) 0.000015 0.9226 JPGE (3) (0.008828) 0.2133 TRFF (3) 0.000083 0.5481 JPGE(-3) 0.002659 0.7239 TRFF(-3) 0.000117 0.3997 Constant 0.004582 0.4351 Constant (0.000024) 0.9970 TRGE** (0.000688) 0.0000 TRJP** (0.000764) 0.0000 TRGE (1) (0.000019) 0.9077 TRJP (1) 0.000030 0.8622 TRGE(-1) 0.000115 0.4846 TRJP(-1) 0.000164 0.3450 TRGE (2) 0.000145 0.3656 TRJP (2) 0.000136 0.4204 TRGE(-2) 0.000015 0.9223 TRJP(-2) 0.000145 0.3973 TRGE (3) 0.000083 0.5491 TRJP (3) 0.000104 0.4843 TRGE(-3) 0.000117 0.3987 TRJP(-3) (0.000075) 0.6083 Constant 0.004504 0.4537 Constant 0.000050 0.9929 TRUK** (0.000634) 0.0003 TRUS** (0.000687) 0.0001 TRUK (1) (0.000077) 0.6472 TRUS (1) (0.000014) 0.9297 TRUK(-1) 0.000060 0.7230 TRUS(-1) 0.000106 0.5257 TRUK (2) 0.000154 0.3551 TRUS (2) 0.000115 0.4803 TRUK(-2) 0.000141 0.4031 TRUS(-2) 0.000078 0.6345 TRUK (3) 0.000112 0.4387 TRUS (3) 0.000106 0.4594 TRUK(-3) 0.000013 0.9245 TRUS(-3) 0.000018 0.8945 58 Table- 4.5.3(A) - DSUR Results for 1999-2007 (continued) 1999-2007(A) β Coefficient Estimated Coefficient P- value β Coefficient Estimated Coefficient P- value Constant (0.001482) 0.5660 Constant (0.000928) 0.7650 UKFF 0.008175 0.1362 UKGE 0.003324 0.3567 UKFF (1) (0.004027) 0.4701 UKGE (1) (0.000326) 0.9266 UKFF(-1) (0.008637) 0.1041 UKGE(-1) (0.003540) 0.3238 UKFF (2) (0.001314) 0.8031 UKGE (2) 0.000355 0.9174 UKFF(-2) (0.002498) 0.6515 UKGE(-2) (0.002073) 0.5831 UKFF (3) 0.002740 0.4976 UKGE (3) (0.000054) 0.9845 UKFF(-3) 0.006137 0.1478 UKGE(-3) 0.002576 0.3991 Table- 4.5.3(B) - DSUR Results for 1999-2007 1999-2007 (B) β Coefficient Estimated Coefficient P- value β Coefficient Estimated Coefficient P- value Constant (0.002564) 0.2942 Constant (0.002718) 0.7153 EUUS (0.005054) 0.5326 JPEU 0.012173 0.5708 EUUS(-1) 0.005348 0.5262 JPEU(-1) (0.024561) 0.2486 EUUS (1) (0.005694) 0.4564 JPEU (1) 0.005155 0.8064 EUUS(-2) (0.001109) 0.9018 JPEU(-2) 0.018780 0.3605 EUUS (2) 0.003378 0.6362 JPEU (2) (0.028275) 0.1544 EUUS(-3) 0.000981 0.8809 JPEU(-3) (0.000180) 0.9886 EUUS (3) 0.002478 0.6238 JPEU (3) 0.018608 0.1178 Constant (0.009372) 0.2092 Constant (0.013185) 0.1004 TREU** (0.001372) 0.0000 TRJP** (0.001420) 0.0000 TREU(-1) 0.000057 0.7971 TRJP(-1) 0.000118 0.6035 TREU (1) (0.000122) 0.5843 TRJP (1) (0.000065) 0.7729 TREU(-2) (0.000014) 0.9504 TRJP(-2) 0.000126 0.5719 TREU (2) 0.000180 0.4167 TRJP (2) 0.000186 0.4076 TREU(-3)** 0.000402 0.0436 TRJP(-3) 0.000185 0.3611 TREU (3) 0.000296 0.1459 TRJP (3) 0.000312 0.1324 61 with the Turkish data. The tests are repeated by using the Euro LIBOR instead of the individual German and French interest rates and the results, presented in Table 4.5.3(B) are the same. In both cases, positive interest rate differentials in favor of Turkey indicate depreciation of the Turkish lira against the corresponding foreign currency. Direction of the effect supports the IFE; however, the magnitude of the effect is found to be smaller than expected. For the 1999-2007 period, significant results are found only between the Turkish lira and other currencies. It is important to note that while interpreting the results, the macroeconomic characteristics of the period should also be evaluated. For instance, at the end of 1999, as part of the stand-by agreement signed with the IMF, Turkey adopted an exchange rate stabilization program11. Although the system collapsed in 2001 and exchange rates were allowed to float free, the monetary authority intervention while the program was in effect might have caused the significant results obtained for the Turkish currency. 11 The stabilization program was based on a crawling peg exchange rate regime in which the exchange rates were allowed to fluctuate within a pre-determined band. This is a system where the central bank intervenes in the market to keep the rates within the previously announced band. 62 CHAPTER 5 CONCLUSION The purpose of this study is to analyze the relationship between interest rates and exchange rates. This relationship is rather important because there are lots of international investment opportunities available in financial markets and the return on these investments is the main concern of investors. The International Fisher Effect is a theory that states nominal interest rate differentials (hence, the differential between the returns that can be earned on investments of an equal risk) between countries are related to and are unbiased predictors of future spot exchange rates. If the theory holds, the advantage that arises from interest rate differentials should be cancelled out by exchange rate adjustments. Throughout this thesis, the tendency of adjustments of exchange rate movements to offset the differences in interest rates across selected countries is questioned. The empirical analyses of the study consist of four main parts. In the first part, time series characteristics of the exchange rate differentials and interest rate differentials are tested with traditional unit root tests. In the second part, variables which are found to be non-stationary as a result of the unit root tests are tested for cointegration within the Johansen framework. The empirical results of this study indicate the presence of at least one cointegration vector for all of the individually modeled equations in all periods. These results imply that exchange rates and interest rates drift together in the long run.12 In the third part of the study, the 12 The presence of one or more cointegration vectors does not necessarily mean that interest rate differentials are precise estimators of changes in the exchange rates. However, the presence of cointegration means that when the two series are linearly modeled, estimation errors will cancel out in time. 63 variables of interest are tested again within the panel framework for the existence of a unit root. Evidence is found for the existence of a panel unit root in the interest rate data. However, none of the test results support the existence of a panel unit root in exchange rates. Failure to find evidence of panel unit root in exchange rates prevents further panel cointegration analysis. Finally, in part four, the individual country IFE equations are estimated as a system within the DSUR framework, an econometric method recently proposed by Mark et al. (2005). Results obtained from the DSUR analysis can be summarized as follows; • Supportive results are found for the 1999-2007 period for equations which include Turkey. The calculated p-values indicate that, at the 5% significance level, changes in exchange rates can be explained by interest rate differentials. Direction of the effect is found as expected and implies that positive interest rate differentials in favor of Turkey cause depreciation of the Turkish currency against the other currencies in the sample. • For the period 1985-1998, supportive results are found only between France and Germany. The results provide evidence of significance in the one- period lead and the three-period lag of the interest rate differentials. However, when the sign and the magnitude of the significant coefficients are analyzed, it is seen that the effects of these lead/lag values cancel each other out almost perfectly. This outcome supports the previous literature which asserts that close trade partners have influence on each other’s financial markets. Also, the results indicate that an over- or under-reaction of the exchange rate change to the interest rate differential is corrected in the market in about four months’ time. The IFE predicts a one-to-one relationship between exchange rate changes and interest rate differentials. For the above mentioned country pairs, supportive results are found within the DSUR framework. However, a crucial point should be emphasized in that supportive estimation results could not be found for the magnitude of the effect. This means that the exchange rate movements are influenced by other factors in addition to the nominal interest rate differentials. 66 BIBLIOGRAPHY Al-Khazali, Osamah M.(2004), “The generalized Fisher hypothesis in the Asian markets”, Journal of Economic Studies, 31 / 2, 144-157. Berument Hakan, Jelassi Mohamed Mehdi (2002), “Fisher Hypothesis: a multi country analysis”, Applied Economics, 34, 1645-1655 Chaboud Alain P., Wright Jonathan H (2005)., “Uncovered Interest Parity: It works, but not for long”, Journal of International Economics , 66, 349-362. Chakrabarti Avik (2006), “Real exchange rates and real interest rates once again: a multivariate panel cointegration anaysis.”, Applied Economics, 38, 1217-1221. Dickey, D.A. and Fuller W.A. (1979), “Distribution of the Estimators for Autoregressive Time Series with a Unit Root,” Journal of the American Statistical Association, 74, 427–431. Evans Martin D.D, Lewis Karen K. (1995), “Do Expected Shifts in Inflation Affect Estimates of the Long Run Fisher Relation?”, The Journal of Finance, 1 / 1, 225-253 Fama E.F.(1975), “Short Term Interest Rates as Predictors of Inflation”, American Economic Review, 65, 269-282. Fisher, I (1930), “The Theory of Interest”, (New York: MacMillan). Gailliot Henry J.(1970), “Purchasing Power Parity As An Explanation of Long Term Changes in Exchange Rates”, Journal of Money Credit and Banking, 2, 348- 357. Granger, C and P. Newbold (1974), “Spurious Regressions in Econometrics”, Journal of Econometrics, 2, 111-120. Hiro Ito and Menzie Chinn (2007); “Price Based Measurement of Financial Globalization: A Cross Country Study of Interest Rate Parity”, Pacific Economic Review , 12 / 4, 419-444. Jacobson Tor, Lyhagen Johan, Larsson Rolf, Nessen MarianneV (2008), “Inflation, exchange rates and PPP in a multivariate panel cointegration model”, Econometrics Journal, 11, 58-79.” Jaebeom Kim (2007); “Real exchange rates and real interest differentials for sectoral data: A dynamic SUR approach”, Economic Letters, 97, 247-252 67 Juntilla Juha (2001), “ Testing an augmented Fisher Hypothesis for small open economy: the case of Findland”, Journal of Macroeconomics, 4, 577-599. Kwiatkowski D., Phillips P.C.B, Schmidt P., Shin.Y (1992); “Testing the Null Hypothesis of Stationary Against the Alternative of a Unir Root.”, Journal of Econometrics, 44 / 1, 105-130. Mark C. Nelson, Ogaki Masao, Donggyu Sul (2005), “ Dynamic Seemingly Unrelated Regressions”, Review of Economic Studies, 72, 797-820. Mishkin Frederic. S (1992), “Is the Fisher Effect for Real? A reexamination of the Relationship between Inflation and Interest Rates”, Journal of Monetary Economics, 30, 195-215. Mishkin Frederic. S, Simon John (1995), “An Empirical Examination of the Fisher Effect in Australia”, The Economic Record, 71 / 214, 217-229. Mishkin S. Frederic (1984); “Are real interest rates equal across countries? An empirical Investigation of International Parity Conditions”, Journal of Finance, 39, 1345-1357. Minford Patrick and Peel David (2007); “On the Equality of Real Interest Rates across Borders in Integrated Capital Markets”, Open Econ Review , 18, 119-125. Özmen Erdal, Gökcan Aysun (2004); “Deviation from PPP and UIP in a financially open economy: the Turkish evidence”, Applied Financial Economics , 14, 779-784. Panopoulou Ekaterini(2005), “A resolution of the fisher effect puzzle: a comparison of estimators”, National University of Ireland Roll Richard (1979), “Violation of Purchasing Power and Their Implications For Efficient International Commodity Markets”, International Finance and Trade, 1, 76-133. Saatçioğlu Cem, Korap Levent (2007), “Does the interest differential explain future exchange rate return? A reexamination of UIP Hypothesis for the Turkish economy”, International Research Journal of Finance and Economics, 10, 120- 128. Said E. and David A. Dickey (1984), 'Testing for Unit Roots in Autoregressive Moving Average Models of Unknown Order', Biometrika, 71, 599–607. Shalishali Maurice K., Ho Johnny C. (2002), “Inflation, Interest Rate and Exchange rate: What is the relationship?”, Journal of Economics and Economic Education Research, 3, 103-111. Taylor Alan M., Taylor Mark P. (2004), “The Purchasing Power Parity Debate”, Journal of Economic Perspectives, 18 / 4, 135-158. 68 Throop W. Adrian (1994), “International Financial Market Integration and Linkages of National Interest Rates”, Federal Reserve Bank of San Francisco, Number 3. Tsung-Wu Ho (2003), “A re-examination of unbiasedness forward rate hypothesis using dynamic SUR model”, The Quarterly Review of Economics and Finance, 43, 542-559. Webster Allan (1987), “Purchasing Power Parity as a theory of international arbitrage in manufactured goods: an empirical view of UK/US prices in the 1970s”, Applied Economics, 19, 1433-1456. Wu Jyh-Lin (1999); “A re-examination of the exchange rate- interest differential relationship: Evidence from Germany and Japan”, Journal of International Money and Finance, 18, 319-336. Zellner, Arnold (1962), "An efficient method of estimating seemingly unrelated regression equations and tests for aggregation bias". Journal of the American Statistical Association 57, 348–368.